Photovoltaic output combined prediction method based on error reciprocal method

Abstract

The invention relates to the technical field of photovoltaic output prediction, discloses a photovoltaic output combined prediction method based on a reciprocal error method, and solves the problems of limitation and low prediction precision of photovoltaic output prediction by using a single algorithm in the prior art. According to the method, a photovoltaic output combined prediction model is constructed, the weight of the sub-models is adjusted according to the prediction error of the sub-models, the photovoltaic output is comprehensively predicted, the method is used for predicting the output of the photovoltaic system, the advantages of the sub-models can be integrated, the prediction model can better adapt to environmental changes, the prediction precision is effectively improved, and the prediction efficiency is improved. Therefore, stable operation of a power grid is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic output prediction, and in particular to a photovoltaic output combination prediction method based on an inverse error method. Background Art

[0002] In recent years, photovoltaic power generation and other new energy power generation methods have developed rapidly. Its clean and environmentally friendly advantages help optimize the energy structure system and alleviate the energy crisis. However, photovoltaic power generation is greatly affected by meteorological factors such as irradiance, which makes its output volatile and random, and has an increasingly significant impact on the stable operation of the power grid. Accurately predicting photovoltaic output will provide reliable information for power grid operation, help improve the stability of power grid operation and promote photovoltaic consumption.

[0003] Nowadays, various time series prediction algorithms have been used to predict photovoltaic output. However, the output prediction effects of a single algorithm in different environments often vary greatly. Therefore, using a single algorithm to predict photovoltaic output has limitations, and its prediction accuracy needs to be improved.

[0004] For example, CN113177652A discloses a photovoltaic output prediction method. First, a cloud-edge collaborative framework is constructed. Then, the cloud computing center uses a Bayesian recurrent neural network to establish an initial prediction module. Then, the edge computing device uses the prediction module to predict the photovoltaic output according to real-time data and uploads it. Finally, the cloud computing center adjusts the parameters according to the prediction deviation to update the prediction module. By introducing the Bayesian idea, the Monte Carlo dropout method is used for approximate inference. The network weights are learned by minimizing the KL divergence between the approximate distribution and the posterior distribution of the network weights. The squared Mahalanobis distance or local density ratio is used to calculate the prediction deviation and adjust the network weights. However, the impact of meteorological conditions on photovoltaic power generation is not taken into account. Therefore, it is necessary to consider realistic factors and improve the photovoltaic output prediction method. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the prediction of photovoltaic output is affected by meteorological conditions in the prior art. To solve the above problem, the present invention provides the following technical solutions: a photovoltaic output combination prediction method based on the inverse error method, comprising the following steps: S1. Collect historical output data of the photovoltaic system and meteorological data such as irradiance, establish a data set, and divide the training data and test data according to a certain ratio; S2. Based on the meteorological data in the data set, cluster the data set by using the self-organizing map algorithm; S3. Build a series of calculation models based on the training data to predict the photovoltaic output and calculate the root mean square error between the predicted results and the actual results; S4. Construct a combined prediction model for photovoltaic output. According to the constructed combined prediction model for photovoltaic output, the weight is calculated by the inverse sampling error method, and the photovoltaic output prediction result is calculated according to meteorological data such as the irradiance of the test data.

[0006] The present invention provides a photovoltaic output combined prediction method based on the inverse error method, wherein the combined prediction is to assign different weights to several prediction methods, thereby forming a combined prediction model with higher prediction accuracy. The present invention adopts the inverse error method to determine the weights of the sub-models and construct a photovoltaic output combined prediction model. Using this method to predict the output of the photovoltaic system can combine the advantages of each sub-model, so that the prediction model can better adapt to environmental changes, effectively improve the prediction accuracy, and thus provide guarantee for the stable operation of the power grid.

