Energy efficiency prediction method and device based on solar photovoltaic panel and medium

Abstract

The embodiment of the invention discloses an energy efficiency prediction method and device based on a solar photovoltaic panel and a medium, belongs to the technical field of solar photovoltaic, and solves the problem that the energy efficiency of the photovoltaic panel is difficult to accurately and reliably predict in the prior art. Comprising the steps of performing feature extraction on historical operation data of a solar photovoltaic panel to be tested, and constructing a multi-dimensional feature vector; clustering and grouping the multi-dimensional feature vectors to obtain feature vector data sets corresponding to different weather conditions respectively; training the fusion model to obtain a trained energy efficiency joint optimization model; inputting the current operation data into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; adjusting the first photovoltaic energy efficiency prediction interval based on the equipment operation state data to obtain a second photovoltaic energy efficiency prediction interval; and determining a spatial dependency relationship with the adjacent solar photovoltaic panel power station through the space-time diagram convolutional network, and detecting the second photovoltaic energy efficiency prediction interval based on the dependency relationship.

Description

Technical Field

[0001] The present application relates to the field of solar photovoltaic technology, and in particular to a method, device and medium for predicting energy efficiency based on solar photovoltaic panels. Background Art

[0002] With the growing global demand for renewable energy, solar photovoltaic panels, as a key component of clean energy, have become a research hotspot for energy efficiency prediction and optimization. Accurately predicting the energy efficiency of solar photovoltaic panels not only helps optimize energy distribution and improve power generation efficiency, but also provides strong support for the stable operation of power systems.

[0003] However, existing technologies for predicting solar photovoltaic panel energy efficiency mostly rely on a single dimension of historical operating data. This limitation makes it difficult for prediction models to fully capture the key factors affecting photovoltaic panel energy efficiency, thereby reducing the accuracy and reliability of predictions. This flaw limits the effectiveness of prediction models in complex and changing environments, making it difficult to accurately and reliably predict photovoltaic panel energy efficiency data. Summary of the Invention

[0004] The embodiments of the present application provide a method, device and medium for energy efficiency prediction based on solar photovoltaic panels, which are used to solve the following technical problems: when predicting the energy efficiency of solar photovoltaic panels, the existing technology mostly relies on historical operating data of a single dimension, making it difficult to accurately and reliably predict the energy efficiency data of photovoltaic panels.

[0005] The embodiments of this application adopt the following technical solutions:

[0006] The embodiment of the present application provides an energy efficiency prediction method based on solar photovoltaic panels. The method comprises the following steps: extracting features from historical operating data corresponding to the solar photovoltaic panel to be tested, and constructing a multidimensional feature vector; clustering and grouping the multidimensional feature vector based on weather data in the historical operating data, and obtaining feature vector data sets corresponding to different weather conditions; training a fusion model of particle swarm optimization and genetic algorithm based on different feature vector data sets, and obtaining a trained energy efficiency joint optimization model; inputting current operating data corresponding to the solar photovoltaic panel to be tested into the trained energy efficiency joint optimization model, and obtaining a first photovoltaic energy efficiency prediction interval; obtaining device operating status data of the solar photovoltaic panel to be tested, and adjusting the first photovoltaic energy efficiency prediction interval based on the device operating status data, and obtaining a second photovoltaic energy efficiency prediction interval; constructing a spatiotemporal graph convolutional network based on the geographical location of the power station, determining the spatial dependency relationship between adjacent solar photovoltaic panel power stations through the spatiotemporal graph convolutional network, detecting the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determining the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

[0007] The embodiment of the present application can more comprehensively capture the various factors that affect the energy efficiency of photovoltaic panels by extracting multidimensional feature vectors from historical operating data, cluster and group the multidimensional feature vectors based on weather data, and construct a more refined feature vector data set for different weather conditions, thereby improving the adaptability and accuracy of the prediction model under different weather conditions. By training with a fusion model of particle swarm optimization and genetic algorithm, it is possible to more effectively explore the multidimensional feature space and find a better energy efficiency optimization strategy, thereby improving the prediction accuracy and generalization ability of the prediction model. By constructing a spatiotemporal graph convolutional network, it is possible to capture the spatial dependency between adjacent solar photovoltaic panel power stations, and detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency, which can further verify the rationality of the prediction results and improve the credibility of the prediction.

[0008] In one implementation of the present application, multidimensional feature vectors are clustered and grouped based on weather data in historical operation data to obtain feature vector data sets corresponding to different weather conditions, specifically including: dividing the historical operation data based on a preset time period to obtain multiple sub-data segments, and determining the weather data corresponding to the multiple sub-data segments; clustering the sub-data segments corresponding to the same weather data to obtain multiple first sub-data segment sets; performing time series alignment on the sub-data segments in the multiple first sub-data segment sets, performing multi-time series similarity processing on each sub-data segment based on the aligned data, screening the first sub-data segment set based on the similarity results to obtain a second sub-data segment set; and determining the feature vector data sets corresponding to the multiple second sub-data segment sets based on the feature vectors corresponding to each sub-data segment.

