Rapid household photovoltaic power station theoretical power prediction method, system and device and storage medium

Abstract

The invention discloses a rapid household photovoltaic power station theoretical power prediction method, system and device and a storage medium, and belongs to the technical field of power station power prediction, and the method comprises the steps: obtaining historical power station data, carrying out the preprocessing of the historical power station data, and obtaining first historical power station data; labeling power stations based on the first historical power station data, and classifying the power stations to obtain each region; selecting a meteorological sampling point for each region, and multiplexing meteorological information of a neighbor region of each region to obtain a power station sample point; performing data processing on the power station sample points to obtain an input sequence, and constructing a neural network prediction model of each region based on the input sequence; and performing power prediction on the neural network prediction model, and optimizing model parameters through online learning to obtain a power prediction result of each region, thereby having better guiding significance for actual production.

Description

Technical Field

[0001] The present invention relates to the technical field of power station power prediction, and in particular to a method, system, device and storage medium for rapid theoretical power prediction of a household photovoltaic power station. Background Art

[0002] The traditional method for predicting the optical power of centralized power stations requires the collection of grid-like numerical weather forecasts within a certain range, and then modeling and machine learning are performed based on these parameters. When applied to household distributed power station scenarios, because a single project is composed of a large number of user power stations, and the power stations are distributed across towns and even across counties and cities, the traditional prediction method requires the collection of a large number of grid-like numerical weather forecast points. Moreover, because the projects cover an extremely wide range of regions, blindly sampling to obtain meteorological data is not particularly representative. The cost of continuous meteorological data samples in multiple regions is huge, and the speed of model training and application is too slow, making online iteration impossible. The deployment cost is too high, and it is difficult to promote in practical applications. Therefore, there is currently no low-cost theoretical power prediction method for household photovoltaic power stations; and traditional power prediction is only deployed on a project (power station) basis. In household distributed scenarios, a management unit often manages a large number of household projects, and each project has thousands of power stations, which are distributed cross-overlappingly. The installation angle, installation method, seasonal shading characteristics, etc. of the power stations are difficult to model household by household, so traditional methods cannot be applied in this scenario. With the rapid growth of household photovoltaic capacity, theoretical power forecasting for household power plants has guiding significance for load management, power trading, and power plant inefficiency analysis. The proposed labeling-aggregation-modeling-classification calculation method significantly reduces model complexity and achieves the following goals: First, rapidity. Through clustering, sampling, and model decomposition, the required model parameters are significantly reduced, enabling rapid theoretical power forecasting for tens of thousands of power plants across a large region. Second, universal accessibility. This method does not require extensive, high-precision gridded numerical weather data; it only requires sampling a subset of numerical prediction points to achieve a conclusion, resulting in low-cost predictions. Furthermore, through clustering and decomposition, the model is often very small and can be deployed on conventional servers or distributed, eliminating the need for large-scale computing power. Third, regional accuracy is achieved. Through weighted calculations, this algorithm, while not able to precisely determine the exact power consumption of a single household, provides relatively accurate predictions for each small load center or region. In practical power grid management, regional load balance is often more important than the performance of individual household power plants, so this approach can meet anticipated needs. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is to solve the problems of high cost, slow speed, inability to iterate online, and difficulty in modeling household-by-household for a large number of distributed power stations in traditional household distributed photovoltaic power station power prediction methods.

[0005] To solve the above technical problems, the present invention provides the following technical solution: a method for rapidly predicting theoretical power of household photovoltaic power stations, comprising: Acquiring historical power plant data, and preprocessing the historical power plant data to obtain first historical power plant data; Labeling the power stations based on the first historical power station data and classifying the power stations to obtain each region; Selecting meteorological sampling points for each area, and reusing meteorological information of neighboring areas of each area to obtain power station sample points; Processing the power station sample points to obtain an input sequence, and constructing a neural network prediction model for each region based on the input sequence; The neural network prediction model is used to perform power prediction, and the model parameters are optimized through online learning to obtain the power prediction results of each area.

