An artificial intelligence-based photovoltaic power station power generation prediction method

Abstract

The present application belongs to the technical field of photovoltaic power generation, and particularly relates to a photovoltaic power station power generation capacity prediction method based on artificial intelligence. First, relevant power generation influence data is collected, historical power generation capacity data and meteorological data of the collected inverters are combined to obtain inverter power generation capacity data with meteorological data factors, and data cleaning and repair and other pretreatments are performed; a prediction model is constructed by using a multi-scale adaptive gated LSTM, multi-scale input sequences are divided, first power generation capacity prediction data is obtained through LSTM network modeling and attention mechanism feature fusion; a correction coefficient is calculated according to inverter aging and temperature data; the preliminary corrected prediction data is secondarily corrected through support vector machine regression to obtain final prediction data. The method improves data quality, solves the problem of insufficient generalization ability of traditional models, reduces long-term prediction deviation, and improves prediction accuracy and reliability in complex scenarios.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of photovoltaic power generation, and particularly relates to a photovoltaic power plant power generation amount prediction method based on artificial intelligence. BACKGROUND

[0002] Photovoltaic power generation is a power generation method that directly converts solar energy into electricity by using the photovoltaic effect of the semiconductor interface. No greenhouse gas emissions are generated in photovoltaic cells, and it has the advantages of being clean, renewable, and easy to maintain. It can be widely used in distributed power generation, large-scale photovoltaic power stations and other scenarios, and can provide power for various types of electrical equipment and power grids. It is an important energy utilization technology to address energy crises and environmental problems.

[0003] In the field of photovoltaic power generation, accurate prediction of power generation is crucial. Accurate power generation prediction can help the power system to reasonably dispatch, optimize the configuration of energy storage according to the price difference in different time periods, and improve economic efficiency.

[0004] However, current photovoltaic power generation prediction faces many challenges. Existing prediction methods are mostly based on historical power generation data to train models, and then use the trained models to predict future power generation. However, photovoltaic power generation is affected by many complex factors, such as light intensity, temperature, and cloud changes in meteorological conditions, changes in light duration and solar radiation angle due to seasonal changes, and aging and failure of photovoltaic equipment. Among them, the changes in meteorological conditions and equipment status have randomness and uncertainty, which makes many prediction models have good prediction results in specific environments, but once applied to different regions or encounter sudden weather changes and equipment abnormalities, the generalization ability of the model is insufficient, and the accuracy and reliability of the prediction results will be significantly reduced, making it difficult to meet the needs of actual production and stable operation of the power system. SUMMARY

[0005] The present application proposes a photovoltaic power plant power generation amount prediction method based on artificial intelligence to solve the technical problems in the background art.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application comprises the following steps:

[0007] S1, collecting historical power generation data of photovoltaic power station inverters, meteorological data, inverter aging data, and inverter temperature data;

[0008] S2, combining the collected historical power generation data of the inverter and the meteorological data to obtain inverter power generation data with meteorological data factors, and performing data preprocessing on the data;

[0009] S3, building a power generation prediction model, inputting the preprocessed inverter power generation data with meteorological data factors to obtain first power generation prediction data;

[0010] The implementation of the power generation amount prediction model in the step S3 is constructed by using a multi-scale adaptive gated LSTM, specifically:

[0011] S31, first, the preprocessed data is divided into multi-scale input sequences according to the time granularity, including at least three types of time sequence feature vectors of hour level, day level and week level;

[0012] S32, independent hour-level LSTM network, day-level LSTM network and week-level LSTM network are constructed to model the time sequence dependence of each scale time sequence feature, and output the hidden state sequence of the corresponding scale

[0013] S33, the weights of the hidden states of each scale are calculated through the attention mechanism The calculation method is Wherein, fj=1, 2, 3 corresponds to hour level, day level and week level, W fj , b fj are learnable parameters, and the integrated hidden state is obtained by fusing multi-scale features

[0014] S4, calculate the power generation correction coefficient according to the collected inverter aging data and inverter temperature data;

[0015] S5, combine the power generation correction coefficient and the first power generation prediction data to obtain the final power generation prediction data.

