Generation power prediction method of photovoltaic power plant

Abstract

The invention relates to the technical field of photovoltaic power generation prediction, and discloses a photovoltaic power plant power generation power prediction method comprising the following steps: obtaining historical photovoltaic output data and corresponding historical meteorological data; performing multi-dimensional feature extraction on the historical meteorological data to generate first-level meteorological features, verifying the first-level meteorological features, calculating the contribution degree of each first-level meteorological feature to photovoltaic power generation, and screening to generate second-level meteorological features; constructing a photovoltaic power generation power prediction model based on the secondary meteorological characteristics and historical photovoltaic output data; and obtaining meteorological data of the target prediction time period, and generating a generation power prediction result of the photovoltaic power plant through the photovoltaic generation power prediction model. In combination with the multi-dimensional analysis and deep learning technology of meteorological data, the accuracy and real-time performance of power generation prediction of the photovoltaic power plant are improved by dynamically adjusting model parameters.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic power generation prediction, and in particular to a method for predicting power generation of a photovoltaic power plant. Background Art

[0002] At present, photovoltaic power generation, as a renewable energy technology, has been rapidly developed and widely used in recent years. With the growth of global demand for clean energy, photovoltaic power generation has become an important force in promoting energy transformation due to its advantages such as modularity, easy deployment and environmental friendliness.

[0003] However, the output power of photovoltaic power generation is affected by many factors, such as weather conditions, seasonal changes, geographical location, etc., which makes the prediction of photovoltaic power generation a technical challenge. Accurate power prediction is crucial for the optimized operation of power stations, the stable operation of power grids, and improving energy utilization. Summary of the invention

[0004] The purpose of the present invention is to provide an accurate photovoltaic power generation prediction method, which combines multi-dimensional analysis of meteorological data with deep learning technology, and improves the accuracy and real-time performance of photovoltaic power plant power generation prediction by dynamically adjusting model parameters.

[0005] In order to achieve the above object, the present invention provides a method for predicting the power generation of a photovoltaic power plant, comprising:

[0006] Obtain historical photovoltaic output data and corresponding historical meteorological data;

[0007] Perform multi-dimensional feature extraction on historical meteorological data to generate primary meteorological features, verify the primary meteorological features, calculate the contribution of each primary meteorological feature to photovoltaic power generation, and screen and generate secondary meteorological features;

[0008] Based on the secondary meteorological characteristics and historical photovoltaic output data, a photovoltaic power generation prediction model is constructed;

[0009] Obtain meteorological data for the target forecast period and generate a forecast result of the photovoltaic power plant's power generation through a photovoltaic power generation forecast model.

[0010] In some embodiments of the present invention, multi-dimensional feature extraction is performed on historical meteorological data to generate primary meteorological features, including:

[0011] Classify the historical meteorological data to generate classified historical meteorological data X', X' = [SR, TM, HU, W, C, P];

[0012] Among them, SR=[sr1,sr2…sr i …sr n ],TM=[tm1,tm2…tmi …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ];

[0013] Among them, SR represents historical solar radiation data, TM represents historical temperature data, HU represents historical humidity data, W represents historical wind speed data, C represents historical cloud cover data, and P represents historical precipitation data; sr i represents the i-th historical solar radiation data, tm i represents the i-th historical temperature data, hu i represents the i-th historical humidity data, w i represents the i-th historical wind speed radiation data, w i represents the ith historical cloud radiation data, p i represents the i-th historical precipitation and radiation data, and n represents the number of historical meteorological data obtained for the last time;

[0014] Construct the covariance matrix Σ[Y ij ];

[0015] Among them, Y ij = Cov(X' i ,X' j ),Y ij Represents the element in the i-th row and j-th column of the covariance matrix;

[0016] Calculate the eigenvalues ​​λ1, λ2, λ3, λ4, λ5, λ6 of the covariance matrix;

[0017] Calculate the cumulative variance contribution

[0018] Calculate the value of i when k ≥ 0.9 and generate the first-level meteorological characteristics.

[0019] In some embodiments of the present invention, the analyzing the relationship between meteorological data and photovoltaic output based on the secondary meteorological characteristics and historical photovoltaic output data includes:

[0020] Calculate the linear correlation between photovoltaic power generation and various meteorological factors;

[0021] Construct a multiple perceptron model of meteorological factors affecting photovoltaic output, train the multiple perceptron model, and capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data;

[0022] Based on historical meteorological data and real-time PV output data, the time offset b of meteorological factors on PV output is determined.

