TW202009803A - Prediction system and method for solar photovoltaic power generation - Google Patents

Abstract

The present invention provides prediction system and method for solar photovoltaic power generation. The method includes establishing a climate sample, calculating the similarity between the historical solar energy output data in the historical database and the climate sample, obtaining multiple weather conditions of predetermined days including a target date from the weather forecast data, and search the historical database for data columns similar to multiple weather conditions. According to the data column, the comparison sample is obtained for standardization processing to establish a training matrix, randomly initializing the number of nodes in the neural network, training the neural network according to the training matrix to obtain an optimized neural network, and performing a solar photovoltaic power generation prediction.

Description

Solar photovoltaic power generation prediction system and method

The invention relates to a solar photovoltaic power generation prediction system and method, in particular to a solar photovoltaic power generation prediction system and method combined with weather forecast.

Electric energy is an important basis for national economic development and an important indicator of economic growth. It is also one of the indispensable important energy sources in human life. Precise solar photovoltaic (PV) prediction in the microgrid can not only provide proper power generation scheduling and planning, but also effectively reduce operating costs and improve power supply reliability. It can also prevent the waste of power resources and effectively use solar photovoltaic to increase system power supply. quality.

The traditional solar photovoltaic prediction method uses environmental information and solar photovoltaic module information to design a prediction model. The relationship between the environment and the module's solar photovoltaic power generation is non-linear. It is difficult to define the relationship between them using traditional prediction methods.

In addition, traditional machine learning requires the addition of weather parameter measurement equipment such as thermometers and illuminance meters, resulting in additional costs for building the system. It is also easy to incorporate huge non-relevant data into the training matrix, which adds additional computational costs.

Therefore, how to improve the accuracy of power generation forecasting methods to improve the accuracy of forecasting the solar photovoltaic system power generation situation, and at the same time reduce the system construction cost and calculation cost required for forecasting, to overcome the above defects, has become the cause of this business. One of the important topics.

The technical problem to be solved by the present invention is to provide a solar photovoltaic power generation prediction method combined with meteorological prediction in view of the deficiencies of the prior art.

In order to solve the above technical problems, one of the technical solutions adopted by the present invention is to provide a solar photovoltaic power generation prediction method combined with meteorological forecasting, including: establishing a climate sample, the climate sample including, corresponding to the first weather parameter and the second weather parameter And multiple solar power output data of the third weather parameter; the similarity calculation of the multiple historical solar power output data in the historical database and the climate sample to determine the corresponding weather parameters of each multiple historical solar power output data, of which Corresponding weather parameters are the first weather parameter, the second weather parameter or the third weather parameter; determine the target date; obtain multiple weather conditions including the predetermined number of days of the target date from the weather forecast data; based on the multiple weather conditions, in the historical database In the process, according to the determined multiple corresponding weather parameters, search for a data row similar to multiple weather conditions; determine whether the data row can be obtained with a predetermined number of days; if not, reduce the predetermined number of days and search again in the historical database, If it is, the comparison sample is obtained according to the data row, where the comparison sample includes multiple historical solar output power data of the corresponding data row; the comparison sample is standardized to establish a training matrix; the number of nodes of the neural network is initialized randomly; Train the neural network according to the training matrix, including: input the forecast date, forecast time and weather forecast corresponding to the forecast date into the neural network, and obtain multiple training results through the neural network calculation; calculate based on the training matrix and multiple training results Training error; judge whether the training error is less than the set target value of the allowable error, if not, then reset the number of nodes, update the number of nodes of the neural network with the optimization algorithm and train the neural network again according to the training matrix, If it is, the neural network is used as the optimized neural network; the optimized neural network is used for solar photovoltaic power generation prediction.