[0006] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, the classification of power stations includes: Divide the power station coordinates into several square areas; Calculate the feature points in each area to get the center of the new area; Traverse all regional center feature points and calculate the geometric distance between any two regional capacities. If the geometric distance is less than the first threshold, perform a weighted merge operation on the new feature points of the two regions and modify the regional label of the power station in the region. The weighted merging operation is repeated until the distances between all region centers are smaller than the first threshold.

[0007] The beneficial effects of this preferred technical solution are that, through regional division and weighted merging, the model complexity is simplified, the cost is reduced, the prediction speed and accuracy are improved, and the practicality and scalability of the method are enhanced.

[0008] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, the meteorological sampling points are selected as follows: For any feature area, take the center point and surrounding sample points; Calculate the distance between the surrounding sample points and the center point. If the distance is greater than a second threshold, remove the corresponding sample point from the neighbor list. Obtain the meteorological information of the nearest area around any feature area, use the meteorological information as the prediction input of the corresponding area, and reuse the meteorological information of the neighboring areas of each area.

[0009] The beneficial effect of this preferred technical solution is that by selecting the center point and surrounding sample points and removing sample points that are too far away, while reusing the meteorological information of neighboring areas, the representativeness of meteorological data is improved, the prediction accuracy is enhanced, and no additional sampling costs are added, thereby reducing the overall cost of the prediction system.

[0010] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, constructing the neural network prediction model for each region includes: Merge the power station sample points into characteristic sample points; For each feature point, meteorological data is collected in the corresponding area and dimensionally transformed to obtain an input sequence; According to the input sequence, an artificial neural network of output units is constructed for each region, and the hidden layer, output layer and neuron data of each layer are set to the first multiple of the input to construct a neural network prediction model; Historical meteorological data is obtained and normalized. The normalized historical meteorological data is used as input, and the power of all power stations in each region and the sum of the power of all power stations in each region at the corresponding time are used as targets to train the neural network prediction model. The training operation is repeated for each region, and the model parameters are adjusted to obtain the neural network prediction model for each region.

[0011] The beneficial effect of this preferred technical solution is that by merging power station sample points, collecting meteorological data and performing dimensionality transformation, a neural network prediction model is constructed for each area, which reduces the complexity of the model and improves the prediction accuracy and speed. At the same time, the model performance is optimized through parameter adjustment, achieving efficient and low-cost power prediction.

[0012] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, wherein: performing power prediction on the neural network prediction model includes: Obtain forecast data for each region, sampling point, and day; According to the neighbor relationship, the prediction data is input into the neural network prediction model to make predictions; The neural network prediction model is used to perform predictions through concurrent computing or distributed deployment on multiple terminals. The prediction results are summarized to a designated central node to obtain the power value of each area at the corresponding time.

[0013] The beneficial effect of this preferred technical solution is that it can quickly and efficiently complete power prediction through concurrent computing or distributed deployment, improve the prediction speed and system responsiveness, and enhance the scalability and practicality of the system.

[0014] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, the optimization of model parameters by online learning includes: Get the actual interval power;

[0015] The actual interval power is fed back to the training operation to enable online learning and optimize the model parameters.

[0016] As a preferred solution of the method for rapid theoretical power prediction of household photovoltaic power stations described in the present invention, the power prediction results include: The power value of each area at the corresponding time, calculate the average historical power of each household power station in any area and the average total power of the area; In any area, calculate the average historical power of each power station in the statistical interval and the average total power of the area; Calculate the ratio of the average historical power of each power station in the statistical interval to the average total power of the region; The predicted output power sequence is obtained based on the ratio calculation, and the power station is labeled to predict the power of the corresponding project.

[0017] Another object of the present invention is to provide a rapid household photovoltaic power station theoretical power prediction system.

[0018] To solve the above technical problems, the present invention provides the following technical solutions: a rapid household photovoltaic power station theoretical power prediction system, comprising: a preprocessing module, configured to obtain historical power plant data, and preprocess the historical power plant data to obtain first historical power plant data; a classification module, configured to label the power stations based on the first historical power station data, and classify the power stations to obtain each region; a multiplexing module, configured to select meteorological sampling points for each area, and multiplex meteorological information of neighboring areas of each area to obtain power station sample points; A model building module is used to process data of power station sample points to obtain an input sequence, and to build a neural network prediction model for each region based on the input sequence; The prediction module is used to perform power prediction on the neural network prediction model and optimize the model parameters through online learning to obtain the power prediction results for each area.