[0023] In some embodiments of the present invention, the calculating of the linear correlation between photovoltaic power generation and various meteorological factors includes:

[0024] Based on the classified historical meteorological data X', the historical photovoltaic power generation data PW at the same time is obtained, PW = [pw1, pw2...pw n ], the correlation analysis of different types of historical meteorological data with photovoltaic power generation data was carried out respectively, and the correlation coefficients between historical meteorological data and photovoltaic power generation were obtained respectively:

[0025] The value of the correlation coefficient lies between -1 and 1.

[0026] In some embodiments of the present invention, the construction of a multiple sensor model of meteorological factors affecting photovoltaic output includes:

[0027] Input layer, hidden layer and output layer

[0028] The input layer includes:

[0029] 6 neurons, corresponding to the input SR = [sr1, sr2…sr i …sr n ],TM=[tm1,tm2…tm i …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ];

[0030] The hidden layer includes:

[0031] Two sub-hidden layers, each with multiple neurons, each neuron is fully connected to all neurons in the next layer, forming a fully connected network;

[0032] The output layer includes 1 neuron and outputs photovoltaic power output data PW.

[0033] In some embodiments of the present invention, the training of multiple perceptron models to capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data includes:

[0034] Use historical data to train a multiple perception machine model of meteorological factors on photovoltaic output, compare the network's predicted output with the actual target output, use mean square error as the loss function, and calculate the mean square error (MSE);

[0035]

[0036] in, PW represents the output data predicted by the multiple perceptron models at the i-th moment. i Indicates the actual photovoltaic power generation output data at the i-th moment;

[0037] The weights and bias of the network are updated according to the value of the mean square error, and the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data is output.

[0038] In some embodiments of the present invention, determining the time offset b of the meteorological factor on the photovoltaic output based on the historical meteorological data and the real-time photovoltaic output data includes:

[0039] Obtain historical meteorological data X' and photovoltaic output data PW from the last s minutes to the current data acquisition time s ,

[0040] The historical meteorological data is divided based on the real-time photovoltaic output data, wherein the time length of the collected real-time photovoltaic processing data is m times the corresponding time length of the historical meteorological data after the division, where m is a positive integer, to generate historical meteorological data at different time stages;

[0041] By comparing the historical meteorological data at different time stages with the photovoltaic output data PW from the last s minutes to the current data acquisition time s , generate a time offset b, and adjust the meteorological data observation period for PV output prediction based on the time offset b.

[0042] In some embodiments of the present invention, the photovoltaic power output data PW from the last s minutes to the current data acquisition time is compared with the historical meteorological data of different time periods. s , generate time offset b, including:

[0043] Set the time window length to m, and calculate the covariance between the historical meteorological data and the real-time photovoltaic output data in different time windows;

[0044] Calculate the standard deviation of historical meteorological data and real-time photovoltaic output data in different time windows respectively;

[0045] The sliding correlation coefficient is calculated based on the covariance of historical meteorological data and real-time photovoltaic output data in different time windows, the standard deviation of historical meteorological data, and the standard deviation of real-time photovoltaic output data;

[0046] Compare the sliding correlation coefficients of different time windows;

[0047] Subtract the sliding correlation coefficients of all time windows from 1, take the absolute value of the result, select the sliding correlation coefficient with the smallest absolute value, and determine the time window corresponding to the selected sliding correlation coefficient as the time offset b.

[0048] In some embodiments of the present invention, the step of constructing a photovoltaic power generation prediction model includes:

[0049] A preliminary model for photovoltaic power generation prediction is established based on the linear and nonlinear correlation between the historical meteorological data and the historical photovoltaic output data.

[0050] The parameters of the preliminary model for photovoltaic power generation prediction are modified based on the time bias b;

[0051] Outputting a photovoltaic power generation power prediction correction model according to the correction result;

[0052] The photovoltaic power generation power prediction correction model is trained to generate a photovoltaic power generation power prediction model.

[0053] Compared with the prior art, the method for predicting power generation of a photovoltaic power plant according to the embodiment of the present invention has the following beneficial effects:

[0054] The first-level meteorological characteristics were determined by the number of eigenvalues ​​k when the cumulative variance contribution rate reached or exceeded 90%, ensuring the balance between the interpretability and accuracy of the model, screening out meteorological characteristics that made significant contributions to photovoltaic power generation, and improving the pertinence and efficiency of model prediction.

[0055] The selected meteorological features are verified to ensure their reliable contribution to the PV power prediction, and a cumulative contribution rate threshold of 90% is set to ensure that the model only includes features that have a significant impact on PV power, avoiding the interference of irrelevant variables.

[0056] By analyzing the Pearson correlation coefficients corresponding to different meteorological factors, the meteorological factors that have a significant impact on photovoltaic power generation can be identified. The identified strongly correlated meteorological factors can be used as key inputs for model construction, which helps to build a more accurate photovoltaic power generation prediction model.