[0007] Preferably, the specific process of step S2 includes: the neural network of the self-organizing mapping algorithm in step S2 is composed of an input layer and a competition layer, the input layer is composed of N neurons, corresponding to an N-dimensional input vector, the competition layer is composed of M=R×C neurons, forming a two-dimensional planar array, the weight vectors of all neurons in the competition layer are compared with the input vector, and the one with the smallest distance is determined as the winning neuron, and the winning neuron is the classification result of the input vector.

[0008] The present invention clusters a data set by a self-organizing map algorithm, the data set includes training data and test data, and a neural network using the self-organizing map algorithm, the neural network of the self-organizing map has an input layer and an output layer. The second layer is usually composed of a two-dimensional grid of M x N neurons. Each neuron in the mapping layer is closely connected to all neurons in the input layer and has different weight values. In competitive learning, neurons in the output layer compete with each other to be activated. The neuron that wins the competition is the only neuron that can be excited; therefore, it is called a "winner takes all" neuron. In the self-organizing map, the competition process is to search for the neuron that is most similar to the input pattern; the winner is called the best matching unit, that is, the winning neuron of the present invention. Therefore, when clustering meteorological data, the clustering result finally generated by using the algorithm has relatively high visualization and interpretability, and because all the centroids are updated in each increment, it is little affected by the initial centroid selection, and the reliability of the data after clustering is high.

[0009] Preferably, the calculation model constructed in step S3 includes a seasonal naive prediction model, a seasonal difference autoregressive moving average model, an exponential smoothing model, a support vector machine regression model, a multivariate linear regression model, a neural network autoregressive model, and a back propagation neural network model, and the photovoltaic output is predicted by the various established models.

[0010] The present invention adopts multiple models to analyze and calculate meteorological data, predicts photovoltaic output under different environments according to different meteorological data, and conducts overall analysis on the results obtained by multiple model analyses to obtain photovoltaic output data closest to actual conditions. The analysis through multiple models improves the authenticity of the prediction, makes it closer to actual data, and improves the reliability of the data.

[0011] Preferably, the seasonal naive prediction model is a simple and effective time series prediction method. The model input is historical photovoltaic output data. Its prediction result is the same as the data at the same time of the previous cycle. Its expression is: Where h is the forecast horizon, m is the seasonal period, and k is the integer part of (h-1) / m.

[0012] The present invention is based on the naive Bayes model, combined with meteorological and seasonal data, and proposes a seasonal naive prediction model. The naive Bayes algorithm is based on the Bayesian principle and uses the knowledge of probability statistics to classify the sample data set. Due to its solid mathematical foundation, the misjudgment rate of the Bayesian classification algorithm is very low. The characteristic of the Bayesian method is the combination of prior probability and posterior probability, which avoids the subjective bias of using only prior probability and the overfitting phenomenon of using sample information alone. The Bayesian classification algorithm shows a higher accuracy rate when the data set is large, and the algorithm itself is relatively simple. Therefore, the seasonal naive prediction model obtained by the present invention based on this model has the advantages of simple algorithm but high accuracy rate in data processing.

[0013] Preferably, the seasonal difference autoregressive moving average model is a time series prediction method, which models the autocorrelation of seasonal data. The model input is historical photovoltaic output data. The seasonal difference autoregressive moving average model is expressed by the following formula: ARIMA(p,d,q)(P,D,Q) m Among them, p, d, q are the non-seasonal autoregression order, non-seasonal difference number, and non-seasonal moving average order respectively; P, D, Q are the seasonal autoregression order, seasonal difference number, and seasonal moving average order respectively; m is the period.

