Photovoltaic power prediction method, device, storage medium and computer equipment based on NPCNN-DDPG

Abstract

An embodiment of the present application discloses a photovoltaic power prediction method, device, storage medium and computer device based on NPCNN-DDPG, which relates to the field of power prediction. The present application is based on a hybrid network model of a non-pooling convolutional neural network and Deep Deterministic Policy Gradient (NPCNN-DDPG). NPCNN is introduced into the actor-critic network to identify the non-linear features of the environment, thereby reducing the impact of partial observability on the prediction performance. Through a new dynamic reward function for imbalanced samples, the policy score of the online actor network can be effectively evaluated, and the prediction accuracy can be effectively improved compared with the existing photovoltaic power prediction methods.

Description

Technical Field

[0001] The present application relates to the field of power prediction, and particularly to a photovoltaic power generation prediction method, device, storage medium and computer device based on NPCNN-DDPG. Background Art

[0002] In recent years, photovoltaic power generation, as a low-carbon renewable energy source, has developed vigorously worldwide. However, photovoltaic power generation has uncertainty and volatility. In order to reduce the negative impacts brought by the grid connection of photovoltaic power generation on the stable operation and economic dispatch of modern power systems, relevant technical researchers have achieved the optimal operation of power systems by designing reasonable and accurate photovoltaic power prediction models. So far, the relevant methods and models for photovoltaic power prediction are roughly divided into three categories: the first category is the physical prediction model based on weather conditions and photovoltaic power generation conversion principles; the second category is the statistical prediction model based on the mapping relationship between historical data and prediction data; the third category is the machine learning prediction model based on the non-linear function structure.

[0003] The first category, the physical model prediction method based on meteorology and photovoltaic power generation conversion principles: Establish a corresponding photovoltaic cell power generation prediction model. This method can intuitively explain the influence of meteorological data on photovoltaic power generation prediction, but this method has the following disadvantages: 1) It is impossible to fully obtain a large number of data sources (the published data is limited), and it is difficult to update in real time; 2) There is an inherent complexity in modeling the photovoltaic power generation prediction, resulting in weak robustness and high operating costs of this method.

[0004] The second category, the statistical model prediction method based on the mapping relationship between historical data and prediction data: The statistical model is established through mathematical statistical methods such as linear regression, exponential smoothing and autoregressive moving average. Compared with the physical model, this method can make the result more effective and more robust through more direct mathematical calculations. The disadvantages of this model are as follows: 1) It is more suitable for relatively smooth and regular time series, and is not very suitable for the non-linear photovoltaic power generation power that changes with natural factors such as weather; 2) The statistical model mainly only considers the historical photovoltaic-related features and ignores the meteorological influence, thus limiting the engineering application of this model.

[0005] In summary, the first category has complex modeling and high data acquisition costs, and the second category does not fully extract the hidden features in the data and is limited in non-stationary meteorology. At present, it is very difficult for these two categories of methods to achieve accurate prediction of photovoltaic power generation power. Summary of the Invention

[0006] The embodiments of the present application provide a photovoltaic power generation prediction method, device, storage medium and computer device based on NPCNN-DDPG, which can solve the problem of low accuracy in predicting the power generation of photovoltaic cells. The technical solutions are as follows:

[0007] In a first aspect, an embodiment of the present application provides a photovoltaic power prediction method based on NPCNN-DDPG. The method includes:

[0008] Obtain a first sample set and a second sample set; wherein, the first sample set is a sample set under extreme environments, and the second sample set is a sample set under normal environments;

[0009] Initialize the network parameters of the NPCNN-DDPG network; wherein, the NPCNN-DDPG network includes: an online actor network, an online critic network, a target actor network, and a target critic network, and the online actor network is a non-pooling convolutional neural network NPCNN;

[0010] Train the NPCNN-DDPG network using the first sample set and the second sample set; wherein, the reward function of the online actor network is a t represents the predicted power value of the photovoltaic cell at time t, y t represents the actual power value of the photovoltaic cell at time t, ζ e represents the training round, c e represents the current training round, S E represents the first sample set, S N represents the second sample set, χ t represents the weight of the first sample set, λ t represents the weight of the second sample set;

[0011] Process the historical time series data according to the trained NPCNN-DDPG network to obtain the predicted power value at the predicted time.

