CN117932345A - Power load data prediction model group training method, device, equipment and medium - Google Patents
Power load data prediction model group training method, device, equipment and medium Download PDFInfo
- Publication number
- CN117932345A CN117932345A CN202410264610.6A CN202410264610A CN117932345A CN 117932345 A CN117932345 A CN 117932345A CN 202410264610 A CN202410264610 A CN 202410264610A CN 117932345 A CN117932345 A CN 117932345A
- Authority
- CN
- China
- Prior art keywords
- sequence
- power load
- load data
- data
- sequences
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 117
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000003860 storage Methods 0.000 claims abstract description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 44
- 230000000694 effects Effects 0.000 claims description 16
- 238000001556 precipitation Methods 0.000 claims description 15
- 238000004891 communication Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 11
- 230000009191 jumping Effects 0.000 claims description 5
- 230000000875 corresponding effect Effects 0.000 description 54
- 230000008859 change Effects 0.000 description 14
- 238000012545 processing Methods 0.000 description 13
- 230000002596 correlated effect Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000005457 optimization Methods 0.000 description 5
- 230000000737 periodic effect Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000001174 ascending effect Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 241000282461 Canis lupus Species 0.000 description 2
- 241000283153 Cetacea Species 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 description 1
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007334 memory performance Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000005406 washing Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the application relates to the technical field of model training, and discloses a power load data prediction model group training method, device, equipment and a computer readable storage medium.
Description
Technical Field
The embodiment of the application relates to the technical field of model training, in particular to a method, a device, equipment and a computer readable storage medium for training a power load data prediction model group.
Background
With the rapid development of the current society, people rely on electric power more and more, electric power is required to be used in production in many fields, electric power driving equipment is required to continuously produce in factories, household equipment such as electric lamps, washing machines and air conditioners is required to be driven by the electric power at home by people, and convenience is brought to life.
The prediction of the power load data is very important for the power plant, the power plant can perform finer management on power supply work according to the predicted power load data, the power plant or the power company can be helped to make a reasonable power generation plan and resource allocation according to the prediction result, and the predicted power load data plays an important role in reasonably distributing power resources for the power plant or the power company, so that the stability of a power system can be improved.
However, the power load data has the characteristics of strong randomness and volatility, so that the current common prediction model for predicting the power load data is difficult to accurately predict the power load data, and the safe operation of the power system and the accurate implementation of the power dispatching plan are affected.
Disclosure of Invention
In view of the above problems, embodiments of the present application provide a method, apparatus, device, and computer readable storage medium for training a power load data prediction model set, which are used to solve the problem that the existing prediction model is not accurate enough for predicting power load data.
According to an aspect of the embodiment of the present application, there is provided a power load data prediction model set training method, including: acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence; performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences; determining a target original data sequence which meets the correlation requirement with any one modal component sequence in a plurality of original data sequences; and combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training, and generating a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
In an alternative manner, determining a target raw data sequence of the plurality of raw data sequences that meets a correlation requirement with any one of the modal component sequences includes: respectively calculating covariance between each original data sequence and an nth modal component sequence to obtain a plurality of covariance data, wherein n=1; respectively calculating the product of the standard deviation of each original data sequence and the standard deviation of the nth modal component sequence to obtain a plurality of standard deviation product data; taking covariance data as divisors, respectively calculating the quotient of each covariance data and each corresponding standard deviation product to obtain a correlation coefficient between each original data sequence and an nth modal component sequence, wherein each covariance data and each corresponding standard deviation product are calculated according to the same original data sequence; setting n as n+1, and jumping to a step of respectively calculating covariance between each original data sequence and an nth modal component sequence until all modal component sequences are traversed; judging whether the correlation coefficient between each original data sequence and any modal component sequence is larger than or equal to a first preset threshold value or not respectively; and determining the original data sequence with the correlation coefficient larger than or equal to a first preset threshold value as a target original data sequence corresponding to the modal component sequence.
In an alternative manner, performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences, including: establishing a first constraint model in which the sum of the plurality of input sequences is equal to the historical power load data sequence; establishing a second constraint model, wherein the sum of bandwidths of the plurality of input sequences is smaller than or equal to a second preset threshold value; decomposing the historical electrical load data sequence into k sub-modal components, wherein k = 1; judging whether the k sub-modal components simultaneously meet a first constraint model and a second constraint model; if yes, taking the k sub-modal components as a plurality of modal component sequences; if not, then set k to k+1 and jump to the step of decomposing the historical electrical load data sequence into k sub-modal components.
In an alternative way, a historical power load data sequence is obtained along with a plurality of raw data sequences, including: at least two data sequences of a plurality of personnel activity related data sequences and a plurality of weather related data sequences are obtained as a plurality of raw data sequences.
In an alternative, the plurality of personnel activity related data sequences includes a flow of people data sequence and the plurality of weather related data sequences includes a precipitation data sequence, an air temperature data sequence, an air pressure data sequence, and a cloud data sequence.
In an optional manner, each modal component sequence is combined with a corresponding target original data sequence respectively and used as a training sample together for model training, and a power load data prediction model set is generated, which comprises the following steps: and respectively combining each modal component sequence with the corresponding target original data sequence and jointly using the modal component sequences as training samples to train the long-short-time memory network model, so as to obtain a power load data prediction model group consisting of a plurality of long-short-time memory network models.
According to an aspect of an embodiment of the present application, there is provided a power load data prediction method including: acquiring a plurality of modal component sequences, wherein the modal component sequences are obtained by carrying out modal decomposition on a historical power load data sequence; obtaining a target original data sequence, wherein the target original data sequence is an original data sequence meeting the correlation requirement with any one modal component sequence; predicting a modal component sequence corresponding to the power load data prediction model group by utilizing a plurality of prediction models in the power load data prediction model group to obtain a plurality of component prediction results, wherein the power load data prediction model group is obtained by training the power load data prediction model group training method according to any one embodiment, and the power load data prediction model group comprises a plurality of prediction models; and linearly adding the plurality of component prediction results to obtain a power load data prediction result.
According to another aspect of the embodiment of the present application, there is provided a power load data prediction model group training apparatus including: the system comprises an acquisition module, a calculation module, a determination module and a model training module. The acquisition module is used for acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence; the calculation module is used for carrying out modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences; the determining module is used for determining a target original data sequence which meets the correlation requirement with any one modal component sequence in the plurality of original data sequences; the model training module is used for combining each modal component sequence with a corresponding target original data sequence respectively and jointly serving as a training sample to carry out model training, and generating a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
According to another aspect of an embodiment of the present application, there is provided an electric load data prediction model group apparatus including: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface are communicated with each other through the communication bus; the memory is used for storing at least one program which causes the processor to execute the operations of the power load data prediction model set training method according to any one of the above.