[0016] As a preferred, the weather data in the step S1 is determined according to the latitude and longitude coordinates of the geographic position of each inverter.

[0017] As a preferred, the data preprocessing in the step S2 includes data cleaning and repair, and the specific implementation steps are:

[0018] S21, first, the dynamic reasonable interval of inverter power generation at each time is calculated by the probability distribution of inverter power generation data with weather data factors in the same period in history, the lower limit is L(t)=μ t-1 ×(1-α1×σ t-1 / μ t-1 ), and the upper limit is U(t)=μ t-1 ×(1+α2×σ t-1 / μ t-1 ), wherein σ t-1 , μ t-1 are the standard deviation and mean value of the historical data in the same period, and α1 and α2 are adjustment coefficients. The power generation data exceeding the interval is determined as out-of-limit data and is removed;

[0019] S22, adopt bidirectional LSTM network, input inverter power generation data sequence with meteorological data factors, train mapping relationship under normal working condition, output predicted power generation at each time, calculate relative error between actual value and predicted value, when relative error is greater than set threshold, determine as abnormal data and eliminate, record position and context features after elimination;

[0020] S23, for time t needing repair, retrieve N normal time periods closest to t in historical data, form set S; according to time interval Δt of each period in S and t i , meteorological similarity q i , calculate repair value, calculation method is: Ensure that the repaired data maintains time sequence coherence.

[0021] As preferred, the step S3 introduces adaptive double-gated mechanism in model construction process to realize dynamic generation of gating parameters and update cell state combined with auxiliary gating, which is specifically realized as:

[0022] First, select multilayer perceptron as main gating parameter generation network, input data at current time, its output includes weight matrix of forgetting gate, input gate and output gate and bias vector

[0023] For forgetting gate, dynamically generated main gating weight matrix and bias vector calculate output f of forgetting gate t , calculation method is: Output i of input gate t is also calculated by dynamically generated weight matrix and bias vector, calculation method is: Output gate is the same;

[0024] Then calculate auxiliary gating, input calculation formula of forgetting gate is wherein, fixed weight matrix and bias vector of auxiliary forgetting gate, m t represent meteorological data; auxiliary input gate is calculated in the same way wherein represent fixed weight matrix and bias vector of auxiliary input gate;

[0025] Combine output of main gating and auxiliary gating to update cell state, specific formula is: wherein represent candidate cell state;

[0026] Then update hidden state of output gate with updated cell state, calculation method is: h t =o t• tanh(C t ), output gate o t Decides how much information in the cell state will be outputted into the hidden state, which will be one of the inputs for the next time step, continuing to participate in the calculation of the model.

[0027] As preferred, the specific determination step of the power generation correction coefficient in the step S4 is:

[0028] S41, first calculate the power generation efficiency after the inverter aging, the specific calculation method is: Wherein, η0 represents the initial power generation efficiency of the inverter, t run represents the running time of the inverter, α age is the aging influence coefficient, which is a parameter obtained by regression analysis on historical aging data and power generation data;

[0029] S42, based on the power generation efficiency η age after aging, calculate the aging correction coefficient k age , the calculation method is:

[0030] S43, analyze the relationship between the collected inverter temperature data and the power generation, and obtain the mapping between the power generation efficiency and the temperature, which is: Wherein, η temp represents the power generation efficiency based on temperature influence, β low is the decline coefficient of the power generation efficiency linearly decreasing with the temperature, β high represents the decline coefficient of the power generation efficiency linearly decreasing with the temperature, β low , β high , T min , T max and η temp are parameters obtained by regression analysis on historical temperature and power generation data;

[0031] S44, based on the power generation efficiency influenced by temperature, calculate the temperature correction coefficient, the calculation method is:

[0032] S45, weight and fuse the aging correction coefficient and the temperature correction coefficient to obtain the final power generation correction coefficient.