[0012] In a second aspect, an embodiment of the present application provides a photovoltaic power prediction device based on NPCNN-DDPG. The device includes:

[0013] An acquisition unit, configured to obtain a first sample set and a second sample set; wherein, the first sample set is a sample set under extreme environments, and the second sample set is a sample set under normal environments;

[0014] An initialization unit, configured to initialize the network parameters of the NPCNN-DDPG network; wherein, the NPCNN-DDPG network includes: an online actor network, an online critic network, a target actor network, and a target critic network, and the online actor network is a non-pooling convolutional neural network NPCNN;

[0015] A training unit for training the NPCNN-DDPG network by using the first sample set and the second sample set; wherein, the reward function of the online actor network is a t represents the predicted power value of the photovoltaic cell at time t, y t represents the actual power value of the photovoltaic cell at time t, ζ e represents the training round, c e represents the current training round, S E represents the first sample set, S N represents the second sample set, χ t represents the weight of the first sample set, λ t represents the weight of the second sample set;

[0016] A prediction unit for processing historical time series data according to the trained NPCNN-DDPG network to obtain the predicted power value at the prediction time.

[0017] In a third aspect, an embodiment of the present application provides a computer storage medium storing multiple instructions adapted to be loaded and executed by a processor to perform the above method steps.

[0018] In a fourth aspect, an embodiment of the present application provides a computer device, which may include: a processor and a memory; wherein, the memory stores a computer program adapted to be loaded and executed by the processor to perform the above method steps.

[0019] The beneficial effects brought by the technical solutions provided by some embodiments of the present application at least include:

[0020] A hybrid network model based on a non-pooling convolutional neural network and deep deterministic policy gradient (NPCNN-DDPG). NPCNN is introduced into the actor-critic network to identify the non-linear features of the environment, thereby reducing the impact of partial observability on the prediction performance. Through a new dynamic reward function for unbalanced samples, the policy score of the online actor network can be effectively evaluated, and the prediction accuracy can be effectively improved compared with the existing photovoltaic power prediction methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1It is a schematic flowchart of the photovoltaic power generation prediction method based on NPCNN-DDPG provided by the embodiments of the present application;

[0023] Figure 2 It is a prediction accuracy curve graph of each NPCNN-DDPG network provided by the embodiments of the present application under different seasons;

[0024] Figure 3 It is a prediction accuracy curve graph of each model on different prediction intervals provided by the embodiments of the present application;

[0025] Figure 4 It is a schematic structural diagram of a photovoltaic power generation prediction device based on NPCNN-DDPG provided by the present application;

[0026] Figure 5 It is a schematic structural diagram of a computer device provided by the present application. Specific embodiments

[0027] To make the objectives, technical solutions, and advantages of the present application clearer, the embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0028] It should be noted that the photovoltaic power generation prediction method based on NPCNN-DDPG provided by the present application is generally executed by a computer device. Correspondingly, the photovoltaic power generation prediction device based on NPCNN-DDPG is generally set in the computer device.

[0029] The present application constructs an NPCNN-DDPG network, then uses a sample set to train the NPCNN-DDPG. After the training is completed, the NPCNN-DDPG network is deployed to the production environment, and then the deployed NPCNN-DDPG network is used to process historical time series data to obtain the predicted power value at the predicted time.

[0030] The computer device of the present application can be a server or a terminal device. The computer device can be implemented as a distributed cluster composed of multiple computer devices, or can also be implemented as a single computer device.

[0031] Below will be combined with the attached Figure 1 , a detailed introduction will be given to the photovoltaic power generation prediction method based on NPCNN-DDPG provided by the embodiments of the present application. Among them, the photovoltaic power generation prediction device based on NPCNN-DDPG in the embodiments of the present application can be Figure 1 the computer device shown.

[0032] Please refer to Figure 1 , which is a schematic flowchart of a photovoltaic power generation prediction method based on NPCNN-DDPG provided by the embodiments of the present application. As Figure 1As shown in the figure, the method of the embodiment of the present application may include the following steps:

[0033] S101. Obtain a first sample set and a second sample set.

