Accuracy optimization method and system for photovoltaic ultra-short-term power prediction

Abstract

The present invention discloses a precision optimization method and system for photovoltaic ultra-short-term power prediction, which relates to the technical field related to photovoltaic power generation management. The method includes: interacting with the target photovoltaic power generation system, extracting historical power data and meteorological data; performing homomorphic division based on photovoltaic scene data, generating photovoltaic scene segmentation data; obtaining the power prediction accuracy requirements of the target photovoltaic power generation system; performing granular division of the prediction input information duration according to the responsiveness requirements, and setting the hierarchical prediction input information depth; constructing a photovoltaic ultra-short-term power prediction depth fluctuation model; and deploying it in the target photovoltaic power generation system to realize photovoltaic ultra-short-term power prediction. The method solves the technical problem that the real-time photovoltaic data acquisition in the existing photovoltaic power prediction is not accurate and reliable enough, resulting in low efficiency and accuracy of the power prediction model, and thus poor stability of the power grid operation.

Description

Technical Field

[0001] The present application relates to the field related to photovoltaic power generation management technology, and specifically to an accuracy optimization method and system for photovoltaic ultra-short-term power prediction. Background Art

[0002] With the rapid development of photovoltaic power generation, photovoltaic ultra-short-term power prediction has become a key issue in optimizing grid operation and energy scheduling. Photovoltaic ultra-short-term power prediction is to predict the power output of photovoltaic power generation systems in the next few minutes to hours to ensure the stable operation and reasonable scheduling of the power grid. It is particularly important to accurately predict the changes in photovoltaic power. However, the existing photovoltaic power output is affected by many factors such as weather conditions, solar radiation, and equipment status, which makes the photovoltaic power fluctuate greatly. Due to the uncertainty of weather conditions, it is difficult to obtain photovoltaic scene data in real time, which reduces the accuracy of the power prediction model and increases the uncertainty of power output.

[0003] Therefore, in the current photovoltaic ultra-short-term power forecasting related technologies, there is a technical problem that the real-time photovoltaic data acquisition is not accurate and reliable enough, resulting in low efficiency and accuracy of the power prediction model and thus poor stability of the power grid operation. Summary of the invention

[0004] The present application provides an accuracy optimization method and system for photovoltaic ultra-short-term power prediction, and adopts technical means such as data homomorphic division, time granularity division, and power model construction to solve the technical problem that the real-time photovoltaic data acquisition in the existing photovoltaic ultra-short-term power prediction is not accurate and reliable enough, resulting in low efficiency and accuracy of the power prediction model and thus poor stability of the power grid operation. The application achieves the technical effect of accurately acquiring real-time photovoltaic data, thereby improving the accuracy of the photovoltaic power prediction model and making the power grid operate more stably.

[0005] The present application provides an accuracy optimization method for photovoltaic ultra-short-term power prediction, the method comprising: interacting with a target photovoltaic power generation system, extracting historical power data and meteorological data, storing and generating photovoltaic scene data; performing homomorphic division based on the photovoltaic scene data, and generating photovoltaic scene segmentation data; obtaining a power prediction accuracy requirement for the target photovoltaic power generation system, wherein the power prediction accuracy requirement includes an accuracy requirement and a responsiveness requirement; performing granular division of the prediction input information duration according to the responsiveness requirement, and setting a hierarchical prediction input information depth; based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, and in combination with the power prediction accuracy requirement, constructing a photovoltaic ultra-short-term power prediction depth fluctuation model; deploying the photovoltaic ultra-short-term power prediction depth fluctuation model in the target photovoltaic power generation system to achieve photovoltaic ultra-short-term power prediction.

[0006] In a possible implementation, based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, combined with the power prediction accuracy requirement, a photovoltaic ultra-short-term power prediction deep fluctuation model is constructed, and the following processing is performed: based on the photovoltaic scene data, data is extracted according to the hierarchical prediction input information time depth to obtain a hierarchical sample data set, wherein the hierarchical sample data set includes a plurality of hierarchical sample data clusters with different input information time depths; according to the hierarchical prediction input information depth, a multi-scale photovoltaic ultra-short-term power prediction network is constructed, wherein the multi-scale photovoltaic ultra-short-term power prediction network includes a plurality of photovoltaic ultra-short-term power prediction networks with different network depths; the hierarchical sample data set is called to perform sample division, and the sample division result is sent to the corresponding photovoltaic ultra-short-term power prediction network to train the multi-scale photovoltaic ultra-short-term power prediction network; the network performance of the trained multi-scale photovoltaic ultra-short-term power prediction network is verified to obtain a deep performance data set, wherein the deep performance data set includes input information depth, output response performance, and output accuracy performance; based on the deep performance data set, the model depth is configured according to the photovoltaic scene segmentation data to obtain a photovoltaic ultra-short-term power prediction deep fluctuation model.

