A power weather data fusion method and system based on multi-source weather forecast and a storage medium

Abstract

The application discloses a power meteorological data fusion method and system based on multi-source meteorological prediction and a storage medium, and combines power meteorological monitoring and multi-source numerical weather prediction data. Through error analysis on the multi-source numerical weather prediction data, the numerical weather prediction is changed from single deterministic prediction to multi-source fusion prediction, which is beneficial to the real application of meteorological information to actual business work of the power grid. The application can provide precision evaluation of different prediction data sources, provide multi-source prediction fusion data, improve prediction precision, and provide technical support for power meteorological fine prediction.

Description

Technical Field

[0001] This invention relates to a power meteorological data fusion method based on multi-source meteorological forecasting, belonging to the field of power grid disaster prevention and mitigation. Background Technology

[0002] The power grid system needs to pay special attention to weather forecasts, especially weather information related to flood prevention, lightning protection, snow prevention, earthquake prevention, and disaster reduction, which directly affects the safe operation of the power grid. Currently, the weather forecasts and warnings issued by domestic meteorological departments lack professional power grid information, cannot be combined with the needs of the power grid system for safe operation, and cannot fully meet the actual needs of the power grid system to defend against various severe weather conditions, leaving the power grid in a passive position in the defense against severe weather.

[0003] Numerical weather forecasting is crucial for power grid disaster prevention and mitigation. With continuous improvements in computing power, numerical model systems have become the main support for weather forecasting. However, the nonlinear motion of the atmosphere itself introduces many uncertainties into the accuracy of numerical forecasts, and a single deterministic forecast cannot provide complete forecast information. Therefore, combining power meteorological monitoring and numerical weather forecasting data, and transforming numerical weather forecasting from single deterministic forecasts to multi-source fusion forecasts, has become a trend.

[0004] Patent 109583467A discloses a method and system for fusing power meteorological data. The method collects power meteorological data from at least two meteorological data generation channels; divides the collected power meteorological data into blocks according to spatial, temporal, and meteorological elements; and fuses the block-based power meteorological data. This method and system, focusing on the heterogeneous nature of power meteorological data, divides and integrates multi-source power meteorological data collected from multiple meteorological data generation channels, directly achieving the assimilation of multi-source heterogeneous observation data. The above method mainly utilizes different data sources to cross-check each other according to spatial, temporal, and meteorological elements. If the values ​​of the checked power meteorological data are not within a preset reasonable range, data interpolation and fusion are performed to form a power meteorological dataset. However, this method does not address the inherent properties of the data sources themselves, especially ignoring the error characteristics of the forecast sources. Currently, it is necessary to solve the error problem of forecast data sources, shifting from single deterministic forecasts to multi-source fusion forecasts to improve the accuracy of forecast data sources. Summary of the Invention

[0005] The technical problem to be solved by this invention is: how to make full use of multiple numerical weather forecast source data in the process of power meteorological monitoring to improve forecast accuracy and provide technical support for refined power meteorological forecasting.

[0006] To address the aforementioned technical problems, this invention provides a method for fusing power meteorological data based on multi-source meteorological forecasts, comprising the following steps:

[0007] Step S1: Collect power meteorological station monitoring data and multiple numerical weather forecast source data for at least one year;

[0008] Step S2: Use downscaling methods to map the multi-source forecast data "point-to-point" to the location of the power meteorological station to form a meteorological forecast and monitoring dataset;

[0009] Step S3: Use the Bayesian model averaging method to fuse multi-source weather forecasts and establish a multi-source weather forecast fusion model, specifically as follows:

[0010]

[0011] In the formula, ω k a represents the weight value of the k-th forecast data source. k b k The constants 1 and 2 of the k-th forecast data source, respectively, f k E(y|f1,...,f) represents the forecast result from the k-th forecast data source. k () represents the forecast fusion result based on k forecast data sources;

[0012] Step S4: Train the parameters of the multi-source forecast fusion model, with constant a k constant two b k Depending on the specific forecast data source, and based on the training period data, it can be calculated using a linear regression equation, as follows:

[0013]

[0014]

[0015] In the formula, T represents the training period length, f k,t O represents the forecast result of the k-th forecast data source at time t. t This represents the measured data during the training period. The average forecast value over the training period of the k-th forecast data source. This represents the average value of the measured data during the training period.