In order to solve the above technical problems, one of the technical solutions adopted by the present invention is to provide a solar photovoltaic power generation prediction system combined with weather forecasting, which includes a data acquisition module, a climate sample creation module, a historical database, and similarity Calculation module, training matrix creation module, neural network training module and solar photovoltaic power generation prediction module. The data retrieval module is used to retrieve the photovoltaic data detected by the photovoltaic module and the weather forecast data generated by the weather forecast module. The climate sample creation module is used to extract multiple solar output power data from the solar photovoltaic data, and correspond them to the first weather parameter, the second weather parameter, and the third weather parameter according to the weather forecast data to create a climate sample . The historical database stores multiple historical solar output power data. The similarity calculation module is used to calculate the similarity between the multiple historical solar power output data in the historical database and the climate sample to determine the corresponding weather parameters of the multiple historical solar power output data, wherein the corresponding The weather parameter is the first weather parameter, the second weather parameter, or the third weather parameter. The training matrix creation module is used to obtain a plurality of weather conditions including a predetermined number of days of the target date from the weather forecast data, and according to the plurality of weather conditions, in the historical database, based on the determined plurality of corresponding weather parameters, Search for data rows similar to the multiple weather conditions, where the training matrix creation module further determines whether the data rows can be obtained with the predetermined number of days, if not, reduce the predetermined number of days and search again in the historical database If yes, the comparison sample is obtained according to the data row, where the comparison sample includes the multiple historical solar output power data corresponding to the data row, and the comparison sample is standardized to establish a training matrix. The neural network training module is used to randomly initialize the number of nodes of the neural network and train the neural network according to the training matrix, including: inputting the forecast date, forecast time, and weather forecast corresponding to the forecast date The neural network calculates a plurality of training results through the neural network calculation; calculates a training error based on the training matrix and the plurality of training results; and judges whether the training error is less than the set target value of the allowable error, if not, then Reset the number of nodes, update the number of nodes of the neural network with an optimization algorithm and train the neural network again according to the training matrix, if so, use the neural network as the optimized neural network . The solar photovoltaic power generation prediction module is used to input the predicted date, the predicted time, and the weather forecast corresponding to the predicted date into the optimized neural network to perform the photovoltaic power generation prediction.

One of the beneficial effects of the present invention is that the solar photovoltaic power generation prediction system and method provided by the present invention screen historical solar output power data to establish climate samples, which can reduce traditional machine learning to measure weather parameters such as thermometers and illuminance meters. The needs of equipment, avoiding the incorporation of huge non-relevant data in each case into the training matrix, and increasing the additional calculation cost, and then through the optimization algorithm to find the most suitable number of neural nodes for the training data, can accurately predict solar photovoltaic System power generation situation.

In order to further understand the features and technical contents of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and explanation only, and are not intended to limit the present invention.

1‧‧‧ Solar photovoltaic power generation prediction system

100‧‧‧Data extraction module

101‧‧‧Climate sample building module

102‧‧‧Historical database

103‧‧‧ Similarity calculation module

104‧‧‧ Training matrix creation module

105‧‧‧Neural Network Training Module

106‧‧‧ Solar photovoltaic power generation prediction module

107‧‧‧Data visualization module

12‧‧‧solar photovoltaic module

14‧‧‧ Meteorological Forecast Module

2‧‧‧Neural Network

221, 222, ..., 22L‧‧‧ hidden layer

Nn‧‧‧Neural Node

Y1, Y2, ..., YN‧‧‧ output results

X1, X2..., XN‧‧‧ input

20‧‧‧ input layer

24‧‧‧ output layer

H1‧‧‧The first hidden layer

H2‧‧‧Second hidden layer

FIG. 1 is a block diagram of a solar photovoltaic power generation prediction system according to an embodiment of the invention.

FIG. 2 is a flowchart of a solar photovoltaic power generation prediction method according to an embodiment of the invention.

FIG. 3 is a schematic diagram of converting historical power generation data into a weather forecast type according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of comparison between climate samples and historical solar power output data according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of a neural network architecture according to an embodiment of the invention.

FIG. 6A is a schematic diagram of a two-layer hidden layer neural network architecture according to an embodiment of the invention, and FIG. 6B is a schematic diagram of an optimized neural network architecture according to an embodiment of the invention.

