A photovoltaic power generation prediction system and method based on smart grid

Abstract

The present invention relates to a photovoltaic power generation prediction system and method based on a smart grid, and belongs to the technical field of photovoltaic power generation prediction. It includes a prediction data acquisition module, a prediction data processing module, a photovoltaic power generation prediction module, and a power generation prediction management module; the prediction data processing module is used to perform standardized processing on the collected multi-party photovoltaic power generation data information, establish a training sample set of multi-party photovoltaic power generation prediction models, and match the photovoltaic array health index corresponding to the training sample set data. The present invention establishes a cloud-based training sample set for each photovoltaic power generation system to screen out the influencing factors of each local training input, selects factors with a higher degree of influence on photovoltaic power generation, effectively reduces the prediction range, and improves the prediction accuracy; at the same time, by jointly training the multi-party photovoltaic power generation prediction model, a global photovoltaic power generation prediction model is obtained, and the photovoltaic power generation characteristics of multiple regions are integrated, thereby improving the generalization ability and adaptability of the model.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power generation prediction, and in particular relates to a photovoltaic power generation prediction system and method based on a smart grid. Background Art

[0002] Photovoltaic power generation technology is the main technology for utilizing solar energy. Due to factors such as weather conditions, solar radiation intensity, temperature, humidity, and cloud cover, photovoltaic power generation technology has intermittent, volatile, and random characteristics. Therefore, it is necessary to study the prediction of photovoltaic power generation and know the photovoltaic power generation curve in advance so that the power department can dispatch the units with economic optimization and obtain the maximum economic benefits.

[0003] However, in actual predictions, the photovoltaic power generation is predicted without obtaining the operating status of the photovoltaic array. Only the factors affecting photovoltaic power generation are considered during the prediction process, which leads to invalid prediction results. At the same time, the data type of the model's training sample set is not screened, and the importance of the factors affecting the prediction results is not distinguished, resulting in low model training efficiency and low model prediction accuracy. Summary of the invention

[0004] In order to solve the above-mentioned problems existing in the prior art, the present invention provides a photovoltaic power generation prediction system and method based on a smart grid. By establishing a cloud-based training sample set for each photovoltaic power generation system, the influencing factors of the local training input are screened out, and the factors with a higher degree of influence on the photovoltaic power generation power are selected, thereby effectively reducing the prediction range and improving the prediction accuracy. At the same time, by jointly training multiple photovoltaic power generation prediction models, a global photovoltaic power generation prediction model is obtained, which integrates the photovoltaic power generation characteristics of multiple regions and improves the generalization ability and adaptability of the model.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A first aspect of the present disclosure provides a photovoltaic power generation prediction system based on a smart grid, comprising a prediction data acquisition module, a prediction data processing module, a photovoltaic power generation prediction module, and a power generation prediction management module;

[0007] The prediction data processing module is used to standardize the collected multi-party photovoltaic power generation data information, establish a training sample set of the multi-party photovoltaic power generation prediction model, and match the photovoltaic array health index corresponding to the training sample set data, including the following steps:

[0008] Screening of factors affecting photovoltaic power generation: setting thresholds by calculating the correlation coefficients of various factors;

[0009] Constructing a training sample set for a multi-party photovoltaic power generation prediction model: by calculating the Euclidean distance between the photovoltaic power generation influencing factor of one party and the photovoltaic power generation influencing factors of multiple parties, historical sample data are selected to establish a training sample set for the prediction model;

[0010] Solve the photovoltaic array health index corresponding to the training sample set data: use the IV curve to evaluate the status of the photovoltaic array, calculate the health index H of one photovoltaic array, and complete the matching of the health index H with the training sample set data.

[0011] As a preferred technical solution of the present invention, the calculation formula of the Euclidean distance is:

[0012] ;

[0013] In the formula, d r To predict the Euclidean distance between the photovoltaic power generation impact factor of one party and the photovoltaic power generation impact factors of multiple parties, n is the total number of influencing factors screened out for photovoltaic power generation, ω j Indicates the impact factor j The weight of x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of x ij,l The cloud cover in the historical records of the multi-party photovoltaic power generation system is i Time Impact Factor j The normalized value of l The data sample number.

[0014] As a preferred technical solution of the present invention, the calculation formula of the health index H of the photovoltaic array is:

[0015] ;

[0016] In the formula, Q i The grey correlation degree of the photovoltaic area sub-array in the photovoltaic array is calculated by grey correlation analysis of the simulated IV curve and the actual IV curve, and L is the power attenuation rate of the photovoltaic area sub-array in the photovoltaic array.

[0017] As a preferred technical solution of the present invention, the prediction data acquisition module is used to monitor the electric energy data and environmental monitoring data of the photovoltaic power generation system in real time, and transmit the data to the prediction data processing module through the communication equipment, wherein the electric energy data includes current, voltage and power, and the environmental monitoring data includes the working environment temperature of the photovoltaic module, the sunshine intensity and the back panel temperature of the photovoltaic module.

[0018] As a preferred technical solution of the present invention, the photovoltaic power generation prediction module is used to construct a multi-party photovoltaic power generation prediction BP neural network model, and train the photovoltaic power generation prediction model of each party by using the photovoltaic power generation training sample set of each party and the corresponding photovoltaic array health index as input, and jointly train the multi-party photovoltaic power generation prediction model using a centralized federal architecture to obtain a global photovoltaic power generation prediction model.