[0014] The present invention proposes a seasonal difference autoregressive moving average model by combining the autoregressive difference moving average model with meteorological and seasonal data. The basic idea of ​​the autoregressive difference moving average model is to use the historical information of the data itself to predict the future. The label value at a time point is affected by both the label value in the past period of time and the accidental events in the past period of time, that is, the autoregressive difference moving average model assumes that: the label value fluctuates around the general trend of time, wherein the trend is formed by the influence of historical labels, and the fluctuation is formed by the influence of accidental events in a period of time, and the general trend itself is not necessarily stable. In short, the autoregressive difference moving average model attempts to extract the time series patterns hidden behind the data through the autocorrelation and difference of the data, and then use these patterns to predict future data. Since the present invention needs to consider meteorological data, it is also necessary to establish a time series model. The established time series model needs to meet the following conditions: the model is simple and easy to implement; only endogenous variables are required, and other variables need to be used to model linear and nonlinear trends; seasonality can be modeled, and the autoregressive difference moving average model fully meets the required conditions. Therefore, the seasonal difference autoregressive moving average model obtained by the present invention based on the autoregressive difference moving average model has the advantages of being simple in model, easy to implement, only endogenous variables are required, other variables need to be used, linear and nonlinear trends can be modeled, seasonality can be modeled, and residuals can be modeled, thereby checking whether there is autocorrelation.

[0015] Preferably, part of the photovoltaic output data predicted by the established model is predicted by an exponential smoothing model. The prediction result of the exponential smoothing model is a weighted average of historical data, and the weight decays exponentially as the historical data becomes older. The exponential smoothing model predicts a time series with both trend characteristics and seasonal characteristics. When predicting the photovoltaic output, the time series is first decomposed by a time series decomposition algorithm STL, the seasonal component is predicted by a seasonal naive prediction model, and the trend component and the remainder are predicted by an exponential smoothing model.

[0016] The present invention predicts the predicted photovoltaic output data through an exponential smoothing model. Since the exponential smoothing model uses a simple weighted average method to process time series data, it does not require complex calculations and statistical inferences. Therefore, the amount of calculation is low while meeting the needs, which can improve the prediction efficiency. The exponential smoothing model can better capture the trend and seasonal changes of the data, which is consistent with the seasonal and meteorological data that the present invention needs to process, and has good accuracy for short-term predictions. Therefore, it is particularly suitable for the stable or slowly changing data series of the present invention. The exponential smoothing model uses a moving average method to smooth the data, which can buffer abnormal values ​​and adaptively adjust factors that have a greater impact on the data, which can improve the accuracy of the prediction. Finally, the exponential smoothing model can also flexibly select smoothing coefficients and seasonal factors according to the characteristics and needs of the data. The present invention selects appropriate coefficients according to the needs to predict meteorological data and seasonal data.

[0017] Preferably, the multivariate linear regression model assumes that the dependent variable has a linear relationship with multiple independent variables, which is a widely used regression method, and its expression is: Among them, x i，t represents the ith independent variable, c i It represents the coefficient corresponding to the ith independent variable. In the photovoltaic output forecast, the model input is historical photovoltaic data and meteorological data; The neural network autoregressive model uses a neural network to solve the regression problem. The back propagation neural network model includes an input layer, a hidden layer and an output layer. The error of the output layer is back propagated to the input layer layer by layer. The vector machine regression model selects a Gaussian kernel function as a kernel function, and the model input is historical photovoltaic data and meteorological data.

[0018] The neural network autoregressive model of the present invention uses a neural network to solve the regression problem. In the photovoltaic output prediction, the input layer of the neural network autoregressive model is the historical photovoltaic output data, there is only one hidden layer, and the output layer is the photovoltaic output prediction value; the back propagation neural network model of the present invention is a classic neural network model, including an input layer, a hidden layer, and an output layer, and the error of the output layer is back-propagated to the input layer layer by layer. When predicting photovoltaic output, its input data includes historical photovoltaic output data and historical meteorological data, and the output data is the photovoltaic output prediction value; the vector machine regression model of the present invention is an application of support vector machines in regression problems, and nonlinear regression is achieved by introducing kernel functions, and the model is optimized according to the principles of maximizing the interval band width and minimizing the total loss. When predicting photovoltaic output, a Gaussian kernel function is selected as the kernel function, and the model input is historical photovoltaic data and meteorological data.

[0019] Preferably, the combined photovoltaic output prediction model in step S4 calculates the weights corresponding to each sub-model based on the root mean square error data of the prediction results of each sub-model and the inverse error method as model parameters.