[0019] The present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method for quickly predicting the theoretical power of a household photovoltaic power station are implemented.

[0020] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method for quickly predicting the theoretical power of a household photovoltaic power station are implemented.

[0021] The beneficial effects of the present invention are as follows: a method of aggregating capacity features of tens of thousands or even hundreds of thousands of household power stations is adopted, which greatly reduces the input of the prediction network, and through multiple high-concurrency regional models, parallel deployment and calculation are performed by reusing input, thereby improving the system prediction response speed, providing the possibility for fast online training iteration and fast prediction results; by reducing the number of samples, reusing sampling results, reducing the model scale and other methods, the method can be used for stacking and deploying tens of thousands of power generation units on low-cost hardware, which can greatly reduce costs; in actual applications, the meteorological information sampling density and neural network structure (for example, adding a recurrent neural network or a graph neural network) can be appropriately adjusted according to the hardware foundation and accuracy requirements to adapt to different needs; although fewer meteorological data samples are used, the user distribution pattern can be found by using capacity as a weight, and the target can be approached as much as possible. Although the accuracy of a single user is affected to a certain extent, the group accuracy is shown in the region, which has good guiding significance for actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 An overall flow chart of a method for rapidly predicting theoretical power of a household photovoltaic power station provided by one embodiment of the present invention; Figure 2 A site distribution diagram of a method for rapidly predicting theoretical power of household photovoltaic power stations provided by one embodiment of the present invention; Figure 3 A new regional center diagram of a method for rapidly predicting theoretical power of household photovoltaic power stations provided by one embodiment of the present invention; Figure 4 A regional center feature point diagram of a method for rapid theoretical power prediction of a household photovoltaic power station provided by one embodiment of the present invention; Figure 5 A sampling point distribution diagram of a method for rapidly predicting theoretical power of household photovoltaic power stations provided by one embodiment of the present invention; Figure 6 A comparison chart of actual power and predicted power of a method for quickly predicting theoretical power of a household photovoltaic power station provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0025] Example 1, reference Figures 1 to 5 , is an embodiment of the present invention, which provides a method for quickly predicting theoretical power of a household photovoltaic power station, including: S100: Acquire historical power station data, pre-process the historical power station data, and obtain first historical power station data; In the embodiment of the present invention, the first historical power station data is pre-processed historical power station data, and the number of a batch is Household power station, any user power station , corner mark Indicates the Power station, value range ; Specifically, the preprocessing includes recording the power output data of each operating photovoltaic power station every 15 minutes over the past period of time. , where the subscript Indicates the A power station, Indicates that at a certain historical time point, the data of the fault and maintenance period should be eliminated; The capacity and coordinate information of each operating power station are recorded as , 、 ,in, 、 It can be obtained by the latitude and longitude of the power station using the haversine method; like Figure 2 The following shows the site distribution, with approximately 150,000 users, each small Representing a power station, the plane coordinates are displayed using the haversine projection method. For data security, noise data is added; Length is Historical meteorological data series Specific meteorological characteristics can be selected as needed, and should at least include information such as irradiation-related parameters, wind speed, temperature, cloud cover, rainfall, snow accumulation, etc. After selection, the length of the meteorological data sequence is consistent, among which the subscript Indicates the coordinates of a specific area. It does not need to correspond to a power station one by one. The specific location is determined by classifying the power stations. A certain point in history.

[0026] It should be noted that by preprocessing historical power plant data, removing data from fault and maintenance periods, retaining effective power information, and combining it with power plant capacity and coordinate information, the present invention enables more accurate power plant classification and modeling. Furthermore, the introduction of historical meteorological data corresponding to power plant classification further enriches the model input, making power forecasts more accurate and reliable. This provides a high-quality data foundation for subsequent model construction and power forecasting, effectively improving the overall performance and practicality of the forecasting system.