[0034] Among them, the first sample set and the second sample set include a plurality of sample data, and the sample data represents the environmental state value and the actual power value of the photovoltaic cell at a certain historical moment. The environmental state value may be the weather state value and the working state value of the photovoltaic cell. For example, the weather state value includes but is not limited to: temperature, humidity, air pressure, irradiance, radiation intensity, and the working state value includes but is not limited to: the installation angle of the photovoltaic cell, the power generation load of the photovoltaic cell, the resistance of the photovoltaic cell, and the current of the photovoltaic cell. The first sample set is generated from the sample data collected under extreme conditions, and the second sample set is generated from the sample data collected under normal conditions. Extreme conditions mean that the climate fluctuates violently, and normal conditions mean that the climate is relatively stable. In this embodiment, the number of the first sample set and the second sample set may be one or more respectively.

[0035] Further, in the embodiment of the present application, the method for obtaining the first sample set and the second sample set includes: collecting a plurality of sample data in a historical time period to generate an original sample set, using a clustering algorithm to cluster the sample data in the original sample set to obtain a plurality of groups, and screening out a preset proportion of sample data in each group to generate the first sample set. The proportion of the sample data screened in each group may be the same or different. For example, the proportion screened in each group is the same and is the reciprocal of the number of groups, and then the sample data in any one group is used to generate the second sample set.

[0036] For example: using a clustering algorithm to cluster the original sample set to obtain 5 groups, and screening out 20% of the sample data in equal proportion in the 5 groups respectively to generate the first sample set. In this way, the first sample set includes sample data belonging to different groups, and then the sample data in any one of the 5 groups is selected to generate the second sample data, that is, the second sample data only contains sample data belonging to the same group. The present application ensures the stability of the prediction result of the artificial neural network by constructing unbalanced samples.

[0037] S102. Initialize the network parameters of the NPCNN-DDPG network.

[0038] Among them, constructing the NPCNN-DDPG network specifically includes the design of the experience pool and the construction of the Actor-Critic network.

[0039] Design of the experience pool: Experience replay mainly has two key steps: "storage" and "replay": "storage" means storing the experience as e t ={s t ,a t, r t , s t+1} is stored in the experience pool in the form of. "Replay" means sampling one or more pieces of experience data from the experience pool according to a certain rule. s t , a t and r t are the environmental state value at time t, the action value (i.e., the predicted power value) output by the online participant network based on s t and the reward value (related to the prediction error); s t+1 is the environmental state value at time t + 1. Advantages of experience replay: 1) It can reuse experiences, with high data utilization rate, which is especially useful for situations where data acquisition is difficult; 2) When training an artificial neural network, it can break the correlation between data, making the data satisfy independent and identically distributed, thereby reducing the variance of parameter updates and improving the convergence speed; 3) It can take the average value of the data distribution of its state, thereby smoothing the learning process and avoiding parameter fluctuations or local minima.

[0040] Construction of the Actor-Critic network: The NPCNN-DDPG network includes: an online actor network, a target actor network, an online critic network, and a target critic network. The network parameters of each network include weights and biases. The network description is as follows:

[0041] a t = μ(s t ) = W 11 ·{σ[W 12 , s t + b 11}+ b 12 ;

[0042] a′ t = μ′(s t+1 ) = W 11 ·{σ[W 22 , s t+1 + b 21}+ b 22 ;

[0043] Q t = φ(s t , a t ) = W 31 ·{σ[W 32 · s t + W 33 · a t + b 31}+ b 32 ;

[0044] Q′ t = φ′(st+1 , a' t ) = W 41 ·{σ[W 42 ·s t+1 + W 43 ·a' t + b 41} + b 42 。

[0045] Among them, in the formula, a t is the action value selected by the online participant network under the environmental state value s t ; Q t is the score value calculated by the online evaluator network under the environmental state value s t and the action value a t . a' t is the action value selected by the target participant network under the environmental state value s t+1 at the next moment; Q' t is the score value calculated by the target evaluator network at the environmental state value s t+1 and the action value a' t . Its θ = {W 11 , W 12 , b 11 , b 12} and θ' = {W 21 , W 22 , b 21 , b 22} are the network parameters (weights and biases) of the online participant network and the target participant network respectively. w = {W 31 , W 32 , W 33 , b 31 , b 32 , b 33} and w' = {W 41 , W 42 , W 43 , b 41 , b 42 , b 43} are the network parameters of the online evaluator network and the target evaluator network respectively.

