Distributed photovoltaic power generation prediction method and system
By adopting empirical modal decomposition, causal convolution and dot product attention mechanisms in recurrent neural networks, combined with the design of activation function, the accuracy and error handling problems of time series data in photovoltaic power generation prediction are solved, and more efficient prediction is achieved and feature extraction improves the overall accuracy and stability of photovoltaic power generation prediction.
Patent Information
- Application Number
- CN202410849999.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-27
- Publication Date
- 2024-11-29
AI Technical Summary
The existing pseudo-measurement method based on recurrent neural networks is not very accurate for excessively long time series data prediction and is not effective when processing erroneous data in timing data.
Empirical modal decomposition is used for data preprocessing, and by introducing causal convolution and dot product attention mechanisms, combined with activation functions, the input and output strategies of recurrent neural networks are optimized to improve prediction accuracy and processing capabilities.
It improves the accuracy and speed of photovoltaic power generation prediction, enhances the feature extraction effect of the model, alleviates the gradient problem, and improves the prediction accuracy.
Smart Images

Figure CN119046782A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic power generation prediction, and in particular to a distributed photovoltaic power generation prediction method and system. Background Art
[0002] Due to the randomness of sunshine intensity, the power generation of photovoltaic power stations is often unstable, and the distribution of power stations is also more dispersed than that of traditional thermal power stations. In view of the randomness of power generation and the dispersed geographical location of photovoltaic power stations, it is necessary to provide them with a control and management method. At present, the state estimation technology has been widely used in the control and management methods of photovoltaic power stations. State estimation is a technology that collects existing data of power stations and analyzes them to derive the operating status of power stations. Existing power station state estimation methods can be divided into transmission system-based measurement, distributed system-based measurement and pseudo-measurement, among which pseudo-measurement is mainly based on learning-based measurement.
[0003] In traditional power plant state estimation, pseudo-measurement has not been widely used due to its accuracy issues. In recent years, the development of deep learning technology has brought new optimization solutions to pseudo-measurement, effectively improving the accuracy of pseudo-measurement. However, the current pseudo-measurement method based on recurrent neural networks still does not meet the prediction accuracy requirements for long time series data, and the processing effect of erroneous data in time series data is not good. These errors may be caused by equipment failure, bad weather and other reasons. Therefore, there is an urgent need for a control method that can process random data with high accuracy. Summary of the invention
[0004] In view of the above existing problems, the present invention is proposed.
[0005] Therefore, the present invention provides a distributed photovoltaic power generation prediction method and system to solve the problem that the current pseudo-measurement method cannot meet the prediction accuracy requirements for time series data that is too long, and has poor effect on processing erroneous data in time series data.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In a first aspect, the present invention provides a distributed photovoltaic power generation prediction method, comprising:
[0008] Obtaining the original power generation data sequence;
[0009] Preprocessing the original power generation data sequence, wherein the preprocessing decomposes the original power generation data sequence by using an empirical mode decomposition principle to obtain a plurality of subsequences;
[0010] The plurality of subsequences are input into a recurrent neural network, the input is convolved on the plurality of subsequences by introducing a causal convolution method, and a prediction component of each subsequence is obtained; the prediction components are aggregated, the aggregation introduces a dot product attention mechanism and uses an activation function as a network connection method to control the aggregation degree of different prediction components and obtain a prediction value;
[0011] The output power is regulated based on the predicted value, and different output strategies are set.
[0012] As a preferred solution of the distributed photovoltaic power generation prediction method of the present invention, the preprocessing decomposes the original power generation data sequence by the empirical mode decomposition principle to obtain several subsequences, including:
[0013] Subtract the original power generation data sequence from the upper and lower envelopes to obtain the candidate photovoltaic power generation characteristic sequence;
[0014] If the candidate power generation characteristic sequence meets the decomposition principle of sub-signal modal stability, it is taken out from the original sequence and the remaining sequence is further decomposed;
[0015] If the alternative power generation characteristics do not satisfy the decomposition principle of the sub-signal modal stability, the upper and lower envelope operations are continued until the decomposition principle is satisfied;
[0016] Several subsequences and residual sequences are obtained, and the residual sequence is eliminated.