[0020] The present invention establishes multiple models to analyze and calculate the meteorological data of the data set clustering processing, and calculates the root mean square error between the predicted results and the actual results. The data calculated by the multiple models need to be combined for processing. For a comprehensive prediction of photovoltaic output, it is not possible to use only the results predicted by a single model. It is necessary to comprehensively consider the data obtained by all models. Therefore, a combined prediction model for photovoltaic output is proposed for integrating and analyzing the data of the above-mentioned multiple models.

[0021] Preferably, the inverse error method is a simple and intuitive weight determination method, which determines the weight of the sub-prediction model in the combined prediction model according to the prediction error of the sub-prediction model. The larger the prediction error of the sub-prediction model, the smaller its weight in the combined prediction. The prediction error is measured by the root mean square error. The calculation formula of the sub-model weight in the combined prediction model is: Among them, ω i is the weight of the i-th seed model, RMSE i is the root mean square error of the i-th seed model.

[0022] The present invention adopts the inverse error method to determine the weight of the data of multiple sub-prediction models of photovoltaic output. It is a simple and efficient method. The weight of the sub-prediction model in the combined prediction model is determined according to the prediction error of the sub-prediction model. In order to improve the accuracy of the final prediction, the weight of the data predicted by the sub-prediction model with higher prediction errors should be reduced, and the weight of the data with lower prediction errors should be increased, so as to comprehensively improve the accuracy of the prediction results of the combined prediction model of photovoltaic output.

[0023] Preferably, the calculation method of the photovoltaic output prediction result in step S4 is: Among them, y i is the predicted value of the i-th seed model.

[0024] The present invention conducts a comprehensive analysis of the weights of the predicted data of each sub-prediction model, and finally obtains a prediction result of photovoltaic output that is more in line with the actual situation and can take into account the influence of meteorological seasonal factors.

[0025] The beneficial effects of the present invention are as follows: the present invention adopts the inverse error method to determine the weights of the sub-models and establishes a photovoltaic output combined prediction model. The combined prediction model combines the advantages of each sub-model and improves the adaptability of the prediction model to environmental changes, thereby effectively improving the prediction accuracy and providing reliable information for the stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the present invention; DETAILED DESCRIPTION

[0027] The specific implementation of the technical solution of the present invention is further described below through examples and in conjunction with the accompanying drawings.

[0028] Embodiment 1: See also Figure 1 The embodiment of the present invention provides a photovoltaic output combination prediction method based on the inverse error method, comprising the following steps: S1. Collecting historical output data of the photovoltaic system and meteorological data such as irradiance, establishing a data set, and dividing training data and test data according to a certain ratio.

[0029] S2. Based on the meteorological data in the data set, the data set is clustered by the self-organizing map algorithm; the neural network of the self-organizing map algorithm consists of an input layer and a competition layer. The input layer consists of N neurons, corresponding to the N-dimensional input vector, and the competition layer consists of M = R × C neurons, forming a two-dimensional planar array. The weight vectors of all neurons in the competition layer are compared with the input vector, and the one with the smallest distance is determined as the winning neuron, and the winning neuron is the classification result of the input vector.

[0030] The present invention clusters a data set by a self-organizing map algorithm, the data set includes training data and test data, and a neural network using the self-organizing map algorithm, the neural network of the self-organizing map has an input layer and an output layer. The second layer is usually composed of a two-dimensional grid of M x N neurons. Each neuron in the mapping layer is closely connected to all neurons in the input layer and has different weight values. In competitive learning, neurons in the output layer compete with each other to be activated. The neuron that wins the competition is the only neuron that can be excited; therefore, it is called a "winner takes all" neuron. In the self-organizing map, the competition process is to search for the neuron that is most similar to the input pattern; the winner is called the best matching unit, that is, the winning neuron of the present invention. Therefore, when clustering meteorological data, the clustering result finally generated by using the algorithm has relatively high visualization and interpretability, and because all the centroids are updated in each increment, it is little affected by the initial centroid selection, and the reliability of the data after clustering is high.

