CN112686456A - Power load prediction system and method combining edge calculation and energy consumption identification - Google Patents
Power load prediction system and method combining edge calculation and energy consumption identification Download PDFInfo
- Publication number
- CN112686456A CN112686456A CN202011642330.2A CN202011642330A CN112686456A CN 112686456 A CN112686456 A CN 112686456A CN 202011642330 A CN202011642330 A CN 202011642330A CN 112686456 A CN112686456 A CN 112686456A
- Authority
- CN
- China
- Prior art keywords
- data
- energy consumption
- identification
- voltage
- layer
- 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
- 238000005265 energy consumption Methods 0.000 title claims abstract description 110
- 238000004364 calculation method Methods 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 32
- 230000005540 biological transmission Effects 0.000 claims abstract description 31
- 238000005259 measurement Methods 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims description 40
- 230000006399 behavior Effects 0.000 claims description 36
- 238000007619 statistical method Methods 0.000 claims description 27
- 238000013528 artificial neural network Methods 0.000 claims description 26
- 238000012545 processing Methods 0.000 claims description 26
- 238000005070 sampling Methods 0.000 claims description 24
- 238000005192 partition Methods 0.000 claims description 21
- 238000010606 normalization Methods 0.000 claims description 15
- 238000013135 deep learning Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000005065 mining Methods 0.000 abstract description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The application discloses a power load prediction system and a power load prediction method combining edge calculation and energy consumption identification, wherein a load identification algorithm is embedded in a terminal chip by means of the edge calculation capacity of an intelligent electric meter, so that non-invasive mining and analysis of high-frequency measurement data of a user are realized, energy consumption behavior and energy consumption rule data of the user are obtained from the data, the calculation capacity of the intelligent electric meter terminal is fully utilized, the problems of network blockage, resource shortage and the like caused by transmission and storage of a large amount of data are avoided while the load prediction precision is improved, and the problems of low load prediction precision and network blockage and resource shortage caused by transmission and storage of a large amount of data in the conventional centralized load prediction mode are solved.
Description
Technical Field
The application relates to the technical field of power system load prediction, in particular to a power load prediction system and method combining edge calculation and energy consumption identification.
Background
The power system load prediction refers to a mathematical method for processing past and future loads under the condition of fully considering some important system operation characteristics, capacity increase decisions, natural conditions and social influences, and determines a load value at a specific future moment in the sense of meeting a certain precision requirement. Improving the accuracy of load prediction plays an important role in guaranteeing the operation safety of the system and improving the economic level of the system.
Most of the existing methods for predicting the power load of the power system adopt a centralized regression analysis or intelligent learning method for prediction, data sources of the methods comprise historical load, meteorological information, economic parameters and the like of a prediction region, and information such as energy utilization behaviors and energy utilization rules of various users in the region is not involved, and the information can be used for further improving the accuracy of load prediction; secondly, the traditional load prediction adopts a centralized method to analyze and calculate, namely, all data are stored, accessed and processed in a centralized manner, the requirement on computing resources is high, and when the high-frequency real-time energy consumption of a user is considered, the traditional load prediction does not have the feasibility in engineering without the help of massive computing and communication resources. Therefore, in order to solve the problems of low load prediction accuracy and network congestion and resource shortage caused by transmission and storage of a large amount of data in a centralized load prediction mode, the application provides a power load prediction system and method combining edge calculation and energy consumption identification.
Disclosure of Invention
The application provides a power load prediction system and method combining edge calculation and energy consumption identification, which are used for solving the problems of low load prediction precision and network congestion and resource shortage caused by transmission and storage of a large amount of data in the conventional centralized load prediction mode.