[0009] In one implementation of the present application, a fusion model of particle swarm optimization and genetic algorithm is trained based on different feature vector data sets to obtain a trained energy efficiency joint optimization model, specifically including: real number encoding of weather parameters in photovoltaic energy efficiency prediction so that the particle position vectors in the particle swarm and the chromosomes in the genetic algorithm are in the same coding space; through a preset information interaction window, after a preset number of iterations, the optimal particle information in the particle swarm is passed to the population of the genetic algorithm as initialization information of some chromosomes, and the chromosome information in the genetic algorithm whose fitness is greater than a preset fitness threshold is fed back to the particle swarm; the inertia weight and learning factor of the particle swarm algorithm, and the crossover probability and mutation probability in the genetic algorithm are adaptively adjusted to obtain a fusion model; multiple preset indicators in photovoltaic energy efficiency prediction are used as optimization targets, and the fusion model is trained using a multi-objective optimization algorithm to obtain a trained energy efficiency joint optimization model; wherein the preset indicators include at least one of prediction accuracy, prediction stability and calculation time.

[0010] In one implementation of the present application, device operating status data of a solar photovoltaic panel to be tested is obtained, and a first photovoltaic energy efficiency prediction interval is adjusted based on the device operating status data to obtain a second photovoltaic energy efficiency prediction interval, specifically including: inputting the device operating status data of the solar photovoltaic panel to be tested into a preset device status prediction model to output a predicted device operating status corresponding to the solar photovoltaic panel to be tested through the preset device status prediction model; dividing the predicted device operating status based on a time series to obtain a plurality of predicted device operating sub-period states; determining status information and operating information corresponding to different components based on each predicted device operating sub-period state; determining a corresponding weight value set in a preset weight adjustment table based on identifications of different components; determining a corresponding weight value in a weight value set based on the status information and operating information; combining the weight values corresponding to each predicted device operating sub-period state based on the time series to construct a dynamic weight value corresponding to the solar photovoltaic panel to be tested; and adjusting the first photovoltaic energy efficiency prediction interval through the dynamic weight value to obtain a second photovoltaic energy efficiency prediction interval.

[0011] In one implementation of the present application, the first photovoltaic energy efficiency prediction interval is adjusted by a dynamic weight value to obtain a second photovoltaic energy efficiency prediction interval, specifically including: based on the function:

[0012]

[0013] Determine the adjusted photovoltaic energy efficiency forecast value corresponding to each forecast equipment operation sub-period;

[0014] Function-based:

[0015]

[0016] Determine the average value of the adjusted photovoltaic energy efficiency prediction value corresponding to the second photovoltaic energy efficiency prediction interval; the second photovoltaic energy efficiency prediction interval is:

[0017]

[0018] Among them, E adj,t represents the adjusted PV energy efficiency forecast value for the t-th sub-period; N Indicates the total number of components; W i,t represents the weight value of the i-th component in the t-th sub-period, E comp,i,t represents the contribution of the i-th component to the overall energy efficiency in the t-th sub-period; E second,avg represents the average value of the photovoltaic energy efficiency prediction value corresponding to the second photovoltaic energy efficiency prediction interval; T is the total number of sub-periods; ΔE is the width of the prediction interval.

[0019] In one implementation of the present application, a spatiotemporal graph convolutional network is constructed based on the geographical location of the power station, and the spatial dependency relationship between adjacent solar photovoltaic panel power stations is determined through the spatiotemporal graph convolutional network, specifically including: obtaining the latitude and longitude coordinates of the adjacent solar photovoltaic panel power stations, and constructing a solar photovoltaic panel power station spatial map through the latitude and longitude coordinates; in the solar photovoltaic panel power station spatial map, determining the initial eigenvector of each node; based on the distance between the solar photovoltaic panel to be tested and the adjacent solar photovoltaic panel power station, constructing a dynamic adjacency matrix to determine the eigenvector update representation corresponding to the solar photovoltaic panel to be tested through the dynamic adjacency matrix; adjusting the eigenvector update representation through a preset spatial prediction error distribution evaluation function to determine the spatial dependency relationship with the adjacent solar photovoltaic panel power station through the adjusted eigenvector representation.

[0020] In one implementation of the present application, the feature vector update representation is adjusted by presetting a spatial prediction error distribution evaluation function, specifically including: using the function:

[0021]

[0022] Adjust the updated representation of the feature vector; where F updated is the updated representation matrix of the adjusted eigenvector; A dynamic is a dynamic adjacency matrix built based on distance; Y true is the spatial dependency index; N is the number of nodes in the spatial graph of solar photovoltaic power plants; is the predicted feature vector; λ is the regularization parameter; KL is the divergence term; σ i is the standard deviation of the prediction error of the eigenvector of node i.

[0023] In one implementation of the present application, the second photovoltaic energy efficiency prediction interval is detected based on the spatial dependency relationship, specifically including: determining the third photovoltaic energy efficiency prediction interval of the solar photovoltaic panel to be tested based on the spatial dependency relationship; determining the photovoltaic energy efficiency influencing parameters based on the equipment operation status data of the solar photovoltaic panel to be tested; adjusting the third photovoltaic energy efficiency prediction interval based on the photovoltaic energy efficiency influencing parameters to obtain an energy efficiency reference interval; comparing the energy efficiency reference interval with the second photovoltaic energy efficiency prediction interval, and determining the detection result based on the comparison result and a preset error threshold.