[0033] As preferred, the specific implementation step of the step S5 is:

[0034] S51, first preliminary correction, multiply the first power generation prediction data by the calculated power generation correction coefficient to obtain the preliminary corrected power generation prediction data;

[0035] S52, a feature vector containing multiple influencing factors is constructed, including inverter aging features, temperature features, weather data features, and hidden state features output by the multi-scale adaptive gated LSTM model;

[0036] S53, the preliminary corrected power generation prediction data is secondarily corrected by support vector machine regression, the actual power generation in the historical data, the preliminary corrected power generation prediction data and the constructed feature vector are used for training, and the parameters of the support vector machine regression are trained by minimizing the prediction error;

[0037] S54, finally, the feature vector of the current time is input into the trained support vector machine regression to obtain a secondary correction value, and the final power generation prediction data is obtained by adding the secondary correction value to the preliminary corrected power generation prediction data.

[0038] Compared with the prior art, the advantages and positive effects of the present application are:

[0039] 1. Through dynamic reasonable interval detection, bidirectional LSTM abnormal data recognition and weather similarity repair, the data quality is improved and the interference of noise on the model is reduced.

[0040] 2. Independent LSTM networks at the hour, day and week levels are constructed, multi-scale time sequence features are fused through the attention mechanism, and key time scales are dynamically focused; an adaptive double-gated mechanism is introduced, gate parameters are dynamically generated based on a multilayer perceptron, and the cell state is updated in combination with an auxiliary gate, thereby enhancing the adaptability of the model to complex time sequences such as weather mutations and solving the problem of insufficient generalization ability of traditional single-scale models.

[0041] 3. The mapping relationship between inverter aging and temperature and power generation efficiency is established through regression analysis, a correction coefficient is calculated for preliminary correction, and support vector machine regression is used in combination with multi-dimensional features for secondary correction, thereby making up for the influence of unmodeled factors, reducing long-term prediction bias, and improving prediction accuracy and reliability in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0043] Figure 1 It is a structure flow chart of a photovoltaic power station power generation prediction method based on artificial intelligence. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described below in conjunction with the accompanying drawings and examples. It should be noted that the examples of the present application and the features in the examples can be combined with each other without conflict.

[0045] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and therefore the present application is not limited to the specific examples disclosed in the following description.

[0046] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and therefore the present application is not limited to the specific examples disclosed in the following description. Figure 1

[0047] First, the data is collected, the historical power generation data of the photovoltaic power station inverter, the weather data, the inverter aging data and the inverter temperature data are collected. The historical power generation data and the weather data of the inverter collected are combined to obtain the inverter power generation data with weather data factors, and the data is preprocessed. Considering that the existing method is rough in processing abnormal data and does not combine weather correlation repair, resulting in distortion of model training data and prediction accuracy affected by noise interference. The present application improves the data quality through dynamic interval detection, bidirectional LSTM abnormality identification and weather similarity repair. Specifically, the data preprocessing includes data cleaning and repair. First, the dynamic reasonable interval of the inverter power generation at each time is calculated through the probability distribution of the historical inverter power generation data with weather data factors at the same period, the lower limit is L(t) = μ t-1 ×(1-α1×σ t-1 / μ t-1 ), the upper limit is U(t) = μ t-1 ×(1+α2×σ t-1 / μ t-1 ), wherein σ t-1 , μ t-1 ​The standard deviation and mean value of the historical data of the same period are alpha1 and alpha2, which are adjustment coefficients. The power generation data exceeding the interval is determined as out-of-limit data and is removed. A bidirectional LSTM network is used to input the inverter power generation data sequence with meteorological data factors, train the mapping relationship under normal working conditions, output the predicted power generation at each time, calculate the relative error between the actual value and the predicted value, and when the relative error is greater than a set threshold, determine that it is abnormal data and remove it. After removal, record the position and context features. For the time t that needs to be repaired, search for the N normal time periods closest to t in the historical data to form a set S. According to the time interval Δt i , the meteorological similarity q i , the repair value is calculated, and the calculation method is: Ensure that the repaired data maintains time sequence coherence.