In view of the above, the present application provides, in a first aspect, a power load prediction system combining edge calculation and energy consumption identification, including an edge calculation layer, a data transmission layer, a statistical analysis layer, and a training prediction layer;
the edge calculation layer comprises an intelligent electric meter, a data acquisition module and a data processing chip are arranged in the intelligent electric meter, the data acquisition module is used for acquiring high-frequency measurement information of an electric power user in real time, the data processing chip is used for carrying out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module by adopting a load identification algorithm to obtain actual energy consumption behavior data of the electric power user, and the actual energy consumption behavior data are uploaded to the data transmission layer;
the data transmission layer comprises a data concentrator, the data concentrator is used for counting and combining the actual energy consumption behavior data to obtain a partition energy consumption sequence, and the partition energy consumption sequence is uploaded to the statistical analysis layer;
the statistical analysis layer is used for carrying out classification statistics on the partition energy consumption sequences uploaded by all the data concentrators in the region according to a preset classification principle to obtain region energy consumption data, and uploading the region energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
the training prediction layer is used for carrying out normalization processing on the region energy consumption data, substituting the data obtained after the normalization processing into a deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
Optionally, the load identification algorithm is an adaptive multi-class learning algorithm based on voltage-current trajectory and multi-feature fusion;
the step of adopting a load identification algorithm to carry out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module comprises the following steps:
acquiring voltage-current instantaneous sampling values of various electrical equipment in starting and stopping states, normalizing the voltage-current instantaneous sampling values, and drawing a voltage-current track graph by taking voltage as a horizontal axis and current as a vertical axis;
extracting characteristic indexes of a voltage-current track in the voltage-current track graph to form characteristic vectors, and constructing samples by combining sampling states of equipment to obtain starting sample sequences of various types of equipment, wherein the characteristic indexes comprise circulation directions, the number of self-intersection points, the slope of the middle section of an average curve, the curvature of the average curve and the track envelope area;
setting iteration times and initializing the weight of each sample;
starting iteration, training and classifying and identifying each sample according to the weight of each sample, calculating the weight sum of the samples identified as non-self type in the k type label samples, judging whether the weight sum of the samples identified as correct type in the k type label samples is greater than the weight sum of the samples identified as wrong type, if so, continuing to carry out circular iteration, and otherwise, returning to recalculate the weight sum of the samples identified as non-self type in the k type label samples;
calculating the error identification rate of the classification function of the classification identification;
updating the weight value, and normalizing the updated weight value to obtain an enhanced classification identifier;
and carrying out energy consumption subdivision identification through an enhanced classification identifier to obtain the actual energy consumption behavior data of the power consumer.
Optionally, the power consumers include industrial power consumers, commercial power consumers, and residential power consumers, and the high frequency measurement information includes voltage waveforms and current waveforms.
Optionally, the sampling frequency of the data acquired by the data acquisition module is not lower than 6000 Hz.
Optionally, the smart meter performs data transmission with the data concentrator through an RS485 bus.
Optionally, the data concentrator is transmitted to the statistical analysis layer via GPRS or 5G private network.
The second aspect of the present application provides a power load prediction method combining edge calculation and energy use identification, including:
the method comprises the steps that an edge calculation layer collects high-frequency measurement information of a power consumer in real time, energy consumption subdivision identification is carried out on the collected high-frequency measurement information by adopting a load identification algorithm to obtain actual energy consumption behavior data of the power consumer, and the actual energy consumption behavior data are uploaded to a data transmission layer;
the data transmission layer counts and combines the actual energy consumption behavior data to obtain a partition energy consumption sequence, and uploads the partition energy consumption sequence to the statistical analysis layer;
the statistical analysis layer classifies and counts the partition energy consumption sequences uploaded by all the data concentrators in an area according to a preset classification principle to obtain area energy consumption data, and uploads the area energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
the training prediction layer is used for carrying out normalization processing on the region energy consumption data, substituting the data obtained after the normalization processing into a deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
Optionally, the load identification algorithm is an adaptive multi-class learning algorithm based on voltage-current trajectory and multi-feature fusion;
the step of using the load identification algorithm to perform energy-consumption subdivision identification on the collected high-frequency measurement information comprises the following steps:
acquiring voltage-current instantaneous sampling values of various electrical equipment in starting and stopping states, normalizing the voltage-current instantaneous sampling values, and drawing a voltage-current track graph by taking voltage as a horizontal axis and current as a vertical axis;
extracting characteristic indexes of a voltage-current track in the voltage-current track graph to form characteristic vectors, and constructing samples by combining sampling states of equipment to obtain starting sample sequences of various types of equipment, wherein the characteristic indexes comprise circulation directions, the number of self-intersection points, the slope of the middle section of an average curve, the curvature of the average curve and the track envelope area;
setting iteration times and initializing the weight of each sample;
starting iteration, training and classifying and identifying each sample according to the weight of each sample, calculating the weight sum of the samples identified as non-self type in the k type label samples, judging whether the weight sum of the samples identified as correct type in the k type label samples is greater than the weight sum of the samples identified as wrong type, if so, continuing to carry out circular iteration, and otherwise, returning to recalculate the weight sum of the samples identified as non-self type in the k type label samples;
calculating the error identification rate of the classification function of the classification identification;
updating the weight value, and normalizing the updated weight value to obtain an enhanced classification identifier;
and carrying out energy consumption subdivision identification through an enhanced classification identifier to obtain the actual energy consumption behavior data of the power consumer.
