A method for constructing a multi-level agent of a regional integrated energy system

Abstract

The application belongs to the technical field of energy intelligence, and proposes a multi-level agent construction method for a regional integrated energy system, which comprises: performing space-time alignment processing on real-time operation data to obtain a synchronous data stream; mapping parameters in the synchronous data stream into specific visual attribute graph elements to obtain a multi-level state semantic image; performing visual semantic segmentation on the multi-level state semantic image to obtain a visual semantic unit and constructing a spatial correlation graph between the visual semantic units; comparing the spatial correlation graph with a historical synchronous spatial correlation graph in terms of visual differences to obtain an abnormal visual semantic unit; performing multi-frame tracking verification on the abnormal visual semantic unit, and integrating an abnormal unit identifier and an abnormal type into an abnormal event report; and performing hierarchical decision-making on the abnormal event report and the spatial correlation graph to obtain a collaborative control instruction. The application can improve the efficiency of collaborative management of a regional integrated energy system.

Description

Technical Field

[0001] This invention belongs to the field of energy intelligence technology, and in particular relates to a method for constructing a multi-level intelligent agent in a regional integrated energy system. Background Technology

[0002] Regional integrated energy systems combine multiple subsystems such as electricity, heat, and gas, serving as a crucial carrier for intelligent energy management. However, during the data acquisition phase, inconsistencies exist in the data acquisition dimensions, time bases, and spatial identifiers among these subsystems. Existing technologies lack efficient spatiotemporal alignment methods for multi-source real-time operational data, making it difficult to integrate scattered data into a unified synchronous data stream. This results in data fragmentation in the perception of system operational status, failing to accurately reflect the overall operational status of the system. Furthermore, existing technologies often employ traditional numerical methods to characterize system operational status, making it difficult to intuitively represent the spatial relationships and energy flow interactions between different levels and units. This poses a significant technical obstacle to system operational status analysis and anomaly identification.

[0003] In the anomaly detection and control phase, existing technologies also have significant shortcomings. Anomaly detection often relies on threshold judgments for single-dimensional parameters, failing to incorporate comprehensive analysis of system topology and inter-unit relationships. This makes it prone to missed or false anomaly detections, and the lack of multi-frame tracking and verification mechanisms for anomalous units hinders accurate identification of anomaly types and persistent characteristics. Furthermore, in the anomaly handling phase, the formulation of control commands lacks hierarchical and collaborative design, making it difficult to achieve precise tiered control based on the anomaly's hierarchical affiliation and propagation path. This results in insufficient system control response efficiency and adaptability, failing to meet the management and control requirements of multi-level collaborative operation in regional integrated energy systems. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method for constructing a multi-level intelligent agent in a regional integrated energy system, which can improve the efficiency of collaborative management of the regional integrated energy system.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for constructing a multi-level intelligent agent in a regional integrated energy system, the method comprising:

[0007] S1. Perform spatiotemporal alignment processing on the real-time operation data of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system.

[0008] S2. According to the preset visual coding rules, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, and specific visual attribute primitives are synthesized according to the system hierarchy of the regional integrated energy system to obtain a multi-level state semantic image of the regional integrated energy system.

[0009] S3. Perform visual semantic segmentation on the multi-level state semantic image to obtain the visual semantic units of the regional integrated energy system, and at the same time construct a spatial association map between the visual semantic units.

[0010] S4. Visually compare the spatial correlation map with the historical spatial correlation map of the regional integrated energy system to obtain the abnormal visual semantic units of the regional integrated energy system.

[0011] S5. Perform multi-frame tracking and verification on abnormal visual semantic units to obtain the abnormal unit identifier and abnormal type of the regional integrated energy system, and integrate the abnormal unit identifier and abnormal type into an abnormal event report of the regional integrated energy system.

[0012] S6. Make hierarchical decisions based on abnormal event reports and spatial correlation maps to obtain coordinated control instructions for the regional integrated energy system.

[0013] Preferably, in S1, the synchronous data stream of the regional integrated energy system is obtained, including:

[0014] By collecting the raw operating data of the power subsystem, heat subsystem, and gas subsystem of the regional integrated energy system, the real-time operating data of the regional integrated energy system can be obtained.

[0015] Extract the collection timestamp and geographic location identifier of real-time operational data;

[0016] Consistency verification is performed on the collected timestamps, and data with time deviations are adjusted to a unified time reference axis to obtain the time-series operation data of the regional integrated energy system.

[0017] Based on geographic location identifiers, time-series operational data is matched to the system spatial grid of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system.

[0018] Preferably, in S2, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, including:

[0019] Parameter identification is performed on the synchronous data stream to obtain the parameter type identifier and parameter value of the synchronous data stream;

[0020] Based on the parameter type identifier, select the corresponding basic graphic shape from the preset visual encoding rules;

[0021] Based on the parameter values, determine the thickness of the outline of the basic shape of the primitive and the density of the internal filling texture;

[0022] Based on the thickness of the outline, the density of the internal filling texture, and the basic shape of the primitives, the synchronous data stream is used to draw images, resulting in specific visual attribute primitives of the regional integrated energy system.

[0023] Preferably, in S2, a multi-level state semantic image of the regional integrated energy system is obtained, including:

[0024] Based on the node level label corresponding to the specific visual attribute primitive, the specific visual attribute primitive is assigned to the equipment level canvas, network level canvas, and energy flow level canvas of the regional integrated energy system.

[0025] By performing de-overlap processing on specific visual attribute primitives in the equipment-level canvas, the equipment status sub-layer of the regional integrated energy system is obtained.

[0026] Using the device status sub-layer as the background, draw lines connecting specific visual attribute primitives in the network layer canvas to obtain the network topology sub-layer of the regional integrated energy system.

[0027] Specific visual attribute primitives in the energy flow hierarchy canvas are drawn on the network topology sublayer in a semi-transparent overlay manner, and the transparency is adjusted according to the fill texture density of the specific visual attribute primitives in the energy flow hierarchy canvas to obtain the energy flow distribution sublayer of the regional integrated energy system.

[0028] By overlaying the equipment status sublayer, network topology sublayer, and energy flow distribution sublayer in a bottom-up order, a multi-level state semantic image of the regional integrated energy system is obtained.

[0029] Preferably, in S3, visual semantic units of the regional integrated energy system are obtained, and a spatial association map between visual semantic units is constructed, including:

[0030] Connectivity analysis is performed on the device state sub-layer in the multi-level state semantic image to obtain candidate device regions of the device state sub-layer. Feature extraction is performed on the filling texture in the candidate device regions to obtain the device visual semantic units of the regional integrated energy system.

[0031] Line segment detection is performed on the network topology sub-layer in the multi-level state semantic image to obtain the connection visual semantic unit of the regional integrated energy system.

[0032] Using pixel transparency and texture density as growth criteria, region growth segmentation is performed on the energy flow distribution sub-layer in the multi-level state semantic image to obtain the energy flow visual semantic unit of the regional integrated energy system.

[0033] Using equipment visual semantic units, connection visual semantic units, and energy flow visual semantic units as nodes, and the connection relationships between equipment and the hierarchical relationships between levels in the regional integrated energy system as edges, a spatial association map between visual semantic units is constructed.