[0007] In a possible implementation, based on the deep performance data set, the model is deeply configured according to the photovoltaic scene segmentation data to obtain a photovoltaic ultra-short-term power prediction deep fluctuation model, and the following processing is performed: the deep performance data set is parsed, data fitting is performed, and the information depth-performance characteristic operator is obtained; the photovoltaic scene segmentation data is called to extract differential photovoltaic scenes; the power prediction accuracy requirements are traversed, and the demand matching based on the differential photovoltaic scenes is performed to obtain multi-scenario accuracy requirements; the multi-scenario accuracy requirements are traversed, and the network depth is solved according to the information depth-performance characteristic operator to obtain hierarchical depth data; based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called to aggregate and construct the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0008] In a possible implementation, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, the photovoltaic ultra-short-term power prediction depth fluctuation model is aggregated and constructed, and the following processing is performed: based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is traversed to perform network depth comparison to obtain a network depth difference coefficient; it is determined whether the network depth difference coefficient satisfies a preset network depth difference constraint; if the network depth difference coefficient satisfies the network depth difference constraint, the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient is embedded in the photovoltaic ultra-short-term power prediction depth fluctuation model.

[0009] In a possible implementation, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, and the photovoltaic ultra-short-term power prediction deep fluctuation model is aggregated and constructed, and the following processing is also performed: if the network depth difference coefficient does not meet the network depth difference constraint, the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient is stored in the alternative network group; based on the network depth difference coefficient, the alternative network group is serialized; according to the serialization result, the best selection is performed to extract the first target alternative network; based on the photovoltaic scene segmentation data, the first target alternative network is strengthened and trained, and the strengthening training result is verified; if the strengthening training result meets the multi-scene accuracy requirements, the strengthening training result is embedded in the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0010] In a possible implementation, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, and the photovoltaic ultra-short-term power prediction deep fluctuation model is aggregated and constructed, and the following processing is also performed: based on the photovoltaic scene segmentation data, a deep classifier is constructed; using the photovoltaic ultra-short-term power prediction network and the enhanced training results, a fluctuation multi-depth parallel prediction channel is constructed, wherein the fluctuation multi-depth parallel prediction channel includes multiple prediction channels of multiple network depths; based on the hierarchical depth data, the deep classifier is connected to the fluctuation multi-depth parallel prediction channel to obtain the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0011] In a possible implementation, the photovoltaic ultra-short-term power prediction deep fluctuation model is deployed in a target photovoltaic power generation system, and the following processing is also performed: a computing power evaluation is performed on the target photovoltaic power generation system to obtain a computing power evaluation result; based on the computing power evaluation result, a deployment evaluation of the target photovoltaic power generation system is performed according to the photovoltaic ultra-short-term power prediction deep fluctuation model to obtain a deployment evaluation result; if the deployment evaluation result does not meet the preset deployment performance constraints, based on the model compression algorithm, a lightweight processing of the prediction deep fluctuation model is performed to obtain a lightweight model of photovoltaic ultra-short-term power prediction deep fluctuation; and the photovoltaic ultra-short-term power prediction deep fluctuation lightweight model is deployed in the target photovoltaic power generation system.

[0012] The present application also provides an accuracy optimization system for photovoltaic ultra-short-term power prediction, including:

[0013] A photovoltaic scene data homomorphic partitioning module, which is used to interact with the target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data, and perform homomorphic partitioning based on the photovoltaic scene data to generate photovoltaic scene segmentation data;

[0014] A module for dividing the duration of the prediction input information into granularity, wherein the module is used to obtain the power prediction accuracy requirement of the target photovoltaic power generation system, wherein the power prediction accuracy requirement includes the accuracy requirement and the responsiveness requirement, and according to the responsiveness requirement, the granularity of the prediction input information duration is divided, and the hierarchical prediction input information depth is set;

[0015] A power prediction depth fluctuation model construction module, the power prediction depth fluctuation model construction module is used to construct a photovoltaic ultra-short-term power prediction depth fluctuation model based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, and in combination with the power prediction accuracy requirement;

[0016] The photovoltaic ultra-short-term power prediction module is used to deploy the photovoltaic ultra-short-term power prediction deep fluctuation model in the target photovoltaic power generation system to realize photovoltaic ultra-short-term power prediction.

[0017] It is intended to use the accuracy optimization method and system for photovoltaic ultra-short-term power prediction proposed in this application, interact with the target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data, perform homomorphic division based on the photovoltaic scene data, generate photovoltaic scene segmentation data, obtain the power prediction accuracy requirements of the target photovoltaic power generation system, including accuracy requirements and responsiveness requirements, perform granular division of the prediction input information time according to the responsiveness requirements, set the hierarchical prediction input information depth, and construct a photovoltaic ultra-short-term power prediction depth fluctuation model based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, combined with the power prediction accuracy requirements. The photovoltaic ultra-short-term power prediction depth fluctuation model is deployed in the target photovoltaic power generation system to realize photovoltaic ultra-short-term power prediction, and solves the technical problem that the real-time photovoltaic data acquisition in the existing photovoltaic ultra-short-term power prediction is not accurate and reliable enough, resulting in low efficiency and accuracy of the power prediction model and thus poor stability of the power grid operation. The real-time photovoltaic data is accurately acquired, thereby improving the accuracy of the photovoltaic power prediction model and making the power grid operate more stably. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solution of the embodiment of the present disclosure, the accompanying drawings of the embodiment of the present disclosure will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0019] Figure 1A schematic flow chart of a method for optimizing the accuracy of photovoltaic ultra-short-term power prediction provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of the structure of a precision optimization system for photovoltaic ultra-short-term power prediction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0022] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0023] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0024] The present application embodiment provides an accuracy optimization method for photovoltaic ultra-short-term power prediction, such as Figure 1 As shown, the method includes:

[0025] Step S100, interact with the target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data. Interactive operation and data processing are performed with the target photovoltaic power generation system, and historical power data and meteorological data can be extracted, and these data are stored and processed to finally generate photovoltaic power generation scene data. Specifically, the historical power data of the target photovoltaic power generation system is extracted, and these data record the power-related data generated by the photovoltaic power generation system in a certain period of history. According to the historical power data, the actual power generation status of the photovoltaic power generation system can be understood. Meteorological data related to the operation of the target photovoltaic power generation system is extracted, such as solar radiation, temperature, wind speed, cloud cover, etc. These data can evaluate the performance of the photovoltaic power generation system and predict future power generation. The extracted historical power data and meteorological data of the target power generation system are stored, and photovoltaic scene data are comprehensively generated, including the power generation of the target photovoltaic power generation system at different times and under different meteorological conditions. By interacting with the target photovoltaic power generation system, the operation of photovoltaic power generation can be monitored and analyzed, and the power of the photovoltaic power generation system can be evaluated and predicted through photovoltaic scene data to ensure stable operation of the power grid.

[0026] After the photovoltaic scene data is generated, step S200 is executed to perform homomorphic division based on the photovoltaic scene data to generate photovoltaic scene segmentation data. The generated photovoltaic scene data is homomorphically divided to generate photovoltaic scene segmentation data. Specifically, the photovoltaic scene data of the target photovoltaic power generation is divided into different subsets, each subset corresponds to a specific scene, and the basis for division can be weather conditions, solar radiation intensity, temperature, time range, etc. For example, the photovoltaic scene data is divided into different subsets based on time periods (morning, noon, night), and each time period corresponds to a scene segmentation data, reflecting the characteristics of the target photovoltaic power generation system in this period; the photovoltaic scene data is divided into different subsets based on weather conditions (sunny, cloudy, overcast), and each subset corresponds to scene segmentation data, reflecting the power characteristics of the target photovoltaic power generation system in this state.

[0027] After generating the photovoltaic scene segmentation data, step S300 is executed to obtain the power prediction accuracy requirement of the target photovoltaic power generation system, wherein the power prediction accuracy requirement includes the accuracy requirement and the responsiveness requirement. The power prediction accuracy requirement of the target photovoltaic power generation system is obtained, that is, the prediction accuracy requirement of the power generation of the target photovoltaic power generation system is obtained, wherein the power prediction accuracy requirement includes the accuracy requirement and the responsiveness requirement. Specifically, the accuracy refers to the accuracy of the target photovoltaic power generation system in predicting the power generation in the future, which is usually expressed as a percentage. For example, the accuracy error of the power generation prediction of the photovoltaic power generation system does not exceed 5%, that is, the deviation between the actual power generation and the predicted value does not exceed 5%; responsiveness refers to the rapid adaptability of the target photovoltaic power generation system to meteorological conditions such as weather conditions and light intensity. The power of the target photovoltaic power generation system is affected by many factors, and it is necessary to quickly update the power prediction results when the weather and light change, so as to adjust and optimize in time. The responsiveness requirement requires the target photovoltaic power generation system to monitor and adapt to the changing environment and meteorological conditions in real time, and update the prediction results in real time. The accurate power prediction of the target photovoltaic power generation system can help optimize energy scheduling and improve the efficiency of the power generation system, and the rapid response can achieve stable operation of the power grid.

[0028] After obtaining the power prediction accuracy requirement of the target photovoltaic power generation system, step S400 is executed to perform granular division of the prediction input information duration according to the responsiveness requirement and set the hierarchical prediction input information depth. Specifically, the hierarchical prediction input information depth refers to the granular division of the depth of the prediction input information according to the different responsiveness requirements and the predicted time range. The deeper the depth corresponding to each level, the longer the predicted time range. For example, the prediction input information is divided into ultra-short-term prediction, short-term prediction and long-term prediction. The prediction input information depth of ultra-short-term prediction is the shallowest, mainly predicting the power within a few minutes to a few hours; the prediction input information depth of short-term prediction is shallow, mainly predicting the power change within the next few hours to a day, which is used for day-ahead scheduling; the prediction input information depth of long-term prediction is deep, which can predict the power change trend within the next day to several days, which is used for long-term planning. By setting the hierarchical prediction input information depth, different levels of prediction information can be selected according to demand to meet the prediction requirements of the target photovoltaic power generation system within different time ranges. At the same time, the deeper the depth of the prediction input information, the accuracy of the power prediction result may be reduced.