Optionally, the sampling frequency of the data collected by the edge calculation layer is not lower than 6000 Hz.
Optionally, the power consumers include industrial power consumers, commercial power consumers, and residential power consumers, and the high frequency measurement information includes voltage waveforms and current waveforms.
According to the technical scheme, the embodiment of the application has the following advantages:
the application provides a power load prediction system combining edge calculation and energy consumption identification, which comprises an edge calculation layer, a data transmission layer, a statistical analysis layer and a training prediction layer; the edge calculation layer comprises an intelligent ammeter, a data acquisition module and a data processing chip are arranged in the intelligent ammeter, the data acquisition module is used for acquiring high-frequency measurement information of the power consumer in real time, the data processing chip is used for carrying out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module by adopting a load identification algorithm to obtain actual energy consumption behavior data of the power consumer, and the actual energy consumption behavior data are uploaded to the data transmission layer; the data transmission layer comprises a data concentrator, the data concentrator is used for counting and combining actual energy consumption behavior data to obtain a partition energy consumption sequence, and the partition energy consumption sequence is uploaded to the statistical analysis layer; the statistical analysis layer is used for classifying and counting the partition energy consumption sequences uploaded by all data concentrators in the region according to a preset classification principle to obtain region energy consumption data, and uploading the region energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position; the training prediction layer is used for carrying out normalization processing on the regional energy data, substituting the data obtained after the normalization processing into the deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system. By means of the edge computing capacity of the intelligent electric meter, the load identification algorithm is embedded in the terminal chip, so that non-invasive mining and analysis of high-frequency measurement data of a user are achieved, energy consumption behavior and energy consumption regular data of the user are obtained, the computing capacity of the intelligent electric meter terminal is fully utilized, the problems of network blockage, resource shortage and the like caused by transmission and storage of a large amount of data are avoided while the load prediction precision is improved, and the problems of low load prediction precision and network blockage and resource shortage caused by transmission and storage of a large amount of data in the existing centralized load prediction mode are solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other related drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic diagram of a physical architecture of a combined edge calculation and utilization identification power load prediction system provided in an embodiment of the present application;
fig. 2 is a schematic diagram of information transfer and processing flow of an identified power load prediction system in combination with edge calculation according to an embodiment of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example 1
For ease of understanding, referring to fig. 1 and 2, the present application provides an embodiment of a system for combining edge calculation with power load prediction with recognition capability, including an edge calculation layer, a data transmission layer, a statistical analysis layer, and a training prediction layer;
the edge calculation layer comprises an intelligent ammeter, a data acquisition module and a data processing chip are arranged in the intelligent ammeter, the data acquisition module is used for acquiring high-frequency measurement information of the power consumer in real time, the data processing chip is used for carrying out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module by adopting a load identification algorithm to obtain actual energy consumption behavior data of the power consumer, and the actual energy consumption behavior data are uploaded to the data transmission layer;
the data transmission layer comprises a data concentrator, the data concentrator is used for counting and combining the actual energy consumption behavior data to obtain a partition energy consumption sequence, and the partition energy consumption sequence is uploaded to the statistical analysis layer;
the statistical analysis layer is used for classifying and counting the partition energy consumption sequences uploaded by all data concentrators in the region according to a preset classification principle to obtain region energy consumption data, and uploading the region energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
the training prediction layer is used for carrying out normalization processing on the regional energy data, substituting the data obtained after the normalization processing into the deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
The physical architecture of the power load prediction system combining edge calculation and identification provided by the embodiment of the application is divided into an edge calculation layer, a data transmission layer, a statistical analysis layer and a training prediction layer. The edge computing layer intelligently senses the energy using behaviors of various users by using the computing power of the intelligent electric meter and extracts valuable electricity using sequence information; the data transmission layer carries out primary combination processing on the use energy recognition results of the connected users through the data concentrator, and then transmits the processed and compressed information to the statistical analysis layer; the statistical analysis layer is responsible for performing statistical classification on the energy use behaviors of all users in the region according to the time sequence, the user types and the electric appliance categories; and the training prediction layer performs repeated training learning according to the overall energy utilization characteristics of the users in the region and other data source information, so as to obtain a trained network and use the trained network for subsequent load prediction.
The power load prediction system combining edge calculation and energy consumption identification provided in the embodiment of the application is characterized in that a load identification algorithm is embedded in a terminal chip by means of the edge calculation capacity of an intelligent electric meter, so that non-invasive mining and analysis of high-frequency measurement data of a user are realized, energy consumption behavior and energy consumption regular data of the user are obtained from the data, the calculation capacity of the intelligent electric meter terminal is fully utilized, the problems of network blockage, resource shortage and the like caused by transmission and storage of a large amount of data are avoided while the load prediction precision is improved, and the problems of low load prediction precision and network blockage and resource shortage caused by transmission and storage of a large amount of data in the existing centralized load prediction mode are solved.
Example 2
For ease of understanding, referring to fig. 1 and 2, the present application provides another embodiment of a system for combining edge calculation with power load prediction with recognition, including an edge calculation layer, a data transmission layer, a statistical analysis layer, and a training prediction layer;
the edge calculation layer comprises an intelligent electric meter arranged at a power consumer, a data acquisition module and a data processing chip are arranged in the intelligent electric meter, high-frequency measurement information of various users (including industry, business and residents) is acquired in real time by using the data acquisition module of the intelligent electric meter, the acquired information comprises voltage and current waveforms, and the sampling frequency is not lower than 6000 Hz. The load identification algorithm embedded in the intelligent electric meter chip is utilized to carry out energy consumption subdivision identification on the collected user measurement information to obtain the actual energy consumption behaviors of the current user, including the start-stop and running sequences of electric appliances such as a water heater, an air conditioner, a refrigerator, an induction cooker and the like, the adopted load identification algorithm is a self-adaptive multi-classification learning algorithm based on voltage-current tracks and multi-feature fusion, the basic idea of the algorithm is that by extracting the implicit characteristics of voltage and current waveforms of various electrical equipment, including the circulating direction of a voltage-current track, the number of self-intersection points, the slope of the middle section of an average curve, the curvature of the average curve, the envelope area of the track and the like, then the characteristics and the labels of the electric appliances are taken as training samples to be substituted into an adaptive multi-classification learning algorithm for training, therefore, an enhanced classifier is obtained, and finally the classifier is built in the intelligent electric meter to identify and sense the user energy using behaviors. The method for carrying out energy-using subdivision identification on the high-frequency measurement information acquired by the data acquisition module by adopting the self-adaptive multi-classification learning algorithm based on the voltage-current track and multi-feature fusion comprises the following steps:
(1) acquiring voltage-current instantaneous sampling values of various electrical equipment (a water heater, an electric lamp, a refrigerator, an air conditioner, an induction cooker and the like) in starting and stopping states, normalizing the voltage and current sampling values, and then drawing a voltage-current track graph by taking the voltage as a horizontal axis and the current as a vertical axis;
(2) extracting 5 indexes of the circulating direction of the voltage-current track, the number of self-intersection points, the slope of the middle section of the average curve, the curvature of the average curve and the track envelope area to form a characteristic vector, and constructing a sample by combining the sampling state of the equipment to obtain a starting sample sequence of various equipmentAnd shutdown sample sequenceIt should be noted that the same type of equipment may have different brands and models, and the voltage-current tracks thereof also have a certain difference, so as to ensure the accuracy of identification, and to sample the characteristics of the equipment of the common brand as much as possible.
For example, the instantaneous values of voltage and current of the water heater (i-th device) in the starting state are collected, and the feature vector is extracted The starting state is indicated by the superscript 1, which corresponds to 5 characteristic indexes of the circulation direction of the voltage-current track, the number of self-intersection points, the slope of the middle section of the average curve, the curvature of the average curve and the track envelope area. Then the label of the electric appliance is combined(representing feature vectors)Class i appliances corresponding to activated state, distinct from other devices and states), a sample can be constructed
(3) Setting iteration times N and initializing the weight of each sampleD is the weight distribution of the samples, i.e.
(4) for N is 1,2, …, N, the following steps are performed:
a. according to weight ωnTraining and class recognition of samples, i.e. hnX → Y, wherein X is a feature space, Y is a label space, hnIs a classification function;
b. for K is 1,2, …, K, the following steps (I) to (II) are performed. Wherein K is the label number of the electric appliance category and the operation state, namely the category number.
(I) Calculating the sum of the weights of the class k labeled samples identified as other class samples, i.e.
for j=1,2,…,K
(II) judging whether the sum of the weight values of the samples which are correctly identified in the K (K is 1,2, …, K) type label is larger than the sum of the weight values of the samples which are incorrectly identified, namely judging whether the sum of the weight values of the samples which are correctly identified in the K (K is 1,2, …, K) type label is larger than theIf yes, continuing to circulate, otherwise, returning to the step (a) for recalculation;
c. calculate hnError recognition rate of (2):
(5) the resulting enhanced classification recognizer is:
according to the above recognition method, the user's energy use behavior can be perceived. For a certain user t-th time period, the result of load identification is the on-off state u and the power p of each device, for example, the on-off state u of lighting in the t-th time period (t is 1,2, …, 24)light,tAnd power plight,tOn-off state u of air conditionerair,tAnd power pair,tStart-stop state u of refrigeratorrefrig,tAnd power prefrig,tOn-off state u of electric water heaterheater,tAnd power pheater,tAnd the like. Then the total power for this user t period can be expressed as:
Pt=ulight,tplight,t+uair,tpair,t+urefrig,tprefrig,t+uheater,tpheater,t+…
the energy consumption of the user is continuously monitored and identified, and the start-stop sequence of various devices, namely the actual energy consumption behavior data of the power user, can be obtained.
The intelligent electric meter uploads the energy consumption identification results of each power consumer to a corresponding data concentrator through an RS485 bus, the data concentrator conducts preliminary statistics and combination on the energy consumption information to form a more simplified energy consumption sequence for subareas (the start and stop states of all equipment in each period), and then the energy consumption sequence is transmitted to an energy consumption center using a statistical analysis layer by means of a GPRS or 5G private network.
The energy utilization information of the whole area is classified and counted by the adoption of the acquisition center, the classification principle comprises user types, time sequences, electric appliance categories, geographic positions and the like, and data with the structure shown in the table 1 are finally formed. The statistical data of each type of equipment in table 1 is the sum of all corresponding equipment in the startup state in the t-period region.
TABLE 1
Transmitting the counted energy utilization rule of the area and other data information (including meteorological data, load data, economic data and the like) to a calculation center of a training prediction layer, and normalizing the parameters according to the following formula by the calculation center;
wherein x is an input parameter sequence, xtIs the t-th sequence parameter, ytIs xtNormalized values.
For example, x is a user lighting statistical data sequence, xtFor all the users in the activated state during the t-th period, i.e. illuminating(superscript i represents the ith user), min (x), max (x) represent the minimum and maximum values of the illumination data sequence, respectively, and then the normalized value y of the t period can be obtained according to the above formulat。
Substituting the normalized data into a deep learning neural network for repeated training to finally obtain a trained neural network structure and parameters;
the neural network is used for predicting future loads, and a prediction result with higher precision can be obtained.
The embodiment of the application provides a power load prediction system combining edge calculation and energy consumption identification, and a load identification algorithm is embedded in a terminal chip by means of the edge calculation capacity of an intelligent electric meter, so that non-invasive mining and analysis of high-frequency measurement data of a user are realized, and energy consumption behaviors and energy consumption rule data of the user are obtained from the non-invasive mining and analysis. And then, transmitting the small amount of data (including the user type, start-stop sequences of various electrical appliances and the like) back to a statistical analysis layer through a data concentrator and a 5G network for storage and analysis, obtaining all-region user energy statistical information of the user type, the time-sharing period and the distributor type, and then transmitting the information and other data (load data, meteorological data, economic data and working day/non-working day data) into a deep learning neural network for repeated training, thereby finally obtaining the neural network with higher prediction precision. The power load prediction system provided by the embodiment of the application can make full use of the computing power of the intelligent terminal, improves the load prediction precision, avoids the problems of network blockage, resource shortage and the like caused by a large amount of data transmission and storage, and has certain economic benefit and practical value.
Example 3
Embodiments of a method for combining edge calculation with power load prediction with energy recognition are provided herein, comprising:
step 101, an edge calculation layer collects high-frequency measurement information of a power consumer in real time, energy consumption subdivision identification is carried out on the collected high-frequency measurement information by adopting a load identification algorithm to obtain actual energy consumption behavior data of the power consumer, and the actual energy consumption behavior data is uploaded to a data transmission layer;
102, the data transmission layer counts and combines the actual energy consumption behavior data to obtain a partition energy consumption sequence, and uploads the partition energy consumption sequence to the statistical analysis layer;
103, classifying and counting the partition energy consumption sequences uploaded by all data concentrators in the area by the statistical analysis layer according to a preset classification principle to obtain area energy consumption data, and uploading the area energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
and 104, performing normalization processing on the regional energy data by the training prediction layer, substituting the data obtained after the normalization processing into the deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
The load identification algorithm is a self-adaptive multi-classification learning algorithm based on voltage-current tracks and multi-feature fusion;
the step of using a load identification algorithm to perform energy consumption subdivision identification on the collected high-frequency measurement information comprises the following steps:
acquiring voltage-current instantaneous sampling values of various electrical equipment in starting and stopping states, normalizing the voltage-current instantaneous sampling values, and drawing a voltage-current track graph by taking voltage as a horizontal axis and current as a vertical axis;
extracting characteristic indexes of a voltage-current track in a voltage-current track graph to form a characteristic vector, and constructing a sample by combining a sampling state of equipment to obtain a starting sample sequence of various equipment, wherein the characteristic indexes comprise a circulation direction, the number of self-intersection points, a slope of a middle section of an average curve, a curvature of the average curve and a track envelope area;
setting iteration times and initializing the weight of each sample;
starting iteration, training and classifying and identifying each sample according to the weight of each sample, calculating the weight sum of the samples identified as non-self type in the k type label samples, judging whether the weight sum of the samples identified as correct type in the k type label samples is greater than the weight sum of the samples identified as wrong type, if so, continuing to carry out circular iteration, and otherwise, returning to recalculate the weight sum of the samples identified as non-self type in the k type label samples;
calculating the error identification rate of the classification function of the classification identification;
updating the weight value, and normalizing the updated weight value to obtain an enhanced classification identifier;
and carrying out energy consumption subdivision identification through an enhanced classification identifier to obtain the actual energy consumption behavior data of the power consumer.
Further, the data concentrator transmits to the statistical analysis layer through a GPRS or 5G private network, and the intelligent electric meter transmits data with the data concentrator through an RS485 bus.
Further, the sampling frequency of the data collected by the edge calculation layer is not lower than 6000 Hz.
Further, the power consumers include industrial power consumers, commercial power consumers and residential power consumers, and the high frequency measurement information includes voltage waveforms and current waveforms.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. A power load prediction system combining edge calculation and energy consumption identification is characterized by comprising an edge calculation layer, a data transmission layer, a statistical analysis layer and a training prediction layer;
the edge calculation layer comprises an intelligent electric meter, a data acquisition module and a data processing chip are arranged in the intelligent electric meter, the data acquisition module is used for acquiring high-frequency measurement information of an electric power user in real time, the data processing chip is used for carrying out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module by adopting a load identification algorithm to obtain actual energy consumption behavior data of the electric power user, and the actual energy consumption behavior data are uploaded to the data transmission layer;
the data transmission layer comprises a data concentrator, the data concentrator is used for counting and combining the actual energy consumption behavior data to obtain a partition energy consumption sequence, and the partition energy consumption sequence is uploaded to the statistical analysis layer;
the statistical analysis layer is used for carrying out classification statistics on the partition energy consumption sequences uploaded by all the data concentrators in the region according to a preset classification principle to obtain region energy consumption data, and uploading the region energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
the training prediction layer is used for carrying out normalization processing on the region energy consumption data, substituting the data obtained after the normalization processing into a deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
2. The combined edge calculation and identifiable power load prediction system of claim 1, wherein the load identification algorithm is an adaptive multi-class learning algorithm based on voltage-current trajectories and multi-feature fusion;
the step of adopting a load identification algorithm to carry out energy consumption subdivision identification on the high-frequency measurement information acquired by the data acquisition module comprises the following steps:
acquiring voltage-current instantaneous sampling values of various electrical equipment in starting and stopping states, normalizing the voltage-current instantaneous sampling values, and drawing a voltage-current track graph by taking voltage as a horizontal axis and current as a vertical axis;
extracting characteristic indexes of a voltage-current track in the voltage-current track graph to form characteristic vectors, and constructing samples by combining sampling states of equipment to obtain starting sample sequences of various types of equipment, wherein the characteristic indexes comprise circulation directions, the number of self-intersection points, the slope of the middle section of an average curve, the curvature of the average curve and the track envelope area;
setting iteration times and initializing the weight of each sample;
starting iteration, training and classifying and identifying each sample according to the weight of each sample, calculating the weight sum of the samples identified as non-self type in the k type label samples, judging whether the weight sum of the samples identified as correct type in the k type label samples is greater than the weight sum of the samples identified as wrong type, if so, continuing to carry out circular iteration, and otherwise, returning to recalculate the weight sum of the samples identified as non-self type in the k type label samples;
calculating the error identification rate of the classification function of the classification identification;
updating the weight value, and normalizing the updated weight value to obtain an enhanced classification identifier;
and carrying out energy consumption subdivision identification through an enhanced classification identifier to obtain the actual energy consumption behavior data of the power consumer.
3. The combined edge calculation and energy use identification power load prediction system of claim 1, wherein the power consumers include industrial power consumers, commercial power consumers, and residential power consumers, and the high frequency measurement information includes voltage waveforms and current waveforms.
4. The system of claim 1, wherein the data acquisition module acquires data at a sampling frequency of no less than 6000 Hz.
5. The system of claim 1, wherein the smart meter is configured to communicate data with the data concentrator via an RS485 bus.
6. The system of claim 1, wherein the data concentrator is transmitted to the statistical analysis layer via GPRS or 5G private network.
7. A method for combining edge calculation with power load prediction with energy recognition, comprising:
the method comprises the steps that an edge calculation layer collects high-frequency measurement information of a power consumer in real time, energy consumption subdivision identification is carried out on the collected high-frequency measurement information by adopting a load identification algorithm to obtain actual energy consumption behavior data of the power consumer, and the actual energy consumption behavior data are uploaded to a data transmission layer;
the data transmission layer counts and combines the actual energy consumption behavior data to obtain a partition energy consumption sequence, and uploads the partition energy consumption sequence to the statistical analysis layer;
the statistical analysis layer classifies and counts the partition energy consumption sequences uploaded by all the data concentrators in an area according to a preset classification principle to obtain area energy consumption data, and uploads the area energy consumption data to the training prediction layer, wherein the preset classification principle comprises a user type, a time sequence, an electric appliance type and a geographic position;
the training prediction layer is used for carrying out normalization processing on the region energy consumption data, substituting the data obtained after the normalization processing into a deep learning neural network for training to obtain a trained neural network, and using the trained neural network for load prediction of the power system.
8. The method for combined edge calculation and identifiable power load prediction according to claim 7, wherein the load identification algorithm is an adaptive multi-class learning algorithm based on voltage-current trajectory and multi-feature fusion;
the step of using the load identification algorithm to perform energy-consumption subdivision identification on the collected high-frequency measurement information comprises the following steps:
acquiring voltage-current instantaneous sampling values of various electrical equipment in starting and stopping states, normalizing the voltage-current instantaneous sampling values, and drawing a voltage-current track graph by taking voltage as a horizontal axis and current as a vertical axis;
extracting characteristic indexes of a voltage-current track in the voltage-current track graph to form characteristic vectors, and constructing samples by combining sampling states of equipment to obtain starting sample sequences of various types of equipment, wherein the characteristic indexes comprise circulation directions, the number of self-intersection points, the slope of the middle section of an average curve, the curvature of the average curve and the track envelope area;
setting iteration times and initializing the weight of each sample;
starting iteration, training and classifying and identifying each sample according to the weight of each sample, calculating the weight sum of the samples identified as non-self type in the k type label samples, judging whether the weight sum of the samples identified as correct type in the k type label samples is greater than the weight sum of the samples identified as wrong type, if so, continuing to carry out circular iteration, and otherwise, returning to recalculate the weight sum of the samples identified as non-self type in the k type label samples;
calculating the error identification rate of the classification function of the classification identification;
updating the weight value, and normalizing the updated weight value to obtain an enhanced classification identifier;
and carrying out energy consumption subdivision identification through an enhanced classification identifier to obtain the actual energy consumption behavior data of the power consumer.
9. The method for integrating edge calculation and identifiable power load prediction according to claim 7, wherein the edge calculation layer collects data at a sampling frequency of not less than 6000 Hz.
10. The combined edge calculation and energy use identification power load prediction method of claim 7, wherein the power consumers include industrial power consumers, commercial power consumers and residential power consumers, and the high frequency measurement information includes voltage waveforms and current waveforms.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011642330.2A CN112686456A (en) | 2020-12-31 | 2020-12-31 | Power load prediction system and method combining edge calculation and energy consumption identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011642330.2A CN112686456A (en) | 2020-12-31 | 2020-12-31 | Power load prediction system and method combining edge calculation and energy consumption identification |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112686456A true CN112686456A (en) | 2021-04-20 |
Family
ID=75456730
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011642330.2A Pending CN112686456A (en) | 2020-12-31 | 2020-12-31 | Power load prediction system and method combining edge calculation and energy consumption identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112686456A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113240184A (en) * | 2021-05-21 | 2021-08-10 | 浙江大学 | Building space unit cold load prediction method and system based on federal learning |
CN113361802A (en) * | 2021-06-28 | 2021-09-07 | 杭州易龙电安科技有限公司 | Power load identification and prediction method based on artificial intelligence |
CN114707737A (en) * | 2022-04-12 | 2022-07-05 | 广州汉光电气股份有限公司 | System and method for predicting power consumption based on edge calculation |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104820825A (en) * | 2015-04-27 | 2015-08-05 | 北京工业大学 | Adaboost algorithm-based face recognition optimization method |
-
2020
- 2020-12-31 CN CN202011642330.2A patent/CN112686456A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104820825A (en) * | 2015-04-27 | 2015-08-05 | 北京工业大学 | Adaboost algorithm-based face recognition optimization method |
Non-Patent Citations (3)
Title |
---|
杨新武等: "基于弱分类器调整的多分类Adaboost算法", 《电子与信息学报》 * |
祁兵等: "居民负荷特征研究及特征库的建立", 《东北电力技术》 * |
钟韬等: "基于决策树的非入侵式负荷分解算法的研究", 《计算机应用研究》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113240184A (en) * | 2021-05-21 | 2021-08-10 | 浙江大学 | Building space unit cold load prediction method and system based on federal learning |
CN113240184B (en) * | 2021-05-21 | 2022-06-24 | 浙江大学 | Building space unit cold load prediction method and system based on federal learning |
CN113361802A (en) * | 2021-06-28 | 2021-09-07 | 杭州易龙电安科技有限公司 | Power load identification and prediction method based on artificial intelligence |
CN114707737A (en) * | 2022-04-12 | 2022-07-05 | 广州汉光电气股份有限公司 | System and method for predicting power consumption based on edge calculation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112686456A (en) | Power load prediction system and method combining edge calculation and energy consumption identification | |
CN109376772B (en) | Power load combination prediction method based on neural network model | |
CN109345409A (en) | A kind of resident's comprehensive energy efficiency management method based on bandwidth carrier | |
CN115796393B (en) | Energy management optimization method, system and storage medium based on multi-energy interaction | |
CN103093288A (en) | Partition power grid bus load prediction system based on weather information | |
CN103577883A (en) | Grid-load intelligent interaction method and device | |
CN109215344B (en) | Method and system for urban road short-time traffic flow prediction | |
CN111751650B (en) | Non-invasive household electric equipment on-line monitoring system and fault identification method | |
CN111563827B (en) | Load decomposition method based on physical characteristics of electrical appliance and electricity utilization behaviors of residents | |
CN108399221A (en) | Indoor electric equipment classifying identification method and system based on big data association analysis | |
CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
CN108346009A (en) | A kind of power generation configuration method and device based on user model self study | |
CN116826710A (en) | Peak clipping strategy recommendation method and device based on load prediction and storage medium | |
CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
CN113361454A (en) | Deep learning non-invasive load monitoring method based on unsupervised optimization | |
CN111144628A (en) | Distributed energy supply type cooling, heating and power load prediction model system and method | |
CN116470491A (en) | Photovoltaic power probability prediction method and system based on copula function | |
CN116298576A (en) | Time-segmentation-considered extensible non-invasive load monitoring method | |
CN112579721B (en) | Method and system for constructing crowd distribution map, terminal equipment and storage medium | |
CN115511218A (en) | Intermittent type electrical appliance load prediction method based on multi-task learning and deep learning | |
CN103970016B (en) | Multipolar magnesium electrolysis bath state recognition control system and recognition control method thereof | |
CN111047465B (en) | Power grid friendly load response method based on power big data | |
Zhang et al. | Load prediction based on depthwise separable convolution model | |
Chavda et al. | Electricity consumption patterns using SOM-based two-level clustering of residential households | |
CN118195279B (en) | Heating management scheduling method and system based on renewable energy intelligent coupling |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210420 |