[0057] The constructed multi-layer perceptron model can effectively capture the complex relationship between meteorological factors and photovoltaic output through its nonlinear processing capabilities and deep learning architecture, thereby improving the accuracy and stability of photovoltaic output prediction.

[0058] By determining the time offset b, the prediction model can more accurately adjust the period of use of meteorological data, significantly improving the accuracy of photovoltaic output prediction. Based on the dynamic adjustment of the sliding correlation coefficient, the model can flexibly adapt to the changes in the relationship between meteorology and photovoltaic output under different environmental conditions, enhancing the adaptability and reliability of the prediction.

[0059] By building a photovoltaic power prediction model based on historical data training and combined with time bias optimization, and ensuring the continued effectiveness and prediction accuracy of the model through regular verification and updating, solid technical support is provided for the efficient operation of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of a method for predicting power generation of a photovoltaic power plant provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0062] In the description of the present application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0063] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0064] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0065] Embodiment 1:

[0066] like Figure 1 As shown, the embodiment of the present invention discloses obtaining historical photovoltaic output data and corresponding historical meteorological data;

[0067] Perform multi-dimensional feature extraction on historical meteorological data to generate primary meteorological features, verify the primary meteorological features, calculate the contribution of each primary meteorological feature to photovoltaic power generation, and screen and generate secondary meteorological features;

[0068] Based on the secondary meteorological characteristics and historical photovoltaic output data, a photovoltaic power generation prediction model is constructed;

[0069] Obtain meteorological data for the target forecast period and generate a forecast result of the photovoltaic power plant's power generation through a photovoltaic power generation forecast model.

[0070] Embodiment 2:

[0071] Perform multi-dimensional feature extraction on historical meteorological data to generate primary meteorological features, including:

[0072] Classify the historical meteorological data to generate classified historical meteorological data X', X' = [SR, TM, HU, W, C, P];

[0073] Among them, SR=[sr1,sr2…sr i …sr n ],TM=[tm1,tm2…tm i …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ];

[0074] Among them, SR represents historical solar radiation data, TM represents historical temperature data, HU represents historical humidity data, W represents historical wind speed data, C represents historical cloud cover data, and P represents historical precipitation data; sr i represents the i-th historical solar radiation data, tm i represents the i-th historical temperature data, hu i represents the i-th historical humidity data, w i represents the i-th historical wind speed radiation data, w i represents the ith historical cloud radiation data, p irepresents the i-th historical precipitation and radiation data, and n represents the number of historical meteorological data obtained for the last time;

[0075] Construct the covariance matrix Σ[Y ij ];

[0076] Among them, Y ij = Cov(X' i ,X' j ),Y ij Represents the element in the i-th row and j-th column of the covariance matrix;

[0077] Calculate the eigenvalues ​​λ1, λ2, λ3, λ4, λ5, λ6 of the covariance matrix;

[0078] Calculate the cumulative variance contribution

[0079] Calculate the value of i when k ≥ 0.9 and generate the first-level meteorological characteristics.

[0080] In this embodiment

[0081] The evidence optimization combination theory is used to verify the main meteorological characteristics and generate the contribution of the main meteorological characteristics to photovoltaic power generation. The secondary meteorological characteristics are generated by screening the meteorological characteristics with a cumulative contribution rate of more than 90%, including:

[0082] A basic probability assignment (BPA) function is constructed to assign a confidence level to each hypothesis (feature). The confidence and likelihood calculation methods in the Dempster-Shafer theory are used to generate the contribution of each feature to photovoltaic power generation.

[0083] In this embodiment, feature extraction and screening: By classifying historical meteorological data and constructing a covariance matrix to extract primary meteorological features, the contribution of each meteorological factor to photovoltaic power generation can be quantitatively evaluated. This method helps to identify the most influential meteorological parameters, thereby establishing a more accurate relationship in the model.

[0084] In this embodiment, by classifying historical meteorological data into six categories, namely solar radiation (SR), temperature (TM), humidity (HU), wind speed (W), cloud cover (C), and precipitation (P), a feature matrix X' is formed, which effectively organizes and prepares the data and provides a clear structure for subsequent analysis.

[0085] Covariance Matrix and Eigenvalue Analysis

[0086] The calculated covariance matrix Σ reveals the statistical relationship between different meteorological factors and is a key step in quantifying the correlation between features. The eigenvalues ​​reveal the principal components of the covariance matrix, that is, the main fluctuation direction of the data, which is crucial for understanding the data structure and information compression.