[0017] As a preferred solution of the distributed photovoltaic power generation prediction method of the present invention, wherein: the input performs convolution operation on several subsequences by introducing a causal convolution method, and obtains the prediction component of each subsequence, including:
[0018] According to the periodic characteristics of photovoltaic power generation data, the sliding window time span of the input length is preset;
[0019] Input several subsequences and introduce causal convolution method to perform convolution operation on the input data;
[0020] The reset state and update state of the gated recurrent unit neurons are used to retain and update the input data, extract the local features of the input data, and obtain the predicted components of each subsequence.
[0021] As a preferred solution of the distributed photovoltaic power generation prediction method described in the present invention, the aggregation introduces a dot product attention mechanism, including:
[0022] Calculate the dot product value of the subsequence and the original power generation data sequence, and normalize it to obtain the weight;
[0023] The aggregation degree of different prediction components is controlled by calculating the product of the weight and the prediction component of the subsequence.
[0024] As a preferred solution of the distributed photovoltaic power generation prediction method described in the present invention, wherein: the activation function is used as the network connection mode, wherein:
[0025] The activation function is set to a Sigmoid function with an output range of (-1, 1).
[0026] As a preferred solution of the distributed photovoltaic power generation prediction method described in the present invention, the upper and lower envelopes are constructed using several cubic functions with equal intersection derivatives, specifically including:
[0027] Obtain all maximum and minimum points of the original power generation data sequence;
[0028] Connect two adjacent maximum points with a cubic function, and solve the cubic function through the coordinates and derivative values of the two points
[0029] Select the constraint that the derivatives at the intersection points are equal to connect adjacent intersecting cubic functions;
[0030] The curve connecting the maximum points is taken as the upper envelope, and the curve connecting the minimum points is taken as the lower envelope.
[0031] As a preferred solution of the distributed photovoltaic power generation prediction method of the present invention, the output power is regulated based on the predicted value, and different output strategies are set, including:
[0032] Design different output power control schemes according to photovoltaic power generation;
[0033] Based on the principle of matching output power with power generation, the control scheme is dynamically changed in combination with the predicted value.
[0034] In a second aspect, the present invention provides a distributed photovoltaic power generation prediction system, comprising:
[0035] An acquisition module, used for acquiring a raw power generation data sequence;
[0036] A preprocessing decomposition module, used for preprocessing the original power generation data sequence, wherein the preprocessing decomposes the original power generation data sequence by using the empirical mode decomposition principle to obtain several subsequences;
[0037] A prediction module is used to input the several subsequences into a recurrent neural network, the input performs convolution operation on the several subsequences by introducing a causal convolution method, and obtains prediction components of each subsequence; the prediction components are aggregated, the aggregation introduces a dot product attention mechanism and uses an activation function as a network connection method to control the aggregation degree of different prediction components and obtain prediction values;
[0038] The output control module is used to control the output power based on the predicted value and set different output strategies.
[0039] In a third aspect, the present invention provides a computing device, comprising:
[0040] Memory and processor;
[0041] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the distributed photovoltaic power generation prediction method are implemented.
[0042] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the distributed photovoltaic power generation prediction method.
