CN-121983922-A - Photovoltaic tracker initiative protection system based on AI meteorological big model

CN121983922ACN 121983922 ACN121983922 ACN 121983922ACN-121983922-A

Abstract

The invention relates to the technical field of meteorological information and photovoltaic trackers, in particular to a photovoltaic tracker active protection system based on an AI meteorological large model, which comprises a meteorological data fusion module, an AI meteorological prediction module, a protection instruction generation module, an instruction issuing and executing module and a cloud operation and maintenance management platform for managing the modules, wherein the cloud operation and maintenance management platform is used for fusing multi-source meteorological data and outputting station-level meteorological data; the method comprises the steps of generating a disaster weather prediction result comprising disaster type, time and intensity through a trained weather large model based on the data, dynamically generating a tracker protection instruction comprising action type, target angle and execution time according to the result, and finally issuing the instruction to an edge side through the Internet of things to control a photovoltaic tracker to complete active protection before disaster occurs. The invention realizes the normal form transition from the hysteresis response to the accurate front protection, constructs a seamless closed loop from prediction to control, and optimizes the power generation benefits while ensuring the safety of equipment.

Inventors

XU LE
WANG GUANGMING
LI SHANGDONG
DING FENGJUN
ZHU JIANJING

Assignees

杭州帷盛科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (10)

1. An AI weather large model-based photovoltaic tracker active protection system, comprising: The meteorological data fusion module is used for fusing multi-source meteorological data and outputting station-level meteorological data; the AI weather prediction module is used for receiving the station-level weather data and outputting disaster weather prediction results; the protection instruction generation module is used for receiving the disaster weather prediction result and generating a tracker protection instruction; the instruction issuing and executing module is used for issuing the tracker protection instruction to the photovoltaic tracker and controlling the photovoltaic tracker to execute the active protection action; the cloud operation and maintenance management platform is used for visually displaying disaster prediction results, protection instruction issuing states, equipment execution feedback and historical event backtracking.
2. The active protection system of a photovoltaic tracker based on AI weather large model according to claim 1, wherein the fusion of multisource weather data and output of station-level weather data specifically comprises: And receiving multi-source meteorological data from a global meteorological data source, regional meteorological monitoring stations and local meteorological sensors of a target station, performing time stamp alignment and space grid matching on the multi-source meteorological data to complete space-time alignment, performing quality calibration through outlier detection and data interpolation, and performing space downscaling processing on the data based on a geographic information system to generate a calibrated station-level meteorological data sequence matched with the micro topography of the target station.
3. The active protection system of a photovoltaic tracker based on AI weather large model according to claim 2, wherein the quality calibration is performed by outlier detection and data interpolation, and the spatial downscaling of the data is performed based on a geographic information system, specifically comprising: Performing abnormal value detection and data interpolation on the multi-source meteorological data for quality calibration, wherein the abnormal value is identified based on a standard difference method or a four-bit distance method, and the data is completed by adopting a time sequence interpolation or a space Kriging interpolation method; And based on the geographic information system, combining the digital elevation model and the land utilization type data to perform space downscaling on the meteorological data.
4. The active protection system of a photovoltaic tracker based on AI weather large model according to claim 2, wherein the steps of receiving the station-level weather data and outputting disaster weather prediction result comprise: Receiving the calibrated station-level meteorological data sequence, and inputting the calibrated station-level meteorological data sequence into a meteorological large model which is trained by historical meteorological data and disaster event records of the target station in advance; The weather large model is a time sequence prediction model based on an attention mechanism, and generates disaster weather prediction results of a target station in a preset time period in the future, wherein the disaster weather prediction results are structured data, and comprise disaster types, expected occurrence time, duration and intensity levels.
5. The photovoltaic tracker active protection system based on an AI weather large model according to claim 4, wherein the time sequence prediction model comprises a short-term prediction sub-model and a long-term prediction sub-model which are arranged in parallel, and prediction results output by the two sub-models are processed by a fusion layer to generate a unified disaster weather prediction result; The short-term prediction sub-model is used for processing a high-frequency meteorological sequence of 0-2 hours in the future, and inputs the high-frequency meteorological sequence and the short-term cloud image radar data, wherein the high-frequency meteorological sequence comprises a calibrated station-level meteorological data sequence and short-term cloud image radar data; the long-term predictor model is used for processing weather trend of 2-24 hours in the future, and the input of the long-term predictor model comprises a calibrated station-level weather data sequence and numerical weather forecast data.
6. The AI weather large model-based photovoltaic tracker active protection system of claim 5, wherein the fusion process of the fusion layer comprises: aligning the first prediction result output by the short-term prediction sub-model with the second prediction result output by the long-term prediction sub-model in the time dimension; respectively evaluating the prediction confidence of the first prediction result and the second prediction result in the overlapping period; distributing fusion weights for the first prediction result and the second prediction result according to the prediction confidence; and carrying out weighted average calculation on the first prediction result and the second prediction result based on the fusion weight, and outputting disaster weather prediction results comprising disaster types, expected occurrence time, duration and intensity levels.
7. The active protection system of a photovoltaic tracker based on AI weather large model of claim 6, wherein receiving the disaster weather prediction result and generating a tracker protection instruction specifically comprises: receiving the disaster weather forecast result, inquiring a preset equipment protection strategy library according to the disaster type and the intensity level in the disaster weather forecast result, and matching the corresponding protection action type and the target angle; Calculating to obtain execution time according to the expected occurrence time and duration in the disaster weather forecast result; And dynamically generating a tracker protection instruction based on the protection action type, the target angle and the execution time.
8. The active protection system of a photovoltaic tracker based on an AI weather large model of claim 7, wherein the protection instruction generation module is further configured to perform protection opportunity optimization based on disaster weather prediction results in combination with real-time tracker state data, and specifically comprises: receiving real-time tracker state data, wherein the real-time tracker state data at least comprises the current angle of the photovoltaic tracker and real-time generation power; and calculating the latest time point at which the protection action can be postponed under the premise of ensuring safety based on the expected occurrence time in the disaster weather forecast result and the real-time power generation, and optimizing the execution time to the latest time point, thereby minimizing the power generation loss.
9. The active protection system of a photovoltaic tracker based on AI weather large model according to claim 8, wherein the tracker protection instruction is issued to the photovoltaic tracker to control the photovoltaic tracker to execute the active protection action, specifically comprising: issuing the tracker protection instruction to an edge computing node in a target station through an Internet of things communication network, wherein the edge computing node performs validity check and safety logic judgment on the received tracker protection instruction; After the verification and judgment are passed, the edge computing node drives a motor of the photovoltaic tracker executing mechanism before the execution time reaches, and the photovoltaic tracker is controlled to complete the protection action corresponding to the protection action type and the target angle.
10. The photovoltaic tracker active protection system based on the AI weather large model according to claim 9, wherein the edge computing node is deployed in a photovoltaic tracker control box, and after the protection action is completed, the edge computing node collects state feedback data after the tracker is executed and uploads the state feedback data to a cloud management platform.