[0048] Then considering that the traditional single-scale model cannot capture the hourly fluctuations, daily cycles and weekly trends at the same time, and the gating mechanism is fixed, it is difficult to adapt to meteorological mutations. The present application realizes multi-granularity feature fusion and adaptive time series modeling through a multi-scale network architecture and a dynamic gating mechanism. First, the preprocessed data is divided into three types of time series feature vectors according to the time granularity, namely hourly, daily and weekly, and is input into independently constructed LSTM networks to model the time series features of each scale and output the corresponding hidden state sequence. Then, the attention mechanism is used to calculate the weights of the hidden states of each scale, and the multi-scale features are fused according to the weights to obtain the comprehensive hidden state, which dynamically focuses on the key time scale features. In the model construction, an adaptive double-gating mechanism is introduced, and a multilayer perceptron is used as the main gating parameter generation network to input the current time data to dynamically generate the weight matrix and bias vector of the forget gate, input gate and output gate. At the same time, the auxiliary gate is calculated, and the main gate and auxiliary gate are combined to output and update the cell state, and the hidden state is updated accordingly. After the model is constructed, the preprocessed data is input to output the first power generation prediction data. Specifically, first, the preprocessed data is divided into multi-scale input sequences according to the time granularity, including at least three types of time series feature vectors, namely hourly, daily and weekly. Independent hourly, daily and weekly LSTM networks are constructed to model the time series features of each scale and output the hidden state sequence of the corresponding scale The weights of the hidden states of each scale are calculated through the attention mechanism The calculation method is Wherein, fj=1, 2, 3 correspond to hourly, daily and weekly, W fj , b fj are learnable parameters, and the comprehensive hidden state is obtained by fusing multi-scale features And in the process of model construction, the adaptive double-gating mechanism is introduced to dynamically generate gating parameters and update the cell state combined with auxiliary gating. The adaptive double-gating mechanism has significant advantages in dynamically generating gating parameters and updating the cell state combined with auxiliary gating. The traditional LSTM has fixed gating parameters, which are difficult to adapt to complex changes. However, this mechanism uses a multi-layer perceptron as the main gating parameter generation network, inputs the current time data, dynamically generates the weight matrix and bias vector of the forget gate, input gate, and output gate, and can adjust the gating parameters in real time to accurately capture data changes, such as quickly adjusting the state when the irradiance changes suddenly, and improving the response capability to high-frequency fluctuations. At the same time, the auxiliary gating is calculated, the meteorological data and other factors are input into the forget gate and auxiliary input gate, and the main gating and auxiliary gating outputs are combined to update the cell state, so that the model can integrate historical sequence information and external influencing factors when processing time series data, enhance the adaptability to multi-scale features, and improve the processing capability and prediction accuracy of the model to complex scenarios. Specifically, first, a multi-layer perceptron is selected as the main gating parameter generation network, and the current time data is input. The output includes the weight matrix of the forget gate, input gate, and output gate and bias vector For the forget gate, the dynamically generated main gating weight matrix and bias vector calculate the output f t of the forget gate, and the calculation method is: The output i t of the input gate is also calculated by the dynamically generated weight matrix and bias vector, and the calculation method is: The output gate is the same; then the auxiliary gating calculation is performed, and the calculation formula of the forget gate is wherein The fixed weight matrix and bias vector of the auxiliary forget gate, m t represent meteorological data; the auxiliary input gate is calculated in the same way wherein represent the fixed weight matrix and bias vector of the auxiliary input gate; the outputs of the main gating and auxiliary gating are combined to update the cell state, and the specific formula is: wherein represents the candidate cell state; then the updated cell state and output gate update the hidden state, and the calculation method is: h t = o t ·tanh(C t ), the output gate o t decides how much information in the cell state will be output to the hidden state, and the hidden state will be one of the inputs to the next time step to continue participating in the model calculation. The preprocessed multi-scale feature vector is input into the constructed model, processed by the multi-scale network and dynamic gating mechanism, the cell and hidden state are updated, and the first power generation prediction data is output.