[0016] For the weight coefficient ω of each data source k Specifically, the error term from each forecast data source and the measured data is normalized and calculated as follows:

[0017]

[0018]

[0019] Among them, G k f represents the error between the k-th forecast data source and the observed value during the training period.k,t For the prediction value at time t of the k-th prediction data source during the training period, OBS t The values ​​are the observations during the training period, and T is the length of the training period.

[0020] Step S5: The multi-source forecast fusion results are evaluated using three indicators: Continuous Rank Probability Score (CRPS), Anomaly Correlation Coefficient (ACC), and Root Mean Square Error (RMSE).

[0021] Continuous Rank Probability Rating (CRPS), using S CRP Represented as:

[0022]

[0023] In the formula, f(y) i H(y) represents the probability density of the predicted value; i -O i ) is the Heaviside function, when y i -O i When ≥0, H(y) i -O i H(y) = 1; otherwise H(y) = 1. i -O i ) = 0, S CRP The smaller the value, the better the forecast result. i Represents the forecast value; O i dx represents the actual value, x has no specific meaning, and dx represents the differential, indicating that the independent variable x is very small, there is no smaller value than it, but it is not equal to zero.

[0024] Anomaly correlation coefficient ACC, using C AC Represented as:

[0025]

[0026] In the formula, m represents the number of monitoring stations; y i Represents the forecast value; O i C represents the actual value. AC The higher the value, the better the forecast result. This represents the average of the forecast values. This represents the average of the actual values;

[0027] Root mean square error (RMSE), expressed in E RMS Represented as:

[0028]

[0029] In the formula, l represents the sum of the data; y i Represents the forecast value; O i Represents the actual value. E RM SThe smaller the value, the better the forecast.

[0030] A power meteorological data fusion system based on multi-source weather forecasting includes the following functional modules:

[0031] Multi-source meteorological data acquisition module: Collects monitoring data from power meteorological stations and data from multiple numerical weather forecast sources for at least one year;

[0032] Dataset module: Using downscaling methods, multi-source forecast data are mapped "point-to-point" to the locations of power meteorological stations to form a meteorological forecast and monitoring dataset;

[0033] Fusion Model Establishment Module: The Bayesian model averaging method is used to fuse multi-source weather forecasts and establish a multi-source weather forecast fusion model.

[0034] Training module: Trains the parameters of the multi-source forecast fusion model;

[0035] Evaluation module: Verify and evaluate the results of multi-source forecast fusion.

[0036] A computer-readable storage medium is provided for storing the above-mentioned method and system for fusing power meteorological data based on multi-source meteorological forecasts.

[0037] The beneficial effects achieved by this invention are as follows: This invention provides a power meteorological data fusion method based on multi-source meteorological forecasts. By introducing a highly centralized probability density statistical post-processing method, the Bayesian model averaging method, multi-source meteorological forecast data is fused and applied. The error characteristics of different data sources are analyzed and trained, and the optimal model parameters are substituted to obtain the fused forecast results. Further forecast verification and evaluation are then conducted, which is beneficial for truly applying meteorological information to actual power grid operations. For extended-range forecasts with longer lead times (10-30 days), deterministic forecasting performance is poor. Using the multi-source forecast fusion algorithm provided by this invention, the forecasting techniques of multiple sources can be combined, resulting in better forecasting performance for extended-range weather. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the principle and flow of the present invention;

[0039] Figure 2 This is a schematic diagram of the horizontal bilinear interpolation algorithm. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] like Figure 1 As shown, this embodiment provides a power meteorological data fusion method based on multi-source meteorological forecasts, including the following steps:

[0042] Step S1: Collect power meteorological station monitoring data and multiple numerical weather forecast source data for at least one year.

