Edge computing-based energy storage cooling system cooperative control method and system

Abstract

The application discloses an energy storage cooling system collaborative control method and system based on edge computing, relates to the technical field of edge computing control, and comprises the following steps: acquiring real-time thermal state data flow, performing distributed feature segmentation and edge node local thermal response mapping, and generating a multi-node temperature control graph; based on the multi-node temperature control graph, performing cross-node cooling resource collaborative allocation to form a hierarchical cooling scheduling sequence; performing space-time arrangement and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node to drive the collaborative action of the cooling unit and form a state offset set; performing edge side time sequence correlation backtracking and local strategy reconstruction processing on the state offset set to generate a feedback control instruction, and writing the feedback control instruction back to the multi-node temperature control graph; and the application solves the problems of uneven distribution of multi-node thermal states in the energy storage system and the lack of dynamic feedback regulation capability of the overall temperature control strategy through edge collaborative regulation and space-time arrangement technology.

Description

Technical Field

[0001] This invention relates to the field of edge computing control technology, and more specifically, to a collaborative control method and system for energy storage and cooling systems based on edge computing. Background Technology

[0002] With the widespread application of energy storage systems in grid peak shaving, renewable energy consumption, and distributed energy scenarios, the scale of battery clusters is constantly expanding, and the problems of internal heat accumulation and uneven spatial distribution are becoming increasingly prominent. During operation, energy storage devices are prone to local hotspots and uncertain heat diffusion paths due to the influence of charging and discharging states, ambient temperature, and structural layout, which in turn affects the stability and safety of system operation. Furthermore, energy storage systems typically adopt a distributed deployment structure, with significant spatial separation and state differences between nodes, leading to uneven distribution of cooling resources and localized control lags. Therefore, how to achieve coordinated scheduling of cooling resources, precise control of the execution process, and continuous feedback optimization of operating status under dynamic changes in the thermal state of multiple nodes, and construct an energy storage cooling system with real-time response and coordinated control capabilities, has become a key technical problem that urgently needs to be solved in the field of energy storage thermal management and intelligent control.

[0003] For example, the invention patent with publication number CN120373115A discloses a real-time thermal flow field control method for power batteries based on edge computing, including the following steps: S1, collecting and preprocessing temperature data to construct a time-series dataset; S2, organizing the data into a Euclidean thermal data structure by space and time; S3, diffusion mapping to non-Euclidean manifold features; S4, constructing a thermal field graph structure model; S5, inputting the graph structure to an adaptive convolutional neural network model for edge nodes; S6, training the model using historical data; S7, optimizing the model training using a sharpness-aware minimization algorithm; S8, predicting future heat flow distribution in real time; S9, generating a thermal management strategy instruction set; S10, controlling the execution module for real-time regulation. This significantly improves the accuracy of thermal anomaly prediction and the adaptive capability of cooling strategies, and can be widely applied in intelligent thermal management scenarios.

[0004] For example, the invention patent CN120725380B discloses a microgrid-based thermal energy management method and system, which includes the following steps: acquiring temperature and power data and establishing path correspondence; extracting thermal response and load change parameters to generate a priority list; embedding the topology diagram to adjust the path activation order; identifying phase changes to form conflict node groups; and writing the data into a control table to generate a path allocation structure. By constructing a path mapping of temperature and power, extracting node response characteristics to generate a priority sequence, updating the path status in conjunction with the topology, identifying interfering nodes based on phase change comparison and setting the control order, the system promotes the maintenance of a control rhythm in thermal energy allocation, enhances the ability to identify dynamic path behavior, improves the coordination level and resource utilization efficiency between nodes, and promotes a smoother thermal energy scheduling process in environments with frequent load fluctuations, reducing the fluctuation impact caused by path conflicts and control lags, and strengthening the support foundation for stable operation.

