Time-space characteristic analysis method for meteorological new energy load big data

Abstract

The invention discloses a time-space characteristic analysis method for meteorological new energy load big data, and belongs to the technical field of new energy power management, and the method comprises the steps of data fusion alignment, physical causal modeling, extreme weather attribution and time-space characteristic analysis. According to the method, time-space characteristic analysis is performed by combining physical causal modeling and extreme weather attribution, the prediction precision of new energy load fluctuation under daily weather is improved, and load abnormity caused by extreme weather can be effectively identified and coped with; a double-flow space-time causal graph network model is adopted to perform physical causal modeling, and through physical constraint expression of wind power and photovoltaic characteristics, causal modeling of load change association between power stations and a dynamic feature fusion prediction mechanism of the wind power and photovoltaic characteristics, the load prediction accuracy, stability and scheduling practicability are improved; extreme weather attribution is carried out by adopting an attribution method combined with an improved disturbance response ratio, the actual influence of the extreme weather on the new energy load is accurately identified, and the interference degree of different types of weather events on the load is quantified.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy power management, and specifically refers to a method for analyzing the spatiotemporal characteristics of meteorological new energy load big data. Background Art

[0002] The spatiotemporal characteristics analysis method of meteorological new energy load big data is a method that uses artificial intelligence technology to integrate a large amount of meteorological data, geographic spatial information and new energy load operation data to conduct spatiotemporal collaborative analysis and explore the temporal evolution trend, spatial propagation law and meteorological driving mechanism of new energy load changes. It aims to improve the accuracy and stability of new energy load forecasts, support load anomaly attribution and emergency response analysis in extreme weather scenarios, and thus provide scientific decision-making support for power grid dispatching and operation safety.

[0003] However, in the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, there are technical problems such as new energy load is driven by meteorological factors and has large volatility, and traditional analysis methods are insufficient in modeling the impact of extreme weather events and the correlation between inter-regional load propagation, resulting in low load forecasting accuracy and delayed early warning response; there is a technical problem that traditional load forecasting ignores the physical operation laws of the new energy equipment itself, which easily leads to forecast deviations and is difficult to effectively identify the propagation of inter-regional power station load disturbances, thereby affecting the overall forecast stability of the dispatching area; there is a technical problem that the impact of extreme weather on new energy load is difficult to quantify, resulting in a significant increase in load forecasting errors during extreme weather periods, thereby affecting the dispatching of the power system. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a method for analyzing the spatiotemporal characteristics of meteorological new energy load big data. In view of the fact that in the spatiotemporal characteristics analysis process of existing meteorological new energy load big data, the new energy load is driven by meteorological factors and has large volatility, and the traditional analysis method is insufficient in modeling the correlation between the impact of extreme weather events and inter-regional load propagation, resulting in low load forecasting accuracy and delayed early warning response. This solution combines physical causal modeling and extreme weather attribution to perform spatiotemporal characteristics analysis, which not only improves the ability to predict new energy load fluctuations under daily meteorological changes, but also can effectively identify and respond to load anomalies caused by extreme weather, significantly improving the efficiency and safety of power grid operation; in view of the fact that in the spatiotemporal characteristics analysis process of existing meteorological new energy load big data, the traditional load forecast ignores the physical operation laws of the new energy equipment itself, which easily leads to prediction deviations and is difficult to effectively identify the propagation of load disturbances of power stations between regions, As for the technical issues that affect the overall forecast stability of the dispatching area, this solution creatively adopts a dual-stream spatiotemporal causal graph network model for physical causal modeling. By expressing the physical constraints of wind power and photovoltaic characteristics, causal modeling of the load change correlation between power stations, and a dynamic fusion prediction mechanism of the two characteristics, the accuracy, stability and dispatch practicality of the new energy load forecast are improved. In the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, there is a technical problem that the impact of extreme weather on new energy load is difficult to quantify, resulting in a significant increase in load forecast errors during extreme weather, which in turn affects the dispatch of the power system. This solution creatively adopts an attribution method combined with an improved disturbance response ratio to attribute extreme weather, accurately identify the actual impact of extreme weather on new energy load, and quantify the degree of interference of different types of weather events on load, providing an important basis for subsequent new energy dispatching decisions, thereby improving the reliability of dispatching decisions and enhancing the stability of power supply.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for analyzing the spatiotemporal characteristics of meteorological new energy load big data, which includes the following steps:

[0006] Step S1: data fusion and alignment;

[0007] Step S2: physical causal modeling;

[0008] Step S3: extreme weather attribution;

[0009] Step S4: Spatiotemporal characteristic analysis.

[0010] Furthermore, in step S1, the data fusion and alignment is used to fuse and align multi-source data in spatiotemporal scales, including the following steps:

[0011] Step S11: multi-source data collection, specifically collecting terrain elevation data, geographic information of new energy power stations, geographic information of weather stations, and spatial information of meteorological satellites to obtain geographic spatial data, and obtaining new energy load data from new energy power stations, meteorological observation data from meteorological stations, and meteorological spatial data from meteorological satellites;

[0012] Step S12: Dynamic weight allocation, specifically, constructing spatiotemporal observation map data based on geographic spatial data, using new energy power stations, meteorological stations, and meteorological satellites as nodes of the spatiotemporal observation map. When the spatial distance between two nodes is less than a set threshold, an edge is established between the two nodes. The edge of the spatiotemporal observation map represents the spatial adjacency relationship between the nodes. By constructing a graph neural network, the spatiotemporal correlation of each node is modeled based on the spatiotemporal observation map data to obtain a spatiotemporal dynamic weight matrix.

[0013] Step S13: Data fusion and alignment, specifically, by constructing a four-dimensional neural radiation field, fusing and aligning geographic space data, new energy load data, meteorological observation data, meteorological space data and spatiotemporal dynamic weight matrix to obtain spatiotemporal fusion and alignment data.

[0014] Furthermore, in step S2, the physical causal modeling is used to predict the new energy load, specifically using a dual-stream spatiotemporal causal graph network model to perform physical causal modeling to obtain new energy load prediction data, including the following steps:

[0015] Step S21: Physical flow modeling is used to construct physical constraints that conform to the operating mechanisms of wind power and photovoltaic equipment. Specifically, based on the spatiotemporal fusion alignment data, hard constraints on wind power are constructed to perform physical modeling on wind power to obtain physically constrained wind power prediction values. Hard constraints on photovoltaic power temperature are constructed to perform physical modeling on photovoltaic power to obtain physically constrained photovoltaic power prediction values.

[0016] Step S22: Causal flow modeling is used to capture the interference propagation relationship of load changes between new energy power stations. Specifically, a spatiotemporal adjacency graph is constructed based on the spatial adjacency relationship and load similarity of the new energy power stations to obtain initial power station spatial adjacency graph data. Then, a standard Granger causality test method is used to perform load interference propagation causal weighted modeling based on the initial power station spatial adjacency graph data to obtain power station load causal graph structure data.

[0017] The power station load causal graph structure data is used to represent the load disturbance propagation path between new energy power stations, including initial power station spatial adjacency graph data and causal weighted power station spatial adjacency graph data;

[0018] The spatiotemporal adjacency graph is constructed by calculating the Euclidean spatial distance between the new energy power stations based on the geographic information of the new energy power stations in the geographic spatial data, constructing an initial spatial adjacency matrix, and calculating the similarity between the power station load curves using the Pearson correlation coefficient based on the new energy load data to obtain an initial time adjacency matrix. The initial spatial adjacency matrix and the initial time adjacency matrix are then weightedly fused to obtain initial power station spatial adjacency graph data.