[0034] Preferably, the visual semantic units of the equipment in the regional integrated energy system include:

[0035] Binarize the equipment status sub-layer, mark the pixels with outlines as foreground pixels and the rest as background pixels to obtain the equipment outline binary image of the regional integrated energy system.

[0036] Eight-neighbor connected component labeling is performed on the binary image of device contour. Interconnected foreground pixels are merged into the same connected component, and the minimum bounding rectangle of the connected component is extracted as the candidate device region of the binary image of device contour.

[0037] Gray value distribution analysis is performed on pixels within the candidate device region to obtain texture contrast parameters, texture correlation parameters, and texture inverse difference moment parameters of the candidate device region;

[0038] The texture contrast parameter, texture correlation parameter, and texture inverse difference moment parameter are compared item by item with the equipment texture feature reference set of the regional integrated energy system to obtain the equipment texture matching degree of the regional integrated energy system.

[0039] Based on the device texture matching degree, cluster analysis is performed on the candidate device regions, and a unique device identifier is assigned to each candidate device region to obtain the device visual semantic unit of the regional integrated energy system.

[0040] Preferably, in S4, the abnormal visual semantic units of the regional integrated energy system are obtained, including:

[0041] Based on spatial correlation maps, the historical spatial correlation maps of the regional integrated energy system are hierarchically analyzed to obtain the historical visual semantic units of the regional integrated energy system.

[0042] The attribute feature vectors in the visual semantic units and the historical attribute feature vectors in the historical visual semantic units are quantified one dimension at a time to obtain the attribute difference degree of the regional integrated energy system.

[0043] Traverse the first-order neighbor nodes of the visual semantic unit, record the unit identifier and the direction of the connecting edge of the first-order neighbor node, and obtain the neighbor adjacency sequence of the visual semantic unit.

[0044] Based on the neighbor adjacency sequence, extract the corresponding historical neighbor adjacency sequence from the historical visual semantic unit;

[0045] By comparing and analyzing the neighbor adjacency sequence with the historical neighbor adjacency sequence, the topological structure difference of the regional integrated energy system can be obtained.

[0046] A global anomaly index of visual semantic units is obtained by nonlinearly combining attribute difference degree and topological structure difference degree.

[0047] By comparing the global anomaly index with the preset anomaly judgment threshold, the abnormal visual semantic units of the regional integrated energy system are obtained.

[0048] Preferably, the formula for calculating the global anomaly index is:

[0049] ;

[0050] Where E represents the global anomaly index, α represents the preset first weighting coefficient, β represents the preset second weighting coefficient, A represents the attribute difference degree, T represents the topological difference degree, and e represents the natural constant. This represents the natural logarithm function.

[0051] Preferably, in step S5, the abnormal unit identifier and abnormal type of the regional integrated energy system are obtained, and the abnormal unit identifier and abnormal type are integrated into an abnormal event report of the regional integrated energy system, including:

[0052] Establish a tracking profile for abnormal visual semantic units, which includes the current frame image coordinates and current visual attribute parameters;

[0053] Acquire subsequent frame multi-level state semantic images of the regional integrated energy system, and locate subsequent visual semantic units with the same image coordinate range and similar visual attribute parameters as the abnormal visual semantic units in the subsequent frame multi-level state semantic images.

[0054] The subsequent visual semantic units are used as tracking continuation units of the abnormal visual semantic units in subsequent frames, and the image coordinates and visual attribute parameters of the tracking continuation units are appended to the tracking archive to obtain the complete tracking archive of the regional integrated energy system.

[0055] Based on the changing trends of visual attribute parameters in the complete tracking file, the anomaly type of the abnormal visual semantic unit is determined, and a unique anomaly unit identifier is assigned to the abnormal visual semantic unit.

[0056] Based on the hierarchical relationship and frame order of abnormal visual semantic units in the spatial association map, the abnormal types and abnormal unit identifiers are sorted out to obtain the abnormal event report of the regional integrated energy system.

[0057] Preferably, in S6, the coordinated control instructions of the regional integrated energy system are obtained, including:

[0058] The abnormal event report is analyzed to obtain the abnormal unit identifier, abnormal type and abnormal occurrence time of the abnormal visual semantic unit. Based on the abnormal unit identifier, the layer level and spatial coordinate position of the abnormal visual semantic unit in the spatial association map are located.

[0059] Based on the layer hierarchy, the abnormal visual semantic units are divided into device-level abnormal units, network-level abnormal units, and power flow-level abnormal units.

[0060] Based on the anomaly type, the equipment-level anomaly units, network-level anomaly units, and energy flow-level anomaly units are clustered and grouped to obtain the hierarchical anomaly grouping set of the regional integrated energy system.

[0061] Based on the hierarchical anomaly group set, the spatial correlation map is used to explore the anomaly propagation path in order to obtain the associated device set of the hierarchical anomaly group set;

[0062] Based on the associated equipment set and anomaly type, preliminary control parameter recommendations for the regional integrated energy system are determined;

[0063] The preliminary control parameter recommendations are merged and deduplicated to obtain the hierarchical control instruction set of the regional integrated energy system.

[0064] The hierarchical control instruction set is encapsulated level by level, and during the encapsulation process, the linkage association identifier between control instructions in the hierarchical control instruction set is established based on the hierarchical unit membership relationship in the spatial correlation map, so as to obtain the coordinated control instructions of the regional integrated energy system.

[0065] The present invention has the following beneficial effects:

[0066] This invention achieves synchronous integration of multi-source data through spatiotemporal alignment processing of real-time operational data. Then, by combining visual encoding rules, the data is mapped into specific visual attribute primitives and synthesized into multi-level state semantic images. Relying on visual semantic segmentation and spatial correlation map construction, the visualization and refined analysis of the system's operational status are realized, which greatly improves the perception efficiency and recognition accuracy of multi-level intelligent agents in the system's state. This allows intelligent agents to quickly and accurately capture the operational characteristics of each level of the regional integrated energy system, effectively improving the construction efficiency and state perception capability of multi-level intelligent agents.

[0067] This invention accurately locates abnormal units and determines the type of abnormality through visual difference comparison and multi-frame tracking verification. Then, it combines hierarchical decision-making to generate collaborative control instructions, realizing intelligent processing of the entire process from anomaly identification to control decision-making. This enables multi-level intelligent agents to have efficient anomaly detection and collaborative control capabilities. At the same time, relying on hierarchical analysis and decision-making logic, it ensures the pertinence and linkage of control instructions, improves the autonomous decision-making and collaborative operation level of multi-level intelligent agents in regional integrated energy systems, and further strengthens the control efficiency of intelligent agents over regional integrated energy systems. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0069] Example 1: As Figure 1 As shown, a method for constructing a multi-level intelligent agent in a regional integrated energy system includes:

[0070] S1. Perform spatiotemporal alignment processing on the real-time operation data of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system.

[0071] S2. According to the preset visual coding rules, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, and specific visual attribute primitives are synthesized according to the system hierarchy of the regional integrated energy system to obtain a multi-level state semantic image of the regional integrated energy system.