[0043] Compared with the prior art, the present invention has the following beneficial effects: the present invention adapts the time series characteristics of photovoltaic power station power generation data through a recurrent neural network, which is easy to capture the long-term and short-term characteristics of photovoltaic data, and is superior to traditional statistical-based methods in prediction speed and accuracy; at the same time, the feature receptive field of the prediction model is expanded by processing the input data through the causal convolution method, and the feature extraction effect of the model is optimized; and the new activation function is optimized and designed to adapt to the fluctuation characteristics of photovoltaic power generation data, and the connection method of the residual mechanism is optimized, which alleviates the gradient problem caused by the stacking of model layers and improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0045] Figure 1 A schematic diagram of the overall process of a distributed photovoltaic power generation prediction method according to an embodiment of the present invention;
[0046] Figure 2 A schematic diagram of the operating process framework of a distributed photovoltaic power generation prediction method according to an embodiment of the present invention;
[0047] Figure 3 It is a schematic diagram of the overall structure of a photovoltaic power generation prediction network in a distributed photovoltaic power generation prediction method according to an embodiment of the present invention;
[0048] Figure 4 A schematic flow chart of a photovoltaic power generation data decomposition method in a distributed photovoltaic power generation prediction method according to an embodiment of the present invention;
[0049] Figure 5 Schematic diagram of the gated recurrent unit structure of the recurrent neural network in the distributed photovoltaic power generation prediction method according to an embodiment of the present invention
[0050] Figure 6 The figure is a schematic diagram of the activation function structure in the distributed photovoltaic power generation prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0051] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0052] Example 1
[0053] Reference Figure 1-Figure 6 , as an embodiment of the present invention, provides a distributed photovoltaic power generation prediction method, including: Figure 1-Figure 3 As shown:
[0054] S100: Acquire original power generation data sequence;
[0055] S200: preprocessing the original power generation data sequence, decomposing the original power generation data sequence by the empirical mode decomposition principle to obtain several subsequences;
[0056] Preferably, the preprocessing decomposes the original power generation data sequence by the empirical mode decomposition principle to obtain several subsequences, including:
[0057] Subtract the original power generation data sequence from the upper and lower envelopes to obtain the candidate photovoltaic power generation characteristic sequence;
[0058] If the candidate power generation characteristic sequence meets the decomposition principle of sub-signal modal stability, it is taken out from the original sequence and the remaining sequence is further decomposed;
[0059] If the alternative power generation characteristics do not satisfy the decomposition principle of the sub-signal modal stability, the upper and lower envelope operations are continued until the decomposition principle is satisfied;
[0060] Several subsequences and residual sequences are obtained, and the residual sequence is eliminated.
[0061] It should be noted that the decomposition principle of sub-signal modal stability is set according to the load decomposition principle and the periodic characteristics of photovoltaic power generation data;
[0062] Specifically, the modally stable sequence should satisfy the following conditions: the number of extreme points and the number of zero-crossing points in the entire time course are equal or differ by at most 1; at any time, the average value of the upper envelope formed by the local maximum points and the lower envelope formed by the local minimum points is zero, that is, the upper and lower envelopes are locally symmetric with respect to the time axis. Take the median of the upper and lower envelopes, and use the original power generation data sequence to subtract the median to obtain an alternative subsequence. If the alternative subsequence satisfies the decomposition principle, decompose the subsequence and continue to decompose the remaining sequence. If the alternative subsequence does not meet the decomposition principle, repeat the upper and lower envelope operations on the sequence until the decomposition principle is met.
[0063] It should also be noted that after the above decomposition operation is completed, the original photovoltaic power generation data will be decomposed into multiple modally stable subsequences and a residual sequence. The residual sequence has a monotonic trend, which will cause the period of the residual sequence to be too long. Within the limited observation length, it cannot represent the periodic characteristics of the original power generation data sequence. Therefore, removing the residual sequence as the noise of the original sequence can reduce the impact of noise.
[0064] The traditional data processing method based on empirical mode decomposition believes that each subsequence can reflect the periodic characteristics of photovoltaic power generation data. However, for the residual sequence, the data change generally shows a monotonic trend, which causes the data signal period of the participating sequence to be too long relative to the original sequence, and actually cannot reflect the periodic characteristics of photovoltaic power generation data. When obtaining the subsequence, the present invention considers the phenomenon that the residual sequence period does not match the original power generation data sequence period, and removes the residual sequence as noise.
[0065] Preferably, the upper and lower envelopes are constructed using several cubic functions with equal intersection derivatives, specifically including:
[0066] Obtain all maximum and minimum points of the original power generation data sequence;
[0067] Connect two adjacent maximum points with a cubic function, and solve the cubic function through the coordinates and derivative values of the two points
[0068] Select the constraint that the derivatives at the intersection points are equal to connect adjacent intersecting cubic functions;
[0069] The curve connecting the maximum points is taken as the upper envelope, and the curve connecting the minimum points is taken as the lower envelope.