[0029] After setting the hierarchical prediction input information depth, execute step S500, and construct a photovoltaic ultra-short-term power prediction depth fluctuation model based on the hierarchical prediction input information time depth, the photovoltaic scene data and the photovoltaic scene segmentation data, and the power prediction accuracy requirement. Based on the hierarchical prediction input information time depth, photovoltaic scene data and photovoltaic scene segmentation data, and combined with the power prediction accuracy requirement, a model for predicting the power fluctuation of the photovoltaic power generation system within a short time range is constructed. Specifically, according to the power prediction accuracy requirement, the model can accurately predict the change trend and specific value of the photovoltaic power generation power in the future time period. The prediction result output by the model should be as close as possible to the actual photovoltaic power generation power to provide reliable power prediction. The photovoltaic ultra-short-term power prediction depth fluctuation model is a prediction model constructed based on the hierarchical prediction input information time depth, photovoltaic scene data and photovoltaic scene segmentation data, combined with the power prediction accuracy requirement, and is used to predict the photovoltaic power change trend within a short time range (usually a few minutes to a few hours) in the future, while considering the power fluctuation characteristics.

[0030] In a possible implementation, step S500 further includes step S510, based on the photovoltaic scene data, data extraction is performed according to the hierarchical prediction input information time depth to obtain a hierarchical sample data set, wherein the hierarchical sample data set includes a plurality of hierarchical sample data clusters with different input information time depths. According to the photovoltaic scene data and the hierarchical prediction input information time depth, data extraction can be performed to obtain a hierarchical sample data set, which includes a plurality of hierarchical sample data clusters with different input information time depths. Specifically, according to the determined hierarchical prediction input information time depth, the photovoltaic scene data is divided into different sample data sets according to different time ranges, each sample data set contains photovoltaic scene data within the corresponding time range, and for each sample data set, based on different time depths, multiple hierarchical sample data clusters with different input information time depths are generated. For example, for a sample data set with a time range of several minutes, a minute-level data cluster can be generated, with data per minute as one sample; for a sample data set with a time range of several hours, data extraction can be performed according to the photovoltaic scene data and the hierarchical prediction input information time depth to obtain a hierarchical sample data set, which includes a plurality of hierarchical sample data clusters with different input information time depths. It also includes step S520, constructing a multi-scale photovoltaic ultra-short-term power prediction network according to the hierarchical prediction input information depth, wherein the multi-scale photovoltaic ultra-short-term power prediction network includes a plurality of photovoltaic ultra-short-term power prediction networks with different network depths. According to the hierarchical prediction input information depth, a multi-scale photovoltaic ultra-short-term power prediction network is constructed, and the multi-scale photovoltaic ultra-short-term power prediction network mainly constructs prediction models of different depths at different time depths, which are used to predict power fluctuations within different time ranges, and each network depth can correspond to a power prediction demand within a time range. It also includes step S530, calling the hierarchical sample data set, performing sample division, sending the sample division result to the corresponding photovoltaic ultra-short-term power prediction network, and training the multi-scale photovoltaic ultra-short-term power prediction network. Specifically, according to the needs and the specific training process, the hierarchical sample data set can be divided into a training set, a verification set, and a test set, etc., and the sample data set is divided according to the proportion to ensure that the samples between the data sets are independent of each other. It also includes step S540, performing network performance verification on the trained multi-scale photovoltaic ultra-short-term power prediction network, and obtaining a deep performance data set, wherein the deep performance data set includes input information depth, output response performance, and output accuracy performance.The network performance of the trained multi-scale photovoltaic ultra-short-term power prediction network is verified to obtain deep performance data sets such as input information depth, output response performance and output accuracy performance. Specifically, the input information depth refers to the time range of key information required by the network when making predictions; the output response performance is to evaluate the response performance of the network tested by the deep performance data set; the output accuracy performance can be evaluated by comparing the predicted value with the true value. It also includes step S550, based on the deep performance data set, the model depth is configured according to the photovoltaic scene segmentation data to obtain the photovoltaic ultra-short-term power prediction deep fluctuation model. Among them, the model depth configuration refers to the model depth configuration according to the analysis results of the deep performance data set and the photovoltaic scene segmentation data. The corresponding network structure, number of layers and parameter settings can be selected for different photovoltaic scene segmentation data to meet the accurate prediction of power changes and fluctuations at different depths. The deep fluctuation model can also adapt to scene changes at different time depths to provide higher prediction accuracy.

[0031] In a possible implementation, step S550 further includes step S551, parsing the deep performance data set, performing data fitting, and obtaining an information depth-performance characteristic operator. The deep performance data set is parsed to extract the input information depth, output response performance, and output accuracy performance. The input information depth is used as an independent variable and the performance index is used as a dependent variable. Data fitting is performed to obtain a fitting curve, and the information depth-performance characteristic operator is extracted, wherein the information depth-performance characteristic operator is used to describe the relationship between the input information depth and the performance index, and to evaluate the impact of the input information depth on the performance index. It also includes step S552, calling the photovoltaic scene segmentation data to extract the difference photovoltaic scene. Among them, the difference photovoltaic scene is a scene of multiple photovoltaic power stations clustered and divided based on weather conditions and seasons. It also includes step S553, traversing the power prediction accuracy requirements, performing demand matching based on the difference photovoltaic scene, and obtaining multi-scene accuracy requirements. Among them, the demand matching of the difference photovoltaic scene refers to determining the requirements and characteristics of each scene according to different difference photovoltaic scenes. For example, a scene with a large change in light intensity may require a higher prediction accuracy, while a scene with a small temperature change may require a smaller range of prediction errors. It also includes step S554, traversing the multi-scenario accuracy requirements, performing network depth solution according to the information depth-performance characteristic operator, and obtaining hierarchical depth data. Based on the multi-scenario accuracy requirements, the network depth solution is performed according to the information depth-performance characteristic operator to determine the most suitable network depth in each scenario, and the hierarchical depth data represents the network depth in each scenario to meet the accuracy requirements. It also includes step S555, based on the hierarchical depth data, calling the multi-scale photovoltaic ultra-short-term power prediction network, and aggregating and constructing the photovoltaic ultra-short-term power prediction depth fluctuation model.