[0072] S3. Perform visual semantic segmentation on the multi-level state semantic image to obtain the visual semantic units of the regional integrated energy system, and at the same time construct a spatial association map between the visual semantic units.

[0073] S4. Visually compare the spatial correlation map with the historical spatial correlation map of the regional integrated energy system to obtain the abnormal visual semantic units of the regional integrated energy system.

[0074] S5. Perform multi-frame tracking and verification on abnormal visual semantic units to obtain the abnormal unit identifier and abnormal type of the regional integrated energy system, and integrate the abnormal unit identifier and abnormal type into an abnormal event report of the regional integrated energy system.

[0075] S6. Make hierarchical decisions based on abnormal event reports and spatial correlation maps to obtain coordinated control instructions for the regional integrated energy system.

[0076] In S1, the synchronous data stream of the regional integrated energy system is obtained, including:

[0077] By collecting the raw operating data of the power subsystem, heat subsystem, and gas subsystem of the regional integrated energy system, the real-time operating data of the regional integrated energy system can be obtained.

[0078] Extract the collection timestamp and geographic location identifier of real-time operational data;

[0079] Consistency verification is performed on the collected timestamps, and data with time deviations are adjusted to a unified time reference axis to obtain the time-series operation data of the regional integrated energy system.

[0080] Based on geographic location identifiers, time-series operational data is matched to the system spatial grid of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system.

[0081] The system comprehensively collects all kinds of raw operating data from the power subsystem, heating subsystem, and gas subsystem within the regional integrated energy system. It integrates and summarizes all data collected during the actual operation of each subsystem to form complete real-time operating data of the regional integrated energy system, ensuring that the data includes all relevant data information during the operation of each subsystem.

[0082] From the real-time operation data of the integrated regional energy system, the collection timestamp and geographic location identifier corresponding to each data point are extracted one by one. The collection timestamp is the specific time information when each subsystem collects data, and the geographic location identifier is the specific spatial location information of the equipment and nodes corresponding to the data collected by each subsystem in the regional energy system, ensuring that the extracted information accurately corresponds to each real-time operation data point.

[0083] All extracted collection timestamps are compared one by one with the preset unified time reference axis to complete the consistency verification. For the real-time operation data corresponding to the collection timestamps that have time deviations from the unified time reference axis, the time dimension is calibrated and adjusted according to the specific duration of the time deviation, so that the time dimension of all data is matched to the unified time reference axis, and finally the time-series operation data of the regional integrated energy system is formed.

[0084] Based on the geographic location identifiers corresponding to each extracted time-series operational data, each time-series operational data is precisely matched to the grid node corresponding to the pre-divided system spatial grid of the regional integrated energy system according to the specific spatial location pointed to by its geographic location identifier. This ensures that each grid node is matched with the time-series operational data within its spatial location range. By integrating all the matched grid node data, a synchronous data stream of the regional integrated energy system is obtained.

[0085] By comprehensively collecting the original operating data of each subsystem of electricity, heat, and gas, the integrity of real-time operating data is ensured. Then, through the precise extraction of collection timestamps and geographic location identifiers, the consistency verification of timestamps, and targeted adjustments, the unified regularization of multi-source data in the time dimension is achieved. Subsequently, the geographic location identifiers are combined to complete the precise matching with the system's spatial grid, enabling the multi-source operating data of the regional integrated energy system to achieve a high degree of synchronization in both time and space dimensions. The resulting synchronized data stream can accurately reflect the actual operating status of each spatial location in the system under a unified time dimension. This provides a spatiotemporally unified and accurate data source foundation for mapping data into specific visual attribute primitives and synthesizing multi-level state semantic images. From the data source, the accuracy and effectiveness of subsequent visualization processing and semantic analysis are guaranteed.

[0086] In S2, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, including:

[0087] Parameter identification is performed on the synchronous data stream to obtain the parameter type identifier and parameter value of the synchronous data stream;

[0088] Based on the parameter type identifier, select the corresponding basic graphic shape from the preset visual encoding rules;

[0089] Based on the parameter values, determine the thickness of the outline of the basic shape of the primitive and the density of the internal filling texture;

[0090] Based on the thickness of the outline, the density of the internal filling texture, and the basic shape of the primitives, the synchronous data stream is used to draw images, resulting in specific visual attribute primitives of the regional integrated energy system.

[0091] By synthesizing specific visual attribute primitives according to the system hierarchy of the regional integrated energy system, a multi-level state semantic image of the regional integrated energy system is obtained, including:

[0092] Based on the node level label corresponding to the specific visual attribute primitive, the specific visual attribute primitive is assigned to the equipment level canvas, network level canvas, and energy flow level canvas of the regional integrated energy system.

[0093] By performing de-overlap processing on specific visual attribute primitives in the equipment-level canvas, the equipment status sub-layer of the regional integrated energy system is obtained.

[0094] Using the device status sub-layer as the background, draw lines connecting specific visual attribute primitives in the network layer canvas to obtain the network topology sub-layer of the regional integrated energy system.

[0095] Specific visual attribute primitives in the energy flow hierarchy canvas are drawn on the network topology sublayer in a semi-transparent overlay manner, and the transparency is adjusted according to the fill texture density of the specific visual attribute primitives in the energy flow hierarchy canvas to obtain the energy flow distribution sublayer of the regional integrated energy system.

[0096] By overlaying the equipment status sublayer, network topology sublayer, and energy flow distribution sublayer in a bottom-up order, a multi-level state semantic image of the regional integrated energy system is obtained.

[0097] Each piece of data in the synchronous data stream is sorted out and identified one by one, clarifying the parameter category and specific numerical information corresponding to each piece of data, forming a parameter type identifier and parameter value for the synchronous data stream that corresponds one-to-one with each parameter, ensuring that the type attribute and specific value of each parameter are accurately defined.

[0098] Based on the parameter type identifiers of the obtained synchronous data stream, precise matching is performed in the pre-set visual coding rules. According to the graphic specifications corresponding to each parameter type identifier in the rules, the graphic style that matches it perfectly is selected as the basic shape of the graphic element, ensuring that the correspondence between the basic shape of the graphic element and the parameter type identifier is without deviation.

[0099] Based on the parameter values ​​corresponding to each parameter in the synchronous data stream, and according to the correspondence between parameter values ​​and visual features in the preset visual coding rules, the specific thickness of the outline matched by the parameter value is determined. At the same time, the specific density arrangement of the filling texture inside the basic shape of the primitive is determined, so that the thickness of the outline and the density of the internal filling texture are accurately correlated with the parameter value.

[0100] Using the selected primitive base shape as the graphic base, the determined outline thickness is applied to the outline drawing of the primitive base shape, and the set internal filling texture density is implemented in the internal filling of the primitive base shape. According to this visual feature, the synchronous data stream is subjected to parameter-by-parameter image drawing operation, and finally the specific visual attribute primitive of the regional integrated energy system is formed.

[0101] Extract the node-level label carried by each specific visual attribute element. According to the system level indicated by the label, classify and place all specific visual attribute elements into the equipment-level canvas, network-level canvas, and energy flow-level canvas specially set by the regional integrated energy system to ensure that the specific visual attribute elements in each canvas match the corresponding system level.