[0049] 7. After obtaining the first power generation prediction data, considering that the existing method ignores the nonlinear impact of inverter aging and temperature on power generation efficiency, resulting in long-term prediction deviation accumulating with equipment running time. The present application uses the collected inverter aging data and inverter temperature data to calculate the power generation correction coefficient. Specifically, the specific determination steps of the power generation correction coefficient are as follows: first, calculate the power generation efficiency after inverter aging, and the specific calculation method is as follows: Where η0 represents the initial power generation efficiency of the inverter, t run represents the running time of the inverter, and α age is the aging impact coefficient, which is a parameter obtained by regression analysis of historical aging data and power generation data; based on the power generation efficiency η age after aging, the aging correction coefficient k age is calculated, and the calculation method is as follows: The relationship between the collected inverter temperature data and the power generation is analyzed to obtain the mapping between the power generation efficiency and the temperature, which is as follows: Where η temp represents the power generation efficiency based on temperature influence, β low is the decline coefficient of the power generation efficiency linearly decreasing with temperature, β high represents the decline coefficient of the power generation efficiency linearly decreasing with temperature, β low , β high , T min , T max and η temp are parameters obtained by regression analysis of historical temperature and power generation data; based on the power generation efficiency influenced by temperature, the temperature correction coefficient is calculated, and the calculation method is as follows: The aging correction coefficient and the temperature correction coefficient are weighted and fused to obtain the final power generation correction coefficient.

[0050] After obtaining the power generation correction coefficient, the first power generation prediction data is multiplied by the calculated power generation correction coefficient to obtain the preliminary corrected power generation prediction data. Then, considering that the preliminary corrected prediction value is still affected by unmodeled factors, further fine calibration needs to be combined with multi-dimensional features. In this scheme, the support vector machine (SVM) regression is used to realize secondary correction. Specifically, a feature vector containing various influencing factors is constructed, including inverter aging features, temperature features, weather data features, and hidden state features output by a multi-scale adaptive gated LSTM model; the support vector machine regression is used to perform secondary correction on the preliminary corrected power generation prediction data, the actual power generation in the historical data, the preliminary corrected power generation prediction data, and the constructed feature vector are used for training, the parameters of the support vector machine regression are trained by minimizing the prediction error; finally, the feature vector of the current time is input into the trained support vector machine regression to obtain the secondary correction value, and the preliminary corrected power generation prediction data is added to the obtained secondary correction value to obtain the final power generation prediction data.

[0051] The above is only a preferred embodiment of the present application, and does not limit other forms of the present application. Any skilled person in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made on the basis of the technical essence of the present application to the above embodiments shall still fall within the protection scope of the present application.

Claims

1. A method for predicting the power generation of a photovoltaic power plant based on artificial intelligence, characterized in that, Includes the following steps: S1. Collect historical power generation data, meteorological data, inverter aging data, and inverter temperature data of the photovoltaic power station inverter; S2. Combine the collected historical power generation data of the inverter with meteorological data to obtain inverter power generation data with meteorological data factors, and perform data preprocessing on this data. S3. Build a power generation prediction model by inputting the preprocessed inverter power generation data with meteorological data factors to obtain the first power generation prediction data. The power generation prediction model in step S3 is constructed using a multi-scale adaptive gated LSTM, specifically: S31. First, the preprocessed data is divided into multi-scale input sequences according to the time granularity, including at least three types of time-series feature vectors: hourly, daily, and weekly. S32. Construct independent hourly, daily, and weekly LSTM networks to model the temporal dependencies of temporal features at each scale, and output the hidden state sequences at the corresponding scales. ; S33. Calculate the weights of hidden states at each scale using an attention mechanism. The calculation method is as follows Where fj=1,2,3 correspond to hourly, daily, and weekly timeframes, respectively. The hidden state is obtained by fusing multi-scale features to obtain learnable parameters. ; S4. Calculate the power generation correction coefficient based on the collected inverter aging data and inverter temperature data; S5. Combining the power generation correction coefficient and the first power generation prediction data, the final power generation prediction data is obtained; In step S3, an adaptive dual-gating mechanism is introduced during the model construction process to dynamically generate gating parameters and combine them with auxiliary gating to update the cell state. Specifically, this is implemented as follows: First, a multilayer perceptron is selected as the main gating parameter generation network. The current data is input, and its output includes the weight matrices of the forget gate, input gate, and output gate. , , and bias vector , , ; For the forget gate, the output of the forget gate is calculated using the dynamically generated master gate weight matrix and bias vector. The calculation method is as follows: ; Output of the input gate It is also calculated using dynamically generated weight matrices and bias vectors, and the calculation method is as follows: The same applies to the output gate; Then, the auxiliary gating calculation is performed, with the forgetting gate as the input. The calculation formula is ,in, The fixed weight matrix and bias vector of the auxiliary forgetting gate. Represents meteorological data; similarly, it calculates auxiliary input gates. ,in The fixed weight matrix and bias vector represent the auxiliary input gate; The cell state is updated by combining the outputs of the main gating and the auxiliary gating, using the following formula: ,in Represents the state of candidate cells; Then, the updated cell state and output gate update the hidden state, calculated as follows: Output gate The amount of information in the cell state is determined and output to the hidden state, which then serves as one of the inputs to the model for the next time step.