7 is a graph of a prediction result according to an embodiment of the invention.

The following are specific specific examples to illustrate the implementation of the "solar power generation prediction system and method" disclosed by the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments. Various details in this specification can also be based on different viewpoints and applications, and various modifications and changes can be made without departing from the concept of the present invention. In addition, the drawings of the present invention are merely schematic illustrations, and are not drawn according to actual sizes, and are declared in advance. The following embodiments will further describe the related technical content of the present invention, but the disclosed content is not intended to limit the protection scope of the present invention.

It should be understood that although terms such as “first”, “second”, and “third” may be used herein to describe various elements or signals, these elements or signals should not be limited by these terms. These terms are mainly used to distinguish one component from another component, or one signal from another signal. In addition, the term "or" as used herein may include any combination of any one or more of the associated listed items, depending on the actual situation.

For clarity of explanation, in some cases, the present technology may be presented as an independent functional block including functional blocks, which include devices, device elements, steps or routes in methods implemented in software, or a combination of hardware and software.

In some embodiments, computer-readable storage devices, media, and memory may include cables or wireless signals that contain bit streams and the like. However, when mentioned, non-transitory computer-readable storage media explicitly exclude media such as energy, carrier signals, electromagnetic waves, and the signal itself.

The method according to the above embodiments is implemented using computer-executable instructions stored or otherwise obtainable from computer-readable media. Such instructions may include, for example, instructions and data that cause or otherwise configure a general purpose computer, a special purpose computer, or a special purpose processing device to perform a certain function or set of functions. Part of the computer resources used can be accessed via the network. The computer executable instructions may be, for example, binary, intermediate format instructions, such as assembly language, firmware, or source code. Examples of computer readable media that can be used to store instructions during the method according to the described embodiments, information used, and/or information created include magnetic or optical disks, flash memory, non-volatile USB devices for sex memory, networked storage devices, etc.

The device implementing the methods according to these disclosures may include hardware, firmware, and/or software, and may take any of various shapes. Typical examples of this form include notebook computers, smart phones, small personal computers, personal digital assistants, and so on. The functions described here can also be implemented in peripheral devices or built-in cards. By way of further example, this function can also be implemented on different chips or on different circuit boards executed on a single device.

The instructions, the medium for transmitting such instructions, the computing resources for executing them, or other structures for supporting such computing resources are means for providing the functions described in these publications.

Please refer to FIGS. 1 and 2 together. FIG. 1 is a block diagram of a solar photovoltaic power generation prediction system according to an embodiment of the present invention. FIG. 2 is a flowchart of a solar photovoltaic power generation prediction method according to an embodiment of the present invention.

As shown in the figure, the first embodiment of the present invention provides a solar photovoltaic power generation prediction system 1 combined with weather forecasting, which includes a data extraction module 100, a climate sample creation module 101, a historical database 102, and a similarity calculation module 103. A training matrix creation module 104, a neural network training module 105, and a solar photovoltaic power generation prediction module 106.

The function of each module will be briefly described below, and each module will be described in more detail in subsequent embodiments. In detail, the data extraction module 100 is used to capture the solar photovoltaic data detected by the solar photovoltaic module 12 and the weather forecast data generated by the weather forecast module 14. Since most weather stations do not There is no illuminance meter, and the station data does not have daily sunshine hours and total sky insolation, and only the weather station data updated daily can obtain, for example, barometric pressure, temperature, humidity, wind speed, wind direction, precipitation, etc. Data, more specifically, the data extraction module 100 can use the actual received data of some solar photovoltaic modules and weather forecast modules to classify and forecast the sunshine according to the weather forecast terms published by the Meteorological Administration, thereby reducing Traditional machine learning needs for weather parameter measurement equipment such as thermometers and illuminance meters.