[0087] Cumulative variance contribution rate and feature screening

[0088] The first-level meteorological characteristics were determined by the number of eigenvalues ​​k when the cumulative variance contribution rate reached or exceeded 90%, ensuring the balance between the interpretability and accuracy of the model, screening out meteorological characteristics that made significant contributions to photovoltaic power generation, and improving the pertinence and efficiency of model prediction.

[0089] Meteorological characteristics verification and contribution quantification

[0090] Model verification: The selected meteorological features are verified to ensure their contribution to the PV power prediction. By setting a cumulative contribution rate threshold of 90%, it is ensured that the model only contains features that have a significant impact on PV power, avoiding the interference of irrelevant variables.

[0091] Embodiment 3:

[0092] The analysis of the relationship between meteorological data and photovoltaic output based on the secondary meteorological characteristics and historical photovoltaic output data includes:

[0093] Calculate the linear correlation between photovoltaic power generation and various meteorological factors;

[0094] Construct a multiple perceptron model of meteorological factors affecting photovoltaic output, train the multiple perceptron model, and capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data;

[0095] Based on historical meteorological data and real-time PV output data, the time offset b of meteorological factors on PV output is determined.

[0096] In this embodiment, through linear correlation analysis, nonlinear relationship construction and model training, and time offset determination, this technology can significantly improve the accuracy and stability of photovoltaic power generation prediction. This not only provides strong technical support for the operation and management of photovoltaic power stations, but also optimizes grid scheduling and improves the operating efficiency and reliability of the overall energy system.

[0097] Embodiment 4:

[0098] The calculation of the linear correlation between photovoltaic power generation and various meteorological factors includes:

[0099] Based on the classified historical meteorological data X', the historical photovoltaic power generation data PW at the same time is obtained, PW = [pw1, pw2...pw n ], the correlation analysis of different types of historical meteorological data with photovoltaic power generation data was carried out respectively, and the correlation coefficients between historical meteorological data and photovoltaic power generation were obtained respectively:

[0100] The value of the correlation coefficient lies between -1 and 1.

[0101] Based on the classified historical meteorological data X', the historical photovoltaic power generation data PW at the same time is obtained, PW = [pw1, pw2...pw n ], the correlation analysis of different types of historical meteorological data with photovoltaic power generation data was carried out respectively, and the correlation coefficients between historical meteorological data and photovoltaic power generation were obtained respectively:

[0102] The value of the correlation coefficient ranges from -1 to 1, with values ​​close to 1 or -1 indicating a strong positive or negative correlation, and values ​​close to 0 indicating no linear correlation.

[0103] In this embodiment, based on the classified historical meteorological data X', the historical photovoltaic power generation data PW at the same time is obtained, PW = [pw1, pw2 ... pw n ], and conducted correlation analysis on different types of historical meteorological data and photovoltaic power generation data, among which:

[0104] r SR Indicates the degree of correlation between historical solar radiation data and photovoltaic power generation.

[0105]

[0106] r TM Indicates the degree of correlation between historical temperature data and photovoltaic power generation,

[0107]

[0108] r HU Indicates the degree of correlation between historical humidity data and photovoltaic power generation,

[0109]

[0110] r W Indicates the degree of correlation between historical wind speed data and photovoltaic power generation,

[0111]

[0112] r C Indicates the degree of correlation between historical cloud cover data and photovoltaic power generation,

[0113]

[0114] r P Indicates the degree of correlation between historical precipitation data and photovoltaic power generation,

[0115]

[0116] In this embodiment, the linear correlation between historical meteorological data and photovoltaic power generation is quantified by calculating the Pearson correlation coefficient. The value of each coefficient is between -1 and 1, which provides an intuitive metric for understanding the relationship between various meteorological factors and photovoltaic power generation.

[0117] In this embodiment, by analyzing the Pearson correlation coefficients corresponding to different meteorological factors, meteorological factors that have a significant impact on photovoltaic power generation can be identified. For example, solar radiation (SR) is usually positively correlated with photovoltaic power generation, while temperature (TM) within a certain range may also affect the efficiency of photovoltaic panels and thus affect power generation. This helps to determine which factors should be prioritized in modeling and prediction.

[0118] In this embodiment, the identified strongly correlated meteorological factors can be used as key inputs for model construction, which helps to build a more accurate photovoltaic power generation prediction model. By excluding factors with weak correlation, the complexity of the model is reduced, while the prediction efficiency and accuracy of the model are improved.

[0119] Embodiment 5:

[0120] The construction of a multiple sensor model of meteorological factors affecting photovoltaic output includes:

[0121] Input layer, hidden layer and output layer

[0122] The input layer includes:

[0123] 6 neurons, corresponding to the input SR = [sr1, sr2…sr i …sr n ],TM=[tm1,tm2…tm i …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ];

[0124] The hidden layer includes:

[0125] Two sub-hidden layers, each with multiple neurons, each neuron is fully connected to all neurons in the next layer, forming a fully connected network;

[0126] The output layer includes 1 neuron and outputs photovoltaic power output data PW.