[0046] For the target actor network and the target critic network, the soft update method is adopted in the DDPG (Deep Deterministic Policy Gradient) algorithm, which can also be called exponential moving average. That is, a learning rate (or called momentum) is introduced, and the old network parameters and the new network parameters of the target network (target actor network or target critic network) are weighted averaged and then assigned to the target network. Since the network parameters in the target network are slowly updated by the soft update method, its output will be more stable, and using the target network to calculate the target value will naturally be more stable, thus further ensuring that the learning processes of the Critic and the target Critic network are smoother.

[0047] Further, in this embodiment, the online actor network is NPCNN (Non-Pooled Convolutional Neural Networks), and NPCCN does not include a pooling layer and is composed of an input layer, a convolutional layer, a fully connected layer, and an output layer. The relationships are as follows: 1) The input layer can provide input for the training of NPCNN: Based on correlation analysis, the sample data set is filtered into one-dimensional state data, and then it is merged and converted into a two-dimensional matrix through a dimension transformation mechanism and supplied to NPCNN. 2) The convolutional layer can effectively extract the relevant features of the previous layer and form the input of the fully connected layer through an activation function, which is based on the tissue principle of the animal visual cortex, that is, neurons in different layers can achieve local connection through weight sharing. 3) The fully connected layer usually maps the two-dimensional feature map output by the convolutional layer to the output layer according to the full connection of neurons. In this application, NPCNN is introduced into the online actor network to identify and learn the non-linear features of the photovoltaic environment to reduce the impact of partial observability on the prediction performance.

[0048] The reasons for designing the online actor network as a non-pooled convolutional neural network are as follows: 1) There are few relevant features, and the pooling operation of the convolutional neural network does not have sufficient support; 2) The weight sharing technology of the convolutional operation of the non-pooled convolutional neural network can reduce the number of training times of the target network parameters; 3) The non-pooled convolutional neural network is more suitable for processing time series data with seasonal change trends and can reduce the negative impact of the pooling layer on the learning environment.

[0049] After the construction of the NPCNN-DDPG network is completed, the network parameters of each network are randomly initialized.

[0050] S103. Train the NPCNN-DDPG network using the first sample set and the second sample set.

[0051] Among them, after randomly initializing the network parameters θ of the online participant network and the network parameters w of the online evaluator network, initialize the network parameters θ′ and w′ of the corresponding target participant network and target evaluator network of the above two networks; initialize the experience pool R; within the number of training rounds T, select the action value a according to the current policy and noise t and execute the action value to obtain the reward value r t and the next environmental state value s t+1 ; store the current environmental state value s t , the current action value a t , the current reward value r t and the next environmental state value s t+1 in the experience pool R; randomly sample a small batch of sample data from the experience pool R for training the NPCNN-DDPG network.

[0052] During the training process, update the network parameter w according to the loss function of the online evaluator network. The loss function is shown in the following formula:

[0053]

[0054] Update the network parameter θ of the online participant network according to the gradient of the online participant network. The gradient function is shown in the following formula:

[0055] Among them, represents the gradient of the online participant network, N is the size of the training batch, is the change amount, Q is the function of the online evaluator network (i.e., the scoring function), w is the network parameter of the online evaluator network, μ is the function of the online participant network, and θ is the network parameter of the online participant network.

[0056] Update the network parameters θ′ of the target participant network and the network parameters w′ of the target evaluator network using the weighted formula. The weighted formula is shown as follows:

[0057] Among them, θ t represents the network parameter of the online participant network at time t, θ′ t represents the network parameter of the target participant network at time t, θ′ t+1 represents the network parameter of the target participant network at time t + 1; w t represents the network parameter of the online evaluator network at time t, w′ t , represents the network parameter of the target evaluator network at time t, w′ t+1 represents the network parameter of the target evaluator network at time t + 1.

[0058] During the training process, after meeting the preset exit conditions, the iteration is stopped. For example, when the number of iterations is equal to the preset minimum number or the prediction error meets the preset accuracy requirement, the iteration is stopped, and finally the final network parameters of each network are obtained, thus completing the training of NPCNN-DDPG.

[0059] Furthermore, this application uses a reward function based on unbalanced samples and training errors to score the online participant network of the DDPG algorithm, so as to ensure the stability of the algorithm prediction results.