[0005] The above-disclosed technical solutions have at least the following technical problems:

[0006] Traditional methods often involve centralized or single-point control, making it difficult to coordinate the thermal status of multiple nodes, resulting in weak dynamic sensing capabilities, delayed cooling scheduling, and a lack of backtracking optimization mechanisms, leading to insufficient overall control accuracy and responsiveness. To address these issues, this invention proposes a solution. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a collaborative control method and system for an energy storage cooling system based on edge computing. By using edge collaborative regulation and spatiotemporal orchestration technology, the method solves the problems of uneven thermal state distribution among multiple nodes in the energy storage system and the lack of dynamic feedback adjustment capability in the overall temperature control strategy.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A collaborative control method and system for an energy storage cooling system based on edge computing includes: acquiring real-time thermal state data streams and performing distributed feature segmentation and mapping with local thermal responses of edge nodes to generate a multi-node temperature control map; based on the multi-node temperature control map, performing cross-node collaborative allocation of cooling resources to form a hierarchical cooling scheduling sequence; performing spatiotemporal orchestration and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node to drive the coordinated action of cooling units and form a state offset set; performing edge-side temporal correlation backtracking and local strategy reconstruction processing on the state offset set to generate feedback control commands, and writing the feedback control commands back to the multi-node temperature control map.

[0010] In a preferred technical solution, the acquisition of real-time thermal state data stream, and the distribution of feature segmentation and edge node local thermal response mapping to generate a multi-node temperature control map, specifically involves the following steps: collecting energy storage unit operation data, marking the collection identifier, and aggregating it to form a real-time thermal state data stream; performing time series rearrangement and spatial aggregation on the real-time thermal state data stream to form a unified spatiotemporal calibration data sequence; performing edge-side self-evolution segmentation on the unified spatiotemporal calibration data sequence to form multi-level thermal evolution fragment clusters; performing multi-dimensional feature decoupling extraction on each thermal evolution fragment cluster to construct a combined feature set and form a fragment feature expression unit; performing cross-node thermal behavior association weaving based on the fragment feature expression unit to form a thermal behavior association chain structure; performing local response projection on the thermal behavior association chain structure to generate a regional thermal response mapping body; performing boundary-guided fusion on each regional thermal response mapping body to form an initial multi-node temperature control map; and performing edge-side scrolling refresh on the initial multi-node temperature control map to output the multi-node temperature control map.

[0011] In a preferred technical solution, the step of performing edge-side self-evolutionary segmentation on the unified spatiotemporal calibration data sequence to form a multi-level thermal evolution fragment cluster is as follows: the unified spatiotemporal calibration data sequence is expanded point by point in chronological order to form a continuous change trajectory sequence; based on the change trajectory sequence, the data that change continuously in the same direction are marked with intervals, and the locations where the change direction reverses and the change rate changes abruptly are marked with boundaries to form a first segment candidate point set; based on the first segment candidate point set, a sliding window scan is performed on adjacent candidate intervals to form a second segment candidate interval; the second segment candidate interval is filtered by interval length and change density to form a primary fragment sequence; the primary fragment sequence is processed by cross-segment trajectory backtracking, and fragments with multi-stage changes are further segmented; the data within each segment after segmentation is reorganized in chronological order, and the segment boundary positions are aligned with neighbors to form stable boundary points; the stable boundary points are sequentially numbered, and the data is collected and encapsulated by time interval to form a multi-level thermal evolution fragment cluster.

[0012] In a preferred technical solution, the step of performing cross-node cooling resource collaborative allocation based on a multi-node temperature control map to form a hierarchical cooling scheduling sequence is as follows: A node thermal state sequence is formed based on the temperature distribution and spatial topology of each node in the multi-node temperature control map; a node thermal state sequence is processed by thermal potential gradient expansion to form a node thermal potential gradient map; a hotspot migration trajectory identification process is performed based on the node thermal potential gradient map to form a set of thermal potential migration trajectories; the set of thermal potential migration trajectories is divided into node affiliations to form path-associated node clusters; a region mapping and reorganization process is performed on the path-associated node clusters to generate a block association index structure; a load mapping and decomposition process is performed based on the cooling response blocks to form a block cooling demand distribution map; the block cooling demand distribution map is chronologically ordered to form a block demand evolution sequence; a cross-block linkage orchestration process is performed on the block demand evolution sequence to form a block collaborative scheduling chain; and a hierarchical mapping and sequence reorganization process is performed based on the block collaborative scheduling chain to form a hierarchical cooling scheduling sequence.