[0031] S3. Construct a series of calculation models based on the training data to predict the photovoltaic output and calculate the root mean square error between the predicted results and the actual results. Specifically, establish a seasonal naive prediction model (SeasonalNaive, SNaive) based on the training data to predict the photovoltaic output, and calculate the root mean square error between the predicted results and the actual results; establish a seasonal autoregressive integrated moving average model (Seasonal Autoregressive Integrated Moving Average Model, SARIMA) based on the training data to predict the photovoltaic output, and calculate the root mean square error between the predicted results and the actual results; establish an exponential smoothing model (Exponential Smoothing, ETS) based on the training data to predict the photovoltaic output, and calculate the root mean square error between the predicted results and the actual results; establish a support vector regression model (SVR) based on the training data to predict the photovoltaic output according to the meteorological data such as the irradiance of the test data, and calculate the root mean square error between the predicted results and the actual results; establish a multiple linear regression model (Multiple Linear Regression, ETS) based on the training data. A neural network autoregression model (NNAR) based on training data is established to predict photovoltaic output, and the root mean square error between the predicted result and the actual result is calculated; a back-propagation neural network model (BPNN) based on training data is established to predict photovoltaic output according to meteorological data such as irradiance of test data, and the root mean square error between the predicted result and the actual result is calculated.

[0032] S4. Construct a combined prediction model for photovoltaic output. According to the constructed combined prediction model for photovoltaic output, the weights are calculated by the inverse sampling error method, and the photovoltaic output prediction results are calculated according to meteorological data such as the irradiance of the test data. Among them, the combined prediction model for photovoltaic output uses the root mean square error data of the prediction results of each sub-model and calculates the weights corresponding to each sub-model based on the inverse error method as model parameters.

[0033] The present invention establishes multiple models to analyze and calculate the meteorological data of the data set clustering processing, and calculates the root mean square error between the predicted results and the actual results. The data calculated by the multiple models need to be combined for processing. For a comprehensive prediction of photovoltaic output, it is not possible to use only the results predicted by a single model. It is necessary to comprehensively consider the data obtained by all models. Therefore, a combined prediction model for photovoltaic output is proposed for integrating and analyzing the data of the above-mentioned multiple models.

[0034] The present invention provides a photovoltaic output combined prediction method based on the inverse error method, wherein the combined prediction is to assign different weights to several prediction methods, thereby forming a combined prediction model with higher prediction accuracy. The present invention adopts the inverse error method to determine the weights of the sub-models and construct a photovoltaic output combined prediction model. Using this method to predict the output of the photovoltaic system can combine the advantages of each sub-model, so that the prediction model can better adapt to environmental changes, effectively improve the prediction accuracy, and thus provide guarantee for the stable operation of the power grid.

[0035] The calculation models constructed in step S3 include seasonal naive prediction model, seasonal difference autoregressive moving average model, exponential smoothing model, support vector machine regression model, multivariate linear regression model, neural network autoregressive model, back propagation neural network model, and the photovoltaic output is predicted by the various established models.

[0036] The present invention adopts multiple models to analyze and calculate meteorological data, predicts photovoltaic output under different environments according to different meteorological data, and conducts overall analysis on the results obtained by multiple model analyses to obtain photovoltaic output data closest to actual conditions. The analysis through multiple models improves the authenticity of the prediction, makes it closer to actual data, and improves the reliability of the data.

[0037] The seasonal naive forecasting model is a simple and effective time series forecasting method. The model input is the historical photovoltaic output data. Its forecast result is the same as the data at the same time in the previous cycle. Its expression is: Where h is the forecast horizon, m is the seasonal period, and k is the integer part of (h-1) / m.