[0019] As a preferred technical solution of the present invention, the photovoltaic power generation prediction module includes a model training module, a model distributed training module and a model joint training module;

[0020] The model training module is used to input a photovoltaic power generation training sample set and a corresponding photovoltaic array health index into a BP neural network for training, so as to construct a photovoltaic power generation prediction model;

[0021] The model distributed training module is used to perform model distributed training through multi-party photovoltaic power generation training sample sets and corresponding photovoltaic array health indexes to obtain a multi-party photovoltaic power generation prediction model;

[0022] The model joint training module is used to jointly train the model parameters of the one-party photovoltaic power generation prediction model and the multiple-party photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model.

[0023] As a preferred technical solution of the present invention, the power generation prediction management module is used to perform power generation prediction management on the target photovoltaic area based on the photovoltaic power generation prediction module, and the collected data is processed according to the input data type and then input into the trained global photovoltaic power generation prediction model to obtain the prediction result, and the maintenance of the photovoltaic components is completed by comparing the predicted value with the actual value.

[0024] A second aspect of the present disclosure provides a photovoltaic power generation prediction method based on a smart grid, which is applied to a photovoltaic power generation prediction system based on a smart grid as described above, and includes the following steps:

[0025] Collect monitoring parameters: Collect power data and environmental monitoring data from the input side of multiple photovoltaic power generation inverters. The power data includes current, voltage and power, and the environmental monitoring data includes the working environment temperature, sunshine intensity and backplane temperature of photovoltaic modules.

[0026] Establish prediction input: Use weighted Euclidean distance to filter out the influencing factors of each prediction input, then establish their own training sample sets based on the filtered influencing factors, and match each data in the training sample set with the PV array health index;

[0027] Constructing a prediction model: Using the multi-party photovoltaic power generation training sample set and the corresponding photovoltaic array health index as input, a multi-party photovoltaic power generation prediction BP neural network model is constructed;

[0028] Joint model training: The centralized federated architecture is used to jointly train multiple photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model.

[0029] Forecast and manage photovoltaic power generation in the target area: the collected data is processed according to the input data type and then input into the trained global photovoltaic power generation prediction model to obtain the prediction results. The maintenance of photovoltaic modules is completed by comparing the predicted values ​​with the actual values.

[0030] Furthermore, the establishing of the prediction input comprises the following steps:

[0031] Screening influencing factors: Calculate the correlation coefficient of each factor:

[0032] ;

[0033] In the formula, r z represents the correlation coefficient of the impact factor z, Nu represents the number of sample sets, a u (z) and P u Respectively represent the corresponding samples u The impact factor and power of represents the average value of the impact factor z in the sample set, Corresponding daily average photovoltaic power generation;

[0034] According to the calculated correlation coefficient, factors with a correlation coefficient greater than 0.5 are selected as the influencing factors of the prediction input;

[0035] Solve the weight value of the impact factor: Calculate the weight of the screened impact factor. The calculation formula is as follows:

[0036] ;

[0037] Where z=1, 2, …, n ;ω z represents the weight of the impact factor z, r z represents the correlation coefficient of the impact factor z;

[0038] Establish a characteristic vector of the factors affecting photovoltaic power generation: For different meteorological conditions, establish a characteristic vector of the factors affecting photovoltaic power generation based on the cloud cover according to the selected factors:

[0039] =[ x i1 , x i2 , …, x ij ];

[0040] In the formula, i =1,2,3,4; when i =1 means the cloud cover is 0~25%. i =2 means the cloud cover is 25~60%. i =3 means the cloud cover is 60~90%. i =4 means the cloud cover is above 90%; j represents the impact factors screened out, x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of

[0041] Constructing a training sample set for a multi-party photovoltaic power generation prediction model: By calculating the Euclidean distance between the photovoltaic power generation influencing factor of one party and the photovoltaic power generation influencing factors of multiple parties, historical sample data is selected to establish a training sample set for the prediction model, including:

[0042] Calculate Euclidean distance d r :

[0043] ;

[0044] In the formula, d r To predict the Euclidean distance between the photovoltaic power generation impact factor of one party and the photovoltaic power generation impact factors of multiple parties, n is the total number of influencing factors screened out for photovoltaic power generation, ω j Indicates the impact factor j The weight of x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of x ij,l The cloud cover in the historical records of the multi-party photovoltaic power generation system is i Time Impact Factor j The normalized value of l Number the data sample;

[0045] Sort the Euclidean distances of the influencing factors in ascending order according to the distance values, select the top N historical samples to construct a training sample set for the photovoltaic power generation prediction model of one party, and in the same way, construct the training sample sets for the photovoltaic power generation prediction models of other parties;

[0046] Solve the photovoltaic array health index corresponding to the training sample set data: Use the IV curve to evaluate the status of the photovoltaic array and calculate the health index H of one photovoltaic array, including the following steps:

[0047] According to the photovoltaic power generation data information of one party matched with the training sample, the IV curve of the photovoltaic array in actual operation, the working temperature T of the photovoltaic array and the sunshine intensity G are obtained;

[0048] The photovoltaic array is simulated to obtain the simulated IV curve under the current working temperature T and the sunshine intensity G, and the simulated IV curve and the actual IV curve are subjected to grey correlation analysis to calculate the grey correlation degree;

[0049] The grey correlation analysis of the simulated IV curve and the actual IV curve includes the following steps:

[0050] (1) Determine the reference sequence and comparison sequence: Determine the reference sequence that reflects the system behavior characteristics and the comparison sequence that affects the system behavior. The data sequence that reflects the system behavior characteristics is the reference sequence, and the data sequence composed of factors that affect the system behavior is the comparison sequence.