[0013] In a preferred technical solution, the process of performing region mapping and reorganization on the path-related node cluster and generating a block association index structure is as follows: Extracting the spatial coordinates, migration path, and time series of each node in the path-related node cluster, and marking overlapping nodes that exist on multiple migration paths; performing dynamic path stripping and assignment optimization on the overlapping nodes, and reassigning them to the region set; dividing the spatially continuous region set with the same thermal potential migration direction into unit segments to form initial region segments; performing cross-path merging on the initial region segments to form cooling response blocks; performing hierarchical structure optimization on the cooling response blocks to generate a block index table; establishing association chains between adjacent blocks or blocks with continuous thermal potential gradients in the cooling response blocks, and outputting the cooling response blocks and the block association index structure.

[0014] In a preferred embodiment, the hierarchical cooling scheduling sequence is spatiotemporally orchestrated and trajectory-constrained at each edge node to drive the cooling units to coordinate actions and form a state offset set. Specifically, this involves: acquiring the scheduling execution segments of the hierarchical cooling scheduling sequence at each edge node to construct a node-level cooling driving primitive set; performing spatiotemporal embedding and rearrangement on the node-level cooling driving primitive set to generate a spatiotemporally aligned driving field; extracting the cooling action propagation path based on the spatiotemporally aligned driving field, and performing continuous expansion and spatial folding mapping on the path to construct a cross-node cooling conduction trajectory field; performing constraint domain embedding processing on the cross-node cooling conduction trajectory field to generate a restricted scheduling trajectory field; mapping trajectory nodes to corresponding cooling execution actions based on the restricted scheduling trajectory field, and performing hierarchical embedding and arrangement of the cooling execution actions to form a node action driving graph; interweaving and combining the action paths of spatially adjacent nodes in the node action driving graph to generate a collaborative driving structure; driving the cooling units at each edge node to execute actions based on the collaborative driving structure to generate a node state evolution graph; and performing spatial reclamation and temporal folding on the node state evolution graph to form a state offset set.

[0015] In a preferred technical solution, the constraint domain embedding process performed on the cross-node cooling conduction trajectory field to generate a constrained scheduling trajectory domain is as follows: The structure of each trajectory path in the cross-node cooling conduction trajectory field is analyzed, and the complete path is decomposed into multiple continuous trajectory segments according to the node connection order; overlapping trajectory segments among the multiple continuous trajectory segments are identified and uniformly marked as a conflict segment set, while trajectory segments with continuous spatial distribution and single connection relationship are marked as a candidate passable segment set; the trajectory segments in the conflict segment set are removed from the path, and their connection positions in the original path are recorded, while the segment combinations that can form a continuous path structure in the candidate passable segment set are retained; based on the segment combinations, each trajectory segment is used as a connection unit in the graph, and according to the segment... Spatial adjacency relationships between segments are used to establish connecting edges, forming a trajectory segment topology graph. Based on this graph, the connecting edges are traversed sequentially, and the corresponding connecting regions are defined as path blocking areas. Simultaneously, connecting path regions that maintain unidirectional extension and do not intersect are defined as priority passage areas. Based on the path blocking areas and priority passage areas, path filtering is performed on the trajectory segment topology graph, retaining connecting paths located within the priority passage areas to form a preliminary path set. Based on this preliminary path set, path branches with interruptions and duplicate coverage are removed, and structurally continuous paths are selected as the preferred path set. Based on the preferred path set, the paths are sequentially reassembled. Compensation connection processing is performed on segments in the reassembled paths that still have local breaks and incomplete connections, generating a restricted scheduling trajectory domain.