[0019] Step S23: Constructing a dual-flow coupling mechanism, specifically constructing a dual-channel graph neural network structure including physical flow and causal flow; obtaining physically constrained wind power prediction values and physically constrained photovoltaic power prediction values as physical features through the physical flow channel, and obtaining power station load causal graph structure data as causal features through the causal flow channel. Furthermore, the importance ratio of the physical features and causal features is regulated through the gating mechanism to obtain fused graph feature data;

[0020] Step S24: Model training, specifically, constructing a dual-flow spatiotemporal causal graph network model through the physical flow modeling, the causal flow modeling, and the dual-flow coupling mechanism, and performing model training by improving the multi-scale loss function to obtain a physical causal improved model for new energy load forecasting;

[0021] The improved multi-scale loss function includes a physical prediction loss function, a causal prediction loss function and a supply-demand balance loss function;

[0022] The physical prediction loss function is specifically a standard mean square error loss function;

[0023] The causal prediction loss function is specifically a standard weighted contrast loss function;

[0024] The calculation formula of the supply and demand balance loss function is:

[0025] ;

[0026] Where, L balance is the supply and demand balance loss function, which is used to balance the deviation between renewable energy supply and load demand. i is the renewable energy power station index, N is the number of renewable energy power stations, and P i is the output load forecast value of the i-th renewable energy power station, j is the load demand point index, M is the number of load demand points, specifically the number of main power loads in the renewable energy dispatch area, L j is the load demand of the jth load demand point;

[0027] Step S25: New energy load forecasting, specifically using the new energy load forecasting physical causal improvement model to forecast new energy load and obtain new energy load forecast data.

[0028] Furthermore, in step S3, the extreme weather attribution is used to quantify the impact of extreme weather on the abnormality of new energy load. Specifically, the extreme weather attribution is performed using an attribution method combined with an improved disturbance response ratio to obtain extreme weather attribution auxiliary data, including the following steps:

[0029] Step S31: extreme weather event determination, specifically using a threshold judgment method to determine extreme weather events on the meteorological observation data in the spatiotemporal fusion alignment data, and constructing an extreme weather event list;

[0030] Step S32: Extracting disturbance response features, specifically extracting load change features during extreme weather events based on the spatiotemporal fusion alignment data and the extreme weather event list, and obtaining extreme weather disturbance response feature data by calculating the deviation between the predicted data in the new energy load forecast data and the actual data in the spatiotemporal fusion alignment data;

[0031] The extreme weather disturbance response characteristic data includes average disturbance characteristics, maximum disturbance characteristics and disturbance change characteristics;

[0032] Step S33: Calculating an improved disturbance response ratio, specifically by combining the extreme weather disturbance response characteristic data and the new energy load forecast data to construct an extreme weather disturbance attribution ratio, and calculating the extreme weather attribution score data based on the extreme weather disturbance attribution ratio and the attribution model;

[0033] Step S34: Generate attribution results. Specifically, based on the extreme weather attribution score data, take the top three extreme weather events with the highest extreme weather attribution scores as the extreme weather attribution results of the load disturbance to obtain extreme weather attribution auxiliary data.

[0034] Furthermore, in step S4, the spatiotemporal characteristic analysis is used to conduct a visual analysis of the spatiotemporal variation characteristics of the new energy load, the physical causal influence relationship and the extreme weather attribution results. Specifically, based on the new energy load forecast data and the extreme weather attribution auxiliary data, a data visualization analysis image is constructed to obtain the spatiotemporal characteristic analysis reference data of the new energy load big data, which is used to assist the new energy load scheduling and abnormal warning.

[0035] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0036] (1) In the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, there are technical problems such as new energy load is driven by meteorological factors and has large fluctuations. Traditional analysis methods are insufficient in modeling the impact of extreme weather events and the load transmission correlation between regions, resulting in low load forecasting accuracy and delayed early warning response. This solution combines physical causal modeling and extreme weather attribution to conduct spatiotemporal characteristics analysis, which not only improves the ability to predict new energy load fluctuations under daily meteorological changes, but also effectively identifies and responds to load anomalies caused by extreme weather, significantly improving the efficiency and safety of power grid operation.