[0051] The reference sequence is denoted as: X 0=[ x 0(1), x 0(2),…, x 0( n )];

[0052] The comparison sequence is recorded as: ;

[0053] (2) Dimensionlessness: unify the data into an approximate range and reduce the difference in the absolute values ​​of the data. The formula is as follows:

[0054] ;

[0055] in, , where i =0, 1, 2… k ;

[0056] (3) Find the difference sequence, maximum difference and minimum difference:

[0057] Find the difference sequence: ;

[0058] Find the maximum difference: ;

[0059] Find the minimum difference: ;

[0060] (4) Calculate the correlation coefficient :The calculation formula is as follows:

[0061] ;

[0062] In the formula, ρ is the resolution coefficient, with a value range of (0, +∞), ρ The smaller the resolution, the greater the ρ When ≤0.5463, the resolution is the best. ρ =0.5;

[0063] (5) Calculation of grey relational degree Q i :

[0064] ;

[0065] The power attenuation rate L of the corresponding PV area subarray in the training sample is calculated by collecting the inverter DC side power and environmental data. The calculation formula is as follows:

[0066] ;

[0067] Where E is the power generation of each subarray during the evaluation period (k·Wh), P0 is the rated power of each subarray (kW), and H is the radiation received by the array surface (k·Wh / m 2 ), G is the irradiance under standard test conditions (1000W / m 2 ), C is the temperature correction coefficient, and the calculation formula is:

[0068] ;

[0069] Where δ is the relative temperature coefficient of power of the photovoltaic module, T C is the backplane temperature of the regional subarray component (°C);

[0070] Define the photovoltaic array health index H and complete the matching of the health index H with the training sample set data:

[0071] ;

[0072] In the formula, Q i is the grey correlation degree of the photovoltaic area sub-array in the photovoltaic array, and L is the power attenuation rate of the photovoltaic area sub-array in the photovoltaic array.

[0073] Furthermore, the model joint training comprises the following steps:

[0074] The multi-party photovoltaic power generation prediction model regularly sends the trained model parameters to the central server through a secure communication channel;

[0075] The central server takes a weighted average of all uploaded parameters and aggregates and updates the global model parameters. The central server sends the updated global model parameters to multiple photovoltaic power generation prediction models as the initial parameter values ​​for the next round of training for each party.

[0076] When the number of global iterations reaches a preset value, the iteration is stopped, the final global photovoltaic power generation prediction model is obtained, and the generated prediction model is evaluated.

[0077] The beneficial effects of the present invention are:

[0078] The present invention collects electric energy data and environmental monitoring data from multiple photovoltaic power generation systems, and screens out the influencing factors of each local training input by establishing a training sample set for each photovoltaic power generation system. The weighted Euclidean distance method is adopted in the screening process to select factors with a higher degree of influence on photovoltaic power generation power, effectively reducing the prediction range. The training samples are established based on cloud cover, which further refines the degree of influence of weather types on prediction results and improves prediction accuracy. At the same time, the photovoltaic power generation prediction models of multiple parties are jointly trained through federated learning to obtain a global photovoltaic power generation prediction model, which effectively protects the original power generation data of each party, integrates the photovoltaic power generation characteristics of multiple regions, and improves the generalization ability and adaptability of the model.

[0079] The present invention adds a photovoltaic array health index to the input of the model prediction. The photovoltaic array health index includes the real-time operating status and performance degradation degree of the photovoltaic components. It is used as a part of feature engineering and combined with other meteorological data and historical power generation data to construct a richer and more comprehensive feature set, thereby improving the prediction model's ability to explain changes in photovoltaic power generation, thereby improving the accuracy of the prediction. At the same time, the photovoltaic array health index is incorporated into the prediction model, and prior knowledge of the photovoltaic array status is introduced to ensure that the model takes into account the health status of the photovoltaic components when generating the prediction results, thereby improving the reliability and stability of the system and reducing the risks caused by failures or abnormal conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0081] Figure 1 A schematic diagram of the structure of a photovoltaic power generation prediction system based on a smart grid provided by an embodiment of the present invention;

[0082] Figure 2 A schematic diagram of a process flow of steps for processing data by a prediction data processing module provided in an embodiment of the present invention;

[0083] Figure 3 A schematic diagram of steps for performing joint training of the model joint training module provided in an embodiment of the present invention;

[0084] Figure 4 A schematic diagram of the steps of a photovoltaic power generation prediction method based on a smart grid provided by an embodiment of the present invention;

[0085] Figure 5 A schematic diagram of the steps of establishing a prediction input provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0086] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0087] Smart grid is based on traditional power grid, highly integrated with modern advanced information technology and computer technology, and improves the reliability, economy and safety of power grid through intelligent AI, manual remote control and other means. Smart grid makes the design and control of energy converters more intelligent and effective by applying electrical engineering technology.

[0088] An energy converter is a device that converts raw energy into electrical energy, such as a solar inverter. Electrical engineering technology designs efficient energy converters by studying and improving the performance and characteristics of power electronic components. These converters can convert raw energy into stable electrical output and have the ability to efficiently utilize energy. Secondly, by setting sensors, the energy converter can be monitored in real time, and various parameters in the energy conversion process can be accurately measured, including voltage, current, power and other information, thereby realizing real-time monitoring and data analysis of the power system. Therefore, this embodiment predicts photovoltaic power generation based on the monitoring data of the smart grid.