According to still another aspect of the embodiments of the present application, there is provided a computer-readable storage medium having stored therein executable instructions for causing a power load data prediction model set training apparatus to perform the operations of the power load data prediction model set training method as any one of the above.
By combining each modal component sequence with the corresponding target original data sequence and jointly using the modal component sequences as training samples to perform model training, each prediction model in the power load data prediction model group can consider the influence of the target original data sequence on the corresponding modal component sequence during training, and because a plurality of modal component sequences are obtained by carrying out modal decomposition on the historical power load data sequence, the influence of each target original data sequence on the corresponding modal component sequence is the influence of the target original data sequence on the historical power load data sequence in one dimension, so that the obtained power load data prediction model group can combine a plurality of target original data sequences to predict a plurality of modal component sequences.
The foregoing description is only an overview of the technical solutions of the embodiments of the present application, and may be implemented according to the content of the specification, so that the technical means of the embodiments of the present application can be more clearly understood, and the following specific embodiments of the present application are given for clarity and understanding.
Drawings
The drawings are only for purposes of illustrating embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
FIG. 1 is a schematic flow chart of a training method for a power load data prediction model set according to an embodiment of the present application;
FIG. 2 is a flow chart of the sub-steps of steps 130 and 140 of the present application;
FIG. 3a is a graph of data points of sequence 1 in an embodiment of the present application;
FIG. 3b is a graph of data points of sequence 2 in an embodiment of the present application;
FIG. 3c is a graph of data points of sequence 3 in an embodiment of the present application;
FIG. 4 is a flow chart of the sub-steps of steps 110 and 120 of the present application;
FIG. 5 is a histogram of correlation coefficients of various raw data sequences and historical power load data sequences in accordance with an embodiment of the present application;
FIG. 6 is a flowchart of a power load data prediction method according to an embodiment of the present application;
FIG. 7 is a functional block diagram of a power load data prediction model set training device according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of a training device for a power load data prediction model set according to an embodiment of the present application.
Detailed Description
Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein.
With the development of modern technology, people cannot leave power in the aspects of production and life, as the power plant cannot directly store electric energy, residual electricity is injected into a power grid through a built-in power grid and an energy storage facility and stored, in a power system, the production, the transmission, the distribution and the use of the electric energy are always in a dynamic balance state, if power supply and demand are unbalanced, power frequency deviation can be caused, and the output of the power system needs to be adjusted to maintain balance of power supply and demand so as to ensure balance and stability of the power system, for example, the power supply capacity and the load demand of the power system are reasonably arranged according to the prediction result of power load data, so that energy waste is reduced and the power supply cost is reduced.
However, the management and the scheduling of the power resources take a certain time to implement, and in the actual situation, the operation condition of the power system changes suddenly, and it may be difficult to meet the requirement of timely scheduling and managing the power resources according to the current power operation condition, so before the power resources are scheduled and managed, a plan is often made in advance, at this time, the power load data needs to be predicted, and then a corresponding management and scheduling plan is formulated according to the predicted future power load data, so as to ensure timeliness and effectiveness of the implementation of the plan.
The core problem of power load data prediction is how to predict future power load data by using existing historical data, and a current common mode of predicting power load data is to establish a prediction model by taking the historical power load data as a training sample, and to realize the prediction of power load data by the prediction model.
However, the inventor of the present application has noted that in practical situations, the power load data may be affected by various data, for example, when the air temperature is high and the weather is hot, people usually turn on the air conditioner or the fan, and these electric devices will increase the load of the power system, so that the power load data is increased, for example, when the flow of people is high, the usage of electric devices of shops such as restaurants and entertainment places is correspondingly increased, and also the power load data is significantly increased, and when the precipitation is high, many shops, shopping centers and other places tend to turn off the equipment located in the open air, and the usage of electric devices is reduced, so that the power load data is reduced. Therefore, when the external data are continuously changed, the power load data are nonlinear, the power load data are difficult to accurately predict by using a prediction model obtained by taking the power load data as a training sample, the obtained prediction result may be different from the actual power load data, and the safe operation of the power system and the accurate implementation of a scheduling plan are seriously affected when the power system is managed and scheduled according to the incorrect prediction result.
Based on the research, the inventor designs a power load data prediction model group training method, by acquiring a historical power load data sequence and a plurality of original data sequences with correlation with the historical power load data sequence, performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences, determining a target original data sequence meeting the correlation requirement with any modal component sequence from the plurality of original data sequence types, and finally combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training to generate the power load data prediction model group, so that the generated power load data prediction model group can be more accurate in predicting power load data.
Referring to fig. 1, fig. 1 is a flow chart illustrating a power load data prediction model set training method according to an embodiment of the present application, in one aspect of the embodiment of the present application, a power load data prediction model set training method is provided, as shown in fig. 1, and the method includes the following steps:
Step 110: acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence;
The historical power load data sequence refers to a sequence formed by power data consumed by power equipment using electric energy in a plurality of continuous time nodes, and the historical power load data sequence refers to a sequence formed by power data consumed by power equipment using electric energy in a plurality of continuous time nodes in a past time period.
The original data sequence refers to a data sequence which has a certain correlation with the historical power load data sequence, the data change of the original data sequence can have a certain influence on the historical power load data sequence, for example, the original data sequence is a sequence formed by equipment opening amount data, when the equipment opening amount data is continuously increased, the load on the power system is increased, the historical power load data sequence is increased along with the increase of the equipment opening amount data, namely, the sequence formed by the equipment opening amount data is positively correlated with the historical power load data sequence, for example, the original data sequence is a sequence formed by precipitation amount data, a plurality of active places pause business with the gradual increase of precipitation amount, a plurality of equipment shut-down operations are carried out, the load on the power system is reduced, and the historical power load data sequence is reduced along with the increase of the precipitation amount data, namely, the sequence formed by the precipitation amount data is negatively correlated with the historical power load data sequence. It will be appreciated that a certain data sequence may be considered as the original data sequence as long as there is a correlation between the data sequence and the historical power load data sequence, whether there is a positive correlation or a negative correlation between them.