[0024] The embodiment of the present application provides an energy efficiency prediction device based on solar photovoltaic panels, comprising: at least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to: extract features from historical operating data corresponding to the solar photovoltaic panel to be tested, and construct a multidimensional feature vector; cluster and group the multidimensional feature vectors based on weather data in the historical operating data, and obtain feature vector data sets corresponding to different weather conditions; and perform particle swarm optimization and genetic algorithm fusion model analysis based on different feature vector data sets. Training is performed to obtain a trained energy efficiency joint optimization model; the current operating data corresponding to the solar photovoltaic panel to be tested is input into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; the equipment operation status data of the solar photovoltaic panel to be tested is obtained, and the first photovoltaic energy efficiency prediction interval is adjusted based on the equipment operation status data to obtain a second photovoltaic energy efficiency prediction interval; a spatiotemporal graph convolutional network is constructed according to the geographical location of the power station, and the spatial dependency relationship between adjacent solar photovoltaic panel power stations is determined through the spatiotemporal graph convolutional network, so as to detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

[0025] An embodiment of the present application provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to: extract features from acquired historical operating data corresponding to a solar photovoltaic panel to be tested to construct a multidimensional feature vector; cluster and group the multidimensional feature vectors based on weather data in the historical operating data to obtain feature vector data sets corresponding to different weather conditions; train a fusion model of particle swarm optimization and genetic algorithm based on different feature vector data sets to obtain a trained energy efficiency joint optimization model; input current operating data corresponding to the solar photovoltaic panel to be tested into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; obtain device operating status data of the solar photovoltaic panel to be tested, and adjust the first photovoltaic energy efficiency prediction interval based on the device operating status data to obtain a second photovoltaic energy efficiency prediction interval; construct a spatiotemporal graph convolutional network based on the geographical location of the power station, determine the spatial dependency relationship between adjacent solar photovoltaic panel power stations through the spatiotemporal graph convolutional network, detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

[0026] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: the embodiments of the present application can more comprehensively capture the various factors affecting the energy efficiency of photovoltaic panels by extracting multidimensional feature vectors from historical operating data, cluster and group the multidimensional feature vectors based on weather data, and construct a more refined feature vector data set for different weather conditions, thereby improving the adaptability and accuracy of the prediction model under different weather conditions. By training using a fusion model of particle swarm optimization and genetic algorithm, it is possible to more effectively explore the multidimensional feature space and find a better energy efficiency optimization strategy, thereby improving the prediction accuracy and generalization ability of the prediction model. By constructing a spatiotemporal graph convolutional network, the spatial dependency between adjacent solar photovoltaic panel power stations can be captured, and the second photovoltaic energy efficiency prediction interval can be detected based on the spatial dependency, which can further verify the rationality of the prediction results and improve the credibility of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present application. For those skilled in the art, other drawings can be obtained based on these drawings without inventive work. In the drawings:

[0028] Figure 1 A flow chart of a method for predicting energy efficiency based on solar photovoltaic panels provided in an embodiment of the present application;

[0029] Figure 2 A schematic structural diagram of an energy efficiency prediction device based on solar photovoltaic panels provided in an embodiment of the present application.

[0030] Reference numerals:

[0031] 200: Energy efficiency prediction device based on solar photovoltaic panels, 201: Processor, 202: Memory. DETAILED DESCRIPTION

[0032] The embodiments of the present application provide a method, device, and medium for predicting energy efficiency based on solar photovoltaic panels.

[0033] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0034] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0035] Figure 1 A flow chart of a method for predicting energy efficiency based on solar photovoltaic panels provided in an embodiment of the present application is shown in FIG. Figure 1 As shown in FIG, the energy efficiency prediction method based on solar photovoltaic panels includes the following steps:

[0036] Step 101: extract features from the acquired historical operating data corresponding to the solar photovoltaic panel to be tested, and construct a multi-dimensional feature vector.

[0037] In one implementation of the present application, historical weather data of the area where the photovoltaic panels are located, such as light intensity, temperature, wind speed, wind direction, humidity, etc., are collected. The historical power generation power, operating voltage, operating current, temperature monitoring data, and fault records of the photovoltaic panels are collected. Features closely related to the power generation efficiency of the photovoltaic panels are extracted from the historical weather data. For example, the daily average light intensity, maximum temperature, minimum temperature, average wind speed, etc. can be calculated. Features reflecting the operating health of the photovoltaic panels are extracted from the historical equipment status data. For example, the average power generation power of the photovoltaic panels, the stability index of the operating voltage and operating current, the range of temperature fluctuation, etc. can be calculated. The extracted weather features and equipment status features are combined to construct a multidimensional feature vector. Each feature vector contains a series of characteristic values related to the energy efficiency of the photovoltaic panels. These characteristic values together describe the operating status and energy efficiency performance of the photovoltaic panels in a specific time period.

[0038] Step 102: Based on the weather data in the historical operation data, cluster and group the multidimensional feature vectors to obtain feature vector data sets corresponding to different weather conditions.

[0039] In one implementation of the present application, historical operating data is divided based on a preset time period to obtain multiple sub-data segments, and weather data corresponding to each of the multiple sub-data segments is determined. The sub-data segments corresponding to the same weather data are clustered and divided to obtain multiple first sub-data segment sets. Time series alignment is performed on the sub-data segments in the multiple first sub-data segment sets. Based on the aligned data, multi-time series similarity processing is performed on each sub-data segment. Based on the similarity results, the first sub-data segment set is screened to obtain a second sub-data segment set. Based on the feature vectors corresponding to each sub-data segment, feature vector datasets corresponding to the multiple second sub-data segment sets are determined.

