A transformer area distributed photovoltaic output prediction method

Abstract

This invention relates to a method for predicting the output of distributed photovoltaic (PV) power in a transformer substation. By abandoning large-scale, coarse-grained meteorological data and employing on-site micro-meteorological monitoring of PV equipment, it accurately captures local irradiance fluctuations, shading, and sudden changes in panel temperature, thereby significantly reducing prediction errors at the data source level and increasing the accuracy of micro-meteorological sensing. Through hierarchical clustering to achieve homogeneous aggregation of scattered PV systems and decoupling of differences, it effectively solves the problem of difficulty in accurately fitting cluster output due to the diverse PV models and operating conditions within the transformer substation. By using prediction residuals to complete adaptive model correction and updates, the parameters of the micro-meteorological-output time-series coupled prediction model can be automatically adjusted according to weather changes, thereby achieving continuous optimization of long-term operational predictions. Furthermore, this invention reduces prediction latency through a lightweight model design and deployment at the edge of the transformer substation. Overall, this invention has the advantages of high accuracy and low latency.

Description

Technical Field

[0001] This invention belongs to the field of distributed renewable energy prediction technology for distribution networks, specifically relating to a method for predicting the output of distributed photovoltaic power in a transformer substation. Background Technology

[0002] With the large-scale grid connection of county-wide and household distributed photovoltaic power, low-voltage distribution areas are characterized by a large number of photovoltaic units, scattered layout, small individual capacity, large regional micro-meteorological differences, and strong randomness in power output. Currently, traditional photovoltaic power output forecasting for distribution areas generally uses general meteorological forecast data from city and county levels. In practice, this method suffers from insufficient meteorological accuracy, poor model adaptability, high forecast latency, and strong cloud dependence, failing to meet the needs of refined operation and management in high-penetration photovoltaic distribution areas. Therefore, to solve the above problems, it is necessary to develop a high-precision, low-latency method for forecasting distributed photovoltaic power output in distribution areas that can achieve on-site forecasting. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a high-precision, low-latency method for predicting distributed photovoltaic power output in a transformer substation that can achieve on-site prediction.

[0004] The objective of this invention is achieved as follows: a method for predicting the output of distributed photovoltaic power in a transformer substation, comprising the following steps: Step 1: Deploy micro-meteorological monitoring terminals at the grid connection points of each distributed photovoltaic power station within the distribution area, and collect local real-time micro-meteorological data and real-time photovoltaic power output data. Then, integrate the collected data to form local micro-meteorological datasets and power output time series datasets for the photovoltaic side. Step 2: Perform outlier removal, wavelet denoising, time-series completion, and spatiotemporal registration on the data within the local micro-meteorological dataset on the photovoltaic side. Based on the processed data, use a correlation screening algorithm to determine the core micro-meteorological features that significantly affect the output of distributed photovoltaic power. Step 3: Based on the photovoltaic installed capacity, equipment model, installation tilt angle, and local micro-meteorological differences of each photovoltaic power station area, the scattered distributed photovoltaic power stations in the power station area are divided into several sub-clusters through three-dimensional indicators such as equipment parameter homogenization, micro-meteorological spatial similarity, and output fluctuation consistency. This completes the hierarchical aggregation of distributed photovoltaic clusters in the power station area, and then constructs a heterogeneous photovoltaic cluster division model for the power station area. Step 4: Construct a micro-meteorological-output time-series coupled prediction model, using the refined micro-meteorological features selected locally as input and the real-time output of each photovoltaic sub-cluster as output, and complete the offline training and online iteration of the model at the edge control terminal of the transformer area; Step 5: Access real-time local micro-meteorological data, realize real-time prediction of photovoltaic cluster output in the power distribution area through the trained micro-meteorological-output time-series coupled prediction model, and complete adaptive correction and update of the model based on the prediction residual.

[0005] Furthermore, the local real-time micro-meteorological data in step 1 includes solar irradiance, photovoltaic panel module temperature, ambient temperature, ambient humidity, near-surface wind speed, and cloud cover and light and shadow variation characteristics.

[0006] Furthermore, the lightweight temporal neural network model of the micro-meteorological-output temporal coupling prediction model in step 4 is deployed at the edge terminal of the transformer area and can adopt an on-site data incremental learning mechanism to achieve on-site training, on-site inference, and on-site updating at the edge.