[0027] S200: Labeling the power stations based on the first historical power station data, and classifying the power stations to obtain each region; S300: Select meteorological sampling points for each area, reuse meteorological information of neighboring areas of each area, and obtain power station sample points; S400: Processing the power station sample points to obtain an input sequence, and building a neural network prediction model for each region based on the input sequence; S500: Use the neural network prediction model to perform power prediction, optimize model parameters through online learning, and obtain power prediction results for each area.

[0028] It should be noted that by preprocessing historical power plant data, labeling and classifying power plants, selecting meteorological sampling points and reusing meteorological information from neighboring areas, building a neural network prediction model, and performing power prediction and online learning optimization, we achieved fast, low-cost, and high-precision power prediction for household photovoltaic power plants. This method reduces model complexity and deployment costs, improves prediction speed and accuracy, and enhances the system's flexibility and practicality, providing important guidance for practical grid management and power trading.

[0029] In the embodiment of the present invention, the above step S200 includes the following sub-steps A1-A4; In A1: Divide the power station coordinates into several square areas; In A2: Calculate the feature points in each region to obtain the new region center; In A3: All regional center feature points are traversed, and the geometric distance between the capacities of any two regions is calculated. If the geometric distance is less than the first threshold, the new feature points of the two regions are weighted merged, and the regional label of the power station in the region is modified. In A4: Repeat the weighted merging operation until all region center distances are smaller than the first threshold.

[0030] Specifically, labeling includes labeling each power station and adding a regional label In the classification process, it is used to record the region to which it belongs and to add other tags for management classification, such as project, province, city, etc. It is used for management classification and can quickly restore regionalized power stations to management statistical caliber, but it has little relevance to the calculation process.

[0031] Specifically, the classification of power stations includes setting a distance parameter , The larger the value, the smaller the model size after segmentation, but the prediction accuracy will be reduced; The smaller the model, the higher the prediction accuracy. However, as the number of regional models increases, the total model size increases. During the application process, you can choose according to your needs. For example, you can choose ; According to the selected distance , divide all power station coordinates into several square areas, and the area boundary spacing is , record the region as , the power station labels falling within this area are modified to a unique characteristic integer; For each area, if you fall into the area, power stations, calculate new feature points ( , 、 ), respectively represent the area The longitude and latitude of the installed capacity and capacity weighted center within are expressed as: , , , in, is the installed capacity of all power stations, For the region The total installed capacity of all power stations in The longitude corresponding to the installed capacity of each power station, For the region The longitude of the capacity-weighted center of all power stations within, The latitude corresponding to the installed capacity of each power station, For the region The latitude of the capacity-weighted center of all power stations within, To fall in the area the number of power stations; like Figure 3As shown, a set of dense new regional centers is obtained ( , 、 ).

[0032] The geometric distance judgment operation includes: the first threshold is , for the calculated Regions, traverse all regional center feature points, calculate the geometric distance between any two regional capacities, and the distance is less than , weightedly merge the new feature points of the region, and modify the region label of the power station in the two regions to the one with the smallest center of the two regions; like Figure 4 As shown, the weighted merging operation is repeated until the distance between all region centers is less than For those with too small a capacity (e.g., 500kW), they are merged into the nearest region. At the same time, the region label of the power station within these two regions is modified to the region with the smallest center. Through continuous iteration, the following regional center feature point map is finally obtained. By classifying power stations, New regions, in the iterative process, It keeps decreasing, but the meaning is the same, so it is used Symbol, each regional feature point is recorded as ( , 、 ) Subscript value ; The regional modeling process only needs to be calculated once. If the regional capacity increases slightly, after a new household power station is added, it is only necessary to calculate the regional feature points closest to it and update the regional features. After a large number of new users are added to the region, the geometric distance judgment operation can be repeated.

[0033] It should be noted that through steps such as plant coordinate segmentation, feature point calculation, and weighted merging, a large number of plants are clustered into several regional centers, simplifying the model complexity and reducing computational costs. Furthermore, dynamic adjustment of regional centers and merging operations improve the model's adaptability and prediction accuracy. Furthermore, labeling facilitates management and statistics, enhancing the system's flexibility and practicality, making it suitable for power forecasting of large-scale residential photovoltaic plants.