[0040] Specifically, historical operating data within a preset time period is divided into multiple sub-segments in chronological order. The length of the sub-segments can be adjusted based on analysis requirements, such as daily, weekly, or monthly. For each sub-segment, corresponding weather data, such as light intensity, temperature, wind speed, and humidity, is extracted. Sub-segments with the same or similar weather data are clustered to form multiple first sub-segment sets.

[0041] Furthermore, since different sub-data segments may be difficult to compare directly due to different time starting and ending points, it is necessary to perform time series alignment on the sub-data segments in the first sub-data segment set after clustering. The alignment method in the embodiment of the present application can be based on time point alignment, such as data at the same time of day, or it can be based on time window alignment, such as the same time period within each sub-data segment. The aligned sub-data segments contain multiple time series variables, such as power generation power, operating voltage, operating current, etc. Euclidean distance processing is performed on different data segments, and multi-time series similarity is determined based on the distance results. Based on the results of the multi-time series similarity processing, the first sub-data segment set is screened, and the sub-data segment set with higher similarity is retained to form the second sub-data segment set. Based on the feature vectors corresponding to each sub-data segment in the second sub-data segment set, a feature vector data set is constructed.

[0042] Step 103: Based on different feature vector data sets, the fusion model of the particle swarm optimization and the genetic algorithm is trained to obtain a trained energy efficiency joint optimization model.

[0043] In one implementation of the present application, the weather parameters in the photovoltaic energy efficiency prediction are encoded with real numbers so that the particle position vectors in the particle swarm and the chromosomes in the genetic algorithm are in the same coding space. Through a preset information interaction window, after a preset number of iterations, the optimal particle information in the particle swarm is passed to the population of the genetic algorithm as the initialization information of some chromosomes, and the chromosome information in the genetic algorithm whose fitness is greater than the preset fitness threshold is fed back to the particle swarm. The inertia weight and learning factor of the particle swarm algorithm, as well as the crossover probability and mutation probability in the genetic algorithm are adaptively adjusted to obtain a fusion model. Multiple preset indicators in the photovoltaic energy efficiency prediction are used as optimization targets, and the fusion model is trained using a multi-objective optimization algorithm to obtain a trained energy efficiency joint optimization model; wherein the preset indicators include at least one of prediction accuracy, prediction stability and calculation time.

[0044] Specifically, the embodiments of the present application encode weather parameters used in photovoltaic energy efficiency prediction, such as light intensity, temperature, and wind speed, as real numbers for use in both the particle swarm algorithm and the genetic algorithm. This encoding method allows the particle position vectors in the particle swarm and the chromosomes in the genetic algorithm to be represented in the same coding space, thereby facilitating information exchange and collaborative optimization between the algorithms. The embodiments of the present application provide a preset information exchange window. During each iteration, when a preset number of iterations is reached, the two algorithms pause their respective iterations and exchange information through this window. Specifically, during each information exchange, the optimal particle information in the particle swarm algorithm—that is, the optimal solution for the current number of iterations—is transmitted to the genetic algorithm population as initialization information for some chromosomes. This helps the genetic algorithm quickly converge to a more optimal solution space. Simultaneously, information about chromosomes in the genetic algorithm whose fitness exceeds a preset fitness threshold is fed back to the particle swarm algorithm to update some of the particle position vectors in the particle swarm. This helps the particle swarm algorithm escape local optimal solutions and improve its global search capabilities.

[0045] Furthermore, in order to further improve the performance of the model, the embodiment of the present application adaptively adjusts the inertia weight and learning factor of the particle swarm algorithm, as well as the crossover probability and mutation probability in the genetic algorithm. Specifically, the inertia weight determines the tendency of the particle to maintain its current speed in the search space. A larger inertia weight helps the particle explore new search areas, while a smaller inertia weight makes the particle more focused on the fine search of the current search area. As the number of iterations increases, the inertia weight is gradually reduced. In the early stage, a larger inertia weight encourages global search; in the later stage, a smaller inertia weight promotes local convergence. When the diversity of the particle swarm is high, a larger inertia weight is used to maintain exploratory power; when the diversity decreases, the inertia weight is reduced to enhance utilization. The learning factors c1 and c2 represent the weights of the particle learning to its own historical best position and the global best position, respectively. When the difference in particle fitness is large, c1 is increased to encourage individual exploration; when the difference is small, c2 is increased to accelerate convergence to the global best position. The crossover probability determines the probability that two parent chromosomes will produce a daughter chromosome. For individuals with low fitness, the crossover probability is increased to increase genetic diversity; for individuals with high fitness, the crossover probability is reduced to protect excellent genes. In the early stages of the algorithm, a higher crossover probability facilitates rapid exploration of the solution space; as the number of iterations increases, the crossover probability is gradually reduced to promote convergence. The mutation probability determines the probability of a gene mutation on a chromosome. During the algorithm's operation, the mutation probability is dynamically adjusted based on population diversity. When diversity is low, the mutation probability is increased to introduce new genotypes; when diversity is high, the mutation probability is reduced to reduce unnecessary search overhead. For individuals with extremely low fitness, the mutation probability is increased to try to escape the local optimal solution; for individuals with high fitness, the mutation probability is kept low to maintain their excellent characteristics.