[0127] In this embodiment, the input layer directly accepts six meteorological factors such as solar radiation (SR), temperature (TM), humidity (HU), wind speed (W), cloud cover (C), and precipitation (P) as input. Through the forward propagation of the neural network, these features are fused and interacted in the hidden layer to form a more abstract feature representation, laying the foundation for more accurate prediction.

[0128] In this embodiment, the output layer has only one neuron for directly predicting the photovoltaic output data (PW). This design enables the output of the model to directly correspond to the prediction target, simplifies the output structure, and facilitates the training and tuning of the model.

[0129] In this embodiment, the MLP model is trained by the back propagation algorithm, which can automatically adjust the weights and biases of each layer of neurons to minimize the prediction error. The use of the ReLU activation function not only speeds up the training speed, but also avoids the gradient vanishing problem, thereby improving the training efficiency and effect of the model.

[0130] In this embodiment, the combination of complex network structure and nonlinear activation function makes the model have stronger generalization ability. Even when processing unseen meteorological data, it can make accurate predictions based on the learned patterns, thus improving the practicality of the model.

[0131] Embodiment 6:

[0132] The multiple perceptron models are trained to capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data, including:

[0133] Use historical data to train a multiple perception machine model of meteorological factors on photovoltaic output, compare the network's predicted output with the actual target output, use mean square error as the loss function, and calculate the mean square error (MSE);

[0134]

[0135] in, PW represents the output data predicted by the multiple perceptron models at the i-th moment. i Indicates the actual photovoltaic power generation output data at the i-th moment;

[0136] The weights and bias of the network are updated according to the value of the mean square error, and the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data is output.

[0137] The multiple perceptron models are trained to capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data, including:

[0138] Use historical data to train a multiple perceptron model of the effect of meteorological factors on photovoltaic output to minimize the mean square error between the predicted output and the actual output:

[0139] Compare the predicted output of the network with the actual target output, use the mean square error as the loss function, and calculate the mean square error MSE;

[0140]

[0141] in, PW represents the output data predicted by the multiple perceptron models at the i-th moment. i Indicates the actual photovoltaic power generation output data at the i-th moment;

[0142] According to the value of the mean square error, the network weights and biases are gradient descended, and the error gradient is calculated and propagated back along the network; according to the gradient calculated by back propagation, the network weights and biases are updated, and iterative optimization is performed, and the training cycle is set to 800-1000 times.

[0143] In this embodiment, the mean square error (MSE) is used as the loss function, which can effectively measure the gap between the model prediction value and the actual value, especially when dealing with continuous numerical prediction problems.

[0144] In this embodiment, by calculating the gradient of the loss function with respect to the network weights and biases and updating the parameters in the opposite direction of the gradient, the gradient descent algorithm ensures that the model can gradually reduce the prediction error and achieve the optimization goal.

[0145] In this embodiment, a training cycle of 800-1000 times is set to provide sufficient training opportunities for the model, avoiding suboptimal solutions caused by stopping training too early, and also preventing overfitting, ensuring that the model has good generalization ability.

[0146] Embodiment 7:

[0147] The method of determining the time offset b of the meteorological factor on the photovoltaic output based on the historical meteorological data and the real-time photovoltaic output data includes:

[0148] Obtain historical meteorological data X' and photovoltaic output data PW from the last s minutes to the current data acquisition time s ,

[0149] The historical meteorological data is divided based on the real-time photovoltaic output data, wherein the time length of the collected real-time photovoltaic processing data is m times the corresponding time length of the historical meteorological data after the division, where m is a positive integer, to generate historical meteorological data at different time stages;

[0150] By comparing the historical meteorological data at different time stages with the photovoltaic output data PW from the last s minutes to the current data acquisition time s , generate a time offset b, and adjust the meteorological data observation period for PV output prediction based on the time offset b.

[0151] Obtain historical meteorological data X' and photovoltaic output data PW for a period of time before the current sampling time s ,

[0152] Among them, PW s The photovoltaic output data from the last s minutes to the current data acquisition time;

[0153] Based on the real-time photovoltaic output data, the historical meteorological data is divided to generate historical meteorological data of different time periods: X'1, X'2, X'3, X'4, X'5, X'6;

[0154] And with PW s Statistical comparison is performed, time offset b is calculated, and the meteorological data observation period for photovoltaic output prediction is adjusted based on time offset b.