[0038] The present invention is based on the naive Bayes model, combined with meteorological and seasonal data, and proposes a seasonal naive prediction model. The naive Bayes algorithm is based on the Bayesian principle and uses the knowledge of probability statistics to classify the sample data set. Due to its solid mathematical foundation, the misjudgment rate of the Bayesian classification algorithm is very low. The characteristic of the Bayesian method is the combination of prior probability and posterior probability, which avoids the subjective bias of using only prior probability and the overfitting phenomenon of using sample information alone. The Bayesian classification algorithm shows a higher accuracy rate when the data set is large, and the algorithm itself is relatively simple. Therefore, the seasonal naive prediction model obtained by the present invention based on this model has the advantages of simple algorithm but high accuracy rate in data processing.

[0039] The seasonal difference autoregressive moving average model is a time series forecasting method that models the autocorrelation of seasonal data. The model input is historical photovoltaic output data. The seasonal difference autoregressive moving average model is expressed as follows: ARIMA(p,d,q)(P,D,Q) m Among them, p, d, q are the non-seasonal autoregression order, non-seasonal difference number, and non-seasonal moving average order respectively; P, D, Q are the seasonal autoregression order, seasonal difference number, and seasonal moving average order respectively; m is the period.

[0040] The present invention proposes a seasonal difference autoregressive moving average model by combining the autoregressive difference moving average model with meteorological and seasonal data. The basic idea of ​​the autoregressive difference moving average model is to use the historical information of the data itself to predict the future. The label value at a time point is affected by both the label value in the past period of time and the accidental events in the past period of time, that is, the autoregressive difference moving average model assumes that: the label value fluctuates around the general trend of time, wherein the trend is formed by the influence of historical labels, and the fluctuation is formed by the influence of accidental events in a period of time, and the general trend itself is not necessarily stable. In short, the autoregressive difference moving average model attempts to extract the time series patterns hidden behind the data through the autocorrelation and difference of the data, and then use these patterns to predict future data. Since the present invention needs to consider meteorological data, it is also necessary to establish a time series model. The established time series model needs to meet the following conditions: the model is simple and easy to implement; only endogenous variables are required, and other variables need to be used to model linear and nonlinear trends; seasonality can be modeled, and the autoregressive difference moving average model fully meets the required conditions. Therefore, the seasonal difference autoregressive moving average model obtained by the present invention based on the autoregressive difference moving average model has the advantages of being simple in model, easy to implement, only endogenous variables are required, other variables need to be used, linear and nonlinear trends can be modeled, seasonality can be modeled, and residuals can be modeled, thereby checking whether there is autocorrelation.

[0041] Part of the photovoltaic output data predicted by the established model is predicted by the exponential smoothing model. The prediction result of the exponential smoothing model is the weighted average of historical data, and the weight decays exponentially as the historical data becomes older. The exponential smoothing model predicts time series with both trend characteristics and seasonal characteristics. When predicting photovoltaic output, the time series is first decomposed by the time series decomposition algorithm STL, the seasonal component is predicted by the seasonal naive prediction model, and the trend component and the remainder are predicted by the exponential smoothing model.

[0042] The present invention predicts the predicted photovoltaic output data through an exponential smoothing model. Since the exponential smoothing model uses a simple weighted average method to process time series data, it does not require complex calculations and statistical inferences. Therefore, the amount of calculation is low while meeting the needs, which can improve the prediction efficiency. The exponential smoothing model can better capture the trend and seasonal changes of the data, which is consistent with the seasonal and meteorological data that the present invention needs to process, and has good accuracy for short-term predictions. Therefore, it is particularly suitable for the stable or slowly changing data series of the present invention. The exponential smoothing model uses a moving average method to smooth the data, which can buffer abnormal values ​​and adaptively adjust factors that have a greater impact on the data, which can improve the accuracy of the prediction. Finally, the exponential smoothing model can also flexibly select smoothing coefficients and seasonal factors according to the characteristics and needs of the data. The present invention selects appropriate coefficients according to the needs to predict meteorological data and seasonal data.