[0102] Spatial position detection is performed on all specific visual attribute primitives in the device-level canvas. For specific visual attribute primitives with overlapping spatial positions, their position coordinates are adjusted according to preset spatial arrangement rules to eliminate the overlap between all primitives. After the adjustment is completed, the device status sub-layer of the regional integrated energy system is formed.

[0103] The generated device status sublayer is kept as a fixed background layer. On top of this background layer, lines are drawn for specific visual attribute primitives in the network layer canvas according to their corresponding system connection relationships. The related primitives are connected by lines. After the drawing is completed, the network topology sublayer of the regional integrated energy system is obtained.

[0104] All specific visual attribute primitives in the energy flow hierarchy canvas are rendered in a semi-transparent form and overlaid on the generated network topology sub-layer. At the same time, based on the fill texture density of each specific visual attribute primitive, its semi-transparency parameters are adjusted according to the preset texture density and transparency correspondence standard. The texture density and transparency have a fixed correlation adjustment relationship. After the rendering and adjustment are completed, the energy flow distribution sub-layer of the regional integrated energy system is obtained.

[0105] The device status sublayer is used as the bottom layer. The network topology sublayer is overlaid on the device status sublayer. Then, the energy flow distribution sublayer is overlaid on top of the network topology sublayer. The image overlay operation of all sublayers is completed in this fixed order from bottom to top, and finally, a multi-level state semantic image of the regional integrated energy system is obtained.

[0106] By accurately identifying parameters of the synchronous data stream and strictly matching visual encoding rules, a precise mapping of data parameters to specific visual attribute primitives is achieved, transforming abstract operational data into concrete visual graphics. Then, through a series of operations including hierarchical canvas allocation, de-overlapping, line drawing, transparent overlay, and sequential merging, various visual primitives are synthesized into multi-level state semantic images according to system hierarchy. This enables the visualization and hierarchical presentation of the operational status of each level of the regional integrated energy system. The multi-level state semantic images clearly reflect the operational characteristics and relationships of system equipment, networks, and energy flows at each level, providing a clear, complete, and hierarchical visual analysis foundation for subsequent visual semantic segmentation and spatial correlation map construction, significantly improving the convenience and accuracy of subsequent visual analysis work.

[0107] In S3, a spatial association map between visual semantic units is constructed simultaneously, including:

[0108] Connectivity analysis is performed on the device state sub-layer in the multi-level state semantic image to obtain candidate device regions of the device state sub-layer. Feature extraction is performed on the filling texture in the candidate device regions to obtain the device visual semantic units of the regional integrated energy system.

[0109] Line segment detection is performed on the network topology sub-layer in the multi-level state semantic image to obtain the connection visual semantic unit of the regional integrated energy system.

[0110] Using pixel transparency and texture density as growth criteria, region growth segmentation is performed on the energy flow distribution sub-layer in the multi-level state semantic image to obtain the energy flow visual semantic unit of the regional integrated energy system.

[0111] Using equipment visual semantic units, connection visual semantic units, and energy flow visual semantic units as nodes, and the connection relationships between equipment and the hierarchical relationships between levels in the regional integrated energy system as edges, a spatial association map between visual semantic units is constructed.

[0112] The visual semantic units of the equipment in the regional integrated energy system are obtained, specifically including:

[0113] Binarize the equipment status sub-layer, mark the pixels with outlines as foreground pixels and the rest as background pixels to obtain the equipment outline binary image of the regional integrated energy system.

[0114] Eight-neighbor connected component labeling is performed on the binary image of device contour. Interconnected foreground pixels are merged into the same connected component, and the minimum bounding rectangle of the connected component is extracted as the candidate device region of the binary image of device contour.

[0115] Gray value distribution analysis is performed on pixels within the candidate device region to obtain texture contrast parameters, texture correlation parameters, and texture inverse difference moment parameters of the candidate device region;

[0116] The texture contrast parameter, texture correlation parameter, and texture inverse difference moment parameter are compared item by item with the equipment texture feature reference set of the regional integrated energy system to obtain the equipment texture matching degree of the regional integrated energy system.

[0117] Based on the device texture matching degree, cluster analysis is performed on the candidate device regions, and a unique device identifier is assigned to each candidate device region to obtain the device visual semantic unit of the regional integrated energy system.

[0118] Pixel values ​​are defined for all pixels in the device status sub-layer of the multi-level state semantic image. Pixels that exhibit contour features are uniformly marked as foreground pixels, and all other pixels that do not have contour features are marked as background pixels. This completes the binarization process of the device status sub-layer, resulting in a binary image of the device contour of the regional integrated energy system.

[0119] An eight-neighbor connected component labeling operation is performed on the binary image of the device contour. Taking each foreground pixel in the image as the center, neighborhood detection is performed on the pixels in the top, bottom, left, right and four diagonal directions. All the detected interconnected foreground pixels are merged into the same connected component. After the division of all connected components is completed, the minimum bounding rectangle of each connected component is extracted. These minimum bounding rectangles are used as candidate device regions of the binary image of the device contour.

[0120] Grayscale values ​​of all pixels within each candidate device region are collected and statistically analyzed. The distribution pattern of pixel grayscale values ​​is analyzed, and the texture contrast parameter, texture correlation parameter, and texture inverse difference moment parameter corresponding to each candidate device region are calculated based on the distribution characteristics of grayscale values ​​to ensure that each parameter can accurately reflect the filling texture characteristics of the candidate device region.

[0121] The texture contrast parameter, texture correlation parameter, and texture inverse difference moment parameter of each candidate device region are compared one by one with the corresponding parameters in the pre-set device texture feature reference set of the regional integrated energy system. Based on the comparison results, the device texture matching degree of the regional integrated energy system corresponding to each candidate device region is calculated.

[0122] Based on the matching degree of device textures in each candidate device region, cluster analysis is performed on all candidate device regions. Candidate device regions with similar matching degrees are grouped into the same category. At the same time, a unique and non-repeating device identifier is assigned to each candidate device region. The clustering results and device identifiers are combined to form the visual semantic unit of the regional integrated energy system.

[0123] A full-image pixel scan is performed on the network topology sub-layer in the multi-level state semantic image to detect the set of pixels in the image that present line segment features. The contours of continuous line segment feature pixels are extracted and their integrity is verified. The verified complete line segments are used as the connection visual semantic units of the regional integrated energy system to ensure that each connection visual semantic unit corresponds to the actual network connection relationship of the system.

[0124] The pixel transparency value and texture density of the energy flow distribution sub-layer in the multi-level state semantic image are used as the core criteria for region growth segmentation. Taking any pixel in the image as the initial growth point, it is detected whether the transparency value and texture density of its neighboring pixels are consistent with the initial growth point. Neighboring pixels that meet the criteria are included in the same growth region. The process continues to expand outward until there are no pixels that meet the criteria. After the segmentation of all growth regions is completed, the energy flow visual semantic unit of the regional integrated energy system is obtained.

[0125] Using the existing visual semantic units of equipment, connection, and energy flow in the regional integrated energy system as all nodes in the spatial association graph, we sort out the actual connection relationships between equipment and the hierarchical relationships between different levels in the regional integrated energy system. We then use these relationships as edges connecting the nodes and construct a spatial association graph between visual semantic units according to the attribute characteristics of the nodes and the association logic of the edges.