[0016] In a preferred technical solution, the step of performing edge-side temporal correlation backtracking and local strategy reconstruction processing on the state offset set to generate feedback control commands, and writing the feedback control commands back to the multi-node temperature control map, specifically includes: performing edge-side data expansion processing on the state offset set to form a node state evolution sequence; performing temporal segment folding reconstruction on the node state evolution sequence, and independently marking segments with abrupt changes and offsets to form state folding units; performing reverse trajectory backtracking based on the state folding units to form a state backtracking path set; and performing path aggregation on the state backtracking path set to form a state evolution correlation structure; Local behavior mapping is performed on the state evolution association structure to form a behavior mapping fragment set; adjacent behavior fragments with continuous control relationships in the behavior mapping fragment set are combined and rearranged, and conflicting and duplicate behavior fragments are filtered out and replaced to form strategy combination units; multi-path fusion is performed on the strategy combination units to generate a cross-node collaborative strategy structure; each control behavior in the cross-node collaborative strategy structure is arranged according to the node and time order to form feedback control instructions; the feedback control instructions are mapped back to the corresponding node positions according to the node spatial position and written into the corresponding node data structure in the multi-node temperature control diagram in time order.

[0017] In a preferred embodiment, the process of performing reverse trajectory backtracking based on state folding units to form a state backtracking path set is as follows: Each folded segment in the state folding unit is obtained, a time inversion index is constructed, and the folded segments are arranged in reverse time order to form a set of reverse time series segments; based on the time inversion index, a segment-by-segment backtracking traversal is performed on the reverse time series segments to form an initial reverse trajectory chain; the time identifiers of each node in the initial reverse trajectory chain are uniformly mapped to the reverse time axis to form a backtracking segment structure; according to the backtracking segment structure, the node states in the current time layer are compared with those in the previous time layer. Node states that are spatially adjacent and have continuous state evolution are matched and connected. In cases where there are multiple candidate predecessor nodes, selection is made according to spatial adjacency, generating multi-branch backtracking paths layer by layer. Backtracking paths with the same starting anchor point and the same starting time are grouped into the same set, and the node sequences are uniformly arranged in reverse chronological order within the set. After the arrangement is completed, node sequences that have shifted or misaligned in adjacent time layers of the multi-branch backtracking paths are locally corrected to obtain backtracking path units. Backtracking path units are summarized and arranged according to node identifier and chronological order to form a state backtracking path set.

[0018] In a preferred embodiment, a collaborative control method for an energy storage cooling system based on edge computing is characterized by comprising a data mapping module, a collaborative allocation module, a scheduling control module, and a feedback optimization module, with interconnections between the modules. The data mapping module is used to acquire real-time thermal state data streams and perform distributed feature segmentation and mapping with local thermal responses of edge nodes to generate a multi-node temperature control map. The collaborative allocation module is used to perform cross-node collaborative allocation of cooling resources based on the multi-node temperature control map to form a hierarchical cooling scheduling sequence. The scheduling control module is used to perform spatiotemporal orchestration and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node to drive the coordinated action of cooling units and form a state offset set. The feedback optimization module is used to perform edge-side temporal correlation backtracking and local strategy reconstruction processing on the state offset set to generate feedback control commands and write the feedback control commands back to the multi-node temperature control map.

[0019] The technical effects and advantages of the collaborative control method and system for energy storage cooling system based on edge computing of the present invention are as follows:

[0020] 1. This invention acquires real-time thermal state data streams and performs distributed feature segmentation and edge node local thermal response mapping technology. Based on the segmentation of thermal state data according to time and spatial characteristics, it maps the thermal changes of each edge node into a local response distribution, forming a multi-node temperature control map with spatial correlation. This enables a refined characterization of the thermal state of the energy storage system, significantly improving the accuracy and dynamic response capability of multi-node thermal state perception. Furthermore, by performing cross-node cooling resource collaborative allocation technology based on the multi-node temperature control map, a hierarchical cooling scheduling sequence is constructed based on the extraction of temperature differences and spatial topological relationships between nodes. This enables the orderly allocation and priority scheduling of cooling resources among multiple nodes, significantly improving the utilization efficiency of cooling resources and the overall control balance.