[0043] The multiple linear regression model assumes that the dependent variable has a linear relationship with multiple independent variables. It is a widely used regression method, and its expression is: Among them, x i，t represents the ith independent variable, c i It represents the coefficient corresponding to the ith independent variable. In the photovoltaic output forecast, the model input is historical photovoltaic data and meteorological data; The neural network autoregressive model uses neural networks to solve regression problems. The back propagation neural network model includes an input layer, a hidden layer, and an output layer. The error of the output layer is back propagated to the input layer layer by layer. The vector machine regression model selects the Gaussian kernel function as the kernel function, and the model input is historical photovoltaic data and meteorological data.

[0044] The neural network autoregressive model of the present invention uses a neural network to solve the regression problem. In the photovoltaic output prediction, the input layer of the neural network autoregressive model is the historical photovoltaic output data, there is only one hidden layer, and the output layer is the photovoltaic output prediction value; the back propagation neural network model of the present invention is a classic neural network model, including an input layer, a hidden layer, and an output layer, and the error of the output layer is back-propagated to the input layer layer by layer. When predicting photovoltaic output, its input data includes historical photovoltaic output data and historical meteorological data, and the output data is the photovoltaic output prediction value; the vector machine regression model of the present invention is an application of support vector machines in regression problems, and nonlinear regression is achieved by introducing kernel functions, and the model is optimized according to the principles of maximizing the interval band width and minimizing the total loss. When predicting photovoltaic output, a Gaussian kernel function is selected as the kernel function, and the model input is historical photovoltaic data and meteorological data.

[0045] The inverse error method used is a simple and intuitive weight determination method. This method determines the weight of the sub-prediction model in the combined prediction model according to the prediction error of the sub-prediction model. The larger the prediction error of the sub-prediction model, the smaller its weight in the combined prediction. The prediction error is measured by the root mean square error. The calculation formula of the sub-model weight in the combined prediction model is: Among them, ω i is the weight of the i-th seed model, RMSE i is the root mean square error of the i-th seed model.

[0046] The present invention adopts the inverse error method to determine the weight of the data of multiple sub-prediction models of photovoltaic output. It is a simple and efficient method. The weight of the sub-prediction model in the combined prediction model is determined according to the prediction error of the sub-prediction model. In order to improve the accuracy of the final prediction, the weight of the data predicted by the sub-prediction model with higher prediction errors should be reduced, and the weight of the data with lower prediction errors should be increased, so as to comprehensively improve the accuracy of the prediction results of the combined prediction model of photovoltaic output.

[0047] The calculation method of the photovoltaic output prediction result in step S4 is: Among them, y i is the predicted value of the i-th seed model.

[0048] The present invention conducts a comprehensive analysis of the weights of the predicted data of each sub-prediction model, and finally obtains a prediction result of photovoltaic output that is more in line with the actual situation and can take into account the influence of meteorological seasonal factors.

[0049] In summary, the present invention adopts the inverse error method to determine the weights of the sub-models and establishes a photovoltaic output combined prediction model. The combined prediction model combines the advantages of each sub-model and improves the adaptability of the prediction model to environmental changes, thereby effectively improving the prediction accuracy and providing reliable information for the stable operation of the power grid.

[0050] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.

Claims

1. A photovoltaic output combination prediction method based on the inverse error method, characterized in that: The following steps are involved: S1. Collect historical output data of the photovoltaic system and meteorological data such as irradiance, establish a data set, and divide the training data and test data according to a certain ratio; S2. Based on the meteorological data in the data set, cluster the data set by using the self-organizing map algorithm; S3. Build a series of calculation models based on the training data to predict the photovoltaic output and calculate the root mean square error between the predicted results and the actual results; S4. Construct a combined prediction model for photovoltaic output. According to the constructed combined prediction model for photovoltaic output, the weight is calculated by the inverse sampling error method, and the photovoltaic output prediction result is calculated according to meteorological data such as the irradiance of the test data.