[0034] In the embodiment of the present invention, the above step S300 includes the following sub-steps B1-B3; In B1: for any feature area, take the center point and surrounding sample points; In B2: Calculate the distance between the surrounding sample points and the center point. If the distance is greater than a second threshold, remove the corresponding sample point from the neighbor list. In B3: obtain the meteorological information of the nearest area around any feature area, use the meteorological information as the prediction input of the corresponding area, and reuse the meteorological information of the neighboring areas of each area.

[0035] Specifically, in any feature area ( , 、 ), take its center point ( , ) and the surrounding four sample points ( , )、( , )、( , )、( , ),in, is the coefficient of the distance to the center, which can be selected between 0.5 and 1; like Figure 5 As shown in the figure, the sampling points with strong correlation are obtained. Different from averaging the sampling points on a large scale, the sampling points obtained are closer to the areas where the installed capacity of power stations is concentrated, which is more representative. Consider any area Some power stations in the region are close to their neighboring areas and have correlation. Meteorological features such as clouds, haze, and rainfall centers around another region will affect the surrounding areas under the influence of atmospheric movement. Reusing the meteorological information of the neighboring areas will significantly improve the prediction accuracy without increasing the sampling points of numerical weather data. Therefore, the region is selected The nearest Regions , and also use their sampled meteorological information as the forecast input for the area; The second threshold is , calculate the distance from the sampling point to the center of the area, if the distance is greater than , it is removed from the neighbor list.

[0036] It should be noted that by selecting the center point and surrounding sample points of the feature area, screening out representative sampling points based on the distance threshold, and reusing the meteorological information of the neighboring areas, the representativeness of the meteorological data is significantly enhanced, which not only improves the prediction accuracy, but also avoids the cost of adding additional sampling points, effectively reduces the overall overhead of the system, and improves the efficiency and economy of the prediction system.

[0037] In the embodiment of the present invention, the above step S400 includes the following sub-steps C1-C5; In C1: merge the power station sample points into characteristic sample points; In C2: for each feature point, meteorological data is collected in the corresponding area and dimensionally transformed to obtain an input sequence; In C3: Based on the input sequence, an artificial neural network with output units is constructed for each region. The hidden layer, output layer, and neuron data of each layer are set to the first multiple of the input to construct a neural network prediction model. In C4: Historical meteorological data is obtained and normalized. The normalized historical meteorological data is used as input, and the power of all power stations in each region and the sum of the power of all power stations in each region at the corresponding time are used as targets to train the neural network prediction model. In C5: Repeat the training operation for each region and adjust the model parameters to obtain the neural network prediction model for each region.

[0038] Specifically, the quantity is (usually tens of thousands to hundreds of thousands) of power station sample points are merged into a number of (usually dozens to hundreds) of feature sample points; For each feature point ( , 、 ), collected within its area Meteorological data of coordinates, Contains the sampling points of this area and several reused sampling points of neighbors, but for an area Once the neighbor relationship is determined, The quantity can be determined and expressed as: , in, For the region The meteorological data series at 0 o'clock is collected at 15-minute intervals and the daily 96-point example formula can be adjusted as needed to achieve ultra-short-term and short-term forecasts. A set of meteorological data samples can be formed. Neighboring sampling points can be added or reduced as needed, but at least the sampling data of the center point coordinates should be included. The matrix is ​​transformed into a dimensional sequence by dimensionally transforming each sampling point. Considering the correlation of weather in time series, the data before sunrise and after sunset are not excluded, and there is no need to consider the one-to-one correspondence with the power generation in time series. This also increases the flexibility of the meteorological data sampling interval. The dimensional sequence is expressed as: , For each area , build For an artificial neural network with 100 output units, 2 to 3 hidden layers are sufficient. The number of neurons in each layer is 1.5 to 2 times the first multiple of the input. The output layer has 96 neurons. If a 1-hour prediction is required, 24 points are sufficient. The hidden layer activation function can be the Gelu function, etc., and the output layer must use the Relu function.