Step 120: performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences;
The step of performing modal decomposition on the historical power load data sequence refers to splitting the historical power load data sequence into a plurality of modal component sequences, so that each modal component sequence can reflect a periodic change contained in the historical power load data sequence, and it should be understood by those skilled in the art that, in an ideal case, a sequence formed by adding data of a plurality of modal component sequences obtained after performing modal decomposition on the historical power load data sequence on a corresponding time node should be consistent with the historical power load data sequence.
It should be noted that, after the historical power load data sequence is subjected to modal decomposition, one or more abnormal sequences with less obvious periodic fluctuation and poor data change regularity may be obtained, in some cases, such abnormal sequences may be used as noise to be separated, only a plurality of sequences with obvious periodic fluctuation and higher data change regularity obtained after the historical power load data sequence is subjected to modal decomposition are reserved as modal component sequences, it may be understood that according to the actual situation of the historical power load data sequence and the difference of the number of the modal component sequences obtained by modal decomposition, different processing modes should be flexibly adopted, for example, when the number of the modal component sequences obtained by modal decomposition of the historical power load data sequence is less, each modal component sequence contains a large amount of data in the historical power load data sequence, and then separation processing of the abnormal sequences therein may possibly result in the loss of important data in the historical power load data sequence, at this time, the noise removal processing may not be performed, and a person skilled in the art may flexibly select the number of the modal component sequences obtained by modal decomposition according to actual needs, so that the number of the noise components obtained by modal decomposition of the power load data sequence is controlled, and the number of the modal components is obtained.
Furthermore, the parameters of the modal decomposition can be optimized through an optimization algorithm, so that the number of decomposition layers selected in modal decomposition of the historical power load data sequence is more reasonable, and the phenomenon that the number of decomposition layers is excessively large to cause excessive decomposition or the number of decomposition layers is excessively small to cause insufficient decomposition is avoided. For example, the selection of parameters for modal decomposition is optimized by a wolf optimization algorithm (GWO, grey Wolf Optimizer) or a whale optimization algorithm (WOA, whale optimization algorithm) to enhance the effect of modal decomposition on the historical power load data sequence.
Step 130: determining a target original data sequence which meets the correlation requirement with any one modal component sequence in a plurality of original data sequences;
In step 130, although the plurality of original data sequences are all correlated with the historical power load data sequence, since the historical power load data sequence has been decomposed into a plurality of modal component sequences, the plurality of modal component sequences reflect different periodic changes and trend information in the historical power load data sequence, the original data sequence correlated with the historical power load data sequence is not necessarily correlated with the plurality of modal component sequences decomposed from the historical power load data sequence, and therefore, a target original data sequence satisfying the correlation requirement with any one modal component sequence is determined in the plurality of original data sequences so as to ensure that the target original data sequence has a certain correlation with at least one modal component sequence.
Step 140: and combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training, and generating a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
Each modal component sequence is combined with a corresponding target original data sequence respectively and is used as a training sample for model training, and a training mode of a neural network or a decision tree and other models can be adopted to obtain a plurality of power load data prediction models and form a power load data prediction model group.
Further, in step 140, in the process of combining each modal component sequence with its corresponding target raw data sequence and jointly performing model training as a training sample to generate a power load data prediction model set, a suitable optimization algorithm may be further used to adjust parameters of the model training to minimize errors of the model, for example, when each prediction model in the power load data prediction model set is trained respectively, the parameters of the prediction model are adjusted by adopting gradient descent and random gradient descent algorithms, so that performance of the model can be gradually improved, and prediction accuracy of each prediction model in the power load data prediction model set can be improved.
In steps 110 to 140, each modal component sequence is combined with a corresponding target original data sequence and used as a training sample together to perform model training, so that each prediction model in the power load data prediction model set can consider the influence of the target original data sequence on the corresponding modal component sequence during training, and because the plurality of modal component sequences are obtained by performing modal decomposition on the historical power load data sequence, the influence of each target original data sequence on the corresponding modal component sequence is the influence of the target original data sequence on the historical power load data sequence in one dimension, and compared with the prediction model obtained by using the historical power load data sequence as the training sample, the prediction model obtained by the power load data prediction model set training method provided by the embodiment of the application can make more accurate predictions.
Referring to fig. 2, fig. 2 is a flow chart illustrating the sub-steps of the steps 130 and 140 according to the present application. As shown in fig. 2, in some embodiments of the application, step 130 further comprises:
Step 131: respectively calculating covariance between each original data sequence and an nth modal component sequence to obtain a plurality of covariance data, wherein n=1;
In this step, the specific step of calculating the covariance between each original data sequence and the nth modal component sequence may be, for example, to use the original data sequence as a variable a, use the corresponding modal component sequence as a variable b, and then use the covariance formula to calculate the covariance:
A plurality of covariance data is calculated, wherein, I-th data point of variable a,/>I-th data point of variable b,/>Is the mean value of the variable a,/>N is the number of data points, which is the mean of the variables b.
The function of the covariance data is to judge the overall error between the two variables, if the covariance between the two variables is positive, it is indicated that the variation trend of the two variables on the whole is the same, if the covariance between the two variables is negative, it is indicated that the variation trend of the two variables on the whole is opposite, if the covariance between the two variables is zero, it is indicated that the variation trend of the two variables is not obviously associated, that is, it can be understood that when the covariance between the two variables is not zero, it is indicated that there is a certain association between the variation trend of the two variables.
Step 132: respectively calculating the product of the standard deviation of each original data sequence and the standard deviation of the nth modal component sequence to obtain a plurality of standard deviation product data;
in this step, the specific step of calculating the product of the standard deviation of each original data sequence and the standard deviation of the nth modality component sequence may be, for example, that the standard deviation of each original data sequence is calculated by a standard deviation calculation formula: Where i is the number of data points in the sequence,/> For the ith data point in the sequence,/>N is the number of data points, and the standard deviation of the nth mode component sequence is calculated according to the standard deviation calculation formulaStandard deviation/>, of each original data sequence is sequentially carried out/>, With the nth modality component sequenceThe multiplication results in a plurality of standard deviation product data.
The standard deviation is one of the most commonly used statistical indicators in probability statistics, and is used to measure the degree of dispersion of a set of data, that is, the degree of difference between each data point in a set of data and the average value of the set of data, the larger the standard deviation of a set of data, the larger the difference between the data points of the set of data, and the smaller the standard deviation, the smaller the difference between the data points of the set of data.