2. The photovoltaic output combination prediction method based on the inverse error method according to claim 1 is characterized in that: The specific process of step S2 includes: the neural network of the self-organizing map algorithm in step S2 is composed of an input layer and a competition layer, the input layer is composed of N neurons, corresponding to an N-dimensional input vector, the competition layer is composed of M=R×C neurons, forming a two-dimensional plane array, the weight vectors of all neurons in the competition layer are compared with the input vector, and the one with the smallest distance is determined as the winning neuron, and the winning neuron is the classification result of the input vector.

3. The photovoltaic output combination prediction method based on the inverse error method according to claim 1 is characterized in that: The calculation models constructed in step S3 include seasonal naive prediction model, seasonal difference autoregressive moving average model, exponential smoothing model, support vector machine regression model, multivariate linear regression model, neural network autoregressive model, back propagation neural network model, and photovoltaic output is predicted by various established models.

4. The photovoltaic output combination prediction method based on the inverse error method according to claim 3 is characterized in that: The seasonal naive prediction model is a simple and effective time series prediction method. The model input is the historical photovoltaic output data. Its prediction result is the same as the data at the same time in the previous cycle. Its expression is: Where h is the forecast horizon, m is the seasonal period, and k is the integer part of (h-1) / m.

5. The photovoltaic output combination prediction method based on the inverse error method according to claim 3 is characterized in that: The seasonal difference autoregressive moving average model is a time series prediction method that models the autocorrelation of seasonal data. The model input is historical photovoltaic output data. The seasonal difference autoregressive moving average model is expressed as follows: ARIMA(p,d,q)(P,D,Q) m Among them, p, d, q are the non-seasonal autoregression order, non-seasonal difference number, and non-seasonal moving average order respectively; P, D, Q are the seasonal autoregression order, seasonal difference number, and seasonal moving average order respectively; m is the period.

6. A photovoltaic output combination prediction method based on the inverse error method according to claim 1 or 3, characterized in that: Part of the photovoltaic output data predicted by the established model is predicted by an exponential smoothing model. The prediction result of the exponential smoothing model is a weighted average of historical data, and the weight decays exponentially as the historical data becomes older. The exponential smoothing model predicts a time series with both trend characteristics and seasonal characteristics. When predicting photovoltaic output, the time series is first decomposed by the time series decomposition algorithm STL, the seasonal component is predicted by the seasonal naive prediction model, and the trend component and the remainder are predicted by the exponential smoothing model.

7. The photovoltaic output combination prediction method based on the inverse error method according to claim 3 is characterized in that: The multivariate linear regression model assumes that the dependent variable has a linear relationship with multiple independent variables. It is a widely used regression method, and its expression is: Among them, x i，t represents the ith independent variable, c i It represents the coefficient corresponding to the ith independent variable. In the photovoltaic output forecast, the model input is historical photovoltaic data and meteorological data; The neural network autoregressive model uses a neural network to solve the regression problem. The back propagation neural network model includes an input layer, a hidden layer and an output layer. The error of the output layer is back propagated to the input layer layer by layer. The vector machine regression model selects a Gaussian kernel function as a kernel function, and the model input is historical photovoltaic data and meteorological data.

8. The photovoltaic output combination prediction method based on the inverse error method according to claim 1 is characterized in that: The combined photovoltaic output prediction model in step S4 calculates the weights corresponding to each sub-model based on the root mean square error data of the prediction results of each sub-model and the inverse error method as model parameters.

9. A photovoltaic output combination prediction method based on the inverse error method according to claim 1 or 8, characterized in that: The inverse error method used is a simple and intuitive weight determination method. This method determines the weight of the sub-prediction model in the combined prediction model according to the prediction error of the sub-prediction model. The larger the prediction error of the sub-prediction model, the smaller its weight in the combined prediction. The prediction error is measured by the root mean square error. The calculation formula of the sub-model weight in the combined prediction model is: Among them, ω i is the weight of the i-th seed model, RMSE i is the root mean square error of the i-th seed model.

10. The photovoltaic output combination prediction method based on the inverse error method according to claim 1, characterized in that: The calculation method of the photovoltaic output prediction result in step S4 is: Among them, y i is the predicted value of the i-th seed model.

Images