[0046] After completing algorithm fusion and information exchange, the fusion model is trained using a multi-objective optimization algorithm, taking multiple preset metrics for PV energy efficiency prediction, such as prediction accuracy, prediction stability, and computation time, as optimization targets. This multi-objective optimization algorithm aims to find a set of solutions that achieve an optimal balance between multiple objectives. Through the training process, the fusion model gradually adapts to the needs of PV energy efficiency prediction and demonstrates excellent performance in terms of prediction accuracy, prediction stability, and computation time. Ultimately, a trained energy efficiency joint optimization model is obtained, which can be used in practical PV energy efficiency prediction and optimization tasks.

[0047] Step 104: Input the current operating data corresponding to the solar photovoltaic panel to be tested into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval.

[0048] In one implementation of the present application, current operating data of the solar photovoltaic panel under test is obtained. This data may include environmental parameters such as light intensity, temperature, humidity, and wind speed, as well as electrical parameters such as the panel's output voltage, current, and power. The obtained data is preprocessed, including steps such as data cleaning and data normalization, to ensure data quality and consistency. The processed data is input into the trained energy efficiency joint optimization model, which then outputs a first photovoltaic energy efficiency prediction interval.

[0049] Step 105: Acquire device operating status data of the solar photovoltaic panel to be tested, and adjust the first photovoltaic energy efficiency prediction interval based on the device operating status data to obtain a second photovoltaic energy efficiency prediction interval.

[0050] In one implementation of the present application, the device operating status data of the solar photovoltaic panel to be tested is input into a preset device status prediction model, so as to output the predicted device operating status corresponding to the solar photovoltaic panel to be tested through the preset device status prediction model. Based on the time series, the predicted device operating status is divided to obtain a plurality of predicted device operating sub-period states. Based on each predicted device operating sub-period state, the status information and operating information corresponding to different components are determined. Based on the identification of different components, a corresponding set of weight values is determined in a preset weight adjustment table. Based on the status information and operating information, a corresponding weight value is determined in the weight value set. Based on the time series, the weight values corresponding to each predicted device operating sub-period state are combined to construct a dynamic weight value corresponding to the solar photovoltaic panel to be tested. The first photovoltaic energy efficiency prediction interval is adjusted by the dynamic weight value to obtain a second photovoltaic energy efficiency prediction interval.

[0051] Specifically, in the embodiments of the present application, a preset device state prediction model is provided. The device operating state data of the solar photovoltaic panel to be tested is input into the preset device state prediction model to predict the device operating state of the solar photovoltaic panel to be tested. The training process of the preset device state prediction model is as follows: using historical device operating state data samples as input and the device state operating data of the next period corresponding to the input samples as output, the preset prediction model is trained to obtain the preset device state prediction model.

[0052] Furthermore, the predicted equipment operating status is divided according to the time series to form a plurality of predicted equipment operating sub-period states. For each predicted equipment operating sub-period state, the status information and operating information of different components in the photovoltaic panel, such as battery cells, inverters, brackets, etc., are further determined. This information includes the temperature, voltage, current, power, etc. of the components. The embodiment of the present application establishes a preset weight adjustment table, which contains a set of weight values of different components in different states. These weight values reflect the degree of influence of different components on the overall energy efficiency prediction interval. According to the component status information and operating information in each predicted equipment operating sub-period state, the corresponding weight value is searched and determined in the weight adjustment table.

[0053] Furthermore, the weight values corresponding to each predicted device operating sub-period are combined in a time series to construct a dynamic weight value for the solar photovoltaic panel under test over the entire prediction period. This dynamic weight value reflects the importance of energy efficiency prediction for the device in different time periods and under different component conditions. Using this constructed dynamic weight value, the different time periods of the initially obtained first photovoltaic energy efficiency prediction interval are adjusted to obtain a more accurate and reliable second photovoltaic energy efficiency prediction interval.

[0054] Furthermore, the first photovoltaic energy efficiency prediction interval is adjusted by the dynamic weight value to obtain the second photovoltaic energy efficiency prediction interval. The specific process is as follows:

[0055] Function-based:

[0056]

[0057] Determine an adjusted photovoltaic energy efficiency prediction value corresponding to each prediction device operation sub-period within the second photovoltaic energy efficiency prediction interval;

[0058] Function-based:

[0059]

[0060] Determine the average value of the adjusted photovoltaic energy efficiency prediction value corresponding to the second photovoltaic energy efficiency prediction interval, and then the second photovoltaic energy efficiency prediction interval is:

[0061]

[0062] Among them, E adj,t represents the adjusted PV energy efficiency forecast value in the t-th sub-period; N represents the total number of components; W i,t represents the weight value of the i-th component in the t-th sub-period, E comp,i,t represents the contribution of the i-th component to the overall energy efficiency in the t-th sub-period; E second,avg represents the average value of the photovoltaic energy efficiency prediction value corresponding to the second photovoltaic energy efficiency prediction interval; T is the total number of sub-periods; ΔE is the width of the prediction interval.

[0063] Step 106: Construct a spatiotemporal graph convolutional network based on the geographical location of the power station, determine the spatial dependency between adjacent solar photovoltaic panel power stations through the spatiotemporal graph convolutional network, detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the test passes.