[0007] Furthermore, the adaptive correction mechanism in step 5 can calculate the residual fluctuation range between the predicted output and the actual output in real time, and dynamically correct the model weight parameters for typical working conditions such as sudden changes in cloud cover, shading, weak light in the morning and evening, and sudden temperature changes.

[0008] The beneficial effects of this invention are as follows: By abandoning large-area coarse-grained meteorological data and adopting on-site micro-meteorological monitoring by photovoltaic equipment, this invention can accurately capture local irradiance fluctuations, shading, and sudden changes in panel temperature, thereby significantly reducing prediction errors at the data source level and increasing the accuracy of micro-meteorological perception. By using hierarchical clustering to aggregate scattered photovoltaic cells into homogeneous clusters and decouple them based on differences, this invention effectively solves the problem of inaccurate fitting of cluster output caused by the diverse photovoltaic models and varying operating conditions in the distribution area. Through predictive residuals to achieve adaptive model correction and updates, the parameters of the micro-meteorological-output time-series coupled prediction model can be automatically corrected for scenarios such as sudden changes in wind and cloud conditions, abrupt changes in sunlight, and seasonal changes, thereby achieving continuous optimization of long-term operational predictions. Furthermore, the invention reduces prediction latency through a lightweight model design and deployment at the edge of the distribution area. In summary, this invention offers the advantages of high accuracy and low latency. Detailed Implementation

[0009] The present invention will now be further described.

[0010] Example: A method for predicting the output of distributed photovoltaic power in a transformer substation, comprising the following steps: Step 1: Deploy micro-meteorological monitoring terminals at the grid connection points of each distributed photovoltaic (PV) system within the distribution area, and collect real-time local micro-meteorological data and real-time PV output data. Then, integrate the collected data to form a local micro-meteorological dataset and an output time-series dataset. The local real-time micro-meteorological data includes solar irradiance, PV panel temperature, ambient temperature, ambient humidity, near-surface wind speed, and cloud cover and shadow variation characteristics. Step 2: Perform outlier removal, wavelet denoising, time-series completion, and spatiotemporal registration on the data within the local micro-meteorological dataset on the photovoltaic side. Based on the processed data, use a correlation screening algorithm to determine the core micro-meteorological features that significantly affect the output of distributed photovoltaic power. Step 3: Based on the photovoltaic installed capacity, equipment model, installation tilt angle, and local micro-meteorological differences of each photovoltaic power station area, the scattered distributed photovoltaic power stations in the power station area are divided into several sub-clusters through three-dimensional indicators such as equipment parameter homogenization, micro-meteorological spatial similarity, and output fluctuation consistency. This completes the hierarchical aggregation of distributed photovoltaic clusters in the power station area, and then constructs a heterogeneous photovoltaic cluster division model for the power station area. Step 4: Construct a micro-meteorological-output time-series coupled prediction model. The model takes the refined micro-meteorological features selected locally as input and the real-time output of each photovoltaic sub-cluster as output. The model is trained offline and iterated online at the edge control terminal of the distribution area. The micro-meteorological-output time-series coupled prediction model is a lightweight time-series neural network model. It is deployed at the edge terminal of the distribution area. With this design, the local data incremental learning mechanism can be used to realize local training, local inference and local update at the edge. Step 5: Access real-time local micro-meteorological data, realize real-time prediction of photovoltaic cluster output in the transformer area through the trained micro-meteorological-output time-series coupled prediction model, complete the model based on the prediction residual, calculate the residual fluctuation range between predicted output and actual output in real time, and dynamically correct the model weight parameters for typical operating conditions such as sudden cloud cover, shading, weak light in the morning and evening, and sudden temperature changes, to complete adaptive correction and update.