[0089] At least one embodiment of the present disclosure provides a photovoltaic power generation prediction system based on a smart grid, such as Figure 1 As shown, it includes a prediction data acquisition module, a prediction data processing module, a photovoltaic power generation prediction module and a power generation prediction management module.

[0090] The prediction data acquisition module is used to monitor the electric energy data and environmental monitoring data of the photovoltaic power generation system in real time, and transmit the data to the prediction data processing module through the communication equipment, wherein the electric energy data includes current, voltage and power, etc., and the environmental monitoring data includes the working environment temperature of the photovoltaic module, the sunshine intensity and the back panel temperature of the photovoltaic module, etc.

[0091] It is understandable that the prediction data acquisition module includes an electric energy data acquisition module and an environmental data monitoring module.

[0092] It should be noted that data acquisition uses sensors and other equipment to obtain various data in the distribution system in real time, providing a basis for subsequent data processing and analysis. During the data acquisition process, the type and installation location of the sensor should be reasonably selected and arranged according to the monitoring requirements and system characteristics. For example, the current sensor is used to measure the current in the circuit, the voltage sensor is used to measure the voltage value in the circuit, the temperature sensor is used to monitor the working temperature change of the photovoltaic array, and the light sensor is used to measure the sunshine intensity of the photovoltaic array. In order to ensure the accuracy and real-time nature of the data, the data acquisition equipment needs to sample and digitize the collected analog signals. During the sampling process, the appropriate sampling frequency and accuracy are selected to fully capture the dynamic changes of the photovoltaic array. After the data acquisition is completed, the collected data needs to be transmitted to the prediction data processing module through the communication network. The communication network adopts wired or wireless methods, such as Ethernet, wireless sensor network, etc.

[0093] The prediction data processing module is used to standardize the collected multi-party photovoltaic power generation data information, establish a training sample set of the multi-party photovoltaic power generation prediction model, and match the photovoltaic array health index corresponding to the training sample set data, such as Figure 2 As shown, the following steps are included:

[0094] Screening the factors affecting photovoltaic power generation: In order to select the factors with a high degree of influence on photovoltaic power generation from the historical records of photovoltaic power generation systems, the threshold is set by calculating the correlation coefficient of each factor;

[0095] ;

[0096] In the formula, r z represents the correlation coefficient of the impact factor z, Nu represents the number of sample sets, a u (z) and P u Respectively represent the corresponding samples u The impact factor and power of represents the average value of the impact factor z in the sample set, Corresponding to the daily average photovoltaic power generation power.

[0097] According to the calculated correlation coefficient, the influencing factors with a correlation coefficient greater than 0.5 are selected as the influencing factors of the prediction input.

[0098] Constructing a training sample set for a multi-party photovoltaic power generation prediction model: By calculating the Euclidean distance between the photovoltaic power generation influencing factor of one party and the photovoltaic power generation influencing factors of multiple parties, historical sample data is selected to establish a training sample set for the prediction model, including:

[0099] Calculate Euclidean distance d r :

[0100] ;

[0101] In the formula, d r To predict the Euclidean distance between the photovoltaic power generation impact factor of one party and the photovoltaic power generation impact factors of multiple parties, n is the total number of influencing factors screened out for photovoltaic power generation, ω j Indicates the impact factor j The weight of x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of x ij,l The cloud cover in the historical records of the multi-party photovoltaic power generation system is i Time Impact Factor j The normalized value of l The data sample number.

[0102] The Euclidean distances of the influencing factors are sorted in ascending order according to the distance values, and the first N historical samples are selected to construct a training sample set of the photovoltaic power generation prediction model for one party. In the same way, the training sample sets of the photovoltaic power generation prediction models for other parties are constructed.

[0103] Solve the photovoltaic array health index corresponding to the training sample set data: Use the IV curve to evaluate the status of the photovoltaic array, calculate the health index H of one photovoltaic array, and complete the matching of the health index H with the training sample set data:

[0104] ;

[0105] In the formula, Q i The grey correlation degree of the photovoltaic area sub-array in the photovoltaic array is calculated by grey correlation analysis of the simulated IV curve and the actual IV curve, and L is the power attenuation rate of the photovoltaic area sub-array in the photovoltaic array.

[0106] The photovoltaic power generation prediction module is used to construct a multi-party photovoltaic power generation prediction BP neural network model, train the photovoltaic power generation prediction model of each party by using the photovoltaic power generation training sample set of each party and the corresponding photovoltaic array health index as input, and jointly train the multi-party photovoltaic power generation prediction model using a centralized federal architecture to obtain a global photovoltaic power generation prediction model.

[0107] The photovoltaic power generation prediction module includes a model training module, a model distributed training module and a model joint training module;

[0108] The model training module is used to input a photovoltaic power generation training sample set and a corresponding photovoltaic array health index into a BP neural network for training, so as to construct a photovoltaic power generation prediction model.

[0109] The model distributed training module is used to perform model distributed training through multi-party photovoltaic power generation training sample sets and corresponding photovoltaic array health indexes to obtain a multi-party photovoltaic power generation prediction model.

[0110] The model joint training module is used to jointly train the model parameters of the one-party photovoltaic power generation prediction model and the multiple-party photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model.