[0043] By utilizing the power meteorological monitoring stations provided in the power grid equipment monitoring network, meteorological monitoring element data for at least one year collected by the meteorological stations are obtained, including temperature, air pressure, humidity, wind speed and direction, rainfall, solar radiation, etc. At the same time, based on gridded forecast data from different numerical weather prediction sources, meteorological forecast gridded data covering the area where the power meteorological monitoring stations are located are obtained for at least one year, with specific forecast elements including temperature, air pressure, humidity, wind speed and direction, rainfall, solar radiation, etc.

[0044] Step S2: Use downscaling methods to map the multi-source forecast data "point-to-point" to the location of the power meteorological station to form a meteorological forecast and monitoring dataset;

[0045] Downscaling calculations were performed on grid point data from different numerical weather forecast sources. Bilinear interpolation was used to calculate numerical forecast data for the locations of all power meteorological monitoring stations, forming a dataset of at least one year of simultaneous labeling of power meteorological monitoring elements and forecast elements.

[0046] Step S3: Perform multi-source meteorological forecast fusion using the Bayesian model averaging method. Select at least one year's worth of power meteorological monitoring and forecast data as training samples. Based on the principle of the Bayesian model averaging method, the forecast quantity y is fused with the data in the training dataset y. T Below, based on K different forecast data sources (M1, ..., M... K The predicted probability density under () is:

[0047]

[0048] p(y|M k () is based on forecast data source M k The predicted probability density, p(M) k |y T M is the forecast data source. k During the training period, the sum of the contributions of all modes is 1, meaning that the relative contribution is equal to the sum of the contributions of all modes. Therefore p(M) k |y T ) is mode M k The weight.

[0049] The probability density of the Bayesian model average prediction is based on p(M) k |y T The Bayesian model averaging method uses y|f1,...,f as the weights to calculate the conditional probability density of all participating forecast models. k Simplified to:

[0050]

[0051] In the formula, f k G represents the forecast result of the k-th model. k (y|f k ) refers to the forecast quantity y in f k Conditional probability density under ω; k This represents the weight value of the k-th pattern, and

[0052] Since meteorological elements roughly follow a normal distribution, the average value is a. k +b k f k The variance is σ 2 Therefore g k (y|f k ) is based on a k +b k f k The normal probability density function is given by σ, where σ is the mean and σ is the variance, and σ is the standard deviation.

[0053] Based on the above model, the multi-source weather forecast fusion model is as follows:

[0054]

[0055] In the formula, ω k a represents the weight value of the k-th forecast data source. k b k The constants 1 and 2 of the k-th forecast data source, respectively, f k E(y|f1,...,f) represents the forecast result from the k-th forecast data source. k () represents the forecast fusion result based on k forecast data sources;

[0056] Step S4: Training parameters for the multi-source forecast fusion model, with constant a k constant two b k Depending on the specific forecast data source, and based on the training period data, it can be calculated using a linear regression equation, as follows:

[0057]

[0058]

[0059] In the formula, T represents the training period length, f k,t O represents the forecast result of the k-th forecast data source at time t. t This represents the measured data during the training period. The average forecast value over the training period of the k-th forecast data source. This represents the average value of the measured data during the training period.

[0060] For the weight coefficient ω of each data source k Specifically, the error term from each forecast data source and the measured data is normalized and calculated as follows:

[0061]

[0062]

[0063] Among them, G k f represents the error between the k-th forecast data source and the observed value during the training period. k,t For the prediction value at time t of the k-th prediction data source during the training period, OBS t The values ​​are the observations during the training period, and T is the length of the training period.

[0064] Step S5: Multi-source forecast fusion verification and evaluation. To fully verify and evaluate the multi-source forecast fusion results, three indicators are used for verification: Continuous Rank Probability Score (CRPS), Anomaly Correlation Coefficient (ACC), and Root Mean Square Error (RMSE).