[0155] In this embodiment, the determination of the time offset b of the meteorological factor on the photovoltaic output based on the historical meteorological data and the real-time photovoltaic output data includes:

[0156] Obtain historical meteorological data X' and real-time photovoltaic output data PW within 30 minutes n , based on the real-time photovoltaic output data, the historical meteorological data are divided to generate historical meteorological data of different time stages: X'1, X'2, X'3, X'4, X'5, X'6;

[0157] Among them, X'1 corresponds to the historical meteorological data of the previous 10 minutes, X'2 corresponds to the historical meteorological data of the previous 20-10 minutes, X'3 corresponds to the historical meteorological data of the previous 30-20 minutes, X'4 corresponds to the historical meteorological data of the previous 40-30 minutes, X'5 corresponds to the historical meteorological data of the previous 50-40 minutes, and X'6 corresponds to the historical meteorological data of the previous 60-50 minutes;

[0158] A statistical comparison is performed with the real-time data to calculate the time offset b, and the meteorological data observation period for photovoltaic output prediction is adjusted based on the time offset b.

[0159] Embodiment 8:

[0160] By comparing the historical meteorological data of different time periods with the photovoltaic output data PW from the last s minutes to the current data acquisition time s , generate time offset b, including:

[0161] Set the time window length to m, and calculate the covariance between the historical meteorological data and the real-time photovoltaic output data in different time windows;

[0162] Calculate the standard deviation of historical meteorological data and real-time photovoltaic output data in different time windows respectively;

[0163] The sliding correlation coefficient is calculated based on the covariance of historical meteorological data and real-time photovoltaic output data in different time windows, the standard deviation of historical meteorological data, and the standard deviation of real-time photovoltaic output data;

[0164] Compare the sliding correlation coefficients of different time windows;

[0165] Subtract the sliding correlation coefficients of all time windows from 1, take the absolute value of the result, select the sliding correlation coefficient with the smallest absolute value, and determine the time window corresponding to the selected sliding correlation coefficient as the time offset b.

[0166] The statistical comparison with the real-time data and calculation of the time offset b include:

[0167] Set the time window to 3, calculate the covariance of historical meteorological data and real-time photovoltaic output data in different time windows, and generate the covariance cov(X'1+X'2+X'3,PW n );

[0168] The covariance cov(X'2+X'3+X'4,PW n );

[0169] The covariance cov(X'3+X'4+X'5,PW n );

[0170] The covariance cov(X'4+X'5+X'6,PW n );

[0171] Measure the common changing trends of historical meteorological data and real-time photovoltaic output data;

[0172] Calculate the standard deviation of historical meteorological data in different time windows, including:

[0173] std(X'1+X'2+X'3) is the standard deviation of the historical meteorological data in the previous 30 minutes;

[0174] std(X'2+X'3+X'4) standard deviation of historical meteorological data from the previous 40 to the previous 10 minutes;

[0175] std(X'3+X'4+X'5) standard deviation of historical meteorological data from the previous 50 to the previous 20 minutes;

[0176] std(X'4+X'5+X'6) standard deviation of historical meteorological data from the previous 60 to the previous 30 minutes;

[0177] And the standard deviation of real-time photovoltaic output data std(PW n ), which measures the discreteness between historical meteorological data and real-time photovoltaic output data;

[0178] The sliding correlation coefficient is calculated based on the covariance of historical meteorological data and real-time photovoltaic output data in different time windows, the standard deviation of historical meteorological data, and the standard deviation of real-time photovoltaic output data:

[0179] The sliding correlation coefficient between the historical meteorological data and the real-time photovoltaic output data in the previous 30 minutes,

[0180] The sliding correlation coefficient between the historical meteorological data of the previous 40-10 minutes and the real-time photovoltaic output data,

[0181] The sliding correlation coefficient between the historical meteorological data of the previous 50-20 minutes and the real-time photovoltaic output data,

[0182] The sliding correlation coefficient between the historical meteorological data of the previous 50-20 minutes and the real-time photovoltaic output data,

[0183] Compare the sliding correlation coefficients of different time windows, select the sliding correlation coefficient closest to 1, and determine the time window corresponding to the current sliding correlation coefficient as the time offset b.

[0184] In this embodiment, the sliding window analysis can accurately identify the time period when meteorological factors have the most significant impact on photovoltaic output, providing key parameters for the prediction model.

[0185] The determination of the time offset b enables the prediction model to more accurately adjust the period of use of meteorological data, significantly improving the accuracy of photovoltaic output prediction.

[0186] Based on the dynamic adjustment of the sliding correlation coefficient, the model can flexibly adapt to the changes in the relationship between meteorology and photovoltaic output under different environmental conditions, enhancing the adaptability and reliability of the prediction.