[0021] 2. This invention employs spatiotemporal orchestration and trajectory constraint control techniques on each edge node to perform hierarchical cooling scheduling sequences. By mapping the scheduling sequences to a unified spatiotemporal structure and constraining and reorganizing the cooling paths, it drives cooling units to coordinate actions according to spatial adjacency and temporal order. This enables synchronous execution and conflict avoidance of multi-node cooling behaviors, significantly improving the coordination and stability of the cooling process. Furthermore, by performing edge-side temporal correlation backtracking and local strategy reconstruction techniques on the state offset set, feedback control commands are generated based on reverse backtracking of node state evolution and mapping state changes to control behaviors. These commands are then written back to the multi-node temperature control diagram, enabling dynamic correction and closed-loop optimization of the control strategy, further enhancing the system's control accuracy and adaptability. Attached Figure Description

[0022] Figure 1This is a flowchart illustrating the collaborative control method for an energy storage cooling system based on edge computing according to the present invention.

[0023] Figure 2 This is a schematic diagram of the system structure of the collaborative control method for energy storage and cooling systems based on edge computing according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1, Figure 1 The present invention provides a collaborative control method for an energy storage cooling system based on edge computing, comprising the following steps:

[0026] S1: Acquire real-time thermal status data stream, perform distributed feature segmentation and edge node local thermal response mapping, and generate multi-node temperature control map;

[0027] In this embodiment, real-time thermal state data stream is acquired, and distributed feature segmentation and local thermal response mapping of edge nodes are performed to generate a multi-node temperature control map, as detailed below:

[0028] The system collects temperature information, cooling equipment operating status, and environmental heat exchange data during the operation of the energy storage unit, marks the collected data, and aggregates them into a continuously updated real-time thermal status data stream.

[0029] The real-time thermal status data stream is subjected to time series rearrangement and spatial aggregation to eliminate the impact of asynchronous sampling of multi-source data and form a unified spatiotemporal calibration data sequence. Specifically, the time series rearrangement and spatial aggregation are as follows: based on the acquisition time identifier carried by each data record in the real-time thermal status data stream, the data from different acquisition nodes are aligned to a unified time axis, and the data in adjacent time windows are mapped to the same time segment. Then, the data with time offset is replaced by proximity processing so that the data of each node forms a complete sequence in a unified time segment. After the time alignment is completed, the data of the same battery cluster, adjacent batteries and their corresponding cooling units are divided into the same spatial set, and the data in the spatial set are rearranged according to the spatial topology order so that the data arrangement order is consistent with the actual physical location. Furthermore, the boundary node data is merged into the adjacent set according to the spatial adjacency relationship to form a continuously spatially distributed data group. Finally, the spatiotemporal calibration data sequence is output under a unified time scale and arranged in a spatial topology order.

[0030] Edge-side self-evolution segmentation is performed on the unified spatiotemporal calibration data sequence, and the data is dynamically segmented according to the thermal change driving trend to form multi-level thermal evolution fragment clusters;