2. The method for predicting photovoltaic power generation based on artificial intelligence according to claim 1, characterized in that, In step S1, the meteorological data is determined based on the latitude and longitude coordinates of each inverter's geographical location.

3. The method for predicting photovoltaic power generation based on artificial intelligence according to claim 1, characterized in that, The data preprocessing in step S2 includes data cleaning and repair, and the specific implementation steps are as follows: S21. First, using the probability distribution of inverter power generation data with meteorological data factors from the same historical period, calculate the dynamic reasonable range of inverter power generation at each moment, with the lower limit being... Upper limit ,in, The standard deviation and mean of the data for the same historical period. As an adjustment factor, power generation data exceeding the range is judged as out-of-limit data and discarded; S22. Using a bidirectional LSTM network, the inverter power generation data sequence with meteorological data factors is input, the mapping relationship under normal operating conditions is trained, the predicted power generation is output at each time, the relative error between the actual value and the predicted value is calculated, and when the relative error is greater than the set threshold, it is judged as abnormal data and removed. After removal, the location and context features are recorded. S23. For the time t that needs to be repaired, retrieve the N normal time periods in historical data that are closest to t, forming a set S; based on the time interval between each time period in S and t... Weather similarity Calculate the repair value, the calculation method is as follows: This ensures that the repaired data maintains temporal continuity.

4. The method for predicting photovoltaic power generation based on artificial intelligence according to claim 1, characterized in that, The specific steps for determining the power generation correction coefficient in step S4 are as follows: S41. First, calculate the power generation efficiency of the inverter after aging. The specific calculation method is as follows: ,in, This represents the initial power generation efficiency of the inverter. This represents the inverter's operating time. This represents the aging impact coefficient, a parameter obtained through regression analysis of historical aging data and power generation data; S42, Power generation efficiency based on aging Calculate the aging correction factor The calculation method is as follows: ; S43. Analyze the relationship between the collected inverter temperature data and power generation to obtain the mapping between power generation efficiency and temperature, specifically: ,in, Represents power generation efficiency based on the effect of temperature. The coefficient representing the linear decrease in power generation efficiency with decreasing temperature. The coefficient representing the linear decrease in power generation efficiency with increasing temperature. , , , and These parameters were obtained through regression analysis of historical temperature and power generation data. S44. Based on the effect of temperature on power generation efficiency, calculate the temperature correction factor. The calculation method is as follows: ; S45. The aging correction coefficient and the temperature correction coefficient are weighted and fused to obtain the final power generation correction coefficient.

5. The method for predicting photovoltaic power generation based on artificial intelligence according to claim 1, characterized in that, The specific implementation steps of step S5 are as follows: S51. First, make a preliminary correction by multiplying the first power generation forecast data by the calculated power generation correction coefficient to obtain the preliminary corrected power generation forecast data. S52. Construct a feature vector containing multiple influencing factors, including inverter aging features, temperature features, meteorological data features, and hidden state features output by the multi-scale adaptive gated LSTM model. S53. The power generation prediction data after the initial correction is corrected by using support vector machine regression. The actual power generation in the historical data, the power generation prediction data after the initial correction, and the constructed feature vector are used for training. The parameters of the support vector machine regression are trained by minimizing the prediction error. S54. Finally, the feature vector at the current moment is input into the trained support vector machine regression to obtain the secondary correction value. The power generation prediction data after the initial correction is added to the obtained secondary correction value to obtain the final power generation prediction data.

Images