[0064] In one implementation of the present application, the longitude and latitude coordinates of adjacent solar photovoltaic panel power stations are obtained, and a spatial map of the solar photovoltaic panel power station is constructed using the longitude and latitude coordinates. In the spatial map of the solar photovoltaic panel power station, the initial eigenvector of each node is determined. Based on the distance between the solar photovoltaic panel to be tested and the adjacent solar photovoltaic panel power station, a dynamic adjacency matrix is constructed to determine the updated eigenvector representation corresponding to the solar photovoltaic panel to be tested using the dynamic adjacency matrix. The updated eigenvector representation is adjusted using a preset spatial prediction error distribution evaluation function to determine the spatial dependency relationship between the solar photovoltaic panel power station and the adjacent solar photovoltaic panel power station using the adjusted eigenvector representation.

[0065] Specifically, the longitude and latitude coordinates of adjacent solar photovoltaic power plants are obtained using a geographic information system or other positioning technology. These coordinates are then used to construct a spatial map of solar photovoltaic power plants. In this map, each power plant represents a node, and the connections between nodes represent the spatial relationships between the plants, such as distance. Within this constructed spatial map, an initial feature vector is determined for each node (i.e., each power plant). These feature vectors can include information such as the plant's power generation, capacity, installation year, and geographic location.

[0066] Furthermore, a dynamic adjacency matrix is used to represent the distance relationship between the solar PV panel under test and its neighboring power plants. The matrix elements are weighted based on the distance between the power plants to reflect spatial proximity. The closer the distance, the larger the corresponding element value, indicating a stronger connection; the farther the distance, the smaller the element value. Using the dynamic adjacency matrix to update the eigenvector representation means adjusting the eigenvector of the solar PV panel under test based on its relationship with the neighboring power plants.

[0067] Specifically, we first obtain the distance between adjacent power stations and construct a Gaussian kernel function mapping based on the distance:

[0068]

[0069] Where σ is the bandwidth parameter, usually 1 / 2 of the average distance; d is the distance between adjacent power stations; distance threshold method: set the threshold D, when d ij When ≤D, w ij =1, otherwise w ij =0.

[0070] Time window adaptation: The adjacency matrix is recalculated every 30 minutes to reflect short-term spatial correlation changes.

[0071] Environmental factor modulation: Introducing the light intensity difference ΔI and temperature difference ΔT, the weight correction formula is:

[0072]

[0073] in, is the revised weight formula; is the weight formula before correction; α is the light intensity difference weight value; β is the temperature difference weight value.

[0074] The eigenvectors can be adjusted through the modified weights and dynamic adjacency matrix.

[0075] For example, some characteristics of adjacent power stations, such as high power generation efficiency, may affect the eigenvector of the power station under test, causing it to be updated in a more optimal direction. Suppose the solar photovoltaic panel under test is D, and its adjacent power stations are A and B. The distance between D and A is 10 kilometers, and the distance between D and B is 15 kilometers. When constructing the dynamic adjacency matrix, it is assumed that the element value is set to 1 for distances within 10 kilometers, 0.5 for distances between 10-20 kilometers, and 0 for distances above 20 kilometers. Then, in the dynamic adjacency matrix, the element value corresponding to D and A is 1, and the element value corresponding to D and B is 0.5. When updating the eigenvector of D, if A has a high power generation efficiency, the eigenvalue of D's power generation efficiency may be closer to A's power generation efficiency value based on the relationship in the adjacency matrix (the connection strength between A and D is strong), thereby obtaining the updated eigenvector.

[0076] In one implementation of the present application, the feature vector update representation is adjusted by presetting a spatial prediction error distribution evaluation function, specifically including:

[0077] Through the function:

[0078]

[0079] Adjust the updated representation of the feature vector;

[0080] Among them, F updated is the updated representation matrix of the adjusted eigenvector; A dynamic is a dynamic adjacency matrix built based on distance; Y true is the spatial dependency index; N is the number of nodes in the spatial graph of solar photovoltaic power plants; is the predicted feature vector; λ is the regularization parameter; KL is the divergence term; σ i is the standard deviation of the prediction error of the eigenvector of node i.

[0081] In one implementation of the present application, a third photovoltaic energy efficiency prediction interval is determined for the solar photovoltaic panel under test based on the spatial dependency relationship. A photovoltaic energy efficiency influencing parameter is determined based on the device operating status data of the solar photovoltaic panel under test. Based on the photovoltaic energy efficiency influencing parameter, the third photovoltaic energy efficiency prediction interval is adjusted to obtain an energy efficiency reference interval. The energy efficiency reference interval is compared with the second photovoltaic energy efficiency prediction interval, and a test result is determined based on the comparison result and a preset error threshold.

[0082] Specifically, the feature vector is adjusted based on the relationship between the solar photovoltaic panel under test and adjacent power plants. The adjusted feature vector is then used for prediction to obtain a third photovoltaic energy efficiency prediction range. The device operating status data contains a wealth of information about the solar photovoltaic panel under test, such as panel temperature, inverter efficiency, and component aging. These factors all affect the panel's energy efficiency. By comparing this device operating status data with a preset parameter adjustment table, the corresponding photovoltaic efficiency influencing parameters can be determined. This preset parameter adjustment table includes multiple device operating states and parameters corresponding to different device operating states. Based on the determined photovoltaic efficiency influencing parameters, the previously obtained third photovoltaic energy efficiency prediction range is revised. By taking these influencing parameters into account, the actual energy efficiency range of the photovoltaic panel can be more accurately estimated, resulting in an energy efficiency reference range. This range comprehensively considers the impact of spatial dependencies and the device's own operating status on energy efficiency.