[0060] Using the rewards feedback from the environment to evaluate the policy score of the target network and thus drive the next behavior, designing the reward function of reinforcement learning is crucial for the training of network parameters. In order to overcome the shortcomings and instability of DDPG and prevent extreme situations or special moments from having an adverse impact on the algorithm model, a new adaptive dynamic reward function for unbalanced samples and training errors is designed and adopted, which can make the convergence and stability of the NPCNN-DDPG network better. The reward function is shown in the following formula:

[0061]

[0062] In the formula, a t and y t are the predicted power value and the actual power value at time t respectively; ζ e and c e are the total number of training rounds and the current training round respectively; S E is the first sample set, that is, the sample set in extreme environments; S N is the second sample set, that is, the sample set in normal environments; χ t is the weight of the second sample set, and λ t is the weight of the first sample set. In order to better learn the environment at extreme moments, the rewards at extreme moments should be more sensitive, thus amplifying the rewards at extreme moments. On the contrary, the rewards at normal moments are reduced. Therefore, the setting of χ t should be greater than λ t

[0063] S104. Process the historical time series data through the trained NPCNN-DDPG network to obtain the predicted power value for future time.

[0064] Among them, after completing the training of the NPCNN-DDPG network, the historical time series data is obtained. The historical time series data is the environmental state values collected at each moment within the historical time (see S101). After inputting the historical time series data into the NPCNN-DDPG network, the predicted power value at the prediction moment is output. The predicted power value is a normalized value between 0 and 1, and the output predicted power value is converted into a power value readable by users.

[0065] See Figure 2 as shown Figure 2 is the prediction accuracy curve of the NPCNN-DDPG network of the present application in different seasons. This curve is drawn by comparing the actual power value and the predicted power value of the photovoltaic cell. Figure 2 The horizontal axis is the predicted power value, and the vertical axis is the actual power value, with the unit of milliwatt (MW). The regression baseline is given in each curve, and it can be seen that the NPCNN-DDPG network has good prediction accuracy.

[0066] Table 1 gives the comparison of the prediction accuracy between the NPCNN-DDPG network and the existing CNN and SVR (Support Vector Regression). The 2 closer the R-index is to 1, the higher the degree of fitting the model to the data. It can be seen from Table 1 that the NPCNN-DDPG network of the present application has better goodness of fit.

[0067]

[0068] See Figure 3 as shown Figure 3 are the prediction accuracy curves of each different model in different prediction intervals. The RMSE (Root Mean Square Error) index is selected. The smaller this index, the higher the model accuracy. Figure 5 It can be seen that the NPCNN-DDPG network has better prediction accuracy than CNN and SVR in the prediction intervals of 30 minutes, 45 minutes, 60 minutes, 75 minutes, and 90 minutes.

[0069] The present application introduces a deep deterministic policy gradient algorithm for photovoltaic prediction: abandoning shallow machine learning methods such as support vector regression, decision tree, K-nearest neighbor, and neural network, and using the proposed NPCNN-DDPG hybrid structure to perform feature training on the daily photovoltaic cell power generation. DDPG can combine the non-linear fitting ability of deep learning and the decision-making ability of reinforcement learning to learn the behavior characteristics in the high-dimensional continuous state space, and can improve the solution ability of the prediction model in complex environments.

[0070] The present application proposes a photovoltaic prediction model of NPCNN-DDPG: abandoning the method of learning behavior strategies based on traditional neural networks, and proposing a hybrid prediction model based on the NPCNN-DDPG algorithm. First, the NPCNN model is introduced into the actor network of DDPG to achieve feature learning in the high-dimensional continuous state space, thereby reducing the impact of the partial observability of DDPG on the prediction performance.

[0071] This application proposes an adaptive dynamic reward function for NPCNN-DDPG, which abandons the reward function of traditional DDPG. The adaptive dynamic reward function can not only adaptively learn the behavioral strategies of problems in complex natural environments, but also effectively evaluate the policy scores of the Actor-network network, making the convergence and stability of NPCNN-DDPG better.

[0072] The following is an apparatus embodiment of this application, which can be used to execute the method embodiment of this application. For the details not disclosed in the apparatus embodiment of this application, please refer to the method embodiment of this application.