[0070] It should be noted that selecting the constraint of equal derivatives at intersection points can smoothly connect adjacent intersecting cubic functions and make the final curve more consistent with the fluctuation characteristics of photovoltaic power generation.
[0071] S300: Input several subsequences into a recurrent neural network, perform convolution operations on the subsequences by introducing a causal convolution method, and obtain prediction components of each subsequence; aggregate the prediction components, introduce a dot product attention mechanism and use an activation function as a network connection method to control the aggregation degree of different prediction components and obtain prediction values;
[0072] It should be noted that according to the time series characteristics of photovoltaic power generation data and the current optimization direction of time series prediction, a recurrent neural network is used to extract features of photovoltaic power generation data and optimize the network.
[0073] Preferably, the input performs convolution operations on several subsequences by introducing a causal convolution method, and obtains the predicted components of each subsequence, such as Figure 4 As shown, including:
[0074] According to the periodic characteristics of photovoltaic power generation data, the sliding window time span of the input length is preset;
[0075] Specifically, the sliding window time span can be set to 1 day or a time suitable for the scenario;
[0076] Input several subsequences and introduce causal convolution method to perform convolution operation on the input data;
[0077] like Figure 5 As shown, the reset state and update state of the gated recurrent unit neuron are used to retain and update the input data, extract the local features of the input data, and obtain the predicted components of each subsequence.
[0078] Specifically, the reset state and the update state both exist in the form of matrices, and by multiplying the information element by element, the effect of controlling the degree of information retention is achieved. Among them, the reset state controls the degree of retention of the previous neuron information, and the update state controls the degree of retention of the new information.
[0079] It should be noted that the introduction of the causal convolution method and the convolution operation on the input power generation data can reduce the dimension of the input data and reduce the parameter overhead of the neural network; and through the reset state and update state of neurons, the degree of information retention can be determined when the recurrent network transmits information.
[0080] Preferably, the aggregation introduces a dot product attention mechanism, including:
[0081] Calculate the dot product value of the subsequence and the original power generation data sequence, and normalize it to obtain the weight;
[0082] The degree of aggregation of different prediction components is controlled by calculating the product of the weight and the prediction component of the subsequence.
[0083] Specifically, the subsequence data corresponds to the Key in the attention mechanism, and the original data corresponds to the Query.
[0084] It should be noted that subsequences of different frequencies have different correlations with the original sequence, and the attention mechanism is used to focus on subsequences with higher correlation; and the attention mechanism can control the degree of aggregation of different prediction components to achieve the effect of focusing on specific prediction components and improve prediction accuracy.
[0085] Regarding the activation function, you can generally choose the Sigmoid function or the Tanh function;
[0086] Preferably, Figure 6 As shown, the activation function is used as the network connection method, where:
[0087] Set the activation function to the Sigmoid function with an output range of (-1,1).
[0088] It should be noted that for the power generation data series of long-term sunny or cloudy days, there are periodic characteristics. For this type of power generation data series, the present invention takes into account the power generation fluctuation phenomenon caused by random illumination of photovoltaic power generation, and designs an activation function with a larger output range to adapt to the random characteristics of photovoltaic power generation data. When inputting photovoltaic power generation type data, it can achieve the effect that the output changes more obviously with the input and the learning effect is better. At the same time, drawing on the structure of the residual mechanism and using the activation function of the present invention as the connection method to design the entire prediction network structure, it can solve the problem of difficulty in updating parameters caused by optimizing the number of stacking layers.
[0089] S400: Regulating output power based on the predicted value and setting different output strategies.
[0090] Preferably, the output power is regulated based on the predicted value, and different output strategies are set, including:
[0091] Design different output power control schemes according to photovoltaic power generation;
[0092] Based on the principle of matching output power with power generation, the control scheme is changed dynamically in combination with the predicted value.
[0093] It should be noted that for the power generation and energy storage system, it is necessary to keep its output power matching the power generation at the corresponding moment to ensure the stability and safety of the system. Therefore, the above-mentioned design and control scheme is adopted.