[0083] Furthermore, the second PV energy efficiency prediction interval is obtained by inputting the currently trained energy efficiency joint optimization model into the current operating data. The energy efficiency reference interval is compared with the second PV energy efficiency prediction interval to examine the difference between the two. The preset error threshold is a pre-set allowable error range. If the difference between the two intervals falls within this threshold, the test is considered passed, indicating a reliable prediction result. If the difference exceeds the threshold, the test fails, and further analysis may be required, such as whether the model needs optimization or whether there are any data anomalies.

[0084] Figure 2This is a schematic diagram of the structure of an energy efficiency prediction device based on solar photovoltaic panels provided in an embodiment of the present application. Figure 2 As shown, the energy efficiency prediction device 200 based on solar photovoltaic panels includes: at least one processor 201; and a memory 202 in communication with the at least one processor 201; wherein the memory 202 stores instructions that can be executed by the at least one processor 201, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: extract features from the acquired historical operating data corresponding to the solar photovoltaic panel to be tested, and construct a multidimensional feature vector; cluster and group the multidimensional feature vectors based on the weather data in the historical operating data to obtain feature vector data sets corresponding to different weather conditions; and perform particle swarm optimization and genetic algorithm based on different feature vector data sets. The fusion model of the method is trained to obtain a trained energy efficiency joint optimization model; the current operation data corresponding to the solar photovoltaic panel to be tested is input into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; the equipment operation status data of the solar photovoltaic panel to be tested is obtained, and the first photovoltaic energy efficiency prediction interval is adjusted based on the equipment operation status data to obtain a second photovoltaic energy efficiency prediction interval; a spatiotemporal graph convolution network is constructed according to the geographical location of the power station, and the spatial dependency relationship between adjacent solar photovoltaic panel power stations is determined through the spatiotemporal graph convolution network, so as to detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

[0085] An embodiment of the present application provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to: extract features from acquired historical operating data corresponding to a solar photovoltaic panel to be tested to construct a multidimensional feature vector; cluster and group the multidimensional feature vectors based on weather data in the historical operating data to obtain feature vector data sets corresponding to different weather conditions; train a fusion model of particle swarm optimization and genetic algorithm based on different feature vector data sets to obtain a trained energy efficiency joint optimization model; input current operating data corresponding to the solar photovoltaic panel to be tested into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; obtain device operating status data of the solar photovoltaic panel to be tested, and adjust the first photovoltaic energy efficiency prediction interval based on the device operating status data to obtain a second photovoltaic energy efficiency prediction interval; construct a spatiotemporal graph convolutional network based on the geographical location of the power station, determine the spatial dependency relationship between adjacent solar photovoltaic panel power stations through the spatiotemporal graph convolutional network, detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

[0086] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0087] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. It will be apparent to those skilled in the art that various modifications and variations may be made to the embodiments of the present application. However, such modifications or substitutions do not deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for predicting energy efficiency based on solar photovoltaic panels, characterized in that: The method comprises: Perform feature extraction on the historical operating data corresponding to the solar photovoltaic panels to be tested and construct a multi-dimensional feature vector; Clustering and grouping the multidimensional feature vectors based on the weather data in the historical operation data to obtain feature vector data sets corresponding to different weather conditions; Based on different feature vector data sets, a fusion model of particle swarm optimization and genetic algorithm is trained to obtain a trained energy efficiency joint optimization model; Inputting the current operating data corresponding to the solar photovoltaic panel to be tested into the trained energy efficiency joint optimization model to obtain a first photovoltaic energy efficiency prediction interval; Acquiring device operating status data of the solar photovoltaic panel to be tested, and adjusting the first photovoltaic energy efficiency prediction interval based on the device operating status data to obtain a second photovoltaic energy efficiency prediction interval; A spatiotemporal graph convolutional network is constructed based on the geographical location of the power station, and the spatial dependency relationship between adjacent solar photovoltaic panel power stations is determined through the spatiotemporal graph convolutional network, so as to detect the second photovoltaic energy efficiency prediction interval based on the spatial dependency relationship, and determine the predicted energy efficiency value corresponding to the solar photovoltaic panel to be tested if the detection passes.

2. The energy efficiency prediction method based on solar photovoltaic panels according to claim 1, characterized in that: The multidimensional feature vectors are clustered and grouped based on the weather data in the historical operation data to obtain feature vector data sets corresponding to different weather conditions, specifically including: Dividing the historical operation data based on a preset time period to obtain a plurality of sub-data segments, and determining weather data corresponding to each of the plurality of sub-data segments; Clustering the sub-data segments corresponding to the same weather data to obtain multiple first sub-data segment sets; performing time series alignment on the sub-data segments in the plurality of first sub-data segment sets, performing multi-time series similarity processing on each of the sub-data segments based on the aligned data, and screening the first sub-data segment sets based on the similarity results to obtain a second sub-data segment set; Based on the feature vectors corresponding to the sub-data segments, feature vector data sets corresponding to a plurality of sets of the second sub-data segments are determined.