[0065] Continuous Rank Probability Rating (CRPS), using S CRP Represented as:

[0066]

[0067] In the formula, f(y) i H(y) represents the probability density of the predicted value; i -O i ) is the Heaviside function, when y i -O i When ≥0, H(y) i -O i H(y) = 1; otherwise H(y) = 1. i -O i ) = 0, S CRP The smaller the value, the better the forecast result. i Represents the forecast value; O i dx represents the actual value, x has no specific meaning, and dx represents the differential, indicating that the independent variable x is very small, there is no smaller value than it, but it is not equal to zero.

[0068] Anomaly correlation coefficient ACC, using C AC Represented as:

[0069]

[0070] In the formula, m represents the number of monitoring stations; y i Represents the forecast value; O iC represents the actual value. AC The higher the value, the better the forecast result. This represents the average of the forecast values. This represents the average of the actual values;

[0071] Root mean square error (RMSE), expressed in E RMS Represented as:

[0072]

[0073] In the formula, l represents the sum of the data; y i Represents the forecast value; O i Represents the actual value. E RM S The smaller the value, the better the forecast.

[0074] A power meteorological data fusion system based on multi-source weather forecasting includes the following functional modules:

[0075] Multi-source meteorological data acquisition module: Collects monitoring data from power meteorological stations and data from multiple numerical weather forecast sources for at least one year;

[0076] Dataset module: Using downscaling methods, multi-source forecast data are mapped "point-to-point" to the locations of power meteorological stations to form a meteorological forecast and monitoring dataset;

[0077] Fusion Model Establishment Module: The Bayesian model averaging method is used to fuse multi-source weather forecasts and establish a multi-source weather forecast fusion model.

[0078] Training module: Trains the parameters of the multi-source forecast fusion model;

[0079] Evaluation module: Verify and evaluate the results of multi-source forecast fusion.

[0080] A computer-readable storage medium is provided for storing the above-mentioned method and system for fusing power meteorological data based on multi-source meteorological forecasts.

[0081] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A power meteorological data fusion method based on multi-source meteorological forecasting, characterized in that, Includes the following steps: Step S1: Collect power meteorological station monitoring data and multiple weather forecast source data for at least one year; Step S2: Use downscaling methods to map the multi-source weather forecast data "point-to-point" to the location of the power meteorological station, forming a dataset of weather forecast elements and monitoring elements; Step S3: Based on the datasets of meteorological forecast elements and monitoring elements, use the Bayesian model averaging method to fuse multi-source meteorological forecast data, establish a multi-source meteorological forecast fusion model, and obtain the meteorological forecast fusion results; Step S4: Train the parameters of the multi-source weather forecast fusion model; Step S5: Verify and evaluate the results of multi-source weather forecast fusion; In step S1, the power meteorological monitoring stations provided in the power grid equipment monitoring network are used to obtain meteorological monitoring element data for at least one year, including temperature, air pressure, humidity, wind speed and direction, rainfall, and solar radiation. At the same time, based on grid forecast data from different numerical weather prediction sources, meteorological forecast grid data covering the area where the power meteorological monitoring stations are located are obtained for at least one year, and the forecast element data includes temperature, air pressure, humidity, wind speed and direction, rainfall, and solar radiation. In step S3: the multi-source weather forecast fusion model is: ； In the formula, a represents the weight value of the k-th forecast data source. k b k The constants one and constant two of the k-th forecast data source, respectively. This represents the forecast result from the k-th forecast data source. This represents the forecast fusion result based on k forecast data sources; In step S4, the constant a k constant two b k The calculation method is as follows: ； ； In the formula, T represents the training period length, f k,t O represents the forecast result of the k-th forecast data source at time t. t This represents the measured data during the training period. The average forecast value over the training period of the k-th forecast data source. This represents the average value of the measured data within the training period. The power meteorological data fusion method based on multi-source meteorological forecasting is characterized by: the weighting coefficients for each data source. The results are obtained by normalizing the error terms between each forecast data source and the actual measurements, as follows: ； ； Among them, G k f represents the error between the k-th forecast data source and the observed value during the training period. k,t This represents the predicted value at time t of the k-th prediction data source during the training period. The values ​​are the observations during the training period, and T is the length of the training period.