Step 133: taking covariance data as divisors, respectively calculating the quotient of each covariance data and each corresponding standard deviation product to obtain a correlation coefficient between each original data sequence and an nth modal component sequence, wherein each covariance data and each corresponding standard deviation product are calculated according to the same original data sequence;
In this step, the objective of calculating the quotient of each covariance data and each corresponding standard deviation product respectively using the covariance data as the dividend is to obtain a correlation coefficient between each original data sequence and the nth modality component sequence, where the obtained correlation coefficient should be able to reflect the influence amplitude of each unit change in the corresponding original data sequence on each unit change in the nth modality component sequence, but not be influenced by the dimensions of the original data sequence or the nth modality component sequence.
Referring to fig. 3a, fig. 3b, and fig. 3c, fig. 3a, fig. 3b, and fig. 3c show graphs of data points of sequence 1, sequence 2, and sequence 3, respectively, in an embodiment, the influence amplitude of each unit change in sequence 3 on sequence 1 and sequence 2 is determined only by covariance data, so as to compare correlations between sequence 3 and sequence 1, and between sequence 3 and sequence 2, and covariance data between sequence 1 and sequence 3 is calculated as follows:
(100-0)×(70-0)+(-100-0)×(-70-0)+(-200-0)×(-200-0)+(-100-0)×(-70-0)+(100-0)×(70-0)+(200-0)×(200-0)+(0-0)×(0-0))/7=15428.57, Covariance data between sequence 2 and sequence 3 is:
(0.01-0)×(70-0)+(-0.01-0)×(-70-0)+(-0.02-0)×(-200-0)+(-0.01-0)×(-70-0)+(0.01-0)×(70-0)+(0.02-0)×(200-0)+(0-0)×(0-0))/7=1.542857, As can be seen from the above calculation, the covariance data of the sequence 3 and the sequence 1 is 15428.57, the covariance data of the sequence 3 and the sequence 2 is 1.542857, the data phase difference is larger, the correlation can only be judged according to the fact that the covariance data are positive numbers, the change trend of the sequence 3 and the sequence 1, the change trend of the sequence 3 and the change trend of the sequence 2 are the same, but as shown in fig. 3a, fig. 3b and fig. 3c, the difference between the sequence 1 and the sequence 2 is 10000 times as compared with the data unit, namely, the dimension difference between the sequence 1 and the sequence 2 is different, when the data point of the sequence 3 is at the lowest peak, the corresponding data point of the sequence 1 and the sequence 2 is at the lowest peak, when the data point of the sequence 3 is at the highest peak, the corresponding data point of the sequence 1 and the sequence 2 is at the highest peak, the degree of influence of the sequence 3 on the sequence 1 and the sequence 2 is the same when the correlation between the sequence 3 and the sequence 3 is concerned only or the correlation between the sequence 3 and the sequence 2 is concerned, namely, the correlation between the sequence 3 and the sequence 2 is not exactly equal, the correlation is difficult to judge when the correlation between the sequence 1 and the sequence 2 is carried out. Therefore, in step 133, the covariance data is taken as the dividend, and the quotient of the standard deviation product corresponding to the covariance data is calculated, so as to eliminate the influence of dimension, so that the obtained correlation coefficient only changes between +1 and-1, and the correlation judgment is facilitated, and the following embodiment is referred to specifically:
in one embodiment of the present application, as shown in fig. 3a, 3b and 3c, the correlation between the sequence 3 and the sequence 1, the correlation between the sequence 3 and the sequence 2 are determined, and then the formula is obtained according to step 133: And according to the formula, the correlation coefficient of the sequence 3 and the sequence 1 is 0.9879, and the correlation coefficient of the sequence 3 and the sequence 2 is 0.9879, the correlation of the sequence 3 and the sequence 1 is consistent with the correlation of the sequence 3 and the sequence 2, and the correlation has extremely high correlation.
Therefore, by the processing of step 132 and step 133, a plurality of correlation coefficients for performing correlation determination can be obtained, and the plurality of correlation coefficients are not affected by dimensions, facilitating correlation determination in subsequent steps.
Step 134: setting n to n+1, and jumping to step 131 until all modal component sequences are traversed;
in this step, after the correlation coefficient between the nth modality sequence and each original data sequence is obtained, n is set to n+1 and the calculation process is started again by jumping to step 131, so that the next modality component sequence will be calculated when each subsequent step is executed until all modality component sequences are calculated, and all the correlation coefficients between each modality component sequence and each original data sequence can be obtained through step 134.
Step 135: judging whether the correlation coefficient between each original data sequence and any modal component sequence is larger than or equal to a first preset threshold value or not respectively;
step 136: and determining the original data sequence with the correlation coefficient larger than or equal to a first preset threshold value as a target original data sequence corresponding to the modal component sequence.
In step 135 and step 136, the correlation coefficients between the multiple original data sequences obtained in step 133 and the modal component sequence are determined, and a target original data sequence corresponding to the correlation coefficient enough to meet the requirement is screened out, where the first preset threshold may be set to 0.6, for example. In practical application, the size of the first preset threshold may be adjusted according to practical situations, for example, when the types of the original data sequences are more, the original data sequence with high correlation with the modal component sequence needs to be selected as the target original data sequence, so as to improve the accuracy of the prediction model obtained by training and reduce the data volume at the same time, save the training time of the model, and then the first preset threshold may be correspondingly increased. It should be noted that, the setting of the first preset threshold should be determined based on the value ranges of the plurality of correlation coefficients obtained in step 133, for example, in step 133, the value ranges of the plurality of correlation coefficients obtained are-1 to +1, and the first preset threshold should be a value in a range greater than-1 and less than 1.
In steps 131 to 136, all correlation coefficients between each modal component sequence and each original data sequence, which are not influenced by the dimension of the sequence, are obtained through calculation, and the target original data sequence corresponding to each modal component sequence is determined by respectively judging whether the correlation coefficient between each original data sequence and any one modal component sequence is larger than or equal to a first preset threshold value, so that the target original sequence with enough correlation with each modal component can be more accurately judged, and the correlation judgment error is not easily caused by the influence of different dimensions among different sequences, thereby improving the accuracy of the training method of the power load data prediction model set provided by the embodiment of the application.