[0011] In use, this invention first involves deploying micro-meteorological monitoring terminals at the grid connection points of each distributed photovoltaic (PV) system within the distribution area. These terminals collect real-time local micro-meteorological data and real-time PV output data. By discarding large-area coarse-grained meteorological data and employing on-site micro-meteorological monitoring of PV equipment, this invention can accurately capture local irradiance fluctuations, shading, and sudden changes in panel temperature, thereby significantly reducing prediction errors and increasing the accuracy of micro-meteorological sensing at the data source level. Next, the collected data is integrated to form a local micro-meteorological dataset for the PV side and a time-series output dataset. Finally, the PV side... The data within the local micro-meteorological dataset undergoes outlier removal, wavelet denoising, time-series completion, and spatiotemporal registration. Based on the processed data, a correlation screening algorithm is used to identify core micro-meteorological features that significantly impact distributed photovoltaic (PV) output. Finally, based on the PV installed capacity, equipment model, installation tilt angle, and local micro-meteorological differences within the distribution area, the scattered distributed PV systems are divided into several sub-clusters using three-dimensional indices: equipment parameter homogenization, micro-meteorological spatial similarity, and output fluctuation consistency. This completes the hierarchical aggregation of distributed PV clusters within the distribution area, thereby constructing a heterogeneous PV cluster for the area. A cluster partitioning model is used; this design enables the homogeneous aggregation and decoupling of scattered photovoltaic (PV) systems through hierarchical clustering, effectively solving the problem of inaccurate fitting of cluster output caused by the diverse PV models and varying operating conditions in the distribution area. After completing the above operations, a micro-meteorological-output time-series coupled prediction model is constructed, using locally selected refined micro-meteorological features as input and the real-time output of each PV sub-cluster as output. Offline training and online iteration of the model are completed at the edge control terminal of the distribution area. The micro-meteorological-output time-series coupled prediction model is a lightweight time-series neural network model, deployed at the edge terminal of the distribution area. The invention utilizes an incremental learning mechanism based on local data to achieve on-site training, inference, and updates at the edge, thereby reducing prediction latency. Finally, it integrates real-time local micro-meteorological data and uses a trained micro-meteorological-output time-series coupled prediction model to achieve real-time prediction of photovoltaic cluster output in the distribution area. Based on the prediction residuals, the model is completed, and the residual fluctuation range between predicted and actual output is calculated in real time. Furthermore, for typical operating conditions such as sudden cloud cover, shading, weak light in the morning and evening, and sudden temperature changes, the model weight parameters are dynamically adjusted to achieve adaptive correction and updates. Overall, this invention has the advantages of high accuracy and low latency.

[0012] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the output of distributed photovoltaic power in a transformer substation, characterized in that, Includes the following steps: Step 1: Deploy micro-meteorological monitoring terminals at the grid connection points of each distributed photovoltaic power station within the distribution area, and collect local real-time micro-meteorological data and real-time photovoltaic power output data. Then, integrate the collected data to form local micro-meteorological datasets and power output time series datasets for the photovoltaic side. Step 2: Perform outlier removal, wavelet denoising, time-series completion, and spatiotemporal registration on the data within the local micro-meteorological dataset on the photovoltaic side. Based on the processed data, use a correlation screening algorithm to determine the core micro-meteorological features that significantly affect the output of distributed photovoltaic power. Step 3: Based on the photovoltaic installed capacity, equipment model, installation tilt angle, and local micro-meteorological differences of each photovoltaic power station area, the scattered distributed photovoltaic power stations in the power station area are divided into several sub-clusters through three-dimensional indicators such as equipment parameter homogenization, micro-meteorological spatial similarity, and output fluctuation consistency. This completes the hierarchical aggregation of distributed photovoltaic clusters in the power station area, and then constructs a heterogeneous photovoltaic cluster division model for the power station area. Step 4: Construct a micro-meteorological-output time-series coupled prediction model, using the refined micro-meteorological features selected locally as input and the real-time output of each photovoltaic sub-cluster as output, and complete the offline training and online iteration of the model at the edge control terminal of the transformer area; Step 5: Access real-time local micro-meteorological data, realize real-time prediction of photovoltaic cluster output in the power distribution area through the trained micro-meteorological-output time-series coupled prediction model, and complete adaptive correction and update of the model based on the prediction residual.

2. The method for predicting the output of distributed photovoltaic power in a transformer substation as described in claim 1, characterized in that: The local real-time micro-meteorological data in step 1 includes solar irradiance, photovoltaic panel module temperature, ambient temperature, ambient humidity, near-surface wind speed, and cloud cover and light and shadow variation characteristics.

3. The method for predicting the output of distributed photovoltaic power in a transformer substation as described in claim 1, characterized in that: The lightweight time-series neural network model of the micro-meteorological-output time-series coupling prediction model in step 4 is deployed at the edge terminal of the transformer area. It can adopt the local data incremental learning mechanism to realize local training, local inference and local update at the edge.

4. The method for predicting the output of distributed photovoltaic power in a transformer substation as described in claim 1, characterized in that: The adaptive correction mechanism in step 5 can calculate the residual fluctuation range between the predicted output and the actual output in real time, and dynamically correct the model weight parameters for typical working conditions such as sudden changes in cloud cover, shadow occlusion, weak light in the morning and evening, and sudden temperature changes.