2. The power meteorological data fusion method based on multi-source meteorological forecasting according to claim 1, characterized in that: In step S2, downscaling calculations are performed on grid point data from different numerical weather forecast sources. Bilinear interpolation is used to calculate numerical forecast data for the locations of all power meteorological monitoring stations, forming a dataset of power meteorological monitoring elements and forecast elements for at least one year.

3. The power meteorological data fusion method based on multi-source meteorological forecasting according to claim 1, characterized in that: In step S5, three indicators are used for testing: Continuous Rank Probability Score (CRPS), Anomaly Correlation Coefficient (ACC), and Root Mean Square Error (RMSE). Continuous Rank Probability Rating (CRPS), using S CRP Represented as: ； In the formula, This represents the probability density of the predicted value; It is the Heaviside function, when hour, ;otherwise , Represents the forecast value; Represents the actual value. To represent the differential; The anomaly correlation coefficient ACC is represented by C. AC Represented as: ； In the formula, m represents the number of monitoring stations; y i Represents the forecast value; O i Represents the actual value. This represents the average of the forecast values. This represents the average of the actual values; Root mean square error (RMSE), expressed as E RMS Represented as: ； In the formula, l represents the sum of the data; y i Represents the forecast value; O i Represents the actual value.

4. A power meteorological data fusion system based on multi-source meteorological forecasting, characterized in that, Includes the following functional modules: Multi-source meteorological data acquisition module: Collects monitoring data from power meteorological stations and multiple weather forecast sources for at least one year; Dataset module: Using downscaling methods, multiple weather forecast source data are mapped "point-to-point" to the locations of power meteorological stations to form a dataset of meteorological element forecasts and monitoring elements; Fusion Model Establishment Module: Based on the datasets of meteorological forecast elements and monitoring elements, the Bayesian model averaging method is used to fuse multi-source meteorological forecast data, establish a multi-source meteorological forecast fusion model, and obtain the meteorological forecast fusion results; Training module: Trains the parameters of the multi-source forecast fusion model; Evaluation module: Verifies and evaluates the results of multi-source forecast fusion; In the multi-source meteorological data acquisition module, the power meteorological monitoring stations provided in the power grid equipment monitoring network are used to acquire meteorological monitoring element data for at least one year, including temperature, air pressure, humidity, wind speed and direction, rainfall, and solar radiation. At the same time, based on gridded forecast data from different numerical weather prediction sources, meteorological forecast gridded data covering the area where the power meteorological monitoring stations are located are acquired for at least one year, with specific forecast elements including temperature, air pressure, humidity, wind speed and direction, rainfall, and solar radiation. In the fusion model building module, the multi-source weather forecast fusion model is as follows: ； In the formula, a represents the weight value of the k-th forecast data source. k b k The constants one and constant two of the k-th forecast data source, respectively. This represents the forecast result from the k-th forecast data source. This represents the forecast fusion result based on k forecast data sources; constant a k constant two b k The calculation method is as follows: ； ； In the formula, T represents the training period length, f k,t O represents the forecast result of the k-th forecast data source at time t. t This represents the measured data during the training period. The average forecast value over the training period of the k-th forecast data source. This represents the average value of the measured data within the training period. The power meteorological data fusion method based on multi-source meteorological forecasting is characterized by: the weighting coefficients for each data source. The results are obtained by normalizing the error terms between each forecast data source and the actual measurements, as follows: ； ； Among them, G k f represents the error between the k-th forecast data source and the observed value during the training period. k,t This represents the predicted value at time t of the k-th prediction data source during the training period. The values ​​are the observations during the training period, and T is the length of the training period.

5. A computer-readable storage medium for storing the power meteorological data fusion method based on multi-source meteorological forecasting as described in any one of claims 1-3.

Images