[0094] For example, considering the fluctuation characteristics of photovoltaic power generation caused by clouds, the photovoltaic power generation forecast is obtained using the power generation prediction module, and different power fluctuation smoothing strategies are set according to different prediction values. For example, when the power generation prediction value is relatively stable, the smoothing strategy is suspended; when the power generation prediction value fluctuates frequently, different smoothing strategies are set according to the degree of fluctuation.
[0095] The above is a schematic scheme of a distributed photovoltaic power generation prediction method of this embodiment. It should be noted that the technical scheme of the distributed photovoltaic power generation prediction system based on the improved recurrent neural network and the technical scheme of the above-mentioned distributed photovoltaic power generation prediction method belong to the same concept. The details of the technical scheme of the distributed photovoltaic power generation prediction system based on the improved recurrent neural network in this embodiment that are not described in detail can all be referred to the description of the technical scheme of the above-mentioned distributed photovoltaic power generation prediction method.
[0096] The distributed photovoltaic power generation prediction system based on the improved recurrent neural network in this embodiment includes:
[0097] An acquisition module, used for acquiring a raw power generation data sequence;
[0098] The preprocessing decomposition module is used to preprocess the original power generation data sequence. The preprocessing decomposes the original power generation data sequence through the empirical mode decomposition principle to obtain several subsequences;
[0099] The prediction module is used to input several subsequences into the recurrent neural network. The input convolution operation is performed on several subsequences by introducing the causal convolution method, and the prediction components of each subsequence are obtained; the prediction components are aggregated, and the dot product attention mechanism is introduced in the aggregation and the activation function is used as the network connection method to control the aggregation degree of different prediction components and obtain the prediction value;
[0100] The output control module is used to control the output power based on the predicted value and set different output strategies.
[0101] This embodiment also provides a computing device, which is applicable to the case of distributed photovoltaic power generation prediction based on an improved recurrent neural network, including:
[0102] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the distributed photovoltaic power generation prediction method proposed in the above embodiment.
[0103] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the distributed photovoltaic power generation prediction method proposed in the above embodiment is implemented.
[0104] The storage medium proposed in this embodiment and the method for implementing distributed photovoltaic power generation prediction proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.
[0105] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.
[0106] Example 2
[0107] Referring to Table 1, based on the previous embodiment, this embodiment provides an application comparison case of a distributed photovoltaic power generation prediction method to illustrate the feasibility and beneficial effects of our solution.
[0108] Five types of models commonly used in power generation forecasting tasks were selected as comparison models, including LSTM, GRU, DARNN, LSTNet, and CNN-LSTM. Among them, LSTM and GRU are traditional recurrent neural network models, and DARNN, LSTNet, and CNN-LSTM are models optimized based on the attention mechanism.
[0109] Table 1 shows the comparison data of the photovoltaic power generation prediction method proposed by the present invention and the currently commonly used prediction methods in terms of RMSE and MAE. Among them, the R2 index reflects the degree of fit of the model to the photovoltaic power generation data. The higher the R2 value, the better the fitting effect. RMSE and MAE reflect the error of the model. The smaller the RMSE and MAE values, the smaller the error. In addition, cost_avg_Epoach is the time loss indicator for model training.
[0110] Table 1 Comparative experimental results
[0111]
[0112]
[0113] The bold data in Table 1 indicate that the corresponding model has the best effect on this indicator. As can be seen from Table 1, the CC-GRU method of the present invention has an advantage in cost_avg_Epoach and is superior to the other five methods. Referring to the cost_avg_Epoach and RMSE indicators, compared with the DARNN method, CC-GRU reduces the time loss while keeping the accuracy of the model in a relatively good state.
[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A distributed photovoltaic power generation prediction method, characterized in that: include: Obtaining the original power generation data sequence; Preprocessing the original power generation data sequence, wherein the preprocessing decomposes the original power generation data sequence by using an empirical mode decomposition principle to obtain a plurality of subsequences; The plurality of subsequences are input into a recurrent neural network, the input is convolved on the plurality of subsequences by introducing a causal convolution method, and a prediction component of each subsequence is obtained; the prediction components are aggregated, the aggregation introduces a dot product attention mechanism and uses an activation function as a network connection method to control the aggregation degree of different prediction components and obtain a prediction value; The output power is regulated based on the predicted value, and different output strategies are set.