[0126] By performing refined binarization, connected component analysis, and texture feature extraction on the device status sub-layer, accurate extraction and identification of device visual semantic units were achieved. Combined with line segment detection and region growing segmentation, connection and energy flow visual semantic units were obtained respectively, so that the operational features of each level of the system were transformed into identifiable visual semantic units. Then, a spatial correlation graph was constructed with various visual semantic units as nodes and the actual correlation relationships of the system as edges, realizing the spatialization and correlation integration of the system's visual semantic units. The graph can clearly present the spatial location and interrelationships of each semantic unit, providing a structured and visualized analysis carrier for subsequent anomaly detection and difference comparison, effectively improving the accuracy and efficiency of subsequent anomaly identification.

[0127] In S4, the abnormal visual semantic units of the regional integrated energy system are obtained, including:

[0128] Based on spatial correlation maps, the historical spatial correlation maps of the regional integrated energy system are hierarchically analyzed to obtain the historical visual semantic units of the regional integrated energy system.

[0129] The attribute feature vectors in the visual semantic units and the historical attribute feature vectors in the historical visual semantic units are quantified one dimension at a time to obtain the attribute difference degree of the regional integrated energy system.

[0130] Traverse the first-order neighbor nodes of the visual semantic unit, record the unit identifier and the direction of the connecting edge of the first-order neighbor node, and obtain the neighbor adjacency sequence of the visual semantic unit.

[0131] Based on the neighbor adjacency sequence, extract the corresponding historical neighbor adjacency sequence from the historical visual semantic unit;

[0132] By comparing and analyzing the neighbor adjacency sequence with the historical neighbor adjacency sequence, the topological structure difference of the regional integrated energy system can be obtained.

[0133] A global anomaly index of visual semantic units is obtained by nonlinearly combining attribute difference degree and topological structure difference degree.

[0134] By comparing the global anomaly index with the preset anomaly judgment threshold, the abnormal visual semantic units of the regional integrated energy system are obtained.

[0135] The formula for calculating the global anomaly index is:

[0136] ;

[0137] Where E represents the global anomaly index, α represents the preset first weighting coefficient, β represents the preset second weighting coefficient, A represents the attribute difference degree, T represents the topological difference degree, and e represents the natural constant. This represents the natural logarithm function.

[0138] Based on the current hierarchical classification standards and node parsing rules of the spatial correlation map, a full-dimensional hierarchical analysis is carried out on the historical spatial correlation map of the regional integrated energy system. The nodes in the historical map are classified and sorted according to the hierarchical categories of equipment, connection, and energy flow. The node information corresponding to each category is extracted and feature annotation is completed to obtain the historical visual semantic units of the regional integrated energy system. The parsing standards ensure that the hierarchical classification and node attributes of the historical visual semantic units are consistent with those of the current visual semantic units.

[0139] Feature vectors are constructed for each attribute feature contained in the current visual semantic unit. At the same time, historical attribute features corresponding to the historical visual semantic units are extracted and historical attribute feature vectors are constructed. The same dimensions of the two feature vectors are compared one by one and numerical differences are calculated. The difference values ​​of each dimension are standardized and integrated to obtain the attribute difference degree of the regional integrated energy system, thereby realizing the quantitative representation of the changes in the attribute features of the visual semantic unit itself.

[0140] Using each visual semantic unit in the spatial association graph as the core node, we comprehensively traverse all first-order neighbor nodes directly connected to the core node according to the connection relationship of the edges in the graph. We record the unique unit identifier of each first-order neighbor node in turn, and mark the direction of the connecting edges between the core node and each neighbor node. We organize the recorded identifiers and direction information in a fixed order to obtain the neighbor adjacency sequence of the visual semantic unit.

[0141] Based on the unit identifiers and connection relationships recorded in the neighbor adjacency sequence of the current visual semantic unit, the corresponding historical core nodes and related historical neighbor nodes are accurately located in the historical visual semantic units. Following the same recording rules and arrangement order as the current neighbor adjacency sequence, the identifiers and connection edge direction information of the related nodes are extracted and organized to obtain the corresponding historical neighbor adjacency sequence.

[0142] The neighbor adjacency sequence of the visual semantic unit is compared and analyzed item by item with the corresponding historical neighbor adjacency sequence. The number of matching unit identifiers and the consistency of the direction of the connecting edge are statistically analyzed. The degree of difference in the topology is quantitatively calculated based on the statistical results, so as to obtain the degree of difference in the topology of the regional integrated energy system and realize the quantitative characterization of the changes in the surrounding associated structure of the visual semantic unit.

[0143] The calculated attribute difference and topological difference are substituted into a preset nonlinear combination logic. The two differences are weighted, fused and processed according to the logic. The dual difference features of attributes and topology are integrated through nonlinear calculation to obtain the global anomaly index corresponding to each visual semantic unit, thereby achieving a comprehensive quantification of the overall anomaly degree of the semantic unit.

[0144] The attribute difference is derived from the dimension-wise difference quantification operation between the attribute feature vector in the visual semantic unit and the historical attribute feature vector in the historical visual semantic unit. The topological difference is derived from the operation of comparing and analyzing the neighbor adjacency sequence of the visual semantic unit with the historical neighbor adjacency sequence corresponding to the historical visual semantic unit. The first weight coefficient and the second weight coefficient are preset values.

[0145] This calculation method integrates the differences between the attribute level and the topological structure level to obtain a quantitative index that can comprehensively reflect the overall degree of anomaly of visual semantic units, thus providing a basis for determining whether a visual semantic unit in a regional integrated energy system is an abnormal visual semantic unit.

[0146] When the attribute difference degree shows an increasing trend, the product of the attribute difference degree and the topological difference degree will increase accordingly. The product of the natural constant raised to the power of the topological difference degree and the product will also increase. At the same time, the increase in the product of the attribute difference degree and the topological difference degree will also increase the result of its natural logarithm operation, ultimately causing the overall comprehensive quantitative index to show an increasing trend.

[0147] When the topological structure difference degree shows an increasing trend, the power of the topological structure difference degree of the natural constant will increase accordingly, and the product of it and the attribute difference degree will also increase. At the same time, the increase in the product of the attribute difference degree and the topological structure difference degree will also increase the result of its natural logarithm operation, ultimately causing the overall comprehensive quantitative index to show an increasing trend.

[0148] As the first weight coefficient increases, the product of its weight, attribute difference degree, and the topological difference degree of the natural constant raised to the power of the weight will also increase, thereby driving the overall comprehensive quantitative index to show an increasing trend.

[0149] As the second weighting coefficient increases, the product of its natural logarithm with the product of attribute difference and topological difference will also increase, thereby driving the overall comprehensive quantitative index to show an increasing trend.

[0150] When the attribute difference shows a decreasing trend, the product of the attribute difference and the topological difference will decrease accordingly. The product of the natural constant raised to the power of the topological difference and the product will also decrease. At the same time, the decrease in the product of the attribute difference and the topological difference will also decrease the result of its natural logarithm operation, ultimately causing the overall comprehensive quantitative index to show a decreasing trend.