3. The energy efficiency prediction method based on solar photovoltaic panels according to claim 1, characterized in that: The method of training the fusion model of particle swarm optimization and genetic algorithm based on different feature vector data sets to obtain a trained energy efficiency joint optimization model specifically includes: The weather parameters in photovoltaic energy efficiency prediction are encoded with real numbers so that the particle position vectors in the particle swarm and the chromosomes in the genetic algorithm are in the same encoding space; Through a preset information interaction window, after a preset number of iterations, the optimal particle information in the particle swarm is transmitted to the population of the genetic algorithm as initialization information of some chromosomes, and the chromosome information whose fitness in the genetic algorithm is greater than a preset fitness threshold is fed back to the particle swarm; Adaptively adjusting the inertia weight and learning factor of the particle swarm algorithm, and the crossover probability and mutation probability of the genetic algorithm to obtain a fusion model; Taking multiple preset indicators in the photovoltaic energy efficiency prediction as optimization targets, the fusion model is trained using a multi-objective optimization algorithm to obtain the trained energy efficiency joint optimization model; wherein the preset indicators include at least one of prediction accuracy, prediction stability and calculation time.

4. The energy efficiency prediction method based on solar photovoltaic panels according to claim 1, characterized in that: The acquiring of the device operating status data of the solar photovoltaic panel to be tested, and adjusting the first photovoltaic energy efficiency prediction interval based on the device operating status data to obtain the second photovoltaic energy efficiency prediction interval specifically includes: Inputting the device operating status data of the solar photovoltaic panel to be tested into a preset device state prediction model, so as to output the predicted device operating status corresponding to the solar photovoltaic panel to be tested through the preset device state prediction model; Based on the time series, the predicted device operating state is divided into multiple predicted device operating sub-period states; Based on each of the predicted device operation sub-period states, determining state information and operation information corresponding to different components; Based on the identifications of different components, a corresponding set of weight values is determined in a preset weight adjustment table; Determining a corresponding weight value in the weight value set based on the state information and the operation information; Based on the time series, the weight values corresponding to the operating sub-period states of each of the predicted devices are combined to construct a dynamic weight value corresponding to the solar photovoltaic panel to be tested; The first photovoltaic energy efficiency prediction interval is adjusted according to the dynamic weight value to obtain a second photovoltaic energy efficiency prediction interval.

5. The energy efficiency prediction method based on solar photovoltaic panels according to claim 4, characterized in that: The step of adjusting the first photovoltaic energy efficiency prediction interval by using the dynamic weight value to obtain the second photovoltaic energy efficiency prediction interval specifically includes: Function-based: Determine the adjusted photovoltaic energy efficiency forecast value corresponding to each forecast equipment operation sub-period; Function-based: Determining an average value of the adjusted photovoltaic energy efficiency prediction values corresponding to the second photovoltaic energy efficiency prediction interval; The second photovoltaic energy efficiency prediction interval is: Among them, E adj,t represents the adjusted PV energy efficiency forecast value in the t-th sub-period; N represents the total number of components; W i,t represents the weight value of the i-th component in the t-th sub-period, E comp,i,t represents the contribution of the i-th component to the overall energy efficiency in the t-th sub-period; E second,avg represents the average value of the photovoltaic energy efficiency prediction value corresponding to the second photovoltaic energy efficiency prediction interval; T is the total number of sub-periods; ΔE is the width of the prediction interval.

6. The energy efficiency prediction method based on solar photovoltaic panels according to claim 1, characterized in that: The step of constructing a spatiotemporal graph convolutional network based on the geographical location of the power stations and determining the spatial dependency between adjacent solar photovoltaic panel power stations through the spatiotemporal graph convolutional network specifically includes: Obtaining the longitude and latitude coordinates of the adjacent solar photovoltaic panel power station, and constructing a spatial map of the solar photovoltaic panel power station using the longitude and latitude coordinates; In the solar photovoltaic panel power station spatial graph, determining the initial eigenvector of each node; Based on the distance between the solar photovoltaic panel to be tested and the adjacent solar photovoltaic panel power station, a dynamic adjacency matrix is constructed to determine the updated representation of the eigenvector corresponding to the solar photovoltaic panel to be tested through the dynamic adjacency matrix; The feature vector update representation is adjusted by a preset spatial prediction error distribution evaluation function, so as to determine the spatial dependency relationship between the feature vector and the adjacent solar photovoltaic panel power station through the adjusted feature vector representation.

7. The method for predicting energy efficiency based on solar photovoltaic panels according to claim 6, characterized in that: The adjusting the updated representation of the feature vector by presetting a spatial prediction error distribution evaluation function specifically includes: Through the function: adjusting the updated representation of the feature vector; Among them, F updated is the updated representation matrix of the adjusted eigenvector; A dynamic is a dynamic adjacency matrix built based on distance; Y true is the spatial dependency index; N is the number of nodes in the spatial graph of solar photovoltaic power plants; is the predicted feature vector; λ is the regularization parameter; KL is the divergence term; σ i is the standard deviation of the prediction error of the eigenvector of node i.

8. The method for predicting energy efficiency based on solar photovoltaic panels according to claim 1, characterized in that: The detecting the second photovoltaic energy efficiency prediction interval based on the spatial dependency specifically includes: Determining a third photovoltaic energy efficiency prediction interval of the solar photovoltaic panel to be tested according to the spatial dependency relationship; Determining photovoltaic energy efficiency influencing parameters based on the device operating status data of the solar photovoltaic panel to be tested; Based on the photovoltaic energy efficiency influencing parameter, adjusting the third photovoltaic energy efficiency prediction interval to obtain an energy efficiency reference interval; The energy efficiency reference interval is compared with the second photovoltaic energy efficiency prediction interval, and a detection result is determined based on the comparison result and a preset error threshold.

9. An energy efficiency prediction device based on solar photovoltaic panels, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 8.

Images