Further, step S100 is performed, and a plurality of pieces of solar output power data are taken from the solar photovoltaic data with the climate sample creation module 101, and corresponding to the first weather parameter, the second weather parameter, and the third weather according to the weather forecast data Parameters to establish climate samples. In detail, the climate sample creation module 101 is mainly used to select a database of adaptive related samples. First, the weather type can be marked as sunny (code: 2), cloudy (code: 1), rainy (code: 0) ), fault (code: -1), at the same time, the historical power generation data is selected by the expert system to select the historical power generation curve of representative weather. Among them, 20 groups can be selected for sunny, cloudy and rainy days, for example, to establish a climate sample.

Please refer to FIG. 3, which is a schematic diagram of converting historical power generation data into a weather forecast type according to an embodiment of the present invention. For example, when there is no shielding or only slight shielding throughout the day, it can be determined as sunny, and in the case of cloudy days, the shielding time is more than 1/2 of the daytime, and the power generation is only 20% to 70% of the sunny day. It is cloudy, and when the amount of power generation is less than 50% of the sunny day, it is determined to be rainy. In addition, when the data is partially missing, there is only one data or no data at all, it is determined to be a fault.

The historical database 101 stores a plurality of historical solar output power data, and is used to perform step S102, and the similarity calculation module 103 compares the plurality of historical solar output power data in the historical database 101 with the climate sample Calculate to determine the corresponding weather parameter for each of the multiple historical solar power output data, where the corresponding weather parameter is the first weather parameter, the second weather parameter, or the third weather parameter, that is, sunny day (Code: 2 ), cloudy days (code: 1), rainy days (code: 0). The similarity calculation performed by the similarity calculation module 103 may include using the Euclidean distance system to calculate multiple similarities between multiple historical solar power output data and multiple solar output power data of the climate sample.

For example, please refer to FIG. 4, which is a schematic diagram of comparison between climate samples and historical solar power output data according to an embodiment of the present invention. Assuming that there is an unknown weather type solar photovoltaic power generation curve in the historical database 101, Euclidean distance calculation is performed with three weather type samples (clear, cloudy and rainy), hereinafter referred to as Euclidean distance. The one with the smallest Euclidean distance is the unknown weather pattern. In Figure 4, the solar photovoltaic power generation curve of an unknown weather type is manually identified as a cloudy power generation curve. After calculating the Euclidean distance from its power generation curve and the three weather pattern samples, the following table 1:

As shown in Table 1, the smallest average Euclidean distance is sunny, so the system determines that it is sunny, and the reason for judging sunny is because the power generation value at the 10th time point in the sample is high, which is very similar to the value of the time period of sunny days, and the European distance at this point is extremely small , Causing misjudgment.

To this end, the similarity calculation performed by the similarity calculation module 103 may further include removing the calculated multiple similarities by extremum, obtaining an average similarity, and determining each historical solar output power data according to the average similarity Corresponding weather parameters.

In this embodiment, if the extreme value is removed and the Euclidean distance is recalculated, the following table 2:

From Table 2, the European distance between the sample and the sunny day rises from 0.1516 to 0.1722; the European distance between cloudy days decreases from 0.1659 to 0.1503; the European distance from rainy days rises from 0.3072 to 0.3219. The cloudy European distance is the smallest, so it is judged as overcast The day is the same as the manual judgment. Then, step S104 is performed, and the historical data is marked by applying the weather pattern of the day with the highest average similarity.

Among them, the above European system specifically includes the Euclidean (Euclidean) distance method to calculate the similarity between the target and each sample, after removing the extremum of the similarity of each group of samples, the average value is obtained, and It is expressed by the following formula (1):

為去除極值後之平均值。如此，將可在建立訓練矩陣前減少因極端狀況造成的誤判。 Where x _i is the Euclidean distance x _ij and y _ij between the target and the sample of the i-th row vector, that is, the Euclidean norm value between the target and the sample matrix,

It is the average value after removing the extreme value. In this way, the misjudgment caused by extreme conditions can be reduced before the training matrix is established.

