CN-121998234-A - Assessment method, system, equipment and medium for power prediction reliability

CN121998234ACN 121998234 ACN121998234 ACN 121998234ACN-121998234-A

Abstract

The invention relates to the technical field of power systems and discloses a power prediction reliability-oriented assessment method, a system, equipment and a medium, wherein the method comprises the steps of responding to a received power prediction result to be assessed, and acquiring weather characteristics of a prediction target day; the method comprises the steps of calculating the similarity between weather features of a predicted target day and weather features corresponding to a plurality of history days in a history database to obtain a plurality of weather similarities, screening similar history days from a plurality of histories Japan and China based on the corresponding weather similarities, obtaining model scores corresponding to each similar history day, wherein the model scores are model historical performance scores obtained by quantifying the prediction accuracy of a power prediction model on the corresponding similar history days, calculating target credibility according to the weather features of each similar history day and the corresponding model scores, calculating penalty factors according to the model scores of each similar history day, and calculating the credibility scores of the power prediction results to be evaluated according to the target credibility and the penalty factors, so that effective evaluation of the power prediction credibility is achieved.

Inventors

Kong Zekun
MA DONG
ZHANG ZHE
Hu Sige
XU SIDA
CHEN YUWEN
NIU SHIKAI
CHEN YINSHENG
LI YUNFEI

Assignees

华电电力科学研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20251217

Claims (10)

1. A power prediction reliability oriented assessment method, the method comprising: responding to a received power prediction result to be evaluated, and acquiring weather characteristics of a prediction target day, wherein the prediction target day is associated with the power prediction result to be evaluated; Calculating the similarity of the weather features of the predicted target day and the weather features corresponding to a plurality of history days in a history database respectively to obtain a plurality of weather similarities; Screening similar historical days from a plurality of histories Japan and China based on corresponding meteorological similarity, and obtaining a model score corresponding to each similar historical day, wherein the model score is a model historical performance score obtained by quantifying the prediction accuracy of a power prediction model on the corresponding similar historical day; Calculating the target credibility according to the meteorological features of each similar historical day and the corresponding model scores; calculating a penalty factor according to the model scores of the similar historical days, wherein the penalty factor is inversely related to the stability of the power prediction model; And calculating the reliability score of the power prediction result to be evaluated according to the target reliability and the penalty factor.
2. The power prediction confidence-oriented assessment method according to claim 1, wherein the calculating the target confidence level based on the weather features and the corresponding model scores of each of the similar history days comprises: And for each similar history day, weighting and fusing the scores of the corresponding models by taking the meteorological features of the similar history day as weight coefficients to obtain the target credibility.
3. The power prediction confidence level oriented assessment method according to claim 1, wherein calculating a penalty factor based on model scores of each of the similar historical days comprises: Calculating standard deviation of scores of corresponding models of all similar history days; And inputting the standard deviation into a preset function to calculate, and outputting a penalty factor, wherein the preset function comprises at least one of a linear function and an exponential function.
4. The power prediction confidence level oriented assessment method according to claim 1, wherein calculating a penalty factor based on model scores of each of the similar historical days comprises: Calculating the average value and standard deviation of the scores of the corresponding models of all similar historical days; Calculating the ratio of the standard deviation to the average value to obtain a stability index; and inputting the stability index into a preset function to calculate, and outputting a penalty factor, wherein the preset function comprises at least one of a linear function and an exponential function.
5. The power prediction reliability oriented assessment method according to claim 1, wherein said screening similar history days from a plurality of histories Japan and China based on respective weather similarity comprises: sequencing the weather similarity of each history day according to the value, and obtaining a sequencing result; And screening a preset number of history days with highest similarity as the similar history days based on the sorting result.
6. The power prediction reliability oriented assessment method according to claim 1, wherein the calculating the reliability score of the power prediction result to be assessed according to the target reliability and the penalty factor comprises: and multiplying the target credibility by the penalty factor to obtain a credibility score of the power prediction result to be evaluated.
7. The power prediction confidence oriented assessment method according to any one of claims 1 to 6, wherein after the calculating of the confidence score of the power prediction result to be assessed according to the target confidence and the penalty factor, the method further comprises: and outputting the reliability score to an electric power production operation system or a decision support terminal so that a corresponding decision maker can judge the reliability of the power prediction result to be evaluated based on the reliability score and make a corresponding decision according to the judgment result.
8. A power prediction reliability oriented evaluation system, the system comprising: The characteristic acquisition module is used for responding to the received power prediction result to be evaluated and acquiring weather characteristics of a prediction target day, wherein the prediction target day is associated with the power prediction result to be evaluated; The similarity calculation module is used for calculating the similarity between the weather features of the predicted target day and the weather features corresponding to a plurality of history days in the history database respectively to obtain a plurality of weather similarities; the scoring acquisition module is used for screening similar historical days from the histories Japan and China based on the corresponding meteorological similarity, and acquiring model scores corresponding to each similar historical day, wherein the model scores are model historical performance scores obtained by quantification of the prediction accuracy of the power prediction model on the corresponding similar historical days; the credibility calculation module is used for calculating target credibility according to the meteorological features of each similar historical day and the corresponding model scores; the penalty factor calculation module is used for calculating penalty factors according to the model scores of the similar historical days, and the penalty factors are in negative correlation with the stability of the power prediction model; and the credibility evaluation module is used for calculating the credibility score of the power prediction result to be evaluated according to the target credibility and the penalty factor.
9. An electronic device comprising a controller, the controller comprising a memory and a processor, the memory and the processor being communicatively coupled to each other, the memory having stored therein computer instructions, the processor executing the computer instructions to perform the power prediction reliability oriented assessment method of any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the power prediction reliability oriented evaluation method of any one of claims 1 to 7.

Description

Assessment method, system, equipment and medium for power prediction reliability Technical Field The invention relates to the technical field of power systems, in particular to a power prediction reliability-oriented evaluation method, a power prediction reliability-oriented evaluation system, power prediction reliability-oriented evaluation equipment and power prediction reliability-oriented evaluation media. Background In an electric power system, especially in relation to power prediction of new energy sources (such as photovoltaic, wind power and the like), a current mainstream scheme learns a mapping relation between weather forecast data and actual power through a machine learning/artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) technology so as to generate a power prediction result of a prediction day, and weather similarity is used as a model evaluation basis to judge suitability of the machine learning/AI model (such as a similar sample required by screening model training and evaluating the applicability of the model under a specific weather scene). In addition, the conventional "similar day" prediction scheme is still widely used, and the core logic of the conventional "similar day" prediction scheme is to select one or more historical days with closest weather conditions by calculating the similarity between the predicted day and weather features (such as irradiance, wind speed, temperature, humidity, etc.) of each day in the historical database, and then generate the power predicted value of the predicted day in a weighted average or other manner based on the historical actual power data or the historical predicted result of the similar days. In practical application, the calculated weather similarity is often used as a weighted basis in the prediction process, and is also simply regarded as a rough reference index of the reliability of the prediction result, and the direct application scene of the reliability reference is to support a power production decision maker (such as a dispatcher, an operation and maintenance person, a manager and the like) to select a reporting result, namely, determine which model of the prediction result should be selected for reporting, further judge whether the prediction result is reliable or not, and make a corresponding decision when the prediction result is reliable, so that risk management and control can be enhanced, and the safety and stability of power production are ensured. However, in the prior art, when reliability evaluation is performed by using the meteorological similarity, for example, the evaluation dimension is single, the reliability of the obtained prediction result is still doubtful, and the evaluation conclusion may be distorted if the actual prediction performance of the model on the history day is poor even if the meteorological conditions are highly similar due to the fact that the power prediction model itself is not included in the history performance on the similarity day due to the fact that the evaluation dimension is single and the meteorological matching degree is excessively depended. Further, when evaluating based on multiple similar days, existing approaches typically only focus on the average level of model historical performance, and cannot quantify its volatility and stability during different days, thus making it difficult to reveal the potential risks involved in model performance instability. In summary, the prior art cannot effectively fuse multiple dimensions such as weather matching degree, model history performance and stability, so that the weather matching degree, model history performance and stability cannot output a unified and quantized comprehensive reliability score, and thus a decision maker cannot easily and intuitively grasp the overall reliability of a prediction result, and the accuracy and efficiency of power dispatching and transaction decision are affected. Therefore, there is a need for a power prediction reliability assessment scheme that can fuse multidimensional information, quantitatively assess risk of fluctuations, and provide intuitive decision support. Disclosure of Invention The invention provides an evaluation method, system, equipment and medium for power prediction reliability, which are used for solving the problems that the comprehensive quantitative evaluation of the power prediction reliability is difficult to realize and the power decision is seriously influenced because the prior art only relies on single weather similarity evaluation to have a plurality of defects. In a first aspect, the present invention provides a method for evaluating reliability of power-oriented prediction, where the method includes: Responding to the received predicted power result to be evaluated, acquiring meteorological features of a predicted target day, wherein the predicted target day is associated with the predicted power result to be evaluated; Calculating the similarity between the weather feature