[0037] (2) In view of the technical problem that in the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, traditional load forecasting ignores the physical operating laws of the new energy equipment itself, which easily leads to prediction deviations and makes it difficult to effectively identify the propagation of load disturbances between power stations in different regions, thereby affecting the overall prediction stability of the dispatching area, this solution creatively adopts a dual-stream spatiotemporal causal graph network model for physical causal modeling. By expressing the physical constraints of wind power and photovoltaic characteristics, causal modeling of the load change correlation between power stations, and a dynamic fusion prediction mechanism of the two characteristics, the accuracy, stability and dispatching practicality of new energy load forecasting are improved;

[0038] (3) In view of the technical problem that the impact of extreme weather on renewable energy load is difficult to quantify during the analysis of the spatiotemporal characteristics of existing meteorological renewable energy load big data, resulting in a significant increase in load forecast errors during extreme weather periods, which in turn affects the dispatch of power systems, this scheme creatively adopts an attribution method combined with an improved disturbance response ratio to attribute extreme weather, accurately identify the actual impact of extreme weather on renewable energy load, and quantify the degree of interference of different types of weather events on load, providing an important basis for subsequent renewable energy dispatch decisions, thereby improving the reliability of dispatch decisions and enhancing the stability of power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic flow chart of a method for analyzing the spatiotemporal characteristics of meteorological new energy load big data provided by the present invention;

[0040] Figure 2 is a schematic diagram of step S1;

[0041] Figure 3 Schematic diagram of the process of step S2;

[0042] Figure 4 Schematic diagram of the process of step S3.

[0043] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0045] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0046] Example 1, see Figure 1 The present invention provides a method for analyzing the spatiotemporal characteristics of meteorological new energy load big data, which includes the following steps:

[0047] Step S1: data fusion and alignment;

[0048] Step S2: physical causal modeling;

[0049] Step S3: extreme weather attribution;

[0050] Step S4: spatiotemporal characteristics analysis;

[0051] By performing the above operations, in the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, there is a technical problem that the new energy load is driven by meteorological factors and has large fluctuations. The traditional analysis method does not adequately model the impact of extreme weather events and the correlation between inter-regional load propagation, resulting in low load forecasting accuracy and delayed early warning response. This solution combines physical causal modeling and extreme weather attribution to perform spatiotemporal characteristics analysis, which not only improves the ability to predict new energy load fluctuations under daily meteorological changes, but also can effectively identify and respond to load anomalies caused by extreme weather, significantly improving the efficiency and safety of power grid operation.

[0052] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the data fusion and alignment is used to fuse and align multi-source data in spatiotemporal scales, including the following steps:

[0053] Step S11: multi-source data collection, specifically collecting terrain elevation data, geographic information of new energy power stations, geographic information of weather stations, and spatial information of meteorological satellites to obtain geographic spatial data, and obtaining new energy load data from new energy power stations, meteorological observation data from meteorological stations, and meteorological spatial data from meteorological satellites;

[0054] The meteorological observation data include wind speed, light, temperature, relative humidity and rainfall;

[0055] The meteorological spatial data includes cloud cover, cloud top temperature and cloud type;

[0056] Step S12: Dynamic weight allocation, specifically, constructing spatiotemporal observation map data based on geographic spatial data, using new energy power stations, meteorological stations, and meteorological satellites as nodes of the spatiotemporal observation map. When the spatial distance between two nodes is less than a set threshold, an edge is established between the two nodes. The edge of the spatiotemporal observation map represents the spatial adjacency relationship between the nodes. By constructing a graph neural network, the spatiotemporal correlation of each node is modeled based on the spatiotemporal observation map data to obtain a spatiotemporal dynamic weight matrix.

[0057] Step S13: Data fusion and alignment, specifically, by constructing a four-dimensional neural radiation field, fusing and aligning geographic space data, new energy load data, meteorological observation data, meteorological space data and spatiotemporal dynamic weight matrix to obtain spatiotemporal fusion and alignment data.