[0073] Please refer to Figure 4 , which shows a schematic structural diagram of a photovoltaic power prediction device based on NPCNN-DDPG provided by an exemplary embodiment of this application, hereinafter referred to as device 4. This device 4 can be implemented as all or part of a computer device through software, hardware, or a combination of both. Device 4 includes: an acquisition unit 401, an initialization unit 402, a training unit 403, and a prediction unit 404.

[0074] The acquisition unit 401 is configured to acquire a first sample set and a second sample set; wherein, the first sample set is a sample set in an extreme environment, and the second sample set is a sample set in a normal environment;

[0075] The initialization unit 402 is configured to initialize the network parameters of the NPCNN-DDPG network; wherein, the NPCNN-DDPG network includes: an online actor network, an online critic network, a target actor network, and a target critic network, and the online actor network is a non-pooling convolutional neural network NPCNN;

[0076] The training unit 403 is configured to train the NPCNN-DDPG network by using the first sample set and the second sample set; wherein, the reward function of the online actor network is a t represents the predicted power value of the photovoltaic cell at time t, y t represents the actual power value of the photovoltaic cell at time t, ζ e represents the number of training rounds, c e represents the current training round, S E represents the first sample set, S N represents the second sample set, χ t represents the weight of the first sample set, λ t represents the weight of the second sample set;

[0077] A prediction unit 404, configured to process historical time series data according to the NPCNN-DDPG network that has completed training, so as to obtain a predicted power value at the predicted time.

[0078] In one or more possible embodiments, the loss function of the online critic network is expressed as:

[0079] where Q represents the evaluation function, and w represents the network parameters of the online critic network.

[0080] In one or more possible embodiments, during the process of training the NPCNN-DDPG network, the following formula is used to update the network parameters of the target critic network and the network parameters of the target actor network:

[0081]

[0082] In one or more possible embodiments, during the process of training the NPCNN-DDPG network, the network parameters of the online actor network are updated according to the gradient of the online actor network.

[0083] In one or more possible embodiments, the first sample set and the second sample set include a plurality of sample data, and the sample data are environmental state values and actual power values of photovoltaic cells at historical moments.

[0084] In one or more possible embodiments, the obtaining of the first sample set and the second sample set includes:

[0085] Collect a plurality of sample data within a historical time period to generate an original sample set, use a clustering algorithm to cluster the sample data in the original sample set to obtain a plurality of groups, screen out a preset proportion of sample data in each group to generate a first sample set, and generate a second sample set from the sample data in any one group.

[0086] It should be noted that when the device 4 provided in the above embodiment executes the photovoltaic power prediction method based on NPCNN-DDPG, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above functions. In addition, the photovoltaic power prediction device based on NPCNN-DDPG provided in the above embodiment and the embodiment of the photovoltaic power prediction method based on NPCNN-DDPG belong to the same concept, and the implementation process is detailed in the method embodiment, which will not be repeated here.

[0087] The serial numbers of the embodiments of the present application above are only for description and do not represent the advantages and disadvantages of the embodiments.

[0088] The embodiment of the present application also provides a computer storage medium, which can store multiple instructions. The instructions are suitable for being loaded and executed by a processor to perform the method steps of the embodiment as described above Figure 1 shown in the embodiment. The specific execution process can be referred to Figure 1 the specific description of the embodiment shown, and will not be elaborated here.

[0089] The present application also provides a computer program product, which stores at least one instruction. The at least one instruction is loaded and executed by the processor to implement the photovoltaic power prediction method based on NPCNN-DDPG described in each of the above embodiments.

[0090] Please refer to Figure 5 , which is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 5 shown, the computer device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0091] Among them, the communication bus 502 is used to realize the connection and communication between these components.

[0092] Among them, the user interface 503 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 503 may further include a standard wired interface and a wireless interface.

[0093] Among them, the network interface 504 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0094] Among them, the processor 501 may include one or more processing cores. The processor 501 connects various parts within the entire computer device 500 through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 505, and by calling the data stored in the memory 505, it executes various functions of the computer device 500 and processes data. Optionally, the processor 501 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 501 may integrate a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 501 and may be implemented separately by a single chip.