Further, when you want to build a training matrix, you must first go to step S106 and enter the target parameters. The target parameters may include the target date (for example, 0000/00/00~0000/00/00), the target time (hour), and the reservation including the target date Multiple weather conditions for days. Next, step S108 is performed, and the training matrix creation module 104 is used to reduce the order of the training matrix. Specifically, the training matrix creation module 104 is based on multiple weather conditions in the historical database and based on the marked multiple corresponding weather Parameters, search for data rows similar to multiple weather conditions. For example, the predetermined number of days is three days to obtain multiple weather conditions on the day, the previous day, and the next day of the target date, for example, sunny, cloudy, and rainy days, search and read each region and each historical time in the historical database 101 Consistent with the target parameters, in other words, the weather data rows, which are also arranged in sunny, cloudy and rainy days, are used to build a training matrix for training neural networks.

Among them, the training matrix creation module 104 will further determine in step S110 whether the data row can be obtained with the predetermined number of days, if not, proceed to step S112 to reduce the predetermined number of days, for example, by 1, and again in the historical database 101 The search is performed in, if it is, the comparison sample is obtained based on this data row. It should be noted that the comparison sample includes multiple historical solar output power data corresponding to this data row.

Next, step S114 is performed, and the training matrix creation module 104 normalizes the comparison sample to create a training matrix. Standardization processing may include power generation standardization processing, time standardization processing, and date standardization processing. Among them, the standardization of power generation includes standardization of daily solar power generation dispersion, as shown in the following formula (2):

分別為原始與更新後之發電量。PV _min與PV _max分別為當日發電量最小值及最大值。 Among them, PV _i and

Respectively, the original and updated power generation. PV _min and PV _max are the minimum and maximum power generation amount of the day, respectively.

The time standardization process is as follows (3), (4):

Among them, T _sin is the sin function of the time component, T _cos is the cos function of the time component, and T _origin is the original data.

The time standardization process is as follows (5), (6):

Among them, Date _sin is the sin function of the time component, Date _cos is the cos function of the time component, and Date _origin is the original data. By standardizing this comparison sample, a standardized training matrix suitable for neural networks can be established.

In a word, proceed to step S116, the neural network training module 105 randomly initializes the number of nodes of the neural network, and trains the neural network according to the training matrix. Please refer to FIG. 5, which is a schematic diagram of a neural network architecture according to an embodiment of the present invention. As shown in the figure, the architecture of the neural network 2 specifically includes an input layer 20, hidden layers 221, 222, ..., 22L and an output layer 24. In the neural network, each neuron Nn is multiplied by the input X1, X2..., XN(input) by a weight w(weight), after the total is added, and a deviation value b(bias) is added to obtain its output Results Y1, Y2, ..., YN (output). During the training process, the weight and deviation of each node are often adjusted through the learning rate η. Common methods to modify the learning rate η include Gradient descent, one-step secant method and Levenberg-Marquardt algorithm.

The neural network training module 105 randomly initializes the number of nodes of the neural network 2 by, for example, adopting the neural network with at least two hidden layers and randomly generating the number of nodes between a maximum number of nodes per layer Lim Multiple nodes between _up and a lower limit of the number of nodes per layer Lim _bottom , and the upper limit of the number of nodes per layer Lim _{up is} less than or equal to twice the number of 20 nodes of the input layer 20 of the neural network 2, the lower limit of the number of nodes per layer Lim _{bottom is} greater than Or equal to half of the number of 20 nodes in the input layer.

Next, proceed to step S118, and use the neural network training module 105 to train the neural network. Specifically, the neural network training module 105 inputs the predicted date, the predicted time, and the weather forecast corresponding to the predicted date to the neural network 2, calculates a plurality of training results via the neural network 2, and proceeds to step S120, based on the training The matrix and multiple training results calculate the training error.

Here, mean absolute error (Mean Absolute Error, MAE) can be used as a training error indicator, and is calculated by the following formula (7):

Where y _{pre is} the training result, y _{act is} the training target included in the training matrix, and n is one dimension of the training matrix, that is, the number of historical solar photovoltaic power generation data.