Referring to fig. 4, fig. 4 is a flow chart illustrating the sub-steps of the steps 110 and 120 according to the present application. As shown in fig. 4, in one embodiment of the present application, step 120 further includes:
step 121: establishing a first constraint model in which the sum of the plurality of input sequences is equal to the historical power load data sequence;
Step 122: establishing a second constraint model, wherein the sum of bandwidths of the plurality of input sequences is smaller than or equal to a second preset threshold value;
In steps 121 and 122, the first constraint model may be, for example, a function:
Wherein s.t. represents constraint conditions, k is the number of layers of modal decomposition, For decomposing the obtained sub-modal component, f is the historical power load data sequence. The new sequence of data points that satisfy the sum of the respective corresponding k sub-modal components of the constraint function is equal to the historical power load data sequence.
The second constraint model may be, for example, a function:
wherein k is the number of layers of modal decomposition, t is a time node, Representing the partial derivative of t,/>For dirac distribution, j is an imaginary number,/>For convolution operation,/>For decomposing the resulting sub-modal components,/>E is a natural constant for the bandwidth of each sub-modal component. The sum of the bandwidths of the k sub-modal components satisfying the constraint function is smallest, where bandwidth refers to the geometric average of the largest data point and the smallest data point on a sub-modal component.
Step 123: decomposing the historical electrical load data sequence into k sub-modal components, wherein k = 1;
step 124: judging whether the k sub-modal components simultaneously meet a first constraint model and a second constraint model;
If yes, executing the following steps:
Step 125: taking k sub-modal components as a plurality of modal component sequences;
If not, executing the following steps:
step 126: let k be k +1 and jump to step 123.
When the sequence is subjected to modal decomposition, the characteristics of excessive original sequences are not lost due to a plurality of decomposition results obtained through decomposition, meanwhile, whether key data in the historical power load data sequence are lost or not is ensured, meanwhile, whether the number of layers of modal decomposition on the historical power load data sequence is enough or not is ensured through a second constraint model, namely, whether the sum of bandwidths of a plurality of sub-modal sequence components is minimum or not is ensured, under-decomposition is avoided, if the obtained k sub-modal components cannot meet the requirements of the first constraint model and the second constraint model at the same time, k+1 is set again to be the same time, and the number of sub-modal components is ensured to be the proper number of the final power load data after each decomposition.
Through the steps 121 to 126, the number of layers of modal decomposition on the historical power load data sequence can be ensured to be more accurate and proper, so that a proper number of sub-modal components can be obtained, each sub-modal component can reflect a periodic change rule and trend information of the historical power load data sequence more accurately, the situation of over-decomposition and under-decomposition is avoided, and a prediction model obtained by performing model training on the proper number of sub-modal components as a data basis can have higher prediction accuracy.
Referring again to fig. 4, in some embodiments of the present application, step 110 further comprises:
Step 111: at least two data sequences of a plurality of personnel activity related data sequences and a plurality of weather related data sequences are obtained as a plurality of raw data sequences.
Wherein, the various data sequences related to personnel activities refer to various data sequences related to personnel activities, such as data sequences of general living population, daily trip number and the like, and the various data sequences related to meteorological changes, such as data sequences of snowfall, air pressure and the like.
Because the power load data in the power system is often mainly influenced by personnel activities, namely the personnel activities change the power utilization state of electric equipment with a high probability, so that the power load data in the power system changes, meanwhile, weather changes can also cause more direct influence on the personnel activities, and the power load data in the power system can be indirectly caused to change, therefore, at least two data sequences are selected from various personnel activity related data sequences and various weather related data sequences to serve as the data basis for model training, the correlation degree of training samples and the power load data can be improved, and the prediction model obtained through training is more accurate.
Further, the plurality of personnel activity related data sequences in step 111 include a personnel flow data sequence, and the plurality of weather related data sequences include a precipitation data sequence, an air temperature data sequence, an air pressure data sequence, and a cloud amount data sequence.
Referring to fig. 5, fig. 5 is a histogram of correlation coefficients of various original data sequences and historical power load data sequences in an embodiment of the present application, by obtaining sequence data of traffic, precipitation, air temperature, air pressure and cloud amount, correlation coefficients of the data and the historical power load data sequences are calculated respectively, and a histogram is generated, so that analysis is facilitated. In various weather related data sequences, when the precipitation data sequence is in an ascending trend, residents are inconvenient to travel, customers in public places such as shops are fewer, and the equipment with parts located in the outfield is often closed due to the influence of precipitation, so that the opening amount of the equipment is reduced, and the historical power load data sequence is in a descending trend. When the temperature data sequence is in an ascending trend, people often use electric fans or air conditioners and other electric appliances to cool down, and the opening amount of equipment is increased, so that the historical power load data sequence is in an ascending trend.
According to the analysis and the correlation coefficient histogram shown in fig. 5, it is known that the human flow data sequence, the precipitation data sequence, the air temperature data sequence, the air pressure data sequence and the cloud amount data sequence all have a certain correlation with the historical power load data sequence, and the human flow data sequence, the precipitation data sequence, the air temperature data sequence, the air pressure data sequence and the cloud amount data sequence have higher feasibility in taking the human flow data sequence, the air temperature data sequence, the air pressure data sequence and the cloud amount data sequence as training samples to participate in training the prediction model so as to improve the prediction accuracy of the prediction model.
Because many meteorological activities are closely related to the laws of people's production and life, the laws of people's production and life often do not leave the participation of electric equipment, so the electric load data of an electric power system is easily influenced, and the data such as precipitation data sequences, air temperature data sequences and the like in the meteorological related data sequences can accurately reflect the change of the meteorological activities, so the precipitation data sequences, the air temperature data sequences, the air pressure data sequences and the cloud data sequences are used as the candidate data of the original data sequences, and the accuracy of the predictive model training can be improved.
Referring back to fig. 2, as shown in fig. 2, in some embodiments of the present application, step 140 further includes:
step 141: and respectively combining each modal component sequence with the corresponding target original data sequence and jointly using the modal component sequences as training samples to train the long-short-time memory network model, so as to obtain a power load data prediction model group consisting of a plurality of long-short-time memory network models.
In this step, the Long-Short-Term Memory network model refers to a model obtained by training a Long-Short-Term Memory network (LSTM), which is a time-cyclic neural network having a certain advantage in Long-sequence processing and a relatively strong sequence modeling capability.
The long-short-time memory network is adopted to train the prediction model in the power load prediction model group, and because the long-short-time memory network is introduced with the gating mechanism, the gradient disappearance problem during long-sequence processing can be solved, the situation that training is difficult to occur is avoided, so that model training can be better carried out according to long-sequence data, meanwhile, in the process of model training, the long-short-time memory network can better capture long-term dependency relationship in sequence data through the cell state and the gating mechanism, better memory performance is achieved, the correlation relationship between an original data sequence and a historical power load data sequence can be better reserved, and further the accuracy of the power load data prediction model group obtained through training is higher.