[0039] Normalize the historical meteorological data and use it as input, using regional The power sum of all power stations is used as the target to train the neural network prediction model; Repeat the training operation in each region and adjust the parameters of the neural network prediction model to obtain A neural network prediction model with a relatively small parameter scale (about tens of thousands of parameters).

[0040] It should be noted that by merging a large number of power station sample points into feature sample points, the model input is simplified and the complexity is reduced; meteorological data is collected for each feature point and dimensionality is transformed to construct a highly targeted neural network prediction model; normalization processing and activation functions are used to improve the model's training efficiency and prediction accuracy; through parameter adjustment, a smaller model is obtained, which is suitable for deployment on low-cost hardware, achieving fast, efficient, and low-cost power prediction, and enhancing the practicality and scalability of the system.

[0041] In the embodiment of the present invention, the above step S500 includes the following sub-steps D1-D3; In D1: obtain data for each region, sampling point, and daily forecast; In D2: Based on the neighbor relationship, the prediction data is input into the neural network prediction model to make predictions; In D3: The neural network prediction model is used for prediction through concurrent computing or distributed deployment on multiple terminals. The prediction results are aggregated to a designated central node to obtain the power value of each area at the corresponding time.

[0042] Specifically, obtain regions, Sampling points, daily forecast data, or 16 points after the current time value, for ultra-short term, according to the established neighbor relationship, the forecast data is fed into the model and the forecast is carried out; the forecast model is composed of hundreds of small models, which can be used for concurrent computing or distributed deployment on multiple low-cost terminals to carry out the forecast, and the results can be aggregated to the designated central node. The aggregation can be obtained for any area At the moment Power value .

[0043] In an embodiment of the present invention, after completing steps D1-D3 in the above step S500, the following steps D4-D5 are further included; In D4: Get the actual interval power; In D5: The actual interval power is sent back to the training operation to achieve online learning and optimize the model parameters.

[0044] Specifically, the neural network prediction model has a small parameter scale, a short single training time, and collects actual 15-minute interval power every time or every day. , Note that abnormal data such as maintenance and failure should be filtered out and sent back to the training phase to achieve online learning, continuously adjust the model accuracy, and obtain regional Real-time power .

[0045] In an embodiment of the present invention, after completing steps D4-D5 in the above step S500, the following steps D6-D9 are further included; In D6: the power value of each area at the corresponding time, the average historical power of each household power station in any area and the average total power of the area are calculated; In D7: In any area, the average historical power of each power station in the statistical interval and the average total power of the area are calculated; In D8: Calculate the ratio of the average historical power of each power station in the statistical interval to the average regional total power; In D9: The predicted output power sequence is obtained based on the ratio calculation, and the power station is labeled to predict the power of the corresponding project.

[0046] Specifically, according to the partition results, calculate any area The statistical interval of each household power station refers to the historical power of the power station in the time interval used by the training data. The installed capacity ratio is not used here because the average value of each user's installation angle and shielding characteristics are different , and the average value of the total regional power ; in any Statistics on all regions and , and calculate and The ratio of ; For any power station , which belongs to When the area is , the predicted power is expressed as: , in, for In the region A power station, Moment power, Indicates area Power at the moment; The predicted output power sequence of all power stations is calculated. Based on the label grouping statistics, the predicted power of the corresponding project, daily predicted power generation and other information can be calculated.

[0047] It should be noted that power forecasting through concurrent computing or distributed deployment improves forecasting efficiency and system response speed; the online learning mechanism uses actual power data to optimize model parameters, thereby improving forecasting accuracy and model adaptability; further, through regional power decomposition and labeling processing, power forecasting from region to single power station is realized, meeting different management needs, enhancing the flexibility and practicality of the system, and is suitable for real-time power forecasting and management of large-scale distributed photovoltaic power stations.