[0187] Embodiment 9:

[0188] The photovoltaic power generation prediction model is constructed, comprising:

[0189] A preliminary model for photovoltaic power generation prediction is established based on the linear and nonlinear correlation between the historical meteorological data and the historical photovoltaic output data.

[0190] The parameters of the preliminary model for photovoltaic power generation prediction are modified based on the time bias b;

[0191] Outputting a photovoltaic power generation power prediction correction model according to the correction result;

[0192] The photovoltaic power generation power prediction correction model is trained to generate a photovoltaic power generation power prediction model.

[0193] The photovoltaic power generation prediction model is constructed, comprising:

[0194] Based on the linear correlation and nonlinear correlation between the acquired historical meteorological data and the historical photovoltaic output data, a preliminary photovoltaic power prediction model is established, and the model parameters are adjusted based on the time bias b. After training the preliminary photovoltaic power prediction model, a photovoltaic power prediction model is generated.

[0195] In this embodiment, linear and nonlinear relationship analysis

[0196] Using historical meteorological data and historical photovoltaic output data, linear and nonlinear relationships are determined through correlation analysis to provide a basis for model establishment. A preliminary model for photovoltaic power generation prediction is constructed, which comprehensively considers the complex relationship between meteorological factors and photovoltaic output. The time bias b is introduced to adjust the parameters of the preliminary model to ensure that the model accurately reflects the time delay of the impact of meteorological changes on photovoltaic output. The preliminary model is trained using historical data to optimize model parameters and improve prediction performance. After training and adjustment, the final photovoltaic power generation prediction model is generated.

[0197] Through historical data training, the model can effectively capture the linear and nonlinear relationship between meteorological factors and photovoltaic output, and improve the prediction accuracy. The time bias b is introduced to optimize the model parameters, so that the model can more accurately reflect the impact of meteorological changes on photovoltaic output.

[0198] The model is regularly verified and updated, including:

[0199] Input real-time meteorological data into the photovoltaic power prediction model and output the power prediction; regularly collect actual photovoltaic output data and compare it with the prediction data to evaluate the prediction accuracy of the model; based on the verification results, retrain the model or adjust the model parameters if necessary to maintain the model performance

[0200] In this embodiment, real-time meteorological data is input into a trained photovoltaic power prediction model to output a predicted power generation value; actual photovoltaic output data is collected regularly and compared with the predicted data to evaluate the model prediction accuracy; based on the verification results, if the model prediction accuracy decreases or there is a large error, the model is retrained or the model parameters are adjusted.

[0201] Regular verification ensures stable model performance and maintains high accuracy of model predictions through timely updates and optimization; the model can be dynamically adjusted according to actual operating conditions to improve adaptability to environmental changes and prediction flexibility.

[0202] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting power generation of a photovoltaic power plant, characterized in that: include: Obtain historical photovoltaic output data and corresponding historical meteorological data; Perform multi-dimensional feature extraction on historical meteorological data to generate primary meteorological features, verify the primary meteorological features, calculate the contribution of each primary meteorological feature to photovoltaic power generation, and screen and generate secondary meteorological features; Based on the secondary meteorological characteristics and historical photovoltaic output data, a photovoltaic power generation prediction model is constructed; Obtain meteorological data for the target forecast period and generate a forecast result of the photovoltaic power plant's power generation through a photovoltaic power generation forecast model.

2. The method for predicting power generation of a photovoltaic power plant according to claim 1, characterized in that: The multi-dimensional feature extraction of historical meteorological data generates primary meteorological features, including: Classify the historical meteorological data to generate classified historical meteorological data X', X' = [SR, TM, HU, W, C, P]; Among them, SR=[sr1,sr2…sr i …sr n ],TM=[tm1,tm2…tm i …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ]; Among them, SR represents historical solar radiation data, TM represents historical temperature data, HU represents historical humidity data, W represents historical wind speed data, C represents historical cloud cover data, and P represents historical precipitation data; sr i represents the i-th historical solar radiation data, tm i represents the i-th historical temperature data, hu i represents the i-th historical humidity data, w i represents the i-th historical wind speed radiation data, w i represents the ith historical cloud radiation data, p i represents the i-th historical precipitation and radiation data, and n represents the number of historical meteorological data obtained for the last time; Construct the covariance matrix Σ[Y ij ]; Among them, Y ij = Cov(X' i ,X' j ),Y ij Represents the element in the i-th row and j-th column of the covariance matrix; Calculate the eigenvalues ​​λ1, λ2, λ3, λ4, λ5, λ6 of the covariance matrix; Calculate the cumulative variance contribution Calculate the value of i when k ≥ 0.9 and generate the first-level meteorological characteristics.