Referring to fig. 6, fig. 6 is a flow chart of a power load data prediction method according to an embodiment of the present application, in one aspect of the embodiment of the present application, a power load data prediction method is provided, as shown in fig. 6, and the method includes the following steps:
Step 210: acquiring a plurality of modal component sequences, wherein the modal component sequences are obtained by carrying out modal decomposition on a historical power load data sequence;
step 220: obtaining a target original data sequence, wherein the target original data sequence is an original data sequence meeting the correlation requirement with any one modal component sequence;
Step 230: predicting a modal component sequence corresponding to the power load data prediction model group by utilizing a plurality of prediction models in the power load data prediction model group to obtain a plurality of component prediction results, wherein the power load data prediction model group is obtained by training the power load data prediction model group training method according to any one embodiment, and the power load data prediction model group comprises a plurality of prediction models;
In the steps 210 to 230, the purpose is to obtain the data after preprocessing and input the data into a model obtained by pre-training for prediction, so as to obtain a plurality of prediction results corresponding to different sequences, specifically, one of a plurality of modal component sequences and one or more original data sequences meeting the correlation requirement are combined and used together as the input of a prediction model obtained by the power load data prediction model training method, so as to obtain a component prediction result, and the operations are repeated until each modal component sequence obtains a corresponding component prediction result. It should be noted that, each prediction model in the power load prediction model set is generated by training a corresponding modal component sequence as a training sample, and the component prediction result obtained by each prediction model when performing prediction also corresponds to a modal component sequence of the prediction model as a training sample when training, that is, the obtained component prediction result is a prediction result of a corresponding modal component sequence.
Step 240: and linearly adding the plurality of component prediction results to obtain a power load data prediction result.
In step 240, the linear addition of the plurality of component predictors corresponding to each modal component sequence in one-to-one mode obtained in steps 210 to 230 refers to summing each data point in the plurality of component predictors corresponding to the time sequence order thereof to obtain a plurality of sum values, and arranging the sum values in the time sequence order to obtain a new sequence, where the obtained new sequence is the power load data predictor. Specifically, for example, the 3 component predictors obtained in the steps 210 to 230 are a predictor sequence 1, a predictor sequence 2, and a predictor sequence 3, and the ordering of the data points in the predictor sequence 1 in the time sequence includes: 3,5,2,4,1; the ordering of the data points in the predicted outcome sequence 2 in time sequence includes: 2,1,6,3,4; the ordering of the data points in the predicted outcome sequence 3 in time sequence includes: 4,3,3,1,4; the power load data prediction results obtained by linearly adding each data point in the prediction result sequence 1, the prediction result sequence 2 and the prediction result sequence 3 according to the time sequence are as follows: 9,9, 11,8,9.
By acquiring a plurality of modal component sequences and a target original data sequence and respectively predicting one modal component sequence corresponding to the modal component sequences by utilizing a plurality of prediction models in the power load data prediction model group, a relatively accurate power load data prediction result can be obtained, and the power can be more stable and accurate when being allocated according to the accurate power load data prediction result.
Fig. 7 is a functional block diagram of a power load data prediction model set training device according to an embodiment of the present application. As shown in fig. 7, the power load data prediction model group training apparatus 700 includes: an acquisition module 701, a calculation module 702, a determination module 703 and a model training module 704. The obtaining module 701 is configured to obtain a historical power load data sequence and a plurality of original data sequences, where the plurality of original data sequences have correlations with the historical power load data sequence, the calculating module 702 is configured to perform modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences, the determining module 703 is configured to determine a target original data sequence, which meets a correlation requirement with any one of the modal component sequences, in the plurality of original data sequences, and the model training module 704 is configured to combine each of the modal component sequences with a corresponding target original data sequence respectively and use each of the modal component sequences together as a training sample to perform model training, so as to generate a power load data prediction model set, where the power load data prediction model set includes a plurality of prediction models respectively corresponding to each of the modal component sequences.
In some embodiments, the determining module 703 further includes a first computing unit, a second computing unit, a third computing unit, a first processing unit, a first judging unit, and a determining unit. The first computing unit is used for respectively computing covariance between each original data sequence and an nth modal component sequence to obtain a plurality of covariance data, n=1, the second computing unit is used for respectively computing products of standard deviation of each original data sequence and standard deviation of the nth modal component sequence to obtain a plurality of standard deviation product data, the third computing unit is used for respectively computing quotient of each covariance data and each corresponding standard deviation product to obtain correlation coefficient between each original data sequence and the nth modal component sequence by taking the covariance data as a divisor, wherein the correlation coefficient between each covariance data and the corresponding standard deviation product are obtained by computing according to the same original data sequence, the first processing unit is used for setting n as n+1, and executing the steps of the first computing unit in a skip mode until all modal component sequences are traversed, and the first judging unit is used for respectively judging whether the correlation coefficient between each original modal data sequence and any modal component sequence is larger than or equal to a first preset threshold value or not, and the determining unit is used for determining that the correlation coefficient is larger than or equal to the first preset threshold value as the original data sequence corresponding target modal component sequence of the original data sequence.
In some embodiments, the computing module 702 further includes a first model building unit, a second model building unit, a fourth computing unit, a second determination unit, a second processing unit, and a third processing unit. The first model building unit is used for building a first constraint model, wherein the sum of a plurality of input sequences is equal to a historical power load data sequence, the second model building unit is used for building a second constraint model, the sum of bandwidths of a plurality of input sequences is smaller than or equal to a second preset threshold value in the second constraint model, the fourth calculation unit is used for decomposing the historical power load data sequence into k sub-modal components, k=1, the second judgment unit is used for judging whether the k sub-modal components simultaneously meet the first constraint model and the second constraint model, the second processing unit is used for taking the k sub-modal components as a plurality of modal component sequences when the k sub-modal components simultaneously meet the first constraint model and the second constraint model, the third processing unit is used for setting k to k+1 when the k sub-modal components do not simultaneously meet the first constraint model and the second constraint model, and executing the step of the fourth calculation unit in a skip mode.
In some embodiments, the acquisition module 701 further comprises an acquisition unit. The acquisition unit is used for acquiring at least two data sequences of a plurality of personnel activity related data sequences and a plurality of weather related data sequences as a plurality of original data sequences.