[0111] The joint training, such as Figure 3 As shown, the following steps are included:

[0112] The multi-party photovoltaic power generation prediction model regularly sends the trained model parameters to the central server through a secure communication channel;

[0113] The central server takes a weighted average of all uploaded parameters and aggregates and updates the global model parameters. The central server sends the updated global model parameters to multiple photovoltaic power generation prediction models as the initial parameter values ​​for the next round of training for each party.

[0114] When the number of global iterations reaches a preset value, the iteration is stopped, the final global photovoltaic power generation prediction model is obtained, and the generated prediction model is evaluated.

[0115] The power generation prediction management module is used to perform power generation prediction management on the target photovoltaic area based on the photovoltaic power generation prediction module, and the collected data is processed according to the input data type and then input into the trained global photovoltaic power generation prediction model to obtain the prediction result, and the maintenance work of the photovoltaic components is completed by comparing the predicted value with the actual value.

[0116] At least one embodiment of the present disclosure also provides a photovoltaic power generation prediction method based on a smart grid, such as Figure 4 As shown, the following steps are included:

[0117] S1. Collect monitoring parameters: Collect power data and environmental monitoring data from the input side of multiple photovoltaic power generation inverters. The power data includes current, voltage and power, and the environmental monitoring data includes the working environment temperature of photovoltaic modules, sunshine intensity and the back panel temperature of photovoltaic modules.

[0118] S2. Establish prediction input: Use weighted Euclidean distance to filter out the influencing factors of each prediction input, and then use the filtered influencing factors to establish their own training sample sets, and match each data in the training sample set with the photovoltaic array health index. Figure 5 As shown, the specific steps include:

[0119] S21. Screening of influencing factors: In order to select factors with a high degree of influence on photovoltaic power generation from the historical records of photovoltaic power generation systems, effectively reduce the prediction range, and improve the prediction accuracy, the correlation coefficients of various factors are calculated:

[0120] ;

[0121] In the formula, r z represents the correlation coefficient of the impact factor z, Nu represents the number of sample sets, a u (z) and P u Respectively represent the corresponding samples u The impact factor and power of represents the average value of the impact factor z in the sample set, Corresponding to the daily average photovoltaic power generation power.

[0122] According to the calculated correlation coefficient, the influencing factors with a correlation coefficient greater than 0.5 are selected as the influencing factors of the prediction input.

[0123] S22. Solve the weight value of the impact factor: Calculate the weight of the screened impact factor. The calculation formula is as follows:

[0124] ;

[0125] Where z=1, 2, …, n ;ω z represents the weight of the impact factor z, r z Represents the correlation coefficient of the impact factor z.

[0126] S23. Establishing a characteristic vector of factors affecting photovoltaic power generation: Weather types seriously affect photovoltaic power generation, but the same weather type has different states at each time point. Therefore, in this embodiment, cloud cover is used to characterize weather types. For different meteorological conditions, a characteristic vector of factors affecting photovoltaic power generation is established based on cloud cover according to the selected influencing factors:

[0127] =[ x i1 , x i2 , …, x ij ];

[0128] In the formula, i =1,2,3,4; when i =1 means the cloud cover is 0~25%. i =2 means the cloud cover is 25~60%. i =3 means the cloud cover is 60~90%. i =4 means the cloud cover is above 90%; j represents the impact factors screened out, x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of .

[0129] It should be noted that the impact factor j It is screened out from the data types of historical records of photovoltaic power generation systems by calculating correlation coefficients and setting thresholds. The data types of historical data include the minimum and maximum ambient temperature, the minimum and maximum humidity, the minimum and maximum temperature of the photovoltaic module backplane, and the minimum and maximum sunshine intensity, and are matched with the power data of the photovoltaic array based on time periods.

[0130] S24. Constructing a training sample set for a multi-party photovoltaic power generation prediction model: by calculating the Euclidean distance between the photovoltaic power generation influencing factor of one party and the photovoltaic power generation influencing factors of multiple parties, selecting historical sample data to establish a training sample set for the prediction model, including:

[0131] Calculate Euclidean distance d r :

[0132] ;

[0133] In the formula, d r To predict the Euclidean distance between the photovoltaic power generation impact factor of one party and the photovoltaic power generation impact factors of multiple parties, nis the total number of influencing factors screened out for photovoltaic power generation, ω j Indicates the impact factor j The weight of x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of x ij,l The cloud cover in the historical records of the multi-party photovoltaic power generation system is i Time Impact Factor j The normalized value of l The data sample number.

[0134] The Euclidean distances of the influencing factors are sorted in ascending order according to the distance values, and the first N historical samples are selected to construct a training sample set of a photovoltaic power generation prediction model for one party. In the same way, the training sample sets of photovoltaic power generation prediction models for other parties are constructed. It can be understood that N is the preset selection threshold for establishing the training sample set.

[0135] It should be noted that, through the above method, based on the historical sample set characteristics of each photovoltaic power generation system, similar sample sets for training are selected from the historical sample set data of multiple photovoltaic power generation systems, which enriches the training sample set data and improves the accuracy of the prediction models of all parties.

[0136] S25, solving the photovoltaic array health index corresponding to the training sample set data: using the IV curve to evaluate the status of the photovoltaic array and calculate the health index H of one photovoltaic array, including the following steps:

[0137] S251. According to the photovoltaic power generation data information matched with the training samples, the IV curve of the photovoltaic array in actual operation and the photovoltaic array operating temperature T and the sunshine intensity G are obtained.