[0048] The above is a schematic diagram of a method for rapidly predicting theoretical power of a household photovoltaic power station according to this embodiment. It should be noted that the technical solution of this system for rapidly predicting theoretical power of a household photovoltaic power station and the technical solution of the aforementioned method for rapidly predicting theoretical power of a household photovoltaic power station are based on the same concept. For details not described in detail in the technical solution of the system for rapidly predicting theoretical power of a household photovoltaic power station in this embodiment, please refer to the description of the technical solution of the aforementioned method for rapidly predicting theoretical power of a household photovoltaic power station.

[0049] In this embodiment, a rapid household photovoltaic power station theoretical power prediction system includes: a preprocessing module, configured to obtain historical power plant data, and preprocess the historical power plant data to obtain first historical power plant data; a classification module, configured to label the power stations based on the first historical power station data, and classify the power stations to obtain each region; a multiplexing module, configured to select meteorological sampling points for each area, and multiplex meteorological information of neighboring areas of each area to obtain power station sample points; A model building module is used to process data of power station sample points to obtain an input sequence, and to build a neural network prediction model for each region based on the input sequence; The prediction module is used to perform power prediction on the neural network prediction model and optimize the model parameters through online learning to obtain the power prediction results for each area.

[0050] This embodiment also provides a computer device suitable for a method for quickly predicting the theoretical power of a household photovoltaic power station, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement a method for quickly predicting the theoretical power of a household photovoltaic power station as proposed in the above embodiment.

[0051] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, a method for rapidly predicting theoretical power of a household photovoltaic power station as proposed in the above embodiment is implemented.

[0052] The storage medium proposed in this embodiment and the method for realizing a rapid theoretical power prediction method of a household photovoltaic power station proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0053] From the above description of the embodiments, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware. Of course, it can also be implemented using hardware, but in many cases the former is the preferred embodiment. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This software product can be stored on a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk, or optical disk, and includes instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods of various embodiments of the present invention.

[0054] Example 2, reference Figure 6 , Table 1 ~ Table 2, This embodiment is different from the first embodiment. It provides a verification test of a rapid household photovoltaic power station theoretical power prediction method to verify and illustrate the technical effects used in this method.

[0055] Taking more than 150,000 distributed household power stations as an example, a fast theoretical power prediction method for household photovoltaic power stations is used for clustering, which can obtain 71 regional center points and 355 meteorological sampling points, and all users are marked by region.

[0056] Take one region, for example, Region 848, which has a capacity of 34,400 and over 1,200 users. Meteorological data is sampled from five sampling points within the region and three surrounding points. Hourly (15-minute intervals are acceptable in practical applications) meteorological data (wind speed, cloud cover, temperature, shortwave radiation, horizontal radiation, etc.) is used as input. Pre-training is performed on one year's data to create a small model for this region. Using this method, training on a quad-core CPU takes only 7 minutes, and the model meets convergence requirements. This simplifies the prediction problem for over 1,200 distributed power generation units in the region to that of a single unit.

[0057] like Figure 6 As shown, the neural network prediction model is used to predict the data for 5 days from January 5 to 10, 2024, and the data is verified using the actual power of the region;

[0058] When the prediction results of 848 areas are known, the power station "Liao Mou" (capacity 33.3kW, no shutdown and maintenance tasks during the prediction data period) is calculated by the historical power ratio. ) The power of the power station can be calculated and compared with the actual average power of the power station as shown in Table 2; Table 2 Comparison with the actual average power of the power station , , As shown in Table 2, in the absence of maintenance or fault interference, the accuracy of a single power station averaged 81.83%, meeting expectations.

[0059] Through clustering, more than 150,000 distributed household power stations were simplified into 71 regional center points, significantly reducing the complexity of the model; using a small number of meteorological sampling points and a neural network model, regional power was predicted quickly and accurately, with an average accuracy of 83.25%, and exceeding 89% after excluding special points; further decomposition into individual power stations, the average accuracy was 81.83%, meeting actual application needs.

[0060] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A rapid household photovoltaic power station theoretical power prediction method, characterized in that: include: Acquiring historical power plant data, and preprocessing the historical power plant data to obtain first historical power plant data; Labeling the power stations based on the first historical power station data and classifying the power stations to obtain each region; Selecting meteorological sampling points for each area, and reusing meteorological information of neighboring areas of each area to obtain power station sample points; Processing the power station sample points to obtain an input sequence, and constructing a neural network prediction model for each region based on the input sequence; The neural network prediction model is used to perform power prediction, and the model parameters are optimized through online learning to obtain the power prediction results of each area.