3. The method for predicting power generation of a photovoltaic power plant according to claim 2, characterized in that: The analysis of the relationship between meteorological data and photovoltaic output based on the secondary meteorological characteristics and historical photovoltaic output data includes: Calculate the linear correlation between photovoltaic power generation and various meteorological factors; Construct a multiple perceptron model of meteorological factors affecting photovoltaic output, train the multiple perceptron model, and capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data; Based on historical meteorological data and real-time PV output data, the time offset b of meteorological factors on PV output is determined.

4. The method for predicting power generation of a photovoltaic power plant according to claim 3, characterized in that: The calculation of the linear correlation between photovoltaic power generation and various meteorological factors includes: Based on the classified historical meteorological data X', the historical photovoltaic power generation data PW at the same time is obtained, PW = [pw1, pw2...pw n ], the correlation analysis of different types of historical meteorological data with photovoltaic power generation data was carried out respectively, and the correlation coefficients between historical meteorological data and photovoltaic power generation were obtained respectively: The value of the correlation coefficient lies between -1 and 1.

5. The method for predicting power generation of a photovoltaic power plant according to claim 4, characterized in that: The construction of a multiple sensor model of meteorological factors affecting photovoltaic output includes: Input layer, hidden layer and output layer; The input layer includes: 6 neurons, corresponding to the input SR = [sr1, sr2…sr i …sr n ],TM=[tm1,tm2…tm i …tm n ],HU=[hu1,hu2…hu i …hu n ],W=[w1,w2…w i …w n ],C=[c1,c2…c i …c n ], P = [p1,p2…p i …p n ]; The hidden layer includes: Two sub-hidden layers, each with multiple neurons, each neuron is fully connected to all neurons in the next layer, forming a fully connected network; The output layer includes 1 neuron and outputs photovoltaic power output data PW.

6. The method for predicting power generation of a photovoltaic power plant according to claim 5, characterized in that: The multiple perceptron models are trained to capture the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data, including: Use historical data to train a multiple perception machine model of meteorological factors on photovoltaic output, compare the network's predicted output with the actual target output, use mean square error as the loss function, and calculate the mean square error (MSE); in, PW represents the output data predicted by the multiple perceptron models at the i-th moment. i Indicates the actual photovoltaic power generation output data at the i-th moment; The weights and bias of the network are updated according to the value of the mean square error, and the nonlinear relationship between different types of historical meteorological data and photovoltaic power generation data is output.

7. The method for predicting power generation of a photovoltaic power plant according to claim 6, characterized in that: The method of determining the time offset b of the meteorological factor on the photovoltaic output based on the historical meteorological data and the real-time photovoltaic output data includes: Obtain historical meteorological data X' and photovoltaic output data PW from the last s minutes to the current data acquisition time s , The historical meteorological data is divided based on the real-time photovoltaic output data, wherein the time length of the collected real-time photovoltaic processing data is m times the corresponding time length of the historical meteorological data after the division, where m is a positive integer, to generate historical meteorological data at different time stages; By comparing the historical meteorological data at different time stages with the photovoltaic output data PW from the last s minutes to the current data acquisition time s , generate a time offset b, and adjust the meteorological data observation period for PV output prediction based on the time offset b.

8. The method for predicting power generation of a photovoltaic power plant according to claim 7, characterized in that: By comparing the historical meteorological data of different time periods with the photovoltaic output data PW from the last s minutes to the current data acquisition time s , generate time offset b, including: Set the time window length to m, and calculate the covariance between the historical meteorological data and the real-time photovoltaic output data in different time windows; Calculate the standard deviation of historical meteorological data and real-time photovoltaic output data in different time windows respectively; The sliding correlation coefficient is calculated based on the covariance of historical meteorological data and real-time photovoltaic output data in different time windows, the standard deviation of historical meteorological data, and the standard deviation of real-time photovoltaic output data; Compare the sliding correlation coefficients of different time windows; Subtract the sliding correlation coefficients of all time windows from 1, take the absolute value of the result, select the sliding correlation coefficient with the smallest absolute value, and determine the time window corresponding to the selected sliding correlation coefficient as the time offset b.

9. The method for predicting power generation of a photovoltaic power plant according to claim 8, characterized in that: The photovoltaic power generation prediction model is constructed, comprising: A preliminary model for photovoltaic power generation prediction is established based on the linear and nonlinear correlation between the historical meteorological data and the historical photovoltaic output data. The parameters of the preliminary model for photovoltaic power generation prediction are modified based on the time bias b; Outputting a photovoltaic power generation power prediction correction model according to the correction result; The photovoltaic power generation power prediction correction model is trained to generate a photovoltaic power generation power prediction model.

Images