[0138] The photovoltaic power generation data information is collected by dividing the area into inverter units, obtaining the inverter operation data and environmental monitoring data, and screening out the current, voltage, power and other electrical energy data and environmental data on the input side of each sub-array for analysis.

[0139] S252, simulating the photovoltaic array to obtain a simulated IV curve under the current operating temperature T and the sunshine intensity G, and performing grey correlation analysis on the simulated IV curve and the actual IV curve to calculate the grey correlation degree.

[0140] The grey correlation analysis of the simulated IV curve and the actual IV curve includes the following steps:

[0141] (1) Determine the reference sequence and comparison sequence: Determine the reference sequence that reflects the system behavior characteristics and the comparison sequence that affects the system behavior. The data sequence that reflects the system behavior characteristics is the reference sequence, and the data sequence composed of factors that affect the system behavior is the comparison sequence.

[0142] The reference sequence is denoted as: X 0=[ x 0(1), x 0(2),…, x 0( n )];

[0143] The comparison sequence is recorded as: ;

[0144] (2) Dimensionlessness: unify the data into an approximate range and reduce the difference in the absolute values ​​of the data. The formula is as follows:

[0145] ;

[0146] in, , where i =0, 1, 2… k ;

[0147] (3) Find the difference sequence, maximum difference and minimum difference:

[0148] Find the difference sequence: ;

[0149] Find the maximum difference: ;

[0150] Find the minimum difference: ;

[0151] (4) Calculation of correlation coefficient: The correlation coefficient is an indicator that describes the degree of correlation between the comparison sequence and the reference sequence at a certain moment. The calculation formula is as follows:

[0152] ;

[0153] In the formula, ρ is the resolution coefficient, with a value range of (0, +∞), ρ The smaller the resolution, the greater the ρ When ≤0.5463, the resolution is the best. ρ =0.5.

[0154] (5) Calculation of grey relational degree Q i :

[0155] .

[0156] Grey relation analysis (GRA) is a multi-factor statistical analysis method, which is a method to measure the degree of correlation between factors based on the similarity or difference of the development trends between factors. In this embodiment, the operating status of the photovoltaic array is characterized by comparing the similarity between the reference data and the measured data.

[0157] S253, calculating the power attenuation rate L of the corresponding photovoltaic area sub-array in the training sample by using the collected inverter DC side power, environmental data, etc., and the calculation formula is as follows:

[0158] ;

[0159] Where E is the power generation of each subarray during the evaluation period (k·Wh), P0 is the rated power of each subarray (kW), and H is the radiation received by the array surface (k·Wh / m 2 ), G is the irradiance under standard test conditions (1000W / m 2 ), C is the temperature correction coefficient, and the calculation formula is:

[0160] ;

[0161] Where δ is the relative temperature coefficient of power of the photovoltaic module, T C is the backplane temperature of the regional subarray component (°C).

[0162] It should be noted that the attenuation rate of photovoltaic modules refers to the attenuation rate of photovoltaic modules after they have been running for a period of time under standard test conditions (AM1.5, module temperature 25°C, irradiance 1000W / m 2 )The ratio of the maximum output power to the initial maximum output power at the time of commissioning. In order to eliminate the influence of temperature difference on the attenuation rate, the temperature of the module backplane is corrected to 25℃ using temperature correction.

[0163] S254, define the photovoltaic array health index H, and complete the matching of the health index H with the training sample set data:

[0164] ;

[0165] In the formula, Q i is the grey correlation degree of the photovoltaic area sub-array in the photovoltaic array, and L is the power attenuation rate of the photovoltaic area sub-array in the photovoltaic array.

[0166] It should be noted that the photovoltaic array health index H defined in this embodiment is composed of the ratio of the grey correlation degree of the photovoltaic array corresponding to the sample data to the power attenuation rate. Therefore, the calculated photovoltaic array health index H is obtained based on the sample data and is matched with the sample data. By defining the photovoltaic array health index, the purpose is to add the correlation degree of the grey correlation analysis and the power attenuation rate of the photovoltaic module to the input of the photovoltaic power generation prediction model. Therefore, there is no need to further analyze the specific fault problems and performance factors of the photovoltaic module, reduce the early data analysis tasks, and improve the accuracy and effectiveness of the model prediction results.

[0167] It is understandable that by evaluating the health index of the photovoltaic array, it is possible to quickly and easily determine whether the photovoltaic array is healthy or has open circuit or short circuit faults. At the same time, by adding the power attenuation rate to provide feedback on the performance status of the photovoltaic array, a comprehensive evaluation of the photovoltaic array health index is obtained to ensure the effectiveness of the output in subsequent photovoltaic power generation forecasts.

[0168] S3, constructing a prediction model: constructing a multi-party photovoltaic power generation prediction BP neural network model by using a multi-party photovoltaic power generation training sample set and a corresponding photovoltaic array health index as input, specifically including the following steps:

[0169] S31, establishing a multi-party photovoltaic power generation prediction model: by collecting the power data and environmental data of the photovoltaic power generation of multiple parties, establishing a training sample set of the multi-party photovoltaic power generation prediction model and solving the corresponding photovoltaic array health index to complete the matching;

[0170] The influencing factors of the training sample set and the corresponding photovoltaic array health index are used as input variables. The number of output layer nodes is the photovoltaic power generation power in the prediction time period, and the number of hidden layer nodes is ,in I is the number of input layer nodes; O is the number of nodes in the output layer; b is a constant ranging from 1 to 10.