2. The distributed photovoltaic power generation prediction method according to claim 1, characterized in that: The preprocessing decomposes the original power generation data sequence by the empirical mode decomposition principle to obtain several subsequences, including: Subtract the original power generation data sequence from the upper and lower envelopes to obtain the candidate photovoltaic power generation characteristic sequence; If the candidate power generation characteristic sequence meets the decomposition principle of sub-signal modal stability, it is taken out from the original sequence and the remaining sequence is further decomposed; If the alternative power generation characteristics do not satisfy the decomposition principle of the sub-signal modal stability, the upper and lower envelope operations are continued until the decomposition principle is satisfied; Several subsequences and residual sequences are obtained, and the residual sequence is eliminated.
3. The distributed photovoltaic power generation prediction method according to claim 1 or 2, characterized in that: The input performs convolution operations on several subsequences by introducing a causal convolution method, and obtains the prediction components of each subsequence, including: According to the periodic characteristics of photovoltaic power generation data, the sliding window time span of the input length is preset; Input several subsequences and introduce causal convolution method to perform convolution operation on the input data; The reset state and update state of the gated recurrent unit neurons are used to retain and update the input data, extract the local features of the input data, and obtain the predicted components of each subsequence.
4. The distributed photovoltaic power generation prediction method according to claim 3, characterized in that: The aggregation introduces the dot product attention mechanism, including: Calculate the dot product value of the subsequence and the original power generation data sequence, and normalize it to obtain the weight; The aggregation degree of different prediction components is controlled by calculating the product of the weight and the prediction component of the subsequence.
5. The distributed photovoltaic power generation prediction method according to claim 4, characterized in that: The activation function is used as the network connection mode, wherein: The activation function is set to a Sigmoid function with an output range of (-1, 1).
6. The distributed photovoltaic power generation prediction method according to claim 5, characterized in that: The upper and lower envelopes are constructed using several cubic functions with equal intersection derivatives, specifically including: Obtain all maximum and minimum points of the original power generation data sequence; Connect two adjacent maximum points with a cubic function, and solve the cubic function through the coordinates and derivative values of the two points Select the constraint that the derivatives at the intersection points are equal to connect adjacent intersecting cubic functions; The curve connecting the maximum points is taken as the upper envelope, and the curve connecting the minimum points is taken as the lower envelope.
7. The distributed photovoltaic power generation prediction method according to claim 6, characterized in that: The output power is regulated based on the predicted value, and different output strategies are set, including: Design different output power control schemes according to photovoltaic power generation; Based on the principle of matching output power with power generation, the control scheme is dynamically changed in combination with the predicted value.
8. A system using the distributed photovoltaic power generation prediction method according to claim 1, characterized in that: include: An acquisition module, used for acquiring a raw power generation data sequence; A preprocessing decomposition module, used for preprocessing the original power generation data sequence, wherein the preprocessing decomposes the original power generation data sequence by using the empirical mode decomposition principle to obtain several subsequences; A prediction module is used to input the several subsequences into a recurrent neural network, the input performs convolution operation on the several subsequences by introducing a causal convolution method, and obtains prediction components of each subsequence; the prediction components are aggregated, the aggregation introduces a dot product attention mechanism and uses an activation function as a network connection method to control the aggregation degree of different prediction components and obtain prediction values; The output control module is used to control the output power based on the predicted value and set different output strategies.
9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the distributed photovoltaic power generation prediction method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the distributed photovoltaic power generation prediction method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Empirical mode decomposition and deep learning hybrid model-based wind speed prediction method and system
CN106126896A
Power load prediction method and device, computer equipment and storage medium
CN111191841A
Photovoltaic power generation power prediction method based on similar day theory and neural network
CN113988391A
Short-term wind power combined prediction method
CN114037143A
Short-term power load prediction method and device, equipment and storage medium
CN114239945A
Cited By
Distributed photovoltaic power generation prediction method and system based on edge calculation
CN119805914A