[0058] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the physical causal modeling is used to predict the new energy load. Specifically, the physical causal modeling is performed using a dual-stream spatiotemporal causal graph network model to obtain new energy load prediction data, including the following steps:

[0059] Step S21: Physical flow modeling is used to construct physical constraints that conform to the operating mechanisms of wind power and photovoltaic equipment. Specifically, based on the spatiotemporal fusion alignment data, hard constraints on wind power are constructed to perform physical modeling on wind power to obtain physically constrained wind power prediction values. Hard constraints on photovoltaic power temperature are constructed to perform physical modeling on photovoltaic power to obtain physically constrained photovoltaic power prediction values.

[0060] The wind power hard constraint is calculated based on the physical working range of the wind turbine generator set. The calculation formula is:

[0061] ;

[0062] Where, P wind is the predicted value of wind power under physical constraints, v is the actual wind speed of the wind turbine, and v cutin is the fan starting wind speed, Pred is the wind power value predicted by the model, v rated is the rated wind speed of the fan, v cutout is the fan cut-out wind speed, P rated is the wind power rated power;

[0063] The photovoltaic power temperature hard constraint is calculated by establishing a correction relationship between the power and the ambient temperature, and the calculation formula is:

[0064] ;

[0065] Where, P pv is the predicted value of the physically constrained photovoltaic power, P std It is the photovoltaic power value predicted by the model under standard test conditions. is the photovoltaic temperature coefficient, T cell is the current photovoltaic cell temperature;

[0066] Step S22: Causal flow modeling is used to capture the interference propagation relationship of load changes between new energy power stations. Specifically, a spatiotemporal adjacency graph is constructed based on the spatial adjacency relationship and load similarity of the new energy power stations to obtain initial power station spatial adjacency graph data. Then, a standard Granger causality test method is used to perform load interference propagation causal weighted modeling based on the initial power station spatial adjacency graph data to obtain power station load causal graph structure data.

[0067] The power station load causal graph structure data is used to represent the load disturbance propagation path between new energy power stations, including initial power station spatial adjacency graph data and causal weighted power station spatial adjacency graph data;

[0068] The spatiotemporal adjacency graph is constructed by calculating the Euclidean spatial distance between the new energy power stations based on the geographic information of the new energy power stations in the geographic spatial data, constructing an initial spatial adjacency matrix, and calculating the similarity between the power station load curves using the Pearson correlation coefficient based on the new energy load data to obtain an initial time adjacency matrix. The initial spatial adjacency matrix and the initial time adjacency matrix are then weightedly fused to obtain initial power station spatial adjacency graph data.

[0069] Step S23: Constructing a dual-flow coupling mechanism, specifically constructing a dual-channel graph neural network structure including physical flow and causal flow; obtaining physically constrained wind power prediction values and physically constrained photovoltaic power prediction values as physical features through the physical flow channel, and obtaining power station load causal graph structure data as causal features through the causal flow channel. The importance ratio of the physical features and causal features is regulated through the gating mechanism to obtain fused graph feature data;

[0070] The calculation formula of the fusion graph feature data is:

[0071] ;

[0072] Where H fused is the fusion graph feature data, G is the gated tensor used to control the importance of physical features and causal features, is the element-wise multiplication symbol, H phys is a physical characteristic, H caus It is a causal characteristic;

[0073] Step S24: Model training, specifically, constructing a dual-flow spatiotemporal causal graph network model through the physical flow modeling, the causal flow modeling, and the dual-flow coupling mechanism, and performing model training by improving the multi-scale loss function to obtain a physical causal improved model for new energy load forecasting;

[0074] The improved multi-scale loss function includes a physical prediction loss function, a causal prediction loss function and a supply-demand balance loss function;

[0075] The physical prediction loss function is specifically a standard mean square error loss function;

[0076] The causal prediction loss function is specifically a standard weighted contrast loss function;

[0077] The calculation formula of the supply and demand balance loss function is:

[0078] ;