[0171] S32, training multi-party photovoltaic power generation prediction model: after normalizing the input variables, the error target is set to 10 -8 The neural networks of multiple photovoltaic power generation prediction models are trained separately.

[0172] S4. Joint model training: The centralized federated architecture is used to jointly train multiple photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model.

[0173] The multi-party photovoltaic power generation prediction model regularly sends the trained model parameters to the central server through a secure communication channel. The central server weights and averages all uploaded parameters and aggregates and updates the global model parameters. The central server sends the updated global model parameters to the multi-party photovoltaic power generation prediction model as the initial parameter values ​​for the next round of training for each party.

[0174] When the number of global iterations reaches a preset value, the iteration is stopped, the final global photovoltaic power generation prediction model is obtained, and the generated prediction model is evaluated.

[0175] In this embodiment, the joint training of multi-party photovoltaic power generation prediction models is completed through federated learning. Federated learning is a distributed machine learning technology that can protect data privacy. Its core idea is that each client trains the model locally without transmitting local original data, and only uploads the updated parameters to the central server. The central server is responsible for aggregating the model parameters of all clients to build a global model.

[0176] S5. Prediction and management of photovoltaic power generation in the target area: the collected data is processed according to the input data type and then input into the trained global photovoltaic power generation prediction model to obtain the prediction result. The maintenance of photovoltaic modules is completed by comparing the predicted value with the actual value.

[0177] The present invention collects electric energy data and environmental monitoring data from multiple photovoltaic power generation systems, and screens out the influencing factors of each local training input by establishing a training sample set for each photovoltaic power generation system. The weighted Euclidean distance method is adopted in the screening process to select factors with a higher degree of influence on photovoltaic power generation power, effectively reducing the prediction range. The training samples are established based on cloud cover, which further refines the degree of influence of weather types on prediction results and improves prediction accuracy. At the same time, the photovoltaic power generation prediction models of multiple parties are jointly trained through federated learning to obtain a global photovoltaic power generation prediction model, which effectively protects the original power generation data of each party, integrates the photovoltaic power generation characteristics of multiple regions, and improves the generalization ability and adaptability of the model.

[0178] The present invention adds a photovoltaic array health index to the input of the model prediction. The photovoltaic array health index includes the real-time operating status and performance degradation degree of the photovoltaic components. It is used as a part of feature engineering and combined with other meteorological data and historical power generation data to construct a richer and more comprehensive feature set, thereby improving the prediction model's ability to explain changes in photovoltaic power generation, thereby improving the accuracy of the prediction. At the same time, the photovoltaic array health index is incorporated into the prediction model, and prior knowledge of the photovoltaic array status is introduced to ensure that the model takes into account the health status of the photovoltaic components when generating the prediction results, thereby improving the reliability and stability of the system and reducing the risks caused by failures or abnormal conditions.

[0179] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A photovoltaic power generation prediction system based on smart grid, characterized by: It includes a prediction data acquisition module, a prediction data processing module, a photovoltaic power generation prediction module and a power generation prediction management module; The prediction data acquisition module is used to monitor the power data and environmental monitoring data of the photovoltaic power generation system in real time, and transmit the data to the prediction data processing module through the communication device; The photovoltaic power generation prediction module is used to construct a multi-party photovoltaic power generation prediction BP neural network model, train the photovoltaic power generation prediction model of each party by using the photovoltaic power generation training sample set of each party and the corresponding photovoltaic array health index as input, and jointly train the multi-party photovoltaic power generation prediction model using a centralized federal architecture to obtain a global photovoltaic power generation prediction model; The power generation prediction management module is used to perform power generation prediction management on the target photovoltaic area based on the photovoltaic power generation prediction module, process the collected data according to the input data type, and input the data into the trained global photovoltaic power generation prediction model to obtain the prediction result, and complete the maintenance of the photovoltaic components by comparing the predicted value with the actual value; The prediction data processing module is used to standardize the collected multi-party photovoltaic power generation data information, establish a training sample set of the multi-party photovoltaic power generation prediction model, and match the photovoltaic array health index corresponding to the training sample set data, including the following steps: Screening of factors affecting photovoltaic power generation: setting thresholds by calculating the correlation coefficients of various factors; Constructing a training sample set for a multi-party photovoltaic power generation prediction model: by calculating the Euclidean distance between the photovoltaic power generation influencing factor of one party and the photovoltaic power generation influencing factors of multiple parties, historical sample data are selected to establish a training sample set for the prediction model; Solve the photovoltaic array health index corresponding to the training sample set data: use the IV curve to evaluate the status of the photovoltaic array, calculate the health index H of one photovoltaic array, and complete the matching of the health index H with the training sample set data; The calculation formula of the Euclidean distance is: ； In the formula, d r To predict the Euclidean distance between the photovoltaic power generation impact factor of one party and the photovoltaic power generation impact factors of multiple parties, n is the total number of influencing factors screened out for photovoltaic power generation, ω j Indicates the impact factor j The weight of x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of x ij,l The cloud cover in the historical records of the multi-party photovoltaic power generation system is i Time Impact Factor j The normalized value of l Number the data sample; The calculation formula of the health index H of the photovoltaic array is: ； In the formula, Q i The grey correlation degree of the photovoltaic area sub-array in the photovoltaic array is calculated by grey correlation analysis of the simulated IV curve and the actual IV curve, and L is the power attenuation rate of the photovoltaic area sub-array in the photovoltaic array.