[0032] In a possible implementation, step S555 further includes the step of traversing the multi-scale photovoltaic ultra-short-term power prediction network to perform network depth comparison based on the hierarchical depth data to obtain a network depth difference coefficient. The depth configuration of each prediction network is compared with the hierarchical data, the depth configuration difference of each network is recorded, and the network depth difference coefficient is calculated, wherein the depth difference coefficient represents the degree of difference between each network depth and the hierarchical depth data. It also includes the step of determining whether the network depth difference coefficient satisfies a preset network depth difference constraint. Specifically, the calculated network depth difference coefficient is compared with the preset network depth difference to determine whether the constraint is met. It also includes the step of, if the network depth difference coefficient satisfies the network depth difference constraint, embedding the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient into the photovoltaic ultra-short-term power prediction depth fluctuation model.

[0033] In a possible implementation, step S555 further includes a step of storing the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient into an alternative network group if the network depth difference coefficient does not satisfy the network depth difference constraint. The alternative network group is used to contain networks that do not satisfy the network depth difference. Based on the network depth difference coefficient, the alternative network group is serialized. Specifically, each network in the alternative network group is serialized and converted into a data format that can be stored or transmitted. It also includes a step of performing optimal selection based on the serialization result and extracting the first target alternative network. It also includes a step of performing intensive training of the first target alternative network based on the photovoltaic scene segmentation data and verifying the intensive training result. It also includes a step of embedding the intensive training result into the photovoltaic ultra-short-term power prediction depth fluctuation model if the intensive training result meets the multi-scene accuracy requirements. When the alternative network after intensive training meets the multi-scene accuracy requirements, the intensive training result is embedded in the photovoltaic ultra-short-term power prediction depth fluctuation model for power prediction of photovoltaic power generation systems.

[0034] In a possible implementation, step S555 further includes the step of constructing a deep classifier based on the photovoltaic scene segmentation data. Specifically, according to the characteristics of meteorological data, weather conditions, solar radiation, etc. in the photovoltaic scene, the corresponding photovoltaic features are extracted, the photovoltaic scene features that clearly distinguish different ones are extracted, and a suitable network model is selected to perform deep classification of the photovoltaic scene. It also includes the step of constructing a fluctuating multi-depth parallel prediction channel with the photovoltaic ultra-short-term power prediction network and the intensive training results, wherein the fluctuating multi-depth parallel prediction channel includes multiple prediction channels of multiple network depths. It also includes the step of connecting the deep classifier with the fluctuating multi-depth parallel prediction channel based on the hierarchical depth data to obtain the photovoltaic ultra-short-term power prediction deep fluctuation model. According to the different depth configurations in the hierarchical depth data, multiple parallel prediction channels with corresponding depth configurations are constructed, each channel adopts a photovoltaic ultra-short-term power prediction model with a corresponding depth configuration, and the deep classifier is connected to the fluctuating multi-depth parallel prediction channel to perform power prediction of photovoltaic power generation according to the depth configuration of the scene.

[0035] After the construction of the photovoltaic ultra-short-term power prediction deep fluctuation model is completed, step S600 is executed to deploy the photovoltaic ultra-short-term power prediction deep fluctuation model in the target photovoltaic power generation system to achieve photovoltaic ultra-short-term power prediction. Specifically, the constructed photovoltaic ultra-short-term power prediction deep fluctuation model is deployed in the target photovoltaic power generation system, and the photovoltaic power is predicted in a short time through the deep fluctuation model, providing more accurate prediction results and optimizing the stable operation of the power grid.

[0036] In one possible manner, step S600 also includes step S610, performing a computing power evaluation on the target photovoltaic power generation system and obtaining a computing power evaluation result. The computing power evaluation of the target photovoltaic power generation system is to determine the computing resources and performance of the system and to ensure the stability and accuracy of the deployed power prediction model. It also includes step S620, based on the computing power evaluation result, performing a deployment evaluation of the target photovoltaic power generation system according to the photovoltaic ultra-short-term power prediction deep fluctuation model, and obtaining a deployment evaluation result. Specifically, based on the computing power evaluation result, the photovoltaic ultra-short-term power prediction deep fluctuation model is used to perform a deployment evaluation of the target photovoltaic power generation system, that is, to evaluate the feasibility of deploying the power prediction model in the target photovoltaic power generation system, and the evaluation result is the deployment evaluation result. It also includes step S630, if the deployment evaluation result does not meet the preset deployment performance constraints, then based on the model compression algorithm, the predicted deep fluctuation model is lightweighted to obtain a photovoltaic ultra-short-term power prediction deep fluctuation lightweight model. Among them, the model compression algorithm is used to compress the power prediction model in a lightweight manner, aiming to reduce the calculation amount and scale of the model while trying to maintain the prediction performance of the model. If the deployment evaluation result does not meet the preset constraints, the model compression algorithm is used to lightweight the prediction depth fluctuation model to obtain a lightweight model for photovoltaic ultra-short-term power prediction depth fluctuation. It also includes step S640, deploying the lightweight model for photovoltaic ultra-short-term power prediction depth fluctuation in the target photovoltaic power generation system. The lightweight power prediction depth fluctuation model is used in the actual photovoltaic power generation system to realize the prediction of photovoltaic ultra-short-term power, ensure the accuracy of the model prediction performance, and improve the accuracy of the predicted photovoltaic power generation power results.

[0037] In the above, refer to Figure 1 The precision optimization method for photovoltaic ultra-short-term power prediction according to an embodiment of the present invention is described in detail. Figure 2 A precision optimization system for photovoltaic ultra-short-term power prediction according to an embodiment of the present invention is described.

[0038] The precision optimization system for photovoltaic ultra-short-term power prediction according to the embodiment of the present invention is used to solve the technical problem that the security monitoring data of the existing urban security management warning is not real-time and accurate, which leads to inaccurate model abnormality recognition and makes the adaptive warning results not timely and accurate, and achieves the technical effect of accurately obtaining security monitoring data, thereby improving the real-time and accuracy of urban security management warning. The precision optimization system for photovoltaic ultra-short-term power prediction includes: a security monitoring network establishment module 10, a security monitoring data set acquisition module 20, an early warning adaptive analysis module 30, and an early warning signal generation module 40.

[0039] A photovoltaic scene data homomorphic partitioning module 10, wherein the photovoltaic scene data homomorphic partitioning module 10 is used to interact with a target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data, and perform homomorphic partitioning based on the photovoltaic scene data to generate photovoltaic scene segmentation data;

[0040] The prediction input information duration granularity division module 20 is used to obtain the power prediction accuracy requirement of the target photovoltaic power generation system, wherein the power prediction accuracy requirement includes the accuracy requirement and the responsiveness requirement, and the prediction input information duration is divided into granularities according to the responsiveness requirement, and the hierarchical prediction input information depth is set;

[0041] A power prediction depth fluctuation model construction module 30, the power prediction depth fluctuation model construction module 30 is used to construct a photovoltaic ultra-short-term power prediction depth fluctuation model based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, combined with the power prediction accuracy requirement;

[0042] The photovoltaic ultra-short-term power prediction module 40 is used to deploy the photovoltaic ultra-short-term power prediction deep fluctuation model in the target photovoltaic power generation system to realize photovoltaic ultra-short-term power prediction.

[0043] The specific configuration of the power prediction deep fluctuation model construction module 30 will be described in detail below. As described above, the photovoltaic ultra-short-term power prediction deep fluctuation model is constructed, and the power prediction deep fluctuation model construction module 30 may further include: based on the photovoltaic scene data, data extraction is performed according to the hierarchical prediction input information time depth to obtain a hierarchical sample data set, wherein the hierarchical sample data set includes a plurality of hierarchical sample data clusters with different input information time depths; according to the hierarchical prediction input information depth, a multi-scale photovoltaic ultra-short-term power prediction network is constructed, wherein the multi-scale photovoltaic ultra-short-term power prediction network includes a plurality of photovoltaic ultra-short-term power prediction networks with different network depths; calling the hierarchical sample data set, performing sample division, sending the sample division result to the corresponding photovoltaic ultra-short-term power prediction network, and training the multi-scale photovoltaic ultra-short-term power prediction network; performing network performance verification on the trained multi-scale photovoltaic ultra-short-term power prediction network to obtain a deep performance data set, wherein the deep performance data set includes input information depth, output response performance, and output accuracy performance; based on the deep performance data set, the model depth configuration is performed according to the photovoltaic scene segmentation data to obtain a photovoltaic ultra-short-term power prediction deep fluctuation model.

[0044] The specific configuration of the power prediction deep fluctuation model construction module 30 will be described in detail below. As described above, based on the deep performance data set, the model is deeply configured according to the photovoltaic scene segmentation data to obtain the photovoltaic ultra-short-term power prediction deep fluctuation model. The power prediction deep fluctuation model construction module 30 further includes: parsing the deep performance data set, performing data fitting, and obtaining the information depth-performance characteristic operator; calling the photovoltaic scene segmentation data to extract the difference photovoltaic scene; traversing the power prediction accuracy requirements, performing demand matching based on the difference photovoltaic scene, and obtaining the multi-scene accuracy requirements; traversing the multi-scene accuracy requirements, performing network depth solution according to the information depth-performance characteristic operator, and obtaining hierarchical depth data; based on the hierarchical depth data, calling the multi-scale photovoltaic ultra-short-term power prediction network, and aggregating and constructing the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0045] The specific configuration of the power prediction depth fluctuation model construction module 30 will be described in detail below. As described above, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, and the photovoltaic ultra-short-term power prediction depth fluctuation model is aggregated and constructed. The power prediction depth fluctuation model construction module 30 further includes: based on the hierarchical depth data, traversing the multi-scale photovoltaic ultra-short-term power prediction network to perform network depth comparison and obtain the network depth difference coefficient; judging whether the network depth difference coefficient meets the preset network depth difference constraint; if the network depth difference coefficient meets the network depth difference constraint, then the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient is embedded in the photovoltaic ultra-short-term power prediction depth fluctuation model.

[0046] The specific configuration of the power prediction deep fluctuation model construction module 30 will be described in detail below. As described above, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, and the photovoltaic ultra-short-term power prediction deep fluctuation model is aggregated and constructed. The power prediction deep fluctuation model construction module 30 further includes: if the network depth difference coefficient does not meet the network depth difference constraint, the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient is stored in the candidate network group; based on the network depth difference coefficient, the candidate network group is serialized; according to the serialization result, the best selection is performed to extract the first target candidate network; based on the photovoltaic scene segmentation data, the first target candidate network is strengthened and trained, and the strengthening training result is verified; if the strengthening training result meets the multi-scene accuracy requirements, the strengthening training result is embedded in the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0047] The specific configuration of the power prediction deep fluctuation model construction module 30 will be described in detail below. As described above, based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called, and the photovoltaic ultra-short-term power prediction deep fluctuation model is aggregated and constructed. The power prediction deep fluctuation model construction module 30 further includes: based on the photovoltaic scene segmentation data, a deep classifier is constructed; using the photovoltaic ultra-short-term power prediction network and the enhanced training results, a fluctuation multi-depth parallel prediction channel is constructed, wherein the fluctuation multi-depth parallel prediction channel includes multiple prediction channels of multiple network depths; based on the hierarchical depth data, the deep classifier is connected with the fluctuation multi-depth parallel prediction channel to obtain the photovoltaic ultra-short-term power prediction deep fluctuation model.

[0048] The specific configuration of the photovoltaic ultra-short-term power prediction module 40 will be described in detail below. As described above, the photovoltaic ultra-short-term power prediction deep fluctuation model is deployed in the target photovoltaic power generation system, and the photovoltaic ultra-short-term power prediction module 40 further includes: performing a computing power evaluation on the target photovoltaic power generation system to obtain a computing power evaluation result; based on the computing power evaluation result, performing a deployment evaluation of the target photovoltaic power generation system according to the photovoltaic ultra-short-term power prediction deep fluctuation model to obtain a deployment evaluation result; if the deployment evaluation result does not meet the preset deployment performance constraints, then based on the model compression algorithm, performing a lightweight processing of the prediction deep fluctuation model to obtain a photovoltaic ultra-short-term power prediction deep fluctuation lightweight model; deploying the photovoltaic ultra-short-term power prediction deep fluctuation lightweight model in the target photovoltaic power generation system.

[0049] The accuracy optimization system for photovoltaic ultra-short-term power prediction provided by the embodiment of the present invention can execute the accuracy optimization method for photovoltaic ultra-short-term power prediction provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0050] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0051] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. An accuracy optimization method for photovoltaic ultra-short-term power prediction, characterized in that: The method comprises: Interact with target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data; Based on the photovoltaic scene data, homomorphic division is performed to generate photovoltaic scene segmentation data; Obtaining a power prediction accuracy requirement of a target photovoltaic power generation system, wherein the power prediction accuracy requirement includes an accuracy requirement and a responsiveness requirement; According to the responsiveness requirement, the granularity of the prediction input information duration is divided and the hierarchical prediction input information depth is set; Based on the time depth of the hierarchical prediction input information, according to the photovoltaic scene data and the photovoltaic scene segmentation data, and in combination with the power prediction accuracy requirement, a photovoltaic ultra-short-term power prediction depth fluctuation model is constructed; Deploy the photovoltaic ultra-short-term power prediction deep fluctuation model in the target photovoltaic power generation system to achieve photovoltaic ultra-short-term power prediction; Wherein, based on the time depth of the hierarchical prediction input information, according to the photovoltaic scene data and the photovoltaic scene segmentation data, combined with the power prediction accuracy requirement, a photovoltaic ultra-short-term power prediction depth fluctuation model is constructed, including: Based on the photovoltaic scene data, data extraction is performed according to the hierarchical prediction input information time depth to obtain a hierarchical sample data set, wherein the hierarchical sample data set includes a plurality of hierarchical sample data clusters with different input information time depths; According to the hierarchical prediction input information depth, a multi-scale photovoltaic ultra-short-term power prediction network is constructed, wherein the multi-scale photovoltaic ultra-short-term power prediction network includes a plurality of photovoltaic ultra-short-term power prediction networks with different network depths; Calling the hierarchical sample data set, performing sample division, sending the sample division result to the corresponding photovoltaic ultra-short-term power prediction network, and training the multi-scale photovoltaic ultra-short-term power prediction network; Performing network performance verification on the trained multi-scale photovoltaic ultra-short-term power prediction network to obtain a deep performance data set, wherein the deep performance data set includes input information depth, output response performance, and output accuracy performance; Based on the deep performance data set, the model is deeply configured according to the photovoltaic scene segmentation data to obtain a photovoltaic ultra-short-term power prediction deep fluctuation model.