[0151] When the topological dissimilarity shows a decreasing trend, the power of the topological dissimilarity of the natural constant will decrease accordingly, and the product of it and the attribute dissimilarity will also decrease. At the same time, the decrease in the product of the attribute dissimilarity and the topological dissimilarity will also decrease the result of its natural logarithm operation, ultimately causing the overall comprehensive quantitative index to show a decreasing trend.

[0152] When the first weight coefficient decreases, the product of its weight, the attribute difference degree, and the topological difference degree of the natural constant raised to the power of the weight will decrease accordingly, thereby causing the overall comprehensive quantitative index to show a decreasing trend.

[0153] When the second weight coefficient decreases, the product of its natural logarithm with the product of attribute difference and topological difference will decrease accordingly, thereby causing the overall comprehensive quantitative index to show a decreasing trend.

[0154] The global anomaly index of each visual semantic unit is compared one by one with the anomaly judgment threshold set in advance for the regional integrated energy system. If the global anomaly index of a certain visual semantic unit exceeds the anomaly judgment threshold, the visual semantic unit is judged as an abnormal unit. The information of all units judged as abnormal is integrated to obtain the abnormal visual semantic units of the regional integrated energy system.

[0155] By performing hierarchical analysis of historical contemporaneous maps based on the current spatial correlation map, the consistency of the analysis of historical and current visual semantic units is ensured, laying a unified foundation for subsequent difference comparison. Then, by quantifying attribute differences dimension by dimension, comparing and analyzing topological structure differences, and nonlinearly combining the two to obtain a global anomaly index, the anomaly degree of visual semantic units in both attribute and topological correlation dimensions is comprehensively quantified. Finally, by comparing the global anomaly index with the threshold, abnormal visual semantic units are accurately located. This allows anomaly identification to take into account both the characteristics of the semantic unit itself and changes in the surrounding correlation structure, greatly improving the comprehensiveness and accuracy of abnormal visual semantic unit identification, and providing accurate anomaly analysis results for subsequent anomaly verification and control decisions.

[0156] In S5, the abnormal unit identifiers and abnormal types of the regional integrated energy system are obtained, and these identifiers and types are integrated into an abnormal event report for the regional integrated energy system, including:

[0157] Establish a tracking profile for abnormal visual semantic units, which includes the current frame image coordinates and current visual attribute parameters;

[0158] Acquire subsequent frame multi-level state semantic images of the regional integrated energy system, and locate subsequent visual semantic units with the same image coordinate range and similar visual attribute parameters as the abnormal visual semantic units in the subsequent frame multi-level state semantic images.

[0159] The subsequent visual semantic units are used as tracking continuation units of the abnormal visual semantic units in subsequent frames, and the image coordinates and visual attribute parameters of the tracking continuation units are appended to the tracking archive to obtain the complete tracking archive of the regional integrated energy system.

[0160] Based on the changing trends of visual attribute parameters in the complete tracking file, the anomaly type of the abnormal visual semantic unit is determined, and a unique anomaly unit identifier is assigned to the abnormal visual semantic unit.

[0161] Based on the hierarchical relationship and frame order of abnormal visual semantic units in the spatial association map, the abnormal types and abnormal unit identifiers are sorted out to obtain the abnormal event report of the regional integrated energy system.

[0162] For each identified abnormal visual semantic unit of the regional integrated energy system, a dedicated tracking file is created. The file contains the precise image coordinate range of the abnormal visual semantic unit in the multi-level state semantic image of the current frame, as well as all the current visual attribute parameters corresponding to the unit, to ensure that the basic information of the tracking file is complete and accurately corresponds to the abnormal visual semantic unit.

[0163] Based on the data acquisition frequency and image generation frequency of the regional integrated energy system, the system continuously acquires each frame of multi-level state semantic image generated by the system. In each subsequent frame image, the system performs precise regional positioning based on the image coordinate range in the established tracking archive. Then, within the positioning area, the system selects subsequent visual semantic units that match the current visual attribute parameter features in the archive, thus completing the subsequent frame matching and positioning of abnormal visual semantic units.

[0164] The subsequent visual semantic units located and matched in the multi-level state semantic images of subsequent frames are identified as the tracking continuation units of the corresponding abnormal visual semantic units in that frame. The actual image coordinates and real-time visual attribute parameters of each tracking continuation unit in the corresponding frame are sequentially added to its dedicated tracking file according to the time sequence of the frames. After the tracking record of the preset number of frames is completed, a complete tracking file of the regional integrated energy system containing information on the entire tracking process is formed.

[0165] A systematic review of all visual attribute parameters recorded in the complete tracking archive in chronological order was conducted. The numerical change trends, fluctuation patterns, and characteristic changes of each parameter during the tracking period were analyzed. Based on the pre-set anomaly type judgment criteria of the regional integrated energy system, the specific anomaly type corresponding to the anomaly visual semantic unit was determined according to the parameter change characteristics. At the same time, a unique and non-repeating anomaly unit identifier was assigned to each anomaly visual semantic unit within the system to achieve accurate identification of anomaly units.

[0166] The specific hierarchical affiliation of each abnormal visual semantic unit to the equipment, network, and energy flow in the spatial association map is sorted out. At the same time, the order of the first frame appearance when each abnormal visual semantic unit is identified as abnormal and the time information of the subsequent tracked frames are confirmed. The units are classified according to their hierarchical affiliation, and then the abnormal unit identifiers and corresponding specific abnormal types are sorted out in each category according to the time order of the frames. All sorted information is standardized and integrated to finally obtain the abnormal event report of the regional integrated energy system.

[0167] By establishing dedicated tracking profiles for abnormal visual semantic units and conducting multi-frame tracking records, the position and attribute parameter changes of abnormal units in different frames are fully preserved. Continuous tracking of abnormal units is achieved through precise positioning and matching in subsequent frames. Furthermore, by analyzing parameter change trends, the abnormality type is accurately determined and a unique abnormal unit identifier is assigned. Finally, information is organized hierarchically and in frame order to form an abnormal event report. This ensures that the recording of abnormal events is accurate, complete, and systematic. It achieves precise identification of abnormal units and accurate determination of abnormal types, and provides a clear, timely, and complete basis for subsequent hierarchical decision-making and collaborative control, effectively guaranteeing the pertinence and effectiveness of subsequent control decisions.

[0168] In S6, coordinated control instructions for the regional integrated energy system are obtained, including:

[0169] The abnormal event report is analyzed to obtain the abnormal unit identifier, abnormal type and abnormal occurrence time of the abnormal visual semantic unit. Based on the abnormal unit identifier, the layer level and spatial coordinate position of the abnormal visual semantic unit in the spatial association map are located.

[0170] Based on the layer hierarchy, the abnormal visual semantic units are divided into device-level abnormal units, network-level abnormal units, and power flow-level abnormal units.

[0171] Based on the anomaly type, the equipment-level anomaly units, network-level anomaly units, and energy flow-level anomaly units are clustered and grouped to obtain the hierarchical anomaly grouping set of the regional integrated energy system.