2. The photovoltaic power generation prediction system based on smart grid according to claim 1, characterized in that: The photovoltaic power generation prediction module includes a model training module, a model distributed training module and a model joint training module; The model training module is used to input a photovoltaic power generation training sample set and a corresponding photovoltaic array health index into a BP neural network for training, so as to construct a photovoltaic power generation prediction model; The model distributed training module is used to perform model distributed training through multi-party photovoltaic power generation training sample sets and corresponding photovoltaic array health indexes to obtain a multi-party photovoltaic power generation prediction model; The model joint training module is used to jointly train the model parameters of the one-party photovoltaic power generation prediction model and the multiple-party photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model.

3. A photovoltaic power generation prediction method based on a smart grid, applied to a photovoltaic power generation prediction system based on a smart grid as claimed in any one of claims 1 to 2, characterized in that: The following steps are involved: Collect monitoring parameters: Collect power data and environmental monitoring data from the input side of multiple photovoltaic power generation inverters. The power data includes current, voltage and power, and the environmental monitoring data includes the working environment temperature, sunshine intensity and backplane temperature of photovoltaic modules. Establish prediction input: Use weighted Euclidean distance to filter out the influencing factors of each prediction input, then establish their own training sample sets based on the filtered influencing factors, and match each data in the training sample set with the PV array health index; Constructing a prediction model: Using the multi-party photovoltaic power generation training sample set and the corresponding photovoltaic array health index as input, a multi-party photovoltaic power generation prediction BP neural network model is constructed; Joint model training: The centralized federated architecture is used to jointly train multiple photovoltaic power generation prediction models to obtain a global photovoltaic power generation prediction model. Forecast and manage photovoltaic power generation in the target area: the collected data is processed according to the input data type and then input into the trained global photovoltaic power generation prediction model to obtain the prediction results. The maintenance of photovoltaic modules is completed by comparing the predicted values ​​with the actual values.

4. The photovoltaic power generation prediction method based on smart grid according to claim 3 is characterized by: The step of establishing the prediction input comprises the following steps: Screening of influencing factors: Calculating the correlation coefficient of each factor; According to the calculated correlation coefficient, factors with a correlation coefficient greater than 0.5 are selected as the influencing factors of the prediction input; Solve the weight value of the impact factor: calculate the weight of the screened impact factor; Establish the characteristic vector of influencing factors of photovoltaic power generation; Construct a multi-party photovoltaic power generation prediction model training sample set; Solve the photovoltaic array health index corresponding to the training sample set data; Among them, the correlation coefficient of each factor is expressed as: ； In the formula, r z represents the correlation coefficient of the impact factor z, Nu represents the number of sample sets, a u (z) and P u Respectively represent the corresponding samples u The impact factor and power of represents the average value of the impact factor z in the sample set, Corresponding to the daily average photovoltaic power generation power.

5. The photovoltaic power generation prediction method based on smart grid according to claim 4 is characterized by: The calculation formula for solving the weight value of the impact factor is as follows: ； Where z=1, 2, …, n ;ω z represents the weight of the impact factor z, r z Represents the correlation coefficient of the impact factor z.

6. The photovoltaic power generation prediction method based on smart grid according to claim 4 is characterized by: The step of establishing a characteristic vector of an influencing factor of photovoltaic power generation of one party includes: establishing a characteristic vector of an influencing factor of photovoltaic power generation of one party based on cloud cover according to the screened influencing factors for different meteorological conditions, expressed as: =[ x i1 ， x i2 ，…， x ij ]； In the formula, i =1,2,3,4; when i =1 means the cloud cover is 0~25%. i =2 means the cloud cover is 25~60%. i =3 means the cloud cover is 60~90%. i =4 means the cloud cover is above 90%; j represents the impact factors screened out, x ij The cloud cover in the forecast period for the photovoltaic array area of ​​the forecasting party is i Time Impact Factor j The normalized value of .

7. The photovoltaic power generation prediction method based on smart grid according to claim 4 is characterized by: The method of solving the photovoltaic array health index corresponding to the training sample set data includes: using the IV curve to evaluate the status of the photovoltaic array and calculating the health index H of one photovoltaic array, which specifically includes the following steps: According to the photovoltaic power generation data information of one party matched with the training sample, the IV curve of the photovoltaic array in actual operation, the working temperature T of the photovoltaic array and the sunshine intensity G are obtained; The photovoltaic array is simulated to obtain the simulated IV curve under the current working temperature T and the sunshine intensity G, and the simulated IV curve and the actual IV curve are subjected to grey correlation analysis to calculate the grey correlation degree; The power attenuation rate L of the corresponding PV area subarray in the training sample is calculated through the collected inverter DC side power and environmental data; Define the photovoltaic array health index H and complete the matching of the health index H with the training sample set data.

8. The photovoltaic power generation prediction method based on smart grid according to claim 3 is characterized by: The model joint training includes the following steps: The multi-party photovoltaic power generation prediction model regularly sends the trained model parameters to the central server through a secure communication channel; The central server takes a weighted average of all uploaded parameters and aggregates and updates the global model parameters. The central server sends the updated global model parameters to multiple photovoltaic power generation prediction models as the initial parameter values ​​for the next round of training for each party. When the number of global iterations reaches a preset value, the iteration is stopped, the final global photovoltaic power generation prediction model is obtained, and the generated prediction model is evaluated.

Images