[0095] Among them, the memory 505 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 505 includes a non-transitory computer-readable storage medium. The memory 505 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 505 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store the data involved in the above-mentioned various method embodiments. Optionally, the memory 505 may also be at least one storage device located far from the aforementioned processor 501. As Figure 5 shown, the memory 505, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and application programs.

[0096] In Figure 5In the computer device 500 shown, the user interface 503 is mainly used to provide an interface for the user to input and obtain the data input by the user; while the processor 501 can be used to call the application programs stored in the memory 505 and specifically execute the method as shown in Figure 2 shown. The specific process can be referred to Figure 1 shown, which will not be elaborated here.

[0097] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above various methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.

[0098] The above-disclosed are only the preferred embodiments of the present application. Of course, the scope of the rights of the present application cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A photovoltaic power prediction method based on NPCNN-DDPG, characterized in that, Including: Obtain a first sample set and a second sample set; wherein, the first sample set is a sample set in an extreme environment, and the second sample set is a sample set in a normal environment; Initialize the network parameters of the NPCNN-DDPG network; wherein, the NPCNN-DDPG network includes: an online actor network, an online critic network, a target actor network, and a target critic network, and the online actor network is a non-pooling convolutional neural network NPCNN; Train the NPCNN-DDPG network using the first sample set and the second sample set; wherein, the reward function of the online actor network is a t represents the predicted power value of the photovoltaic cell at time t, y t represents the actual power value of the photovoltaic cell at time t, ζ e represents the training round, c e represents the current training round, S E represents the first sample set, S N represents the second sample set, x t represents the weight of the first sample set, λ t represents the weight of the second sample set; Process the historical time series data according to the trained NPCNN-DDPG network to obtain the predicted power value at the predicted time.

2. The method according to claim 1, wherein The loss function of the online critic network is expressed as: Among them, Q represents the evaluation function, and w represents the network parameters of the online evaluator network.

3. The method according to claim 1 or 2, characterized in that, During the process of training the NPCNN-DDPG network, use the following formula to update the network parameters of the target critic network and the network parameters of the target actor network:

4. The method according to claim 3, characterized in that During the process of training the NPCNN-DDPG network, update the network parameters of the online actor network according to the gradient of the online actor network.

5. The method according to claim 1 or 2 or 4, characterized in that The first sample set and the second sample set include multiple sample data, and the sample data are the environmental state values and the actual power values of the photovoltaic cells at historical moments.

6. The method according to claim 5, characterized in that, The obtaining of the first sample set and the second sample set includes: Collect multiple sample data within a historical time period to generate an original sample set, use a clustering algorithm to cluster the sample data in the original sample set to obtain multiple groups, screen out a preset proportion of the sample data in each group to generate the first sample set, and generate the second sample set from the sample data in any one group.

7. The method according to claim 6, characterized in that, The preset proportion is equal to the reciprocal of the number of groups.

8. A photovoltaic power generation power prediction device based on NPCNN-DDPG, characterized in that, Including: An obtaining unit, configured to obtain a first sample set and a second sample set; wherein, the first sample set is a sample set in an extreme environment, and the second sample set is a sample set in a normal environment; An initializing unit, configured to initialize the network parameters of the NPCNN-DDPG network; wherein, the NPCNN-DDPG network includes: an online actor network, an online critic network, a target actor network, and a target critic network, and the online actor network is a non-pooling convolutional neural network NPCNN; A training unit for training the NPCNN-DDPG network using the first sample set and the second sample set; wherein, the reward function of the online actor network is a t represents the predicted power value of the photovoltaic cell at time t, y t represents the actual power value of the photovoltaic cell at time t, ζ e represents the training round, c e represents the current training round, S E represents the first sample set, S N represents the second sample set, χ t represents the weight of the first sample set, λ t represents the weight of the second sample set; A predicting unit, configured to process the historical time series data according to the trained NPCNN-DDPG network to obtain the predicted power value at the predicted time.

9. A computer storage medium, characterized in that, The computer storage medium stores multiple instructions, and the instructions are suitable for being loaded and executed by a processor to perform the method steps of any one of claims 1 to 7.

10. A computer device, characterized in that, Including: A processor and a memory; wherein, the memory stores a computer program, and the computer program is suitable for being loaded and executed by the processor to perform the method steps of any one of claims 1 to 7.

Images