2. The method according to claim 1, characterized in that Based on the deep performance data set, the model is deeply configured according to the photovoltaic scene segmentation data to obtain a photovoltaic ultra-short-term power prediction deep fluctuation model, including: Analyzing the depth performance data set, performing data fitting, and obtaining an information depth-performance characteristic operator; Calling the photovoltaic scene segmentation data to extract differential photovoltaic scenes; Traversing the power prediction accuracy requirements, performing demand matching based on the different photovoltaic scenarios, and obtaining multi-scenario accuracy requirements; Traversing the multi-scenario accuracy requirements, performing network depth solution according to the information depth-performance characteristic operator, and obtaining hierarchical depth data; Based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called to aggregate and construct the photovoltaic ultra-short-term power prediction depth fluctuation model.

3. The method according to claim 2, characterized in that Based on the hierarchical depth data, the multi-scale photovoltaic ultra-short-term power prediction network is called to aggregate and construct the photovoltaic ultra-short-term power prediction depth fluctuation model, including: Based on the hierarchical depth data, traverse the multi-scale photovoltaic ultra-short-term power prediction network to perform network depth comparison and obtain a network depth difference coefficient; Determining whether the network depth difference coefficient satisfies a preset network depth difference constraint; If the network depth difference coefficient satisfies the network depth difference constraint, the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient is embedded in the photovoltaic ultra-short-term power prediction depth fluctuation model.

4. The method according to claim 3, characterized in that Based on the hierarchical depth data, calling the multi-scale photovoltaic ultra-short-term power prediction network, and aggregating and constructing the photovoltaic ultra-short-term power prediction depth fluctuation model, further comprising: If the network depth difference coefficient does not satisfy the network depth difference constraint, storing the photovoltaic ultra-short-term power prediction network corresponding to the network depth difference coefficient into the candidate network group; Based on the network depth difference coefficient, serializing the candidate network group; According to the serialization results, the best selection is made to extract the first target candidate network; Based on the photovoltaic scene segmentation data, performing enhanced training of the first target candidate network and verifying the enhanced training result; If the enhanced training results meet the multi-scenario accuracy requirements, the enhanced training results are embedded in the photovoltaic ultra-short-term power prediction deep fluctuation model.

5. The method according to claim 4, characterized in that Based on the hierarchical depth data, calling the multi-scale photovoltaic ultra-short-term power prediction network, and aggregating and constructing the photovoltaic ultra-short-term power prediction depth fluctuation model, further comprising: Based on the photovoltaic scene segmentation data, a deep classifier is constructed; Using the photovoltaic ultra-short-term power prediction network and the enhanced training results, a fluctuation multi-depth parallel prediction channel is constructed, wherein the fluctuation multi-depth parallel prediction channel includes multiple prediction channels of multiple network depths; Based on the hierarchical depth data, the depth classifier is connected to the fluctuation multi-depth parallel prediction channel to obtain the photovoltaic ultra-short-term power prediction depth fluctuation model.

6. The method according to claim 1, characterized in that The photovoltaic ultra-short-term power prediction deep fluctuation model is deployed in the target photovoltaic power generation system, including: Evaluate the computing power of the target photovoltaic power generation system and obtain the computing power evaluation results; Based on the computing power evaluation result, performing a deployment evaluation of a target photovoltaic power generation system according to the photovoltaic ultra-short-term power prediction deep fluctuation model, and obtaining a deployment evaluation result; If the deployment evaluation result does not meet the preset deployment performance constraints, the prediction depth fluctuation model is lightweighted based on the model compression algorithm to obtain a photovoltaic ultra-short-term power prediction depth fluctuation lightweight model; The photovoltaic ultra-short-term power prediction deep fluctuation lightweight model is deployed in the target photovoltaic power generation system.

7. An accuracy optimization system for photovoltaic ultra-short-term power prediction, characterized in that: The system is used to implement the accuracy optimization method for photovoltaic ultra-short-term power prediction according to any one of claims 1 to 6, and the system comprises: A photovoltaic scene data homomorphic partitioning module, which is used to interact with the target photovoltaic power generation system, extract historical power data and meteorological data, store and generate photovoltaic scene data, and perform homomorphic partitioning based on the photovoltaic scene data to generate photovoltaic scene segmentation data; A module for dividing the duration of the prediction input information into granularity, wherein the module is used to obtain the power prediction accuracy requirement of the target photovoltaic power generation system, wherein the power prediction accuracy requirement includes the accuracy requirement and the responsiveness requirement, and according to the responsiveness requirement, the granularity of the prediction input information duration is divided, and the hierarchical prediction input information depth is set; A power prediction depth fluctuation model construction module, the power prediction depth fluctuation model construction module is used to construct a photovoltaic ultra-short-term power prediction depth fluctuation model based on the hierarchical prediction input information time depth, according to the photovoltaic scene data and the photovoltaic scene segmentation data, and in combination with the power prediction accuracy requirement; The photovoltaic ultra-short-term power prediction module is used to deploy the photovoltaic ultra-short-term power prediction deep fluctuation model in the target photovoltaic power generation system to realize photovoltaic ultra-short-term power prediction.

Images