Next, proceed to step S122 to determine whether the training error is less than the set allowable error target value d, that is, to determine | MAE |<| d |? If not, it means that the current node settings of neural network 2 cannot converge the training error, and the number of nodes needs to be reset. If so, it means that the neural network is fully trained and can be predicted. The neural network is used as an optimized neural network.

If it is determined that the training error is not less than the set target value d of the allowable error, step S124 is executed to update the number of nodes of the neural network 2 with the optimization algorithm, and the neural network is trained again according to the training matrix. More specifically, updating the number of nodes of the neural network with the optimization algorithm includes updating the number of neural nodes Nn of the hidden layers 221, 222, ..., 22L of the neural network 2, and the optimization algorithm Commonly used in the nonlinear function solving process, iteratively updating the number of neural nodes Nn of each hidden layer 221, 222, ..., 22L to solve. Common optimization algorithms are genetic algorithm, particle swarm algorithm and leapfrog algorithm.

If it is judged that the training error is less than the set allowable error target value d, the neural network 2 is used as an optimized neural network, and step S126 is executed.

Please refer to FIGS. 6A and 6B, which are schematic diagrams of a two-layer hidden layer neural network architecture and optimized neural network architecture according to an embodiment of the present invention. As shown in the figure, suppose that there is a 2-layer hidden layer neural network, and the upper limit of the number of nodes in each hidden layer is 3 nodes, as shown in Table 3:

After inputting the desired date (2018/01/01), prediction time (8 o'clock), and weather forecast (clear weather) through the input layer 20, nine kinds of power generation training results are obtained from the output layer 24. It is known that the actual power generation is 30kW, and it is known from the training results in Table 3 that when the number of nodes of the first hidden layer H1 and the second hidden layer H2 is (2,1), (2,3), (3,3) The training result is closest to the actual power generation of 30kW. These three combinations are the neural network architecture suitable for the situation. The output power of the remaining combinations is very different from the actual power generation. Therefore, the use of the neural network in this embodiment is not considered.

Next, take the smallest sum of the first hidden layer H1 and the second hidden layer H2 as the hidden layer neural network node architecture of this embodiment, as shown in FIG. 6B, (2,1) is the optimal algorithm The calculated hidden layer neural network architecture can obtain an optimized neural network.

In step S126, the solar photovoltaic power generation prediction module 106 inputs the predicted date, the predicted time, and the weather forecast corresponding to the predicted date into the optimized neural network to perform the photovoltaic power generation prediction. Please refer to FIG. 7, which is a graph of a prediction result according to an embodiment of the present invention. As shown in the figure, by adopting the solar photovoltaic power generation prediction system and method of the present invention, the prediction curve is compared with the actual curve, and the result of obtaining an average absolute error MAE of about 1.0040 is obtained, which proves that the photovoltaic power generation amount can be accurately predicted.

In addition, the solar photovoltaic power generation prediction system 1 may further include a data visualization module 107 for visualizing the photovoltaic power generation prediction data generated by the photovoltaic power generation prediction module 106 for photovoltaic power generation prediction. Presenting data through an intuitive interface can further assist power plant maintenance personnel to quickly predict the power generation status of the solar photovoltaic system.

One of the beneficial effects of the present invention is that the solar photovoltaic power generation prediction system and method provided by the present invention screen historical solar output power data to establish climate samples, which can reduce traditional machine learning to measure weather parameters such as thermometers and illuminance meters. The needs of equipment, avoiding the incorporation of huge non-relevant data in each case into the training matrix, and increasing the additional calculation cost, and then through the optimization algorithm to find the most suitable number of neural nodes for the training data, can accurately predict solar photovoltaic System power generation situation.

The content disclosed above is only a preferred and feasible embodiment of the present invention, and therefore does not limit the scope of the patent application of the present invention, so any equivalent technical changes made by using the description and drawings of the present invention are included in the application of the present invention. Within the scope of the patent.