In some embodiments, model training module 704 further includes a model training unit. The model training unit is used for combining each modal component sequence with the corresponding target original data sequence and jointly using the modal component sequences as training samples to train the long-short-term memory network model, so as to obtain a power load data prediction model group consisting of a plurality of long-short-term memory network models.
According to another aspect of an embodiment of the present application, there is provided a power load data predictive model set training apparatus. Referring specifically to fig. 8, fig. 8 is a schematic block diagram of a power load data prediction model set training apparatus according to an embodiment of the present application, and the embodiment of the present application does not limit the implementation of the power load data prediction model set training apparatus.
As shown in fig. 8, the power load data prediction model group training apparatus may include: a processor 802, a memory 806, a communication interface 804, and a communication bus 808.
Wherein the processor 802, the memory 806, and the communication interface 804 perform communication with each other via a communication bus 808. The memory 806 is configured to store at least one program 810, and the program 810 causes the processor 802 to perform relevant steps in the power load data prediction model set training method embodiment as described above.
In particular, program 810 may include program code including computer-executable instructions.
The processor 802 may be a central processing unit CPU, or an Application-specific integrated Circuit ASIC (Application SPECIFIC INTEGRATED Circuit), or one or more integrated circuits configured to implement embodiments of the present application. The one or more processors included in the power load data prediction model set training apparatus may be the same type of processor, such as one or more CPUs; but may also be different types of processors such as one or more CPUs and one or more ASICs.
Memory 806 for storing a program 810. The memory 806 may include high-speed RAM memory or may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
Program 810 may be specifically invoked by processor 802 to cause a power load data prediction model set training device to: acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence; performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences; determining a target original data sequence which meets the correlation requirement with any one modal component sequence in a plurality of original data sequences; and combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training, and generating a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
The embodiment of the application also provides a computer readable storage medium, wherein executable instructions are stored in the storage medium, and when the executable instructions run on the power load data prediction model group training device, the power load data prediction model group training device executes the power load data prediction model group training method in any embodiment.
The executable instructions may be specifically configured to cause the electrical load data prediction model set training apparatus to: acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence; performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences; determining a target original data sequence which meets the correlation requirement with any one modal component sequence in a plurality of original data sequences; and combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training, and generating a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
The algorithms or displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general-purpose systems may also be used with the teachings herein. The required structure for a construction of such a system is apparent from the description above. In addition, embodiments of the present application are not directed to any particular programming language. It will be appreciated that the teachings of the present application described herein may be implemented in a variety of programming languages, and the above description of specific languages is provided for disclosure of enablement and best mode of the present application.
In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the above description of exemplary embodiments of the application, various features of the embodiments of the application are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects.
Those skilled in the art will appreciate that the modules in the apparatus of the embodiments may be adaptively changed and disposed in one or more apparatuses different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component, and they may be divided into a plurality of sub-modules or sub-units or sub-components. Any combination of all of the features disclosed in this specification (including the accompanying abstract and drawings), and all of the processes or units of any method or apparatus so disclosed, may be used in combination, except insofar as at least some of such features and/or processes or units are mutually exclusive. Each feature disclosed in this specification (including the accompanying abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. The use of the words first, second, third, etc. do not denote any order. These words may be interpreted as names. The steps in the above embodiments should not be construed as limiting the order of execution unless specifically stated.
Claims (10)
1. A method for training a power load data prediction model set, comprising:
Acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence;
Performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences;
determining a target original data sequence which meets the correlation requirement with any one modal component sequence in the plurality of original data sequences;
and combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to perform model training, so as to generate a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
2. The method of claim 1, wherein determining a target raw data sequence of the plurality of raw data sequences that meets a correlation requirement with any one of the modal component sequences comprises:
Respectively calculating covariance between each original data sequence and an nth modal component sequence to obtain a plurality of covariance data, wherein n=1;
respectively calculating the product of the standard deviation of each original data sequence and the standard deviation of the nth modal component sequence to obtain a plurality of standard deviation product data;
Taking the covariance data as a dividend, respectively calculating the quotient of each covariance data and each corresponding standard deviation product to obtain a correlation coefficient between each original data sequence and an nth modal component sequence, wherein each covariance data and the corresponding standard deviation product are calculated according to the same original data sequence;
Setting n as n+1, and jumping to the step of calculating covariance between each original data sequence and the nth modal component sequence respectively until all modal component sequences are traversed;
judging whether the correlation coefficient between each original data sequence and any modal component sequence is larger than or equal to a first preset threshold value or not respectively;
and determining the original data sequence with the correlation coefficient larger than or equal to the first preset threshold value as a target original data sequence corresponding to the modal component sequence.
3. The method for training a power load data prediction model set according to claim 1, wherein performing modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences comprises:
establishing a first constraint model in which a sum of a plurality of input sequences is equal to the historical power load data sequence;
Establishing a second constraint model, wherein the sum of bandwidths of the plurality of input sequences is smaller than or equal to a second preset threshold value;
Decomposing the historical electrical load data sequence into k sub-modal components, wherein k = 1;
Judging whether the k sub-modal components simultaneously meet the first constraint model and the second constraint model;
if yes, the k sub-modal components are used as the multi-modal component sequences;
If not, setting k to k+1, and jumping to the step of decomposing the historical power load data sequence into k sub-modal components.
4. The method of claim 1, wherein the obtaining historical power load data sequences and the plurality of raw data sequences comprises:
And acquiring at least two data sequences of a plurality of personnel activity related data sequences and a plurality of weather related data sequences as the plurality of original data sequences.
5. The method of claim 4, wherein the plurality of personnel activity related data sequences comprises a flow of people data sequence and the plurality of weather related data sequences comprises a precipitation data sequence, an air temperature data sequence, an air pressure data sequence, and a cloud data sequence.
6. The method for training a power load data prediction model set according to claim 1, wherein the step of combining each modal component sequence with a corresponding target raw data sequence and jointly performing model training as a training sample to generate the power load data prediction model set includes:
And respectively combining each modal component sequence with the corresponding target original data sequence and jointly using the modal component sequences as training samples to train the long-short-time memory network model, so as to obtain the power load data prediction model group consisting of a plurality of long-short-time memory network models.