[0079] Where, L balance is the supply and demand balance loss function, which is used to balance the deviation between renewable energy supply and load demand. i is the renewable energy power station index, N is the number of renewable energy power stations, and P i is the output load forecast value of the i-th renewable energy power station, j is the load demand point index, M is the number of load demand points, specifically the number of main power loads in the renewable energy dispatch area, L j is the load demand of the jth load demand point;

[0080] Step S25: New energy load forecasting, specifically using a new energy load forecasting physical causal improvement model to forecast new energy load and obtain new energy load forecast data;

[0081] By performing the above operations, in order to address the technical problem that in the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, traditional load forecasting ignores the physical operating laws of the new energy equipment itself, which easily leads to prediction deviations and makes it difficult to effectively identify the propagation of load disturbances between power stations in different regions, thereby affecting the overall prediction stability of the scheduling area, this solution creatively adopts a dual-stream spatiotemporal causal graph network model for physical causal modeling. By expressing the physical constraints of wind power and photovoltaic characteristics, causal modeling of the load change correlation between power stations, and a dynamic fusion prediction mechanism of the two characteristics, the accuracy, stability and scheduling practicality of new energy load forecasting are improved.

[0082] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, the extreme weather attribution is used to quantify the impact of extreme weather on the abnormality of new energy load. Specifically, the extreme weather attribution is performed using an attribution method combined with an improved disturbance response ratio to obtain extreme weather attribution auxiliary data, including the following steps:

[0083] Step S31: extreme weather event determination, specifically using a threshold judgment method to determine extreme weather events on the meteorological observation data in the spatiotemporal fusion alignment data, and constructing an extreme weather event list;

[0084] A list of the extreme weather events, including event types and event times;

[0085] The event types include extreme wind weather, extreme sunlight weather, extreme cloud weather, extreme rain weather, extreme temperature weather and combined extreme weather;

[0086] Step S32: Extracting disturbance response features, specifically extracting load change features during extreme weather events based on the spatiotemporal fusion alignment data and the extreme weather event list, and obtaining extreme weather disturbance response feature data by calculating the deviation between the predicted data in the new energy load forecast data and the actual data in the spatiotemporal fusion alignment data;

[0087] The extreme weather disturbance response characteristic data includes average disturbance characteristics, maximum disturbance characteristics and disturbance change characteristics;

[0088] Step S33: Calculating an improved disturbance response ratio, specifically by combining the extreme weather disturbance response characteristic data and the new energy load forecast data to construct an extreme weather disturbance attribution ratio, and calculating the extreme weather attribution score data based on the extreme weather disturbance attribution ratio and the attribution model;

[0089] The calculation formula for the extreme weather disturbance attribution ratio is:

[0090] ;

[0091] Where, is the disturbance attribution ratio of the i-th new energy power station under extreme weather conditions, is the actual load disturbance vector of the i-th renewable energy power station during extreme weather, is the physical prediction residual of the i-th new energy power station, is the causal prediction residual of the i-th new energy power station, is a stabilizing term used to avoid the denominator being zero, and ||·||2 is the L2 norm operator;

[0092] Step S34: extreme weather attribution, specifically, taking the top three extreme weather events with the highest extreme weather attribution scores as the extreme weather attribution results of the load disturbance based on the extreme weather attribution score data, and obtaining extreme weather attribution auxiliary data;

[0093] By performing the above operations, in order to address the technical problem that in the process of analyzing the spatiotemporal characteristics of existing meteorological new energy load big data, the impact of extreme weather on new energy load is difficult to quantify, resulting in a significant increase in load forecast errors during extreme weather, which in turn affects the dispatch of the power system, this solution creatively adopts an attribution method combined with an improved disturbance response ratio to attribute extreme weather, accurately identify the actual impact of extreme weather on new energy load, and quantify the degree of interference of different types of weather events on load, providing an important basis for subsequent new energy dispatch decisions, thereby improving the reliability of dispatch decisions and enhancing the stability of power supply.