Description

Photovoltaic tracker initiative protection system based on AI meteorological big model Technical Field The invention relates to the technical field of meteorological information and photovoltaic trackers, in particular to an active protection system of a photovoltaic tracker based on an AI meteorological large model. Background The weather protection system of the traditional photovoltaic tracker generally adopts a monitoring-responding passive mode, relies on sensors such as on-site wind speed, hail and the like to monitor in real time, and triggers protection action after the threshold exceeds the standard. The mode has inherent defects that firstly, physical delay exists for about 10 minutes from monitoring and execution, equipment is damaged before protection is finished under weather conditions of rapid abrupt change such as strong convection weather, secondly, risk cannot be prejudged depending on single-point real-time data, protection action is always delayed from disaster occurrence, thirdly, fixed threshold triggering is adopted, and differentiation and self-adaptive protection cannot be carried out according to disaster type, intensity and duration. The existing weather forecast technology can provide short-term weather information, but has two general pain points, namely a forecast result is a large-range and generalized weather parameter, such as the regional average wind speed, the forecast result cannot be accurately matched with the micro-topography and micro-climate of a specific photovoltaic station, the forecast precision is insufficient on the scale of the station, a forecast system and an equipment control system are mutually independent, the weather information cannot be directly converted into an executable control instruction, and a data island and a semantic gap exist, so that the forecast result is difficult to be used for active risk intervention. Therefore, how to realize high-precision and short-cut station-level weather risk prediction and how to seamlessly, automatically and accurately convert the prediction result into a protection instruction which can be executed by equipment in the lead time becomes a key technical bottleneck for improving the safety and the operation efficiency of the photovoltaic asset. Disclosure of Invention The invention aims to provide a photovoltaic tracker active protection system based on an AI weather large model, which aims to solve the problems of equipment damage and power generation loss caused by weather protection hysteresis and decision stiffness of a traditional photovoltaic tracker in the prior art. In order to solve the technical problems, the invention specifically provides the following technical scheme: an AI weather large model-based active protection system for a photovoltaic tracker, comprising: The meteorological data fusion module is used for fusing multi-source meteorological data and outputting station-level meteorological data; the AI weather prediction module is used for receiving the station-level weather data and outputting disaster weather prediction results; the protection instruction generation module is used for receiving the disaster weather prediction result and generating a tracker protection instruction; the instruction issuing and executing module is used for issuing the tracker protection instruction to the photovoltaic tracker and controlling the photovoltaic tracker to execute the active protection action; The cloud operation and maintenance management platform is used for visually displaying disaster prediction results, protection instruction issuing states, equipment execution feedback and historical event backtracking, and further supporting continuous iterative optimization of a weather large model and a protection strategy library based on historical disaster event and protection effect data. As a preferable scheme of the invention, the method for fusing multisource meteorological data and outputting station-level meteorological data specifically comprises the following steps: And receiving multi-source meteorological data from a global meteorological data source, regional meteorological monitoring stations and local meteorological sensors of a target station, performing time stamp alignment and space grid matching on the multi-source meteorological data to complete space-time alignment, performing quality calibration through outlier detection and data interpolation, and performing space downscaling processing on the data based on a geographic information system to generate a calibrated station-level meteorological data sequence matched with the micro topography of the target station. As a preferable scheme of the invention, the quality calibration is carried out by outlier detection and data interpolation, and the data is subjected to spatial downscaling based on a geographic information system, which comprises the following steps: performing outlier detection and data interpolation on the multi-source me