[0172] Based on the hierarchical anomaly group set, the spatial correlation map is used to explore the anomaly propagation path in order to obtain the associated device set of the hierarchical anomaly group set;

[0173] Based on the associated equipment set and anomaly type, preliminary control parameter recommendations for the regional integrated energy system are determined;

[0174] The preliminary control parameter recommendations are merged and deduplicated to obtain the hierarchical control instruction set of the regional integrated energy system.

[0175] The hierarchical control instruction set is encapsulated level by level, and during the encapsulation process, the linkage association identifier between control instructions in the hierarchical control instruction set is established based on the hierarchical unit membership relationship in the spatial correlation map, so as to obtain the coordinated control instructions of the regional integrated energy system.

[0176] A comprehensive analysis of abnormal event reports from the regional integrated energy system is conducted. The abnormal unit identifier, specific abnormal type, and the time of first occurrence of the abnormality are extracted from the reports for all abnormal visual semantic units. Based on the extracted abnormal unit identifiers, a precise search is performed in the spatial correlation map to locate the specific layer level of each abnormal visual semantic unit in the map. At the same time, the precise spatial coordinates of the unit in the map are obtained to ensure accurate matching of the hierarchical and spatial information of the abnormal unit.

[0177] Based on the layer hierarchy results obtained from the spatial association map, all abnormal visual semantic units are hierarchically divided. Abnormal units belonging to the device level are classified as device-level abnormal units, abnormal units belonging to the network level are classified as network-level abnormal units, and abnormal units belonging to the power flow level are classified as power flow-level abnormal units, thus completing the hierarchical classification of abnormal units.

[0178] Based on the predefined equipment-level, network-level, and energy flow-level anomaly units, and according to the specific anomaly type corresponding to each unit, the anomaly units at the same level are clustered. Anomaly units at the same level with the same anomaly type are grouped into the same group, while units with different anomaly types are grouped into different groups. By integrating the clustering results of all levels, a hierarchical anomaly grouping set of the regional integrated energy system is obtained.

[0179] Taking the hierarchical anomaly group set as the core, and relying on the connection and affiliation relationships of each unit in the spatial correlation map, the map is explored in its entirety to investigate the anomaly propagation path, track the possible anomaly impact range of each unit within the anomaly group, retrieve all equipment units that are directly or indirectly related to the anomaly group, and integrate all retrieved equipment unit information to obtain the associated equipment set of the hierarchical anomaly group set.

[0180] Based on the operational attributes of all equipment units included in the associated equipment set, combined with the specific anomaly types corresponding to each anomaly group, and referring to the operation and control specifications of the regional integrated energy system, appropriate control parameter adjustment suggestions are formulated for each associated equipment unit. By integrating the control parameter suggestions of all equipment units, preliminary control parameter suggestions for the regional integrated energy system are obtained.

[0181] A comprehensive review and verification of the preliminary control parameter recommendations was conducted. Multiple control parameter recommendations for the same equipment unit were merged and integrated, and duplicate parameter adjustments were removed. Only unique and suitable control parameter recommendations for each equipment unit were retained. The recommendations were then classified and organized according to the levels of equipment, network, and energy flow to obtain the hierarchical control instruction set of the regional integrated energy system.

[0182] The control commands in the hierarchical control command set are encapsulated level by level according to the order of equipment, network, and energy flow. During the encapsulation process, based on the hierarchical relationship between cross-level units in the spatial correlation map, the correlation logic between control commands at different levels is sorted out. A unified linkage correlation identifier is added to the control commands with correlation relationships, the linkage execution relationship between commands is clarified, and all the encapsulated control commands with added linkage identifiers are integrated to obtain the coordinated control commands of the regional integrated energy system.

[0183] By analyzing abnormal event reports and combining them with spatial correlation maps, the hierarchical and spatial positioning of abnormal units is accurately located, laying a precise foundation of abnormal information for hierarchical decision-making. Further hierarchical division, clustering, and exploration of abnormal propagation paths clarify the hierarchical distribution, type characteristics, and impact range of abnormalities. Subsequently, control recommendations are formulated based on associated equipment sets and abnormal types, and deduplication is completed to make control instructions more targeted and concise. Finally, by encapsulating at each level and establishing cross-layer linkage association identifiers, the hierarchical and collaborative design of control instructions is realized. This enables the generated collaborative control instructions to take into account the abnormal control needs of each level, while relying on linkage identifiers to ensure the collaborative execution of cross-layer instructions. This significantly improves the scientific, targeted, and collaborative nature of abnormal control in the regional integrated energy system, effectively ensuring the stable operation of the system.

[0184] Example 2: A multi-level intelligent agent construction device for a regional integrated energy system, comprising:

[0185] One or more processors;

[0186] Memory, used to store one or more computer programs;

[0187] When one or more programs are executed by one or more processors, the one or more processors execute the method in Example 1.

[0188] Example 3: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method in Example 1.

[0189] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can make equivalent substitutions or modifications based on the technical solution and concept disclosed in the present invention within the scope of the technology disclosed in the present invention, and such modifications should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for constructing a multi-level intelligent agent in a regional integrated energy system, characterized in that the method... include: S1. Perform spatiotemporal alignment processing on the real-time operation data of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system. S2. According to the preset visual coding rules, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, and specific visual attribute primitives are synthesized according to the system hierarchy of the regional integrated energy system to obtain a multi-level state semantic image of the regional integrated energy system. S3. Perform visual semantic segmentation on the multi-level state semantic image to obtain the visual semantic units of the regional integrated energy system, and at the same time construct a spatial association map between the visual semantic units. S4. Visually compare the spatial correlation map with the historical spatial correlation map of the regional integrated energy system to obtain the abnormal visual semantic units of the regional integrated energy system. S5. Perform multi-frame tracking and verification on abnormal visual semantic units to obtain the abnormal unit identifier and abnormal type of the regional integrated energy system, and integrate the abnormal unit identifier and abnormal type into an abnormal event report of the regional integrated energy system. S6. Make hierarchical decisions based on abnormal event reports and spatial correlation maps to obtain coordinated control instructions for the regional integrated energy system.

2. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S1, the synchronous data stream of the regional integrated energy system is obtained, including: By collecting the raw operating data of the power subsystem, heat subsystem, and gas subsystem of the regional integrated energy system, the real-time operating data of the regional integrated energy system can be obtained. Extract the collection timestamp and geographic location identifier of real-time operational data; Consistency verification is performed on the collected timestamps, and data with time deviations are adjusted to a unified time reference axis to obtain the time-series operation data of the regional integrated energy system. Based on geographic location identifiers, time-series operational data is matched to the system spatial grid of the regional integrated energy system to obtain the synchronous data stream of the regional integrated energy system.

3. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S2, the parameters in the synchronous data stream are mapped to specific visual attribute primitives of the regional integrated energy system, including: Parameter identification is performed on the synchronous data stream to obtain the parameter type identifier and parameter value of the synchronous data stream; Based on the parameter type identifier, select the corresponding basic graphic shape from the preset visual encoding rules; Based on the parameter values, determine the thickness of the outline of the basic shape of the primitive and the density of the internal filling texture; Based on the thickness of the outline, the density of the internal filling texture, and the basic shape of the primitives, the synchronous data stream is used to draw images, resulting in specific visual attribute primitives of the regional integrated energy system.

4. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S2, a multi-level state semantic image of the regional integrated energy system is obtained, including: Based on the node level label corresponding to the specific visual attribute primitive, the specific visual attribute primitive is assigned to the equipment level canvas, network level canvas, and energy flow level canvas of the regional integrated energy system. By performing de-overlap processing on specific visual attribute primitives in the equipment-level canvas, the equipment status sub-layer of the regional integrated energy system is obtained. Using the device status sub-layer as the background, draw lines connecting specific visual attribute primitives in the network layer canvas to obtain the network topology sub-layer of the regional integrated energy system. Specific visual attribute primitives in the energy flow hierarchy canvas are drawn on the network topology sublayer in a semi-transparent overlay manner, and the transparency is adjusted according to the fill texture density of the specific visual attribute primitives in the energy flow hierarchy canvas to obtain the energy flow distribution sublayer of the regional integrated energy system. By overlaying the equipment status sublayer, network topology sublayer, and energy flow distribution sublayer in a bottom-up order, a multi-level state semantic image of the regional integrated energy system is obtained.

5. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S3, visual semantic units of the regional integrated energy system are obtained, and a spatial relationship map between visual semantic units is constructed, including: Connectivity analysis is performed on the device state sub-layer in the multi-level state semantic image to obtain candidate device regions of the device state sub-layer. Feature extraction is performed on the filling texture in the candidate device regions to obtain the device visual semantic units of the regional integrated energy system. Line segment detection is performed on the network topology sub-layer in the multi-level state semantic image to obtain the connection visual semantic unit of the regional integrated energy system. Using pixel transparency and texture density as growth criteria, region growth segmentation is performed on the energy flow distribution sub-layer in the multi-level state semantic image to obtain the energy flow visual semantic unit of the regional integrated energy system. Using equipment visual semantic units, connection visual semantic units, and energy flow visual semantic units as nodes, and the connection relationships between equipment and the hierarchical relationships between levels in the regional integrated energy system as edges, a spatial association map between visual semantic units is constructed.

6. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 5, characterized in that, The visual semantic units of the equipment in the regional integrated energy system are obtained, including: Binarize the equipment status sub-layer, mark the pixels with outlines as foreground pixels and the rest as background pixels to obtain the equipment outline binary image of the regional integrated energy system. Eight-neighbor connected component labeling is performed on the binary image of device contour. Interconnected foreground pixels are merged into the same connected component, and the minimum bounding rectangle of the connected component is extracted as the candidate device region of the binary image of device contour. Gray value distribution analysis is performed on pixels within the candidate device region to obtain texture contrast parameters, texture correlation parameters, and texture inverse difference moment parameters of the candidate device region; The texture contrast parameter, texture correlation parameter, and texture inverse difference moment parameter are compared item by item with the equipment texture feature reference set of the regional integrated energy system to obtain the equipment texture matching degree of the regional integrated energy system. Based on the device texture matching degree, cluster analysis is performed on the candidate device regions, and a unique device identifier is assigned to each candidate device region to obtain the device visual semantic unit of the regional integrated energy system.

7. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S4, the abnormal visual semantic units of the regional integrated energy system are obtained, including: Based on spatial correlation maps, the historical spatial correlation maps of the regional integrated energy system are hierarchically analyzed to obtain the historical visual semantic units of the regional integrated energy system. The attribute feature vectors in the visual semantic units and the historical attribute feature vectors in the historical visual semantic units are quantified one dimension at a time to obtain the attribute difference degree of the regional integrated energy system. Traverse the first-order neighbor nodes of the visual semantic unit, record the unit identifier and the direction of the connecting edge of the first-order neighbor node, and obtain the neighbor adjacency sequence of the visual semantic unit. Based on the neighbor adjacency sequence, extract the corresponding historical neighbor adjacency sequence from the historical visual semantic unit; By comparing and analyzing the neighbor adjacency sequence with the historical neighbor adjacency sequence, the topological structure difference of the regional integrated energy system can be obtained. A global anomaly index of visual semantic units is obtained by nonlinearly combining attribute difference degree and topological structure difference degree. By comparing the global anomaly index with the preset anomaly judgment threshold, the abnormal visual semantic units of the regional integrated energy system are obtained.

8. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 7, characterized in that, The formula for calculating the global anomaly index is: ； Where E represents the global anomaly index, α represents the preset first weighting coefficient, β represents the preset second weighting coefficient, A represents the attribute difference degree, T represents the topological difference degree, and e represents the natural constant. This represents the natural logarithm function.

9. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S5, the abnormal unit identifiers and abnormal types of the regional integrated energy system are obtained, and these identifiers and types are integrated into an abnormal event report for the regional integrated energy system, including: Establish a tracking profile for abnormal visual semantic units, which includes the current frame image coordinates and current visual attribute parameters; Acquire subsequent frame multi-level state semantic images of the regional integrated energy system, and locate subsequent visual semantic units with the same image coordinate range and similar visual attribute parameters as the abnormal visual semantic units in the subsequent frame multi-level state semantic images. The subsequent visual semantic units are used as tracking continuation units of the abnormal visual semantic units in subsequent frames, and the image coordinates and visual attribute parameters of the tracking continuation units are appended to the tracking archive to obtain the complete tracking archive of the regional integrated energy system. Based on the changing trends of visual attribute parameters in the complete tracking file, the anomaly type of the abnormal visual semantic unit is determined, and a unique anomaly unit identifier is assigned to the abnormal visual semantic unit. Based on the hierarchical relationship and frame order of abnormal visual semantic units in the spatial association map, the abnormal types and abnormal unit identifiers are sorted out to obtain the abnormal event report of the regional integrated energy system.

10. The method for constructing a multi-level intelligent agent in a regional integrated energy system as described in claim 1, characterized in that, In S6, coordinated control instructions for the regional integrated energy system are obtained, including: The abnormal event report is analyzed to obtain the abnormal unit identifier, abnormal type and abnormal occurrence time of the abnormal visual semantic unit. Based on the abnormal unit identifier, the layer level and spatial coordinate position of the abnormal visual semantic unit in the spatial association map are located. Based on the layer hierarchy, the abnormal visual semantic units are divided into device-level abnormal units, network-level abnormal units, and power flow-level abnormal units. Based on the anomaly type, the equipment-level anomaly units, network-level anomaly units, and energy flow-level anomaly units are clustered and grouped to obtain the hierarchical anomaly grouping set of the regional integrated energy system. Based on the hierarchical anomaly group set, the spatial correlation map is used to explore the anomaly propagation path in order to obtain the associated device set of the hierarchical anomaly group set; Based on the associated equipment set and anomaly type, preliminary control parameter recommendations for the regional integrated energy system are determined; The preliminary control parameter recommendations are merged and deduplicated to obtain the hierarchical control instruction set of the regional integrated energy system. The hierarchical control instruction set is encapsulated level by level, and during the encapsulation process, the linkage association identifier between control instructions in the hierarchical control instruction set is established based on the hierarchical unit membership relationship in the spatial correlation map, so as to obtain the coordinated control instructions of the regional integrated energy system.

Images