[0031] Multidimensional feature decoupling extraction is performed on each thermal evolution fragment cluster to construct a combined feature set of temperature gradient features, trend features, and cross-node correlation features, thereby forming a fragment feature expression unit. Specifically, the multidimensional feature decoupling extraction involves: expanding the data within each thermal evolution fragment cluster according to a time series; extracting the temperature difference between adjacent sampling points to form a temperature change sequence; calculating the change direction sequence based on continuous differences; performing a horizontal comparison of the temperature data of each node within the same fragment to obtain the temperature difference between different nodes in the same time segment, forming a spatial gradient distribution sequence; and performing cross-combination on the data of each node based on the time change sequence and the spatial gradient distribution sequence to separate and represent time changes and spatial distribution, obtaining independent trend features and gradient features. Furthermore, the change sequences of different nodes within the fragment are synchronized and aligned to extract data groups with consistent change directions, forming... The process involves generating cross-node correlation feature sequences, numbering and organizing each type of feature sequence, and encapsulating them according to segment boundaries to form corresponding segment feature expression units. Specifically, the combined feature set is constructed by aligning the extracted temperature gradient features, trend features, and cross-node correlation features according to a unified time segment, establishing correspondences for each type of feature within the same segment, arranging the temperature gradient features according to spatial order, and concatenating the trend features according to chronological order to form a trend sequence describing the temperature evolution process. Furthermore, the cross-node correlation features are grouped and identified based on the correlation between nodes to form a sequence of inter-node linkage relationships. Finally, the corresponding gradient sequences, trend sequences, and correlation sequences within the same time segment are bound together to generate a unified data structure unit, and a segment identifier is added to each data structure unit to form a corresponding segment feature expression unit.

[0032] Based on the segment feature expression unit, cross-node thermal behavior association weaving is performed to form a thermal behavior association chain structure. Specifically, the association identifiers in each segment feature expression unit are extracted, and the association features in different segments are matched. Feature sequences with the same or adjacent node relationships are aligned and connected. Within the same time segment, the change trend features of multiple nodes are synchronously compared. Node sequences with the same change direction or similar change magnitude are connected in series to form continuous node association segments. The association segments across time segments are extended and connected in chronological order. Node sequences that maintain continuous change relationships in the preceding and following segments are spliced ​​together. During the connection process, the spatial position and chronological order identifiers of each node are preserved. Finally, through the hierarchical concatenation of multiple association sequences, a thermal behavior association chain structure covering multiple nodes and multiple time segments is formed.

[0033] Perform local response projection on the thermal behavior correlation chain structure to map the thermal change relationship in the correlation chain into a propagable local thermal response field, generating a regional thermal response mapping volume;

[0034] Boundary-guided fusion is performed on the thermal response maps of each region to form a continuous initial multi-node temperature control map. Specifically, the spatial coverage and boundary node set of the thermal response maps of each region are extracted, the contact boundary between adjacent regions is identified, and the local thermal response data of different regions are aligned at the boundary nodes. The response values ​​at the same location are matched accordingly. For response data with differences at the boundary locations, adjacent node data are selected according to spatial adjacency and interpolated to complete the data. Then, the response changes of adjacent regions are gradually connected to eliminate abrupt changes. After the boundary transition is completed, the data inside each region and the boundary fusion data are uniformly stitched together and rearranged according to the overall spatial topology. At the same time, the regional identifiers of the fused data are updated to form an initial multi-node temperature control map that is continuously distributed in space.

[0035] An edge-side rolling refresh is performed on the initial multi-node temperature control map to output an optimized multi-node temperature control map that can be collaboratively scheduled and invoked. Specifically, the newly collected thermal state data is mapped to the target node position in the temperature control map, and the updated node data is replaced point by point to form the latest time segment. Then, the data of the latest time segment is written to the end of the original data sequence, and the data of the earliest time segment is removed simultaneously. During the replacement process, the local nodes that have changed are updated, and the time order of the updated node data is reorganized to maintain the continuous arrangement of the data of each node in the time dimension. For node updates involving cross-regions, the data positions of adjacent nodes are adjusted synchronously according to their spatial adjacency relationship. After the local update is completed, the updated nodes and the unupdated nodes are re-stitched according to the original spatial topology to form a new multi-node temperature control map data structure. Then, this data structure overwrites the original temperature control map to complete one rolling refresh.

[0036] In this embodiment, edge-side self-evolutionary segmentation is performed on the unified spatiotemporal calibration data sequence. The data is dynamically segmented according to the thermal change-driven trend to form multi-level thermal evolution fragment clusters, as detailed below:

[0037] The unified spatiotemporal calibration data sequence is expanded point by point in chronological order, and the temperature change amplitude and direction between adjacent data points are extracted to form a continuous change trajectory sequence.

[0038] Based on the change trajectory