2. A rapid household photovoltaic power station theoretical power prediction method according to claim 1, characterized in that: Classification of power stations includes: Divide the power station coordinates into several square areas; Calculate the feature points in each area to get the center of the new area; Traverse all regional center feature points and calculate the geometric distance between any two regional capacities. If the geometric distance is less than the first threshold, perform a weighted merge operation on the new feature points of the two regions and modify the regional label of the power station in the region. The weighted merging operation is repeated until the distances between all region centers are smaller than the first threshold.

3. A rapid household photovoltaic power station theoretical power prediction method according to claim 2, characterized in that: The meteorological sampling points selected include: For any feature area, take the center point and surrounding sample points; Calculate the distance between the surrounding sample points and the center point. If the distance is greater than a second threshold, remove the corresponding sample point from the neighbor list. Obtain the meteorological information of the nearest area around any feature area, use the meteorological information as the prediction input of the corresponding area, and reuse the meteorological information of the neighboring areas of each area.

4. A rapid household photovoltaic power station theoretical power prediction method according to claim 3, characterized in that: Constructing the neural network prediction model for each region includes: Merge the power station sample points into characteristic sample points; For each feature point, meteorological data is collected in the corresponding area and dimensionally transformed to obtain an input sequence; According to the input sequence, an artificial neural network of output units is constructed for each region, and the hidden layer, output layer and neuron data of each layer are set to the first multiple of the input to construct a neural network prediction model; Historical meteorological data is obtained and normalized. The normalized historical meteorological data is used as input, and the power of all power stations in each region and the sum of the power of all power stations in each region at the corresponding time are used as targets to train the neural network prediction model. The training operation is repeated for each region, and the model parameters are adjusted to obtain the neural network prediction model for each region.

5. A rapid household photovoltaic power station theoretical power prediction method according to claim 4, characterized in that: Using the neural network prediction model to perform power prediction includes: Obtain forecast data for each region, sampling point, and day; According to the neighbor relationship, the prediction data is input into the neural network prediction model to make predictions; The neural network prediction model is used to perform predictions through concurrent computing or distributed deployment on multiple terminals. The prediction results are summarized to a designated central node to obtain the power value of each area at the corresponding time.

6. A rapid household photovoltaic power station theoretical power prediction method according to claim 5, characterized in that: Optimizing model parameters through online learning includes: Get the actual interval power; The actual interval power is fed back to the training operation to enable online learning and optimize the model parameters.

7. A rapid household photovoltaic power station theoretical power prediction method according to claim 6, characterized in that: Power prediction results include: The power value of each area at the corresponding time, calculate the average historical power of each household power station in any area and the average total power of the area; In any area, calculate the average historical power of each power station in the statistical interval and the average total power of the area; Calculate the ratio of the average historical power of each power station in the statistical interval to the average total power of the region; The predicted output power sequence is obtained based on the ratio calculation, and the power station is labeled to predict the power of the corresponding project.

8. A rapid household photovoltaic power station theoretical power prediction system, using a rapid household photovoltaic power station theoretical power prediction method according to any one of claims 1 to 7, characterized in that: a preprocessing module, configured to obtain historical power plant data, and preprocess the historical power plant data to obtain first historical power plant data; a classification module, configured to label the power stations based on the first historical power station data, and classify the power stations to obtain each region; a multiplexing module, configured to select meteorological sampling points for each area, and multiplex meteorological information of neighboring areas of each area to obtain power station sample points; A model building module is used to process data of power station sample points to obtain an input sequence, and to build a neural network prediction model for each region based on the input sequence; The prediction module is used to perform power prediction on the neural network prediction model and optimize the model parameters through online learning to obtain the power prediction results for each area.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for quickly predicting theoretical power of a household photovoltaic power station according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for quickly predicting theoretical power of a household photovoltaic power station according to any one of claims 1 to 7 are implemented.

Images