[0094] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the spatiotemporal characteristic analysis is used to perform a visual analysis of the spatiotemporal variation characteristics of the new energy load, the physical causal influence relationship, and the extreme weather attribution results. Specifically, based on the new energy load forecast data and the extreme weather attribution auxiliary data, a data visualization analysis image is constructed to obtain the spatiotemporal characteristic analysis reference data of the new energy load big data, which is used to assist in new energy load scheduling and abnormal warning.

[0095] The reference data for the analysis of the spatiotemporal characteristics of the new energy load big data includes a visualization diagram of load time series changes, a visualization heat map of load interference spatial propagation, and an extreme weather attribution impact map;

[0096] The extreme weather attribution impact diagram includes a weather type attribution pie chart and a load response trajectory line chart.

[0097] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0098] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0099] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for analyzing the spatiotemporal characteristics of meteorological new energy load big data, characterized by: The method comprises the following steps: Step S1: Data fusion and alignment, used to fuse and align multi-source data in spatiotemporal scales, including the following steps: Step S11: multi-source data collection; Step S12: dynamic weight allocation; Step S13: data fusion and alignment; Step S2: Physical causal modeling, used to predict renewable energy load, specifically using a dual-flow spatiotemporal causal graph network model to perform physical causal modeling to obtain renewable energy load forecast data, including the following steps: Step S21: Physical flow modeling, used to construct physical constraints that conform to the operating mechanisms of wind power and photovoltaic equipment; Step S22: Causal flow modeling, used to capture the interference propagation relationship of load changes between renewable energy power stations; Step S23: Dual-flow coupling mechanism construction; Step S24: Model training; Step S25: New energy load forecasting; Step S3: extreme weather attribution, used to quantify the impact of extreme weather on abnormal new energy loads. Specifically, extreme weather attribution is performed using an attribution method combined with an improved disturbance response ratio to obtain extreme weather attribution auxiliary data, including the following steps: Step S31: extreme weather event determination; Step S32: disturbance response feature extraction; Step S33: calculation of the improved disturbance response ratio; Step S34: generation of attribution results; Step S4: Spatiotemporal characteristic analysis, which is used to conduct a visual analysis of the spatiotemporal variation characteristics of new energy loads, physical causal influence relationships, and extreme weather attribution results.

2. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 1 is characterized by: In step S21, the physical flow modeling is specifically performed by constructing a hard constraint on wind power based on the time-space fusion alignment data to perform physical modeling on the wind power to obtain a physically constrained wind power prediction value, and by constructing a hard constraint on photovoltaic power temperature to perform physical modeling on the photovoltaic power to obtain a physically constrained photovoltaic power prediction value; In step S22, the causal flow modeling is specifically to construct a spatiotemporal adjacency graph based on the spatial adjacency relationship and load similarity of the new energy power stations to obtain initial power station spatial adjacency graph data, and to perform load interference propagation causal weighted modeling based on the initial power station spatial adjacency graph data by adopting the standard Granger causality test method to obtain power station load causal graph structure data; In step S23, the dual-flow coupling mechanism is constructed, specifically by constructing a dual-channel graph neural network structure including physical flow and causal flow; through the physical flow channel, the physically constrained wind power prediction value and the physically constrained photovoltaic power prediction value are obtained as physical features, and through the causal flow channel, the power station load causal graph structure data is obtained as causal features, and the importance ratio of the physical features and the causal features is regulated through the gating mechanism to obtain fused graph feature data; In step S24, the model training is specifically to construct a dual-flow spatiotemporal causal graph network model through the physical flow modeling, the causal flow modeling and the dual-flow coupling mechanism, and perform model training by improving the multi-scale loss function to obtain a physical causal improved model for new energy load forecasting; In step S25, the new energy load forecasting is specifically to use the new energy load forecasting physical causal improvement model to perform new energy load forecasting and obtain new energy load forecasting data.

3. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 2 is characterized by: In step S22, the power station load causal graph structure data is used to represent the load disturbance propagation path between new energy power stations, including initial power station spatial adjacency graph data and causal weighted power station spatial adjacency graph data; the spatiotemporal adjacency graph is constructed, specifically based on the geographic information of the new energy power stations in the geographic spatial data, calculating the Euclidean spatial distance between the new energy power stations, constructing an initial spatial adjacency matrix, and calculating the similarity between the power station load curves through the Pearson correlation coefficient based on the new energy load data to obtain the initial time adjacency matrix, and then performing graph structure weighted fusion on the initial spatial adjacency matrix and the initial time adjacency matrix to obtain the initial power station spatial adjacency graph data.

4. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 3 is characterized by: In step S24, the improved multi-scale loss function includes a physical prediction loss function, a causal prediction loss function, and a supply-demand balance loss function; The physical prediction loss function is specifically a standard mean square error loss function; The causal prediction loss function is specifically a standard weighted contrast loss function; The calculation formula of the supply and demand balance loss function is: ； Where, L balance is the supply and demand balance loss function, which is used to balance the deviation between renewable energy supply and load demand. i is the renewable energy power station index, N is the number of renewable energy power stations, and P i is the output load forecast value of the i-th renewable energy power station, j is the load demand point index, M is the number of load demand points, specifically the number of main power loads in the renewable energy dispatch area, L j is the load demand of the jth load demand point.

5. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 4 is characterized by: In step S31, the extreme weather event determination is specifically to use a threshold determination method to determine the extreme weather event of the meteorological observation data in the spatiotemporal fusion alignment data, and to construct an extreme weather event list; In step S32, the disturbance response feature extraction is specifically performed by extracting the load change characteristics during the extreme weather period based on the spatiotemporal fusion alignment data and the extreme weather event list, and obtaining the extreme weather disturbance response feature data by performing deviation calculation on the predicted data in the new energy load forecast data and the actual data in the spatiotemporal fusion alignment data; The extreme weather disturbance response characteristic data includes average disturbance characteristics, maximum disturbance characteristics and disturbance change characteristics; In step S33, the improved disturbance response ratio is calculated by combining the extreme weather disturbance response characteristic data and the new energy load forecast data to construct an extreme weather disturbance attribution ratio, and based on the extreme weather disturbance attribution ratio and the attribution model, the extreme weather attribution score data is calculated; In step S34, the attribution result is generated by taking the top three extreme weather events with the highest extreme weather attribution scores as the extreme weather attribution results of the load disturbance based on the extreme weather attribution score data to obtain extreme weather attribution auxiliary data.

6. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 5 is characterized by: In step S4, the spatiotemporal characteristic analysis is specifically based on the new energy load forecast data and the extreme weather attribution auxiliary data, by constructing a data visualization analysis image to obtain the spatiotemporal characteristic analysis reference data of the new energy load big data, which is used to assist the new energy load scheduling and abnormal warning.

7. The method for analyzing the spatiotemporal characteristics of meteorological new energy load big data according to claim 6, characterized in that: In step S11, the multi-source data collection specifically collects terrain elevation data, geographic information of new energy power stations, geographic information of weather stations, and spatial information of meteorological satellites to obtain geographic spatial data, and obtains new energy load data from new energy power stations, meteorological observation data from meteorological stations, and meteorological spatial data from meteorological satellites; In step S12, the dynamic weight allocation is specifically to construct spatiotemporal observation map data based on geographic spatial data, and use new energy power stations, meteorological stations and meteorological satellites as nodes of the spatiotemporal observation map respectively. When the spatial distance between two nodes is less than a set threshold, an edge is established between the two nodes. The edge of the spatiotemporal observation map represents the spatial adjacency relationship between the nodes; by constructing a graph neural network, the spatiotemporal correlation of each node is modeled based on the spatiotemporal observation map data to obtain a spatiotemporal dynamic weight matrix; In step S13, the data fusion and alignment is specifically to fuse and align the geographic space data, new energy load data, meteorological observation data, meteorological space data and spatiotemporal dynamic weight matrix by constructing a four-dimensional neural radiation field to obtain spatiotemporal fusion and alignment data.

Images