The designated representative diagram is a flowchart, so there is no symbol for a simple explanation

Claims (17)

A solar photovoltaic power generation prediction method combined with meteorological forecasting includes: establishing a climate sample including multiple solar output power data corresponding to a first weather parameter, a second weather parameter and a third weather parameter; A plurality of pieces of historical solar power output data in a historical database and a climatic sample are subjected to a similarity calculation to determine a corresponding weather parameter of each of the plurality of historical solar power output data, wherein the corresponding weather parameter is the first weather Parameters, the second weather parameter or the third weather parameter; determine a target date; obtain a plurality of weather conditions including a predetermined number of days of the target date from a weather forecast data; according to the plurality of weather conditions, in the historical data In the library, according to the determined multiple corresponding weather parameters, search a data row similar to the multiple weather conditions; determine whether the data row can be obtained with the predetermined number of days, if not, then reduce the predetermined number of days and again Perform a search in the historical database, and if so, obtain a comparison sample based on the data row, where the comparison sample includes the multiple historical solar output power data corresponding to the data row; perform a standardization process on the comparison sample To create a training matrix; randomly initialize the number of nodes in a neural network; train the neural network according to the training matrix, including: inputting a forecast date, a forecast time, and a weather forecast corresponding to the forecast date The neural network calculates a plurality of training results through the neural network calculation; calculates a training error based on the training matrix and the plurality of training results; judges whether the training error is less than a set target error tolerance value, if not, Then reset the number of nodes, update the number of nodes of the neural network with an optimization algorithm and train the neural network again according to the training matrix, if so, use the neural network as an optimization Neural network; use this optimized neural network to predict solar photovoltaic power generation. The solar photovoltaic power generation prediction method combined with meteorological forecasting as described in claim 1, wherein the similarity calculation includes: calculating the plurality of historical solar power output data and the plurality of solar power of the climate sample by the Euclidean distance method Multiple similarities between output power data. The solar photovoltaic power generation prediction method combined with meteorological forecasting as described in claim 2, wherein the similarity calculation further includes: after removing the extreme values of the calculated multiple similarities, an average similarity is obtained; based on the average similarity To determine the corresponding weather parameter of each of the multiple historical solar power output data. The solar photovoltaic power generation prediction method combined with meteorological forecasting as described in claim 1, wherein the step of randomly initializing the number of nodes of the neural network includes: using the neural network with at least two hidden layers and randomly generating the node Multiple nodes with a number between a maximum number of nodes per layer and a minimum number of nodes per layer. The solar photovoltaic power generation prediction method combined with meteorological forecasting as described in claim 4, wherein the upper limit of the number of nodes in each layer is less than or equal to twice the number of nodes in an input layer of the neural network, and the lower limit of the number of nodes in each layer is greater than or equal to The number of nodes in the input layer is half. The solar photovoltaic power generation prediction method combined with weather forecasting as described in claim 1, wherein the standardization processing includes a power generation standardization processing, a time standardization processing, and a date standardization processing. The solar photovoltaic power generation prediction method combined with weather forecasting as described in claim 1, wherein the training error includes a mean absolute error (Mean Absolute Error, MAE), and is calculated by the following equation:

Where y _{pre is} the training result, y _{act is} the training target included in the training matrix, and n is one of the dimensions of the training matrix. The photovoltaic power generation prediction method combined with meteorological prediction according to claim 1, wherein the step of updating the number of nodes of the neural network with the optimization algorithm includes: updating the neural network with the optimization algorithm The number of a neural node in each hidden layer, and the optimization algorithm includes genetic algorithm, particle swarm algorithm and leapfrog algorithm. A solar photovoltaic power generation prediction system combined with meteorological forecasting includes: a data extraction module for acquiring a solar photovoltaic data detected by a solar photovoltaic module and a weather forecast data generated by a weather forecast module; A climate sample creation module for extracting multiple pieces of solar output power data from the solar photovoltaic data and corresponding them to a first weather parameter, a second weather parameter and a third weather parameter according to the weather forecast data To create a climate sample; a historical database that stores multiple pieces of historical solar power output data; a similarity calculation module for the multiple historical solar power output data and the climate sample in the historical database Perform a similarity calculation to determine a corresponding weather parameter for each of the multiple historical solar power output data, where the corresponding weather parameter is the first weather parameter, the second weather parameter, or the third weather parameter; a training matrix Establish a module for obtaining a plurality of weather conditions of a predetermined number of days including a target date from the weather forecast data, according to the plurality of weather conditions, in the historical database, according to the determined plurality of corresponding weather parameters , Searching for a data row similar to the multiple weather conditions, where the training matrix creation module further determines whether the data row can be obtained with the predetermined number of days, if not, reducing the predetermined number of days and again in the historical database Perform a search, and if so, obtain a comparison sample based on the data row, where the comparison sample includes the multiple historical solar power output data corresponding to the data row, and perform a standardization process on the comparison sample to establish a Training matrix; a neural network training module, used to randomly initialize the number of a node of a neural network, and train the neural network according to the training matrix, including: a prediction date, a prediction time and A weather forecast corresponding to the predicted date is input to the neural network, and multiple training results are obtained through the neural network calculation; a training error is calculated based on the training matrix and the multiple training results; and it is determined whether the training error is less than the set A target value of tolerance, if not, reset the number of nodes, update the number of nodes of the neural network with an optimization algorithm and train the neural network again according to the training matrix, if yes, then The neural network is used as an optimized neural network; and a solar photovoltaic power generation prediction module for inputting a predicted date, a predicted time, and a weather forecast corresponding to the predicted date into the optimized neural network , To make solar photovoltaic power generation forecast. The solar photovoltaic power generation prediction system combined with weather forecasting as described in claim 9, wherein the similarity calculation performed by the similarity calculation module includes: using the Euclidean distance system to calculate the multiple historical solar output power data and Multiple similarities between the multiple solar output power data of the climate sample. The solar photovoltaic power generation prediction system combined with meteorological forecasting as described in claim 10, wherein the similarity calculation performed by the similarity calculation module further includes: after removing the extreme values of the calculated similarities, a Average similarity; according to the average similarity, the corresponding weather parameters of each of the multiple historical solar power output data are determined. The solar photovoltaic power generation prediction system combined with meteorological prediction according to claim 9, wherein the neural network training module randomly initializes the number of nodes of the neural network further includes: using the neural network with at least two hidden layers And randomly generate multiple nodes with the number of nodes between an upper limit of the number of nodes per layer and a lower limit of the number of nodes per layer. The solar photovoltaic power generation prediction system combined with meteorological forecast according to claim 12, wherein the upper limit of the number of nodes per layer is less than or equal to twice the number of nodes of an input layer of the neural network, and the lower limit of the number of nodes per layer is greater than or equal to The number of nodes in the input layer is half. The solar photovoltaic power generation prediction system combined with weather forecasting as described in claim 9, wherein the standardization processing includes a power generation standardization processing, a time standardization processing, and a date standardization processing. The solar photovoltaic power generation prediction system combined with weather forecasting as described in claim 9, wherein the training error includes a mean absolute error (Mean Absolute Error, MAE) and is calculated by the following equation:

Where y _{pre is} the training result, y _{act is} the training target included in the training matrix, and n is one of the dimensions of the training matrix. The solar photovoltaic power generation prediction system combined with meteorological forecasting as described in claim 9, wherein the step of updating the number of nodes of the neural network with the optimization algorithm includes: updating the neural network with the optimization algorithm The number of a neural node in each hidden layer, and the optimization algorithm includes genetic algorithm, particle swarm algorithm and leapfrog algorithm. The solar photovoltaic power generation prediction system combined with weather forecast as described in claim 9 further includes a data visualization module for carrying out data on the photovoltaic power generation prediction data generated by the photovoltaic power generation prediction module of the photovoltaic power generation prediction visualize.

Images