7. A method of predicting electrical load data, comprising:
Acquiring a plurality of modal component sequences, wherein the modal component sequences are obtained by carrying out modal decomposition on a historical power load data sequence;
Obtaining a target original data sequence, wherein the target original data sequence is an original data sequence meeting the correlation requirement with any one modal component sequence;
Predicting a modal component sequence corresponding to the power load data prediction model set by using a plurality of prediction models in the power load data prediction model set to obtain a plurality of component prediction results, wherein the power load data prediction model set is obtained by training a power load data prediction model set training method according to any one of claims 1-6, and the power load data prediction model set comprises the plurality of prediction models;
And linearly adding the plurality of component prediction results to obtain a power load data prediction result.
8. A power load data predictive model set training apparatus, comprising:
The acquisition module is used for acquiring a historical power load data sequence and a plurality of original data sequences, wherein the plurality of original data sequences have correlation with the historical power load data sequence;
The calculation module is used for carrying out modal decomposition on the historical power load data sequence to obtain a plurality of modal component sequences;
The determining module is used for determining a target original data sequence which meets the correlation requirement with any one modal component sequence in the plurality of original data sequences;
The model training module is used for combining each modal component sequence with the corresponding target original data sequence respectively and jointly serving as a training sample to carry out model training to generate a power load data prediction model group, wherein the power load data prediction model group comprises a plurality of prediction models respectively corresponding to each modal component sequence.
9. A power load data predictive model set training apparatus, comprising: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete communication with each other through the communication bus;
The memory is configured to store at least one program that causes the processor to perform the operations of the power load data predictive model set training method of any one of claims 1-6.
10. A computer readable storage medium, characterized in that the storage medium has stored therein executable instructions that, when run on a power load data prediction model set training device, cause the power load data prediction model set training device to perform the operations of the power load data prediction model set training method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410264610.6A CN117932345A (en) | 2024-03-08 | 2024-03-08 | Power load data prediction model group training method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410264610.6A CN117932345A (en) | 2024-03-08 | 2024-03-08 | Power load data prediction model group training method, device, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117932345A true CN117932345A (en) | 2024-04-26 |
Family
ID=90770399
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410264610.6A Pending CN117932345A (en) | 2024-03-08 | 2024-03-08 | Power load data prediction model group training method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117932345A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230096258A1 (en) * | 2019-06-21 | 2023-03-30 | Siemens Aktiengesellschaft | Power load data prediction method and device, and storage medium |
| CN116561567A (en) * | 2022-12-06 | 2023-08-08 | 国网浙江省电力有限公司丽水供电公司 | Short-term photovoltaic power prediction model based on variation modal decomposition, construction method and application method |
| CN116663746A (en) * | 2023-06-29 | 2023-08-29 | 神木县隆德矿业有限责任公司 | Power load prediction method and device, computer equipment and storage medium |
| CN117371510A (en) * | 2023-10-31 | 2024-01-09 | 南方电网数字电网研究院有限公司 | Power load prediction model training method and device based on domain generalization |
| CN117543544A (en) * | 2023-10-31 | 2024-02-09 | 国网浙江省电力有限公司杭州供电公司 | Load prediction method, device, equipment and storage medium |
-
2024
- 2024-03-08 CN CN202410264610.6A patent/CN117932345A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230096258A1 (en) * | 2019-06-21 | 2023-03-30 | Siemens Aktiengesellschaft | Power load data prediction method and device, and storage medium |
| CN116561567A (en) * | 2022-12-06 | 2023-08-08 | 国网浙江省电力有限公司丽水供电公司 | Short-term photovoltaic power prediction model based on variation modal decomposition, construction method and application method |
| CN116663746A (en) * | 2023-06-29 | 2023-08-29 | 神木县隆德矿业有限责任公司 | Power load prediction method and device, computer equipment and storage medium |
| CN117371510A (en) * | 2023-10-31 | 2024-01-09 | 南方电网数字电网研究院有限公司 | Power load prediction model training method and device based on domain generalization |
| CN117543544A (en) * | 2023-10-31 | 2024-02-09 | 国网浙江省电力有限公司杭州供电公司 | Load prediction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kadhem et al. | Computational techniques for assessing the reliability and sustainability of electrical power systems: A review | |
| CN112633316B (en) | Load prediction method and device based on boundary estimation theory | |
| CN106899660A (en) | Cloud data center energy-saving distribution implementation method based on trundle gray forecast model | |
| CN115796393B (en) | Energy management optimization method, system and storage medium based on multi-energy interaction | |
| CN111680841B (en) | Short-term load prediction method, system and terminal equipment based on principal component analysis | |
| CN113222403B (en) | Big data-based power regulation method and device, storage medium and electronic equipment | |
| CN103957261A (en) | Cloud computing resource distributing method based on energy consumption optimization | |
| KR20210125363A (en) | method for forecasting power demanding and apparatus adopting the method | |
| Wu et al. | A deadline-aware estimation of distribution algorithm for resource scheduling in fog computing systems | |
| CN108984830A (en) | A kind of building efficiency evaluation method and device based on FUZZY NETWORK analysis | |
| Man-Im et al. | Multi-objective economic dispatch considering wind power penetration using stochastic weight trade-off chaotic NSPSO | |
| CN107977741B (en) | Energy scheduling optimization method of combined cooling heating and power micro-grid system | |
| Dong et al. | Improving the accuracy of wind speed statistical analysis and wind energy utilization in the Ningxia Autonomous Region, China | |
| CN116611589B (en) | Power failure window period prediction method, system, equipment and medium for main network power transmission and transformation equipment | |
| CN112330044A (en) | Support vector regression model based on iterative aggregation grid search algorithm | |
| Cai et al. | An improved sequential importance sampling method for reliability assessment of renewable power systems with energy storage | |
| CN111652759A (en) | Comprehensive evaluation method and device for elastic load quick response and regulation demonstration engineering | |
| CN117932345A (en) | Power load data prediction model group training method, device, equipment and medium | |
| CN116345565A (en) | New energy and energy storage capacity combined optimization method, system, equipment and medium | |
| Hartmann et al. | CSAR: The cross-sectional autoregression model | |
| CN114611805A (en) | Net load prediction method and device, electronic equipment and storage medium | |
| Dal Pont et al. | MPC with machine learning applied to resource allocation problem using lambda architecture | |
| KR20220161849A (en) | Method and Apparatus for Predicting Appliance Power Usage State Using Non Intrusive Load Monitoring Based on Long-Short Term Memory | |
| CN112801415A (en) | Ultra-short-term load prediction method and system based on Markov chain distribution model | |
| Anagnostopoulos et al. | Sustainable supply chain management of electric grid power consumption load for smart cities based on second-order exponential smoothing algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |