CN117973644B - Distributed photovoltaic power virtual acquisition method considering optimization of reference power station - Google Patents
Distributed photovoltaic power virtual acquisition method considering optimization of reference power station Download PDFInfo
- Publication number
- CN117973644B CN117973644B CN202410389240.9A CN202410389240A CN117973644B CN 117973644 B CN117973644 B CN 117973644B CN 202410389240 A CN202410389240 A CN 202410389240A CN 117973644 B CN117973644 B CN 117973644B
- Authority
- CN
- China
- Prior art keywords
- virtual
- distributed photovoltaic
- reference power
- power station
- optimization
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000005457 optimization Methods 0.000 title claims abstract description 48
- 230000006870 function Effects 0.000 claims abstract description 84
- 230000007246 mechanism Effects 0.000 claims abstract description 40
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000012546 transfer Methods 0.000 claims abstract description 17
- 230000002787 reinforcement Effects 0.000 claims abstract description 11
- 238000005259 measurement Methods 0.000 claims abstract description 6
- 230000000739 chaotic effect Effects 0.000 claims abstract description 4
- 238000013507 mapping Methods 0.000 claims abstract description 4
- 230000009471 action Effects 0.000 claims description 36
- 238000012549 training Methods 0.000 claims description 35
- 238000009826 distribution Methods 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 30
- 239000003795 chemical substances by application Substances 0.000 claims description 27
- 238000004364 calculation method Methods 0.000 claims description 25
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 9
- 238000002790 cross-validation Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 6
- 230000003993 interaction Effects 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000006399 behavior Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000005728 strengthening Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 2
- 241000282346 Meles meles Species 0.000 abstract description 5
- 238000012545 processing Methods 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 2
- 238000012423 maintenance Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 13
- 230000000694 effects Effects 0.000 description 9
- 238000004590 computer program Methods 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 5
- 238000003860 storage Methods 0.000 description 4
- 241000288105 Grus Species 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000009434 installation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 241000282461 Canis lupus Species 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 241000287127 Passeridae Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention provides a distributed photovoltaic power virtual acquisition method considering reference power station optimization, which relates to the technical field of data processing and comprises the following steps: step 1: the gating circulation unit is used as a virtual collector to capture the time dependence of the distributed photovoltaic. A loss function and an attention mechanism based on Wasserstein distance measurement are introduced into the GRU to assist the virtual collector to learn complex mapping relations between different power stations; step 2: introducing a time-varying binary transfer function and a chaotic initialization strategy into a traditional badger optimization algorithm to search a reference power station set with highest virtual acquisition precision; step 3: and adopting model-free reinforcement learning depth DQN to adaptively and dynamically adjust super parameters of the virtual collector. The invention solves the problems of high acquisition cost and low acquisition reliability caused by mass distributed photovoltaic construction, and has important significance for better developing fine management and fine operation and maintenance of future distributed photovoltaic power stations.
Description
Technical Field
The invention belongs to the technical field of electric power, relates to distributed photovoltaic operation data acquisition, and discloses a distributed photovoltaic power virtual acquisition method considering optimization of a reference power station.
Background
With the increase in energy shortage and environmental pollution, photovoltaic systems have become one of the important means for solving energy and environmental problems. Compared with the centralized photovoltaic, the distributed photovoltaic has the advantages of flexible installation, high energy utilization rate and the like. However, the increase in the number of distributed photovoltaics makes monitoring the operational status of the distributed photovoltaics require a large number of sensors and communication devices. Distributed photovoltaic data transmission in remote areas requires high communication costs, and large data volumes require extensive servers, databases, and data monitoring platforms, resulting in many users not willing to use the service for privacy or cost reasons. In addition, with the complex and changeable geographical environment and weather conditions of the distributed photovoltaic installation location, the transmission process often has the problems of data missing, transmission blockage, equipment failure and the like. Therefore, it is necessary to develop an economical and robust data collection method for distributed photovoltaic clusters.
Disclosure of Invention
The invention aims at: the distributed photovoltaic operation data virtual acquisition method considering the reference power station and the super-parameter optimization is provided to solve the problems of high acquisition cost and low reliability of mass distributed photovoltaic operation data, and the main invention content is as follows: based on the fact that each distributed photovoltaic in the area has spatial correlation, a gating circulation unit GRU is adopted as a virtual collector to capture the time correlation of the distributed photovoltaic. Loss functions and attention mechanisms based on wasperstein distance metrics were introduced into the GRU to mine complex mappings between different power stations. Secondly, a time-varying binary transfer function and a chaos initialization strategy are introduced into a traditional badger optimization algorithm (HBA) to search a reference power station set with highest virtual acquisition precision. Finally, in order to improve the performance of the virtual collector under the complex weather condition, the super-parameters of the virtual collector are adaptively and dynamically adjusted by adopting a model-free reinforcement learning Deep Q Network (DQN).
The technical scheme adopted by the invention is as follows:
a distributed photovoltaic power virtual acquisition method considering optimization of a reference power station comprises the following steps:
step S1: designing a virtual collector AL-GRU, adopting the GRU to excavate space-time correlation characteristics among distributed photovoltaics, dynamically adjusting the weight of a reference power station at different moments by an attention mechanism, and improving the fitting performance of the GRU by a new loss function based on Wasserstein distance measurement;
Step S2: optimizing a reference power station in an area by using the virtual collector AL-GRU as a data reasoning model so as to collect operation data of all distributed photovoltaic power stations in the area through the optimized reference power station, introducing a mel optimization algorithm HBA of a time-varying binary transfer function and a chaos initialization strategy, and optimizing by using the highest K-fold cross validation score as an objective function;
step S3: the super-parameter change in the DQN learning history scene in the reinforcement learning is guided by the design state space, the action space and the reward function, and the super-parameter can be adaptively adjusted according to the output change trend of the distributed photovoltaic in the off-line application stage.
Further, the GRU structure comprises an update gate and a reset gate, the output of the reset gateControlling the fusion degree between the input of the current moment and the history moment memory of the network, and updating the output/>The ratio of the historical output power information of the reference power station is reserved is determined, and the input of GRU at the moment t is assumed to be/>The output/>, at the time t, can be obtained by combining the reset gate and the update gateThe specific calculation formula is as follows:
；
；
；
；
Wherein: 、/> The weight matrix corresponding to the reset gate is obtained; w zy、Wzh is a weight matrix corresponding to the update gate; /(I) 、/>Outputting a corresponding weight matrix for the GRU; /(I)Representing Hadamard operations; b represents a bias vector; /(I)The output of GRU at time t-1; /(I)To update candidate values; /(I)Representing a sigmoid activation function.
Further, the attention mechanism takes the historical output state into consideration to construct an input sliding window, so as to provide more reference information for the attention mechanism module, and the input of the attention mechanism module at the moment t is assumed to beWherein/>For the number of reference stations selected in the area,/>For a time window/>, of length T 1 In order to obtain the importance degree of each reference power station to the current distributed photovoltaic to be collected at the moment, the method comprises the following steps of/>Three layers of attention mechanism neural network ANN are constructed as input, and the output of the ANN is normalized through a Softmax layer, and the specific calculation process is as follows:
；
；
；
Wherein: And/> The output of the attention mechanism module input layer and the hidden layer are respectively; /(I)And/>The activation functions of the input layer and the hidden layer respectively; /(I)And/>The weight matrix and the bias term of the neural network are respectively; /(I)The weight coefficient of the ith reference power station;
reference plant attention distribution matrix to be obtained Multiplying the input of the current moment to obtain the input/>, after the importance degree evaluation of the distributed photovoltaic to be collected, of different reference power stations on the current photovoltaic to be collectedThe virtual collector will fit the output of the station to be collected according to the input, the calculation process is as follows:
；
；
dividing a daily variation curve of the distributed photovoltaic to be collected into K modes by a K-shape clustering algorithm, wherein the calculation method comprises the following steps:
；
；
；
Wherein: SBD is shape-based distance; ，/> Representing two distributed photovoltaic power time series for comparison; since the sampling time is 15min, the length of the time series t=96; /(I) Representing cross-correlation sequences/>Is a length of (2); k is the relative sliding distance of the two sequences.
Further, a sample weight of a K-type curve is obtained, the difference between the K-type curve and the overall fluctuation condition is measured by using probability distribution, the probability distribution and the overall probability distribution of the obtained K-type curve are measured by using Wasserstein distance, and the distribution of two curve categories is assumed to be P and Q respectively, wherein the specific calculation formula is as follows:
；
Wherein: Representing a set of all joint distributions that combine the P and Q distributions; x and y represent the slave joint distribution/>, respectively A sample obtained by middle sampling; /(I)Representing the distance between samples; in all possible joint distributions, the lower bound of the expected value of the sample pair distance is Wasserstein distance;
The weight occupied by each type of curve is obtained through a Softmax function, different weights are given to samples belonging to different types of curves in the offline training process, and a specific calculation formula of the Wasserstein distance measurement loss function guiding virtual collector is as follows:
；
；
Wherein: Wasserstein distance representing the i-th class curve from the overall probability distribution; /(I) Representing the number of samples of the i-th class curve in the training set; /(I)Sample weights representing class i curves; /(I)Is the virtual collection result of the kth sample belonging to the ith class curve; /(I)Is the real power.
Further, the reference power station selection process in step S2 includes:
Step S201: defining a reference plant selection problem, assumed to contain Binary state variable set of individual distributed photovoltaics is/>Indicating that the distributed photovoltaic is selected as a reference power station when the element in the distributed photovoltaic is 1, and indicating that the distributed photovoltaic is selected as the distributed photovoltaic to be collected when the element in the distributed photovoltaic is 0;
step S202: introducing the K-fold cross validation score into an objective function, and optimizing on the premise of meeting constraint conditions, wherein the specific description is as follows:
；
；
；
Wherein: representing the number of samples of the verification set; /(I) Representing the number of training set samples; /(I)And/>Respectively representing an ith virtual acquisition value and a true value of the mth distributed photovoltaic to be acquired; k V is the cross-validated fold number; n cpv represents the number of distributed photovoltaics to be collected,
The calculation formula of the number of the reference power stations in the optimization process is as follows:
；
the constraints on the number of reference stations are:
；
Wherein: Is the maximum value of the number of reference power stations;
Step S203: the method comprises the steps of improving the HBA, introducing a time-varying binary transfer function to enable the HBA to adapt to a binary optimization task, introducing a chaotic initialization strategy, and improving the quality of an initial solution set of the HBA, wherein the method is specifically described as follows:
the time-varying binary transfer function calculation formula is:
；
Wherein: control parameters T is the current iteration number,/>And/>Respectively an upper limit and a lower limit of a control parameter, wherein x represents the position of an HBA individual; e is a natural base number;
the transformation process of the HBA individual's location through the time-varying binary transfer function is as follows:
；
Wherein: represents the d dimension of the mth search agent of the HBA; rand represents a random number from 0 to 1,
The chaos initialization strategy adopts tent mapping to generate an initial population sequence, and the expression is as follows:
；
Step S204: based on the objective function and the constraint conditions set in the step S202, the model developed in the step S1 is used as a virtual collector, the optimization strategy proposed in the step S203 is used for optimizing to obtain an optimal reference power station set, then the selected reference power station is used as input, other distributed photovoltaics to be collected are used as output training virtual collectors, and the collection of the operation data of the whole distributed photovoltaic system in the area is realized.
Further, in the design state space in step S3, the variables influencing the control operation decision are set as the state quantity of the system, and the state space is set
Wherein/>Representing the current period, W representing weather, DR and DP representing the first order difference of irradiance and output power in the time step of T 2, R and P representing irradiance at the current time and output power of all reference stations, and H representing the super-parameter set of the virtual harvester, respectively.
Further, in the step S3, the intelligent agent in the DQN adjusts the super parameters of the virtual collector according to the current state space, and the combination of the super parameters and the changesConstitutes the action space of the agent, wherein/>Represents the/>Variation of individual superparameters,/>，/>The motion space size of each super parameter is/>; G represents the granularity vector discretizing the continuous space.
Further, the reward function in the step S3 excites the agent to take an action for improving the accuracy of virtual acquisition, which is defined as follows:
；
Wherein: And/> And respectively representing the mean square error of the virtual acquisition in the current state in the step length of 2T 2 and after the action of the intelligent agent.
Further, the adaptive adjustment of the super parameter in step S3 includes the following steps:
step S301: in order to improve the virtual acquisition performance under complex weather conditions, a virtual acquisition robustness strengthening strategy based on DQN is provided, dynamic adjustment of super parameters is converted into action selection of an agent through DQN algorithm, attenuation epsilon-greedy strategy is adopted to balance random search and greedy behavior in the training process of the agent, and actions are selected in the strategy The probability of random action is epsilon and the action selection process is expressed as:
；
；
Wherein: representing the action corresponding to the highest cost function at the moment t; /(I) And/>A lower limit and an upper limit of epsilon, respectively; Representing the maximum iteration number of reinforcement learning training, the probability of random action is gradually reduced along with the gradual maturity of network training; e is a natural base number;
step S302: introducing an empirical playback mechanism to increase the efficiency of agent-to-environment interactions and reduce sample-to-sample correlation and dependency, wherein the empirical sample data is obtained for each time step of agent-to-environment interactions Stored in the experience pool, in a subsequent training process, the DQN will update the weights of the target network and the current value network based on the following loss function:
；
Wherein: Output for the target value network; /(I) Output for the current value network; /(I)Representing the number of DQN training samples; /(I)And/>Parameters representing the current value network and the target value network, respectively; /(I)Representing a rebate factor in return; max represents the maximum function.
Advantageous effects
The invention has the following beneficial effects:
1. the invention provides a concept of virtual acquisition, and solves the data acquisition challenges such as insufficient data transmission equipment, reduced data transmission reliability and the like brought in the process of mass distributed photovoltaic construction.
2. Aiming at the virtual acquisition problem, a virtual acquisition device oriented to space-time feature extraction is developed. A loss function based on Wasserstein distance measurement is provided to enhance the fitting capability of GRU under different weather fluctuation scenes. Furthermore, attention mechanisms are integrated to provide different levels of attention to the reference power station at different times.
3. In order to solve the selection problem of a reference power station, a time-varying binary transfer function is introduced into a traditional badger optimization algorithm to adapt to binary decision variable optimization, and the quality of an initial solution of the badger optimization algorithm is improved through a chaos initialization strategy.
4. The invention combines the deep reinforcement learning and the virtual acquisition technology, and dynamically adjusts the super-parameters of the virtual acquisition device according to the change of the trend of the distributed photovoltaic output, thereby enhancing the robustness of the virtual acquisition technology.
Drawings
Fig. 1 is a flow chart of the distributed photovoltaic operation data virtual acquisition method considering the optimization of the reference power station and the super parameters.
Fig. 2 is a schematic diagram of a virtual harvester in accordance with the present invention.
Fig. 3 is a flow chart of a reference power station selection method involved in the present invention.
Fig. 4 is a schematic diagram of K-fold cross-validation as referred to in the present invention.
Fig. 5 is a diagram of virtual acquisition error bins for the loss function and conventional loss functions MAE, MSE according to the present invention.
Fig. 6 (a) - (d) are graphs comparing virtual acquisition results of the attention adding mechanism and the attention not adding mechanism on typical day 1-typical day 4, respectively.
FIG. 7 is a graph of error versus virtual acquisition of different power stations using different virtual collectors.
Fig. 8 is a graph of the variation of virtual acquisition accuracy at different iteration times.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
As shown in fig. 1, the invention provides a distributed photovoltaic power virtual collection method considering optimization of a reference power station, which comprises the following steps:
Step S1: the virtual collector AL-GRU is designed. And adopting a gating circulating unit (GRU) as a virtual collector to mine the time-space correlation characteristic between the distributed photovoltaics. The attention mechanism is used to dynamically adjust the weights of the reference plants at different times, and a new loss function based on the wasperstein distance metric is used to improve the GRU fitting performance.
Step S2: and optimizing the reference power station in the area by taking the developed virtual collector as an inference model so as to realize that the installation sensor of the selected site collects the photovoltaic operation data of the whole system. In order to improve the quality of a reference power station, a mel optimization algorithm introducing a time-varying binary transfer function and a chaos initialization strategy is provided, and the optimization is performed by taking the highest K-fold cross validation score as an objective function.
Step S3: to improve the performance of the model under different weather conditions, the state space, the action space and the reward function are designed to guide the hyper-parameter change in the DQN learning history scene in reinforcement learning. The super parameters can be self-adaptively adjusted according to the output change trend of the distributed photovoltaic in the off-line application stage, so that the virtual acquisition precision is improved.
The interpretation of the virtual collector in the step S1 is as follows: the data reasoning of the distributed photovoltaic virtual collection takes collected data of a reference power station as input, and the running data of all distributed photovoltaic in the area are estimated in real time through a regression model. The virtual harvester is thus a regression model developed for performing virtual harvesting tasks.
The explanation of the optimization of the reference power station in the step S2 is as follows:
The invention aims to select a reference power station based on an improved badger optimization algorithm, and real-time power data is input into an intelligent computing model as multi-dimensional characteristics so as to estimate the output of all distributed photovoltaics in an area. However, randomly selecting reference power stations within an area is unreliable. Reasonable reference power station combinations are selected in the power station area, so that the mutual supplementing effect can be achieved, and the virtual acquisition precision is further effectively improved.
Further, the virtual collector AL-GRU in step S1 includes a loss function of the base GRU, the attention mechanism module and the Wasserstein distance metric:
step S101: the structure of the GRU includes an update gate and a reset gate. Reset gate output And controlling the fusion degree between the input of the current moment of the network and the history moment memory. Update gate output/>The proportion of the historical output power information of the reference power station is reserved. Let the input of the GRU unit at time t be/>The output/>, at the time t, can be obtained by combining the reset gate and the update gate. The specific calculation formula is as follows:
；
；
；
；
Wherein: 、/> The weight matrix corresponding to the reset gate is obtained; w zy、Wzh is a weight matrix corresponding to the update gate; /(I) 、/>Outputting a corresponding weight matrix for the GRU unit; /(I)Representing Hadamard operations; b represents a bias vector; /(I)The output of the GRU unit at the time t-1 is shown; /(I)To update candidate values; /(I)Representing a sigmoid activation function.
Step S102: the introduction of the attention mechanism improves the fitting ability of the GRU. The attention mechanism builds an input sliding window taking into account the historical force states, thereby providing more reference information to the attention mechanism module. Assume that the input of the attention mechanism module at time t is. Wherein/>For the number of reference stations selected in the area,/>For a time window/>, of length T 1 . In order to obtain the importance degree of each reference power station on the current photovoltaic to be collected at the moment, so as to/>A three-layer attention-mechanism neural network (ANN) was constructed as an input and the output of the ANN was normalized by the Softmax layer. The specific calculation process is as follows:
；
；
；
Wherein: And/> The output of the attention mechanism module input layer and the hidden layer are respectively; /(I)And/>The activation functions of the input layer and the hidden layer respectively; /(I)And/>The weight matrix and the bias term of the neural network are respectively; /(I)The weight coefficient of the ith reference power station;
Step S103: reference plant attention distribution matrix to be obtained Multiplying the input of the current moment to obtain the input/>, after the importance degree of different reference power stations on the current photovoltaic to be collected is evaluated. The virtual collector will fit the output of the distributed photovoltaic to be collected according to the input. The calculation process is as follows:
；
；
The reasons why the attention mechanism is introduced in step S102 include: each power station in the region has a unique geospatial location, and each reference power station at different times is not of the same importance to the distributed photovoltaic to be collected. This motivates us to use the attention mechanism to mine the degree of spatiotemporal correlation between different photovoltaics. The attention mechanism mimics the resource allocation mechanism of human brain attention, and at a specific moment, attention is focused on the areas which are important to pay attention to. Therefore, attention mechanisms are introduced into the virtual collector, and the schematic diagram of the virtual collector related to the invention is combined with fig. 2, so that different attention degrees can be given to different reference power stations, and the dynamic allocation of the weights of the reference power stations is realized.
Step S104: and dividing the daily change curve of the photovoltaic to be collected into K modes by a K-shape clustering algorithm. The calculation method comprises the following steps:
；
；
；
Wherein: SBD is shape-based distance; ，/> Representing an input time sequence of k-shape; since the sampling time is 15min, the length of the time series t=96; /(I) Representing cross-correlation sequences/>Is a length of (2); k is the relative sliding distance of the two sequences.
Step S105: in order to obtain the sample weight of the K-type curve, the difference between the K-type curve and the overall fluctuation condition is measured by adopting probability distribution. This has the advantage that it is possible to avoid that a certain outlier affects the overall weight. The Wasserstein distance does not need to be the same as the length of the 2 sets of data, and can effectively measure the similarity even when the two probability distributions do not overlap. The probability distribution and the overall probability distribution of the resulting K-class curve are thus measured using the wasperstein distance. The specific calculation formula is as follows:
；
Wherein: Representing a set of all joint distributions that combine the P and Q distributions; x and y represent the slave joint distribution/>, respectively A sample obtained by middle sampling; /(I)Representing the distance between samples; in all possible joint distributions, the lower bound of the expected value of the sample versus distance is the Wasserstein distance.
Step S105: the weight of each type of curve is obtained through a Softmax function. In the offline training process, samples belonging to different category curves are given different weights. The loss function of the Wasserstein distance metric directs the specific calculation formula of the virtual collector as follows:
；
；
Wherein: Wasserstein distance representing the i-th class curve from the overall probability distribution; /(I) Representing the number of samples of the i-th class curve in the training set; /(I)Sample weights representing class i curves; /(I)Is the virtual collection result of the kth sample belonging to the ith class curve; /(I)Is the real power.
Further, referring to fig. 3 in combination, the reference power station selection process in step S2 includes:
Step S201: defining a reference plant selection problem. It is assumed to contain Binary state variable set of individual distributed photovoltaics is/>. Indicating that the distributed photovoltaic is selected as a reference power station when the element in the power station is 1, and indicating that the distributed photovoltaic is selected as the distributed photovoltaic to be collected when the element in the power station is 0;
step S202: the optimization process needs to consider the performance of the virtual collector, and the training loss cannot reflect the fitting capacity of the model under unknown conditions. Therefore, in order to fully utilize the existing data and reflect the comprehensive virtual acquisition errors when different distributed photovoltaics are used as reference power stations in the optimization process, the K-fold cross validation score is introduced into an objective function, optimization is performed on the premise of meeting constraint conditions, and the validation set effect is shown in a schematic diagram of the K-fold cross validation in the invention in fig. 4, and the specific description is as follows:
；
；
；
Wherein: representing the number of samples of the verification set; /(I) Representing the number of training set samples; /(I)And/>Respectively representing an ith virtual acquisition value and a true value of the mth distributed photovoltaic to be acquired; k V is the cross-validated fold number; /(I)Representing the number of photovoltaics to be collected.
The optimization algorithm may favor increasing the number of reference stations to improve the accuracy of the virtual acquisition. If sparsity is not required, the purpose of reducing acquisition cost in virtual acquisition cannot be achieved. The calculation formula of the number of the reference power stations is as follows:
；
The constraints on the number of reference stations are:
；
Wherein: is the maximum value of the number of reference stations.
Step S203: the HBA is improved, and a time-varying binary transfer function is introduced to adapt the HBA to a binary optimization task. Compared with other binary transfer functions, the time-varying transfer function can improve the exploration capability of the algorithm in different search stages, and avoid sinking into local optimum. In addition, a chaos initialization strategy is introduced, so that the quality of HBA initial solution sets is improved. The specific description is as follows:
the time-varying binary transfer function calculation formula is:
；
Wherein: x represents the position of the individual HBA; e is a natural base number; control parameters ；
T is the number of current iterations and,And/>The upper and lower limits of the control parameter, respectively.
The transformation process of the HBA individual's location through the binary transfer function is as follows:
；
Wherein: represents the d dimension of the mth search agent of the HBA; rand represents a random number of 0 to 1.
The chaotic initialization strategy is described as follows:
The traditional sparrow optimization algorithm initializes the population through random numbers, and the random generation mode possibly causes uneven distribution of generated individuals, so that population diversity and optimizing speed are reduced. Thus, the tent map was used to generate the initial population sequence, which was expressed as follows:
；
Step S204: based on the objective function and the constraint conditions set in the step S202, the model developed in the step S1 is used as a virtual collector, and optimization is performed through the optimization strategy proposed in the step S203, so that the optimal reference power station set is obtained. And then taking the selected reference power station as input and other distributed photovoltaics to be collected as output training virtual collectors to realize virtual collection of the running data of the whole distributed photovoltaic system in the area.
Further, the dynamic adjustment of the super parameters of the virtual collector in the step S3 includes the following steps:
Step S301: in order to improve the virtual acquisition performance under the complex weather condition, a virtual acquisition robustness strengthening strategy based on DQN is provided. The method is based on state information and acquisition errors in historical scenes, and the robust performance of virtual acquisition is improved by dynamically adjusting model super parameters, so that the virtual acquisition errors in specific scenes are reduced. Dynamic adjustment of super parameters is converted into action selection of the intelligent agent through the DQN algorithm, and the attenuation epsilon-greedy strategy is adopted to balance random search and greedy behaviors in the intelligent agent training process. Selecting an action in the policy The probability of random action is epsilon and the probability of random action is 1 epsilon.
；
；
Wherein: representing the action corresponding to the highest cost function at the moment t; /(I) And/>A lower limit and an upper limit of epsilon, respectively; representing the maximum number of iterations of reinforcement learning training. As the network training matures, the probability of random action will decrease gradually.
The agent in step S301 is explained as: the intelligent agent in reinforcement learning is composed of a neural network and is an entity for executing super-parameter adjustment action in the process of data acquisition by the virtual acquisition device so as to minimize virtual acquisition errors and complete training.
Step S302: an empirical playback mechanism is introduced to improve the efficiency of agent interactions with the environment and to reduce inter-sample correlation and dependency. In this mechanism, empirical sample data is obtained for each time step of agent and environment interactionIs stored in an experience pool. In a subsequent training process, the DQN will update the weights of the target network and the current value network based on the following loss function.
；
Wherein: Output for the target value network; /(I) Representing the number of DQN training samples; /(I)And/>Representing parameters of the current value network and the target value network, respectively.
Step S303: for the virtual acquisition task, the description of the agent state space, action space, and rewarding function is as follows:
And (3) designing a state space:
the variables that affect the control action decisions are set as state quantities of the system. Our goal is to improve the performance of virtual acquisition in complex weather scenarios. The state space of the agent needs to not only exhibit the DPV output trend, but also need to contain the hyper-parametric state. Thus, the state space is set . Wherein/>Representing the current period, W representing weather, DR and DP representing the first order difference of irradiance and output power in the time step of T 2, R and P representing irradiance at the current time and output power of all reference stations, and H representing the super-parameter set of the virtual harvester, respectively.
And (3) designing an action space:
The intelligence in the DQN will adjust the super parameters of the virtual collector according to the current state. Thus, combinations of hyper-parametric variations The action space of the intelligent body is formed. Wherein/>Represents the/>The variation of the individual super-parameters,，/>The motion space size of each super parameter is/>; G represents the granularity vector discretizing the continuous space.
And (3) bonus function design:
the rewarding function stimulates the agent to take the action of improving the virtual acquisition precision, and the definition is as follows:
；
Wherein: And/> And respectively representing the mean square error of the virtual acquisition in the current state in the step length of 2T 2 and after the action of the intelligent agent.
Best mode for carrying out the invention for specific applications:
to realize virtual data collection of distributed photovoltaic power, a proper distributed photovoltaic power data set needs to be found, and the collected data set is preprocessed to provide data support for subsequent calculation analysis. The invention selects 29 Distributed Photovoltaic (DPV) in Jiang Ning areas of Nanjing, jiangsu, china for virtual acquisition test. Data collected from month 1 of 2018 2 to month 31 of 2018 was collected over 15 minute intervals of 07:15 to 17:00. To facilitate training and verification of the network, we use data from 2 months 1 day to 10 months 31 days as training set and verification set, and the rest data is used for validity verification of DPV virtual acquisition. The virtual acquisition accuracy is assessed by error index Mean Absolute Error (MAE), root Mean Square Error (RMSE) and Mean Absolute Percent Error (MAPE):
；/>
；
；
Wherein: representing the number of test samples; - And/>Respectively representing a virtual acquisition value and a real value of the distributed photovoltaic output power; the sample point of 0 of real power needs to be removed when evaluating virtual acquisition performance by MAPE.
In order to prove the effectiveness of IHBA developed by the invention, a mel optimization algorithm (HBA), a suburban wolf algorithm (COA), a gray wolf algorithm (GWO) and a Harris eagle optimization algorithm (HHO) are respectively selected to optimize the reference power station. The objective function values of each optimization algorithm at different reference plant numbers are shown in table 1. It can be seen that the IHBA optimization algorithm used in the present invention results in a reference plant with an optimal objective function value.
TABLE 1 objective function values of optimization algorithms for different reference plant numbers (W 2)
In order to verify the effect of the virtual collector, the total number of reference power stations is 14, and optimization is carried out through IHBA algorithm, so that the reference power stations are obtained with the following numbers: 2. 4, 7, 10, 11, 12, 14, 15, 16, 18, 20, 21, 24, 27. Firstly, the superiority of the loss function proposed by the invention is demonstrated, and the classical loss functions MAE and MSE are compared with the accuracy of virtual acquisition under the proposed loss function. The experiment is repeated for 20 times without setting random number seeds, so that the error condition shown in an error box diagram of the loss function and the traditional loss functions MAE and MSE virtual acquisition error box diagram in the invention is obtained in FIG. 5. It can be seen that the virtual acquisition error through the proposed loss function is minimal and has a low standard deviation. In addition, MSE reflects more widely-deviated sampling points than MAE. The loss function provided by the invention can further reflect the virtual acquisition errors under different fluctuation weather conditions on the basis of the MSE loss function, and has more competitive power in the virtual acquisition problem compared with other loss functions. For convenience of presentation we will hereinafter refer to GRUs with attention mechanisms as A-GRUs and GRUs with improved loss functions as L-GRUs.
We selected four typical days of different trends in the test set, and represented the effect of verifying the add-on mechanism with station No. 1. The virtual acquisition results are shown in fig. 6, and the comparison graphs of the virtual acquisition results with and without the attention mechanism added under 4 typical days are closer to each other in the whole, which indicates that the virtual acquisition can be completed with higher precision by the reference power station selected by the invention. Furthermore, the A-GRU with added attention mechanisms is superior to GRU in performance. This is because the attention mechanism dynamically assigns weights to the reference power stations based on the data of the sliding window over the period of time, increasing the weights of the reference power stations that are more closely related to the trend of distributed photovoltaic output to be collected. In summary, the proposed method has good virtual acquisition performance in both stationary and fluctuating weather conditions.
The virtual acquisition accuracy under different modifications is shown in table 2. The L-GRU has more obvious effect on reducing MAE and RMSE errors. This shows that the loss function developed by the invention can improve the fitting performance when the actual output is larger, and MAPE is often influenced by the sampling point when the actual output is smaller. MAE and RMSE lowering effects of A-GRU are not as good as L-GRU, but MAPE lowering effects are better. In addition, the attention mechanism and the GRU share a loss function, and the proposed loss function can assist the attention mechanism to give more differentiated weights to each power station under different weather conditions, so that the two power stations are combined to have optimal virtual acquisition precision.
Table 2 virtual acquisition errors under different improvements
Further, to embody the superiority of the proposed virtual collector, the performances of different virtual collectors are compared, as shown in an error comparison chart of virtual collection of different power stations by adopting different virtual collectors in fig. 7. From the model perspective, the MAE and the RMSE of the BP neural network are the highest, and the virtual acquisition task is difficult to finish with higher precision. The ensemble learning has good fitting performance. Wherein XGBoost has an RMSE above LightGBM, but MAPE below LightGBM, since the outliers of XGBoost occur mostly at the sampling points where the actual power is large, and LightGBM is the opposite. The two models can be fused later to improve the acquisition accuracy. The AL-GRU provided by the invention has the optimal virtual acquisition precision. In addition, GRUs with time-series memory function are also superior to BP neural networks and RF in ensemble learning. Therefore, the timing analysis function is very important for the task of virtual acquisition, and subsequent studies should be focused on this feature. From the power station point of view, the virtual acquisition accuracy of the power stations No. 1 and No. 13 is lower than that of other power stations. Since this may affect the virtual acquisition accuracy of all plants, IHBA will act as a non-reference plant, reflecting IHBA the rationality of selecting a reference plant.
The output trend of the distributed photovoltaic is not always standard like beta distribution, and the trend change is obvious under different weather conditions. In order to verify the effectiveness of the DQN dynamic adjustment super-parameter method provided by the invention, simulation verification is carried out based on 6 virtual collectors. In order to clearly show the variation of the model performance under different iteration times, we select AL-GRU as a typical display, as shown in fig. 8, and in order to make the display clearer, only the variation of RMSE is shown. It can be seen that the initial performance of the AL-GRU model is severely degraded by inadequate DQN experience. This illustrates that the super-parametric settings of the model severely affect the performance of the virtual acquisition. With the increase of iteration times, the DQN effectively improves the capacity of super-parameter dynamic adjustment by learning more interactive information. When the maximum iteration number exceeds 1000, ALGRU virtual acquisition performance with DQN auxiliary adjustment super-parameters tends to be stable. Therefore, we selected max_iter=1200 to experiment, and repeated the experiment 15 times, observing the virtual acquisition errors of different VCMs with DQN assistance.
The virtual acquisition accuracy of each model is shown in table 3. It can be seen that the virtual acquisition accuracy of each VCM is improved with the aid of DQN. As LightGBM, XGBoost in the integrated learning have more super parameters and larger super parameter adjustment space, the precision is improved more obviously. Also, the standard deviation of the ensemble learning model is large because more iterations are required to make the model more stable.
Table 3 virtual acquisition error after adaptive adjustment of superparameter for different virtual collectors
/>
Further, we observe the virtual acquisition effect of dynamic adjustment of the super-parameters of DQN under different training set proportions. The average virtual acquisition errors for the training set ratio from 40% to 80%, respectively, are shown in table 4. It can be seen that the parameter boosting effect improves with the increase of the training set, which means that reinforcement learning requires more history status scenes for learning. Therefore, the technology improvement should focus on the collection of diversified power scenes in the following.
TABLE 4 virtual acquisition error at different training set ratios
An electronic device comprising a memory, wherein the processors are in communication connection with each other, the memory stores computer instructions, and the processors execute the distributed photovoltaic operation data virtual acquisition method taking reference power stations and super-parameter optimization into consideration by executing the computer instructions.
A computer storage medium having stored thereon computer instructions for causing a computer to perform the distributed photovoltaic operational data virtual collection method of any of the above, taking into account reference power station and hyper-parameter optimization.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The scheme in the embodiment of the invention can be realized by adopting various computer languages, such as object-oriented programming language Java, an transliteration script language JavaScript and the like.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the applicant has described embodiments of the present invention in detail and with reference to the drawings, the present invention is better understood by the reader in the field, and is not limited to the scope of the invention, but any improvement or modification based on the spirit of the invention should fall within the scope of the invention.
Claims (7)
1. A distributed photovoltaic power virtual acquisition method considering optimization of a reference power station is characterized in that: the method comprises the following steps:
step S1: designing a virtual collector AL-GRU, adopting the GRU to excavate space-time correlation characteristics among distributed photovoltaics, dynamically adjusting the weight of a reference power station at different moments by an attention mechanism, and improving the fitting performance of the GRU by a new loss function based on Wasserstein distance measurement;
Step S2: optimizing a reference power station in an area by using the virtual collector AL-GRU as a data reasoning model so as to collect operation data of all distributed photovoltaic power stations in the area through the optimized reference power station, introducing a mel optimization algorithm HBA of a time-varying binary transfer function and a chaos initialization strategy, and optimizing by using the highest K-fold cross validation score as an objective function;
the reference power station selection process includes:
Step S201: defining a reference plant selection problem, assuming a binary set of state variables containing N DPV distributed photovoltaics as Indicating that the distributed photovoltaic is selected as a reference power station when the element in the power station is 1, and indicating that the distributed photovoltaic is selected as the distributed photovoltaic to be collected when the element in the power station is 0;
step S202: introducing the K-fold cross validation score into an objective function, and optimizing on the premise of meeting constraint conditions, wherein the specific description is as follows:
wherein: n V represents the verification set sample number; n T represents the number of training set samples; And y m,i represents the ith virtual collection value and the true value of the mth distributed photovoltaic to be collected respectively; k V is the cross-validated fold number; n cpv represents the number of distributed photovoltaics to be collected,
The calculation formula of the number of the reference power stations in the optimization process is as follows:
the constraints on the number of reference stations are:
Wherein: Is the maximum value of the number of reference power stations;
Step S203: the method comprises the steps of improving the HBA, introducing a time-varying binary transfer function to enable the HBA to adapt to a binary optimization task, introducing a chaotic initialization strategy, and improving the quality of an initial solution set of the HBA, wherein the method is specifically described as follows:
the time-varying binary transfer function calculation formula is:
Wherein: control parameters T is the current iteration number,/>And/>Respectively an upper limit and a lower limit of a control parameter, wherein x represents the position of an HBA individual; e is a natural base number;
the transformation process of the HBA individual's location through the time-varying binary transfer function is as follows:
Wherein: represents the d dimension of the mth search agent of the HBA; rand represents a random number from 0 to 1,
The chaos initialization strategy adopts tent mapping to generate an initial population sequence, and the expression is as follows:
Step S204: based on the objective function and the constraint conditions set in the step S202, the model developed in the step S1 is used as a virtual collector, the optimization strategy proposed in the step S203 is used for optimizing to obtain an optimal reference power station set, then the selected reference power station is used as input, other distributed photovoltaics to be collected are used as output training virtual collectors, and the collection of the running data of the whole distributed photovoltaic system in the area is realized;
step S3: the super-parameter change in the DQN learning history scene in reinforcement learning is guided by the design state space, the action space and the reward function, and the super-parameter can be adaptively adjusted according to the output change trend of the distributed photovoltaic in the off-line application stage;
The attention mechanism takes the historical output state into consideration to construct an input sliding window so as to provide more reference information for the attention mechanism module, and the input of the attention mechanism module at the moment t is assumed to be Wherein N RPV is the number of selected reference stations in the region, and P i is the time window/>, with length T 1 In order to acquire the importance degree of each reference power station to the current distributed photovoltaic to be acquired at the moment, a three-layer attention mechanism neural network ANN is constructed by taking P RPV (t) as input, and the output of the ANN is normalized through a Softmax layer, wherein the specific calculation process is as follows:
H1＝f1(W1PRPV(t)+b1)
H2＝f2(W2H1+b2)
Wherein: h 1 and H 2 are the outputs of the attention mechanism module input layer and hidden layer, respectively; f 1 (·) and f 2 (·) are activation functions of the input layer and the hidden layer, respectively; w and b are respectively a weight matrix and a bias term of the neural network; alpha i is the weight coefficient of the ith reference power station;
Multiplying the obtained attention distribution matrix alpha of the reference power station with the input of the current moment to obtain the input of different reference power stations after the importance degree evaluation of the current distributed photovoltaic to be acquired The virtual collector will fit the output of the station to be collected according to the input, the calculation process is as follows:
dividing a daily variation curve of the distributed photovoltaic to be collected into K modes by a K-shape clustering algorithm, wherein the calculation method comprises the following steps:
Wherein: SBD is shape-based distance; Representing two distributed photovoltaic power time series for comparison; since the sampling time is 15min, the length of the time series t=96; ω ε {1,2, …,2T-1} represents the length of the cross-correlation sequence CC ω; k is the relative sliding distance of the two sequences.
2. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: the GRU structure comprises an update gate and a reset gate, wherein the output r t of the reset gate controls the fusion degree between the input of the current moment of the network and the memory of the historical moment, the output z t of the update gate determines the proportion of the historical output power information of the reserved reference power station, the input of the GRU at the moment t is x t, the output h t at the moment t can be obtained by combining the reset gate and the update gate, and the specific calculation formula is as follows:
rt＝σ(Wryxt+ht-1Wrh+br)
zt＝σ(Wzyxt+ht-1Wzh+bz)
Wherein: w ry、Wrh is a weight matrix corresponding to the reset gate; w zy、Wzh is a weight matrix corresponding to the update gate; w hy、Whh is a weight matrix corresponding to GRU output; the ". If represents Hadamard operation; b represents a bias vector; h t-1 represents the output of the GRU at time t-1; To update candidate values; sigma (·) represents a sigmoid activation function.
3. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: obtaining sample weight of a K-type curve, measuring the difference between the K-type curve and the overall fluctuation condition by using probability distribution, measuring the probability distribution and the overall probability distribution of the obtained K-type curve by using Wasserstein distance, and assuming that the distribution of two curve types is P and Q respectively, wherein the specific calculation formula is as follows:
Wherein: pi (P, Q) represents the set of all joint distributions where P and Q distributions are combined; x and y represent samples sampled from the joint distribution γ, respectively; the ||x-y|| represents the distance between samples; in all possible joint distributions, the lower bound of the expected value of the sample pair distance is Wasserstein distance;
The weight occupied by each type of curve is obtained through a Softmax function, different weights are given to samples belonging to different types of curves in the offline training process, and a specific calculation formula of the Wasserstein distance measurement loss function guiding virtual collector is as follows:
Wherein: w i represents the Wasserstein distance of the i-th class curve from the overall probability distribution; n i represents the number of samples of the i-th class curve in the training set; beta i represents the sample weight of the i-th class curve; Is the virtual collection result of the kth sample belonging to the ith class curve; /(I) Is the real power.
4. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: in the design state space in the step S3, variables affecting the decision of the control action are set as state quantities of the system, and the state spaces S V＝{tC, W, DR, DP, R, P, H are set, where T C e {1, … } represents the current period, W represents weather, DR and DP represent the first difference of irradiance and output power in the time step T 2, R and P represent irradiance at the current moment and output power of all reference power stations, and H represents the super-parameter set of the virtual collector, respectively.
5. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: the motion space design in step S3 is that the intelligent agent in the DQN adjusts the super parameters of the virtual collector according to the current state space, and the combination of the super parameters changesConstitutes the action space of the agent, wherein/>The variation of the xi super-parameters is expressed, and the action space size of the xi E {1,2, …, N H},NH super-parameters is/>G represents the granularity vector discretizing the continuous space.
6. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: the rewarding function in the step S3 stimulates the agent to take the action for improving the virtual acquisition precision, and the definition is as follows:
Wherein: And/> And respectively representing the mean square error of the virtual acquisition in the current state in the step length of 2T 2 and after the action of the intelligent agent.
7. The distributed photovoltaic power virtual collection method taking reference power station optimization into consideration as claimed in claim 1, wherein: the adaptive adjustment of the super parameter in the step S3 includes the following steps:
step S301: in order to improve the virtual acquisition performance under complex weather conditions, a virtual acquisition robustness strengthening strategy based on DQN is provided, dynamic adjustment of super parameters is converted into action selection of an agent through DQN algorithm, attenuation epsilon-greedy strategy is adopted to balance random search and greedy behavior in the training process of the agent, and actions are selected in the strategy The probability of random action is epsilon and the action selection process is expressed as:
Wherein: Representing the action corresponding to the highest cost function at the moment t; epsilon end and epsilon start are the lower and upper limits of epsilon, respectively; it max represents the maximum number of iterations of reinforcement learning training, as the network training matures gradually, the probability of random action will decrease gradually; e is a natural base number;
Step S302: an empirical playback mechanism is introduced to increase the efficiency of agent and environment interactions and reduce the correlation and dependency between samples, in which empirical sample data (s t,at,rt,st+1) obtained for each time step of agent and environment interactions is stored in an empirical pool, and in a subsequent training process, the DQN will update the weights of the target network and the current value network based on a loss function:
wherein: q '(s t+1,at+1, ω') is the output of the target value network; q (s t,at, ω) is the output of the current value network; n B represents the number of DQN training samples; omega and omega' represent parameters of the current value network and the target value network, respectively; gamma represents a return discount factor; max represents the maximum function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410389240.9A CN117973644B (en) | 2024-04-02 | 2024-04-02 | Distributed photovoltaic power virtual acquisition method considering optimization of reference power station |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410389240.9A CN117973644B (en) | 2024-04-02 | 2024-04-02 | Distributed photovoltaic power virtual acquisition method considering optimization of reference power station |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117973644A CN117973644A (en) | 2024-05-03 |
| CN117973644B true CN117973644B (en) | 2024-06-14 |
Family
ID=90863099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410389240.9A Active CN117973644B (en) | 2024-04-02 | 2024-04-02 | Distributed photovoltaic power virtual acquisition method considering optimization of reference power station |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117973644B (en) |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111898812B (en) * | 2020-07-17 | 2022-10-04 | 国网江苏省电力有限公司南京供电分公司 | Distributed photovoltaic data virtual acquisition method |
| CN112100904B (en) * | 2020-08-12 | 2022-08-23 | 国网江苏省电力有限公司南京供电分公司 | ICOA-BPNN-based distributed photovoltaic power station active power virtual acquisition method |
| CN112052913B (en) * | 2020-09-27 | 2022-08-23 | 国网江苏省电力有限公司南京供电分公司 | Distributed photovoltaic power station power data virtual acquisition method |
| WO2022236064A2 (en) * | 2021-05-06 | 2022-11-10 | Strong Force Iot Portfolio 2016, Llc | Quantum, biological, computer vision, and neural network systems for industrial internet of things |
| CN113962357A (en) * | 2021-09-14 | 2022-01-21 | 天津大学 | GWO-WNN-based distributed photovoltaic power data virtual acquisition method |
| CN114186475B (en) * | 2021-10-28 | 2023-05-12 | 南京工业大学 | Slewing bearing service life prediction method based on Attention-MGRU |
| CN116796194A (en) * | 2023-06-27 | 2023-09-22 | 国网宁夏电力有限公司石嘴山供电公司 | IDBO-KELM-BiGRU neural network-based active power virtual collection method for distributed photovoltaic power station |
| CN117134334A (en) * | 2023-08-28 | 2023-11-28 | 淮阴工学院 | Mechanism-data driving hybrid integration-based photovoltaic power generation power prediction method |
-
2024
- 2024-04-02 CN CN202410389240.9A patent/CN117973644B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| A Virtual Data Collection Model of Distributed PVs Considering Spatio-Temporal Coupling and Affine Optimization Reference;Leijiao Ge等;《IEEE》;20230731;第3939-3951页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117973644A (en) | 2024-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liang et al. | A novel wind speed prediction strategy based on Bi-LSTM, MOOFADA and transfer learning for centralized control centers | |
| CN110705743B (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
| Bai et al. | Double-layer staged training echo-state networks for wind speed prediction using variational mode decomposition | |
| Wang et al. | Wind speed forecasting based on multi-objective grey wolf optimisation algorithm, weighted information criterion, and wind energy conversion system: A case study in Eastern China | |
| Li et al. | Probabilistic charging power forecast of EVCS: Reinforcement learning assisted deep learning approach | |
| Jasmin et al. | Reinforcement learning approaches to economic dispatch problem | |
| CN112529283A (en) | Comprehensive energy system short-term load prediction method based on attention mechanism | |
| CN112614009A (en) | Power grid energy management method and system based on deep expected Q-learning | |
| CN104636985A (en) | Method for predicting radio disturbance of electric transmission line by using improved BP (back propagation) neural network | |
| CN112884236B (en) | Short-term load prediction method and system based on VDM decomposition and LSTM improvement | |
| Li et al. | Short term prediction of photovoltaic power based on FCM and CG-DBN combination | |
| CN115759458A (en) | Load prediction method based on comprehensive energy data processing and multi-task deep learning | |
| CN116894504A (en) | Wind power cluster power ultra-short-term prediction model establishment method | |
| CN115310782A (en) | Power consumer demand response potential evaluation method and device based on neural turing machine | |
| CN116796194A (en) | IDBO-KELM-BiGRU neural network-based active power virtual collection method for distributed photovoltaic power station | |
| Luo et al. | A cascaded deep learning framework for photovoltaic power forecasting with multi-fidelity inputs | |
| Lin et al. | Deep reinforcement learning and LSTM for optimal renewable energy accommodation in 5G internet of energy with bad data tolerant | |
| CN117543537A (en) | Agent electricity purchasing user electric quantity prediction method, device and storage medium | |
| CN117973644B (en) | Distributed photovoltaic power virtual acquisition method considering optimization of reference power station | |
| CN115577647B (en) | Power grid fault type identification method and intelligent agent construction method | |
| CN116565876A (en) | Robust reinforcement learning distribution network tide optimization method and computer readable medium | |
| Sabri et al. | A comparative study of LSTM and RNN for photovoltaic power forecasting | |
| CN115374998A (en) | Load prediction method based on attention mechanism and CNN-BiGRU | |
| CN114372418A (en) | Wind power space-time situation description model establishing method | |
| CN113890112A (en) | Power grid prospective scheduling method based on multi-scene parallel learning |
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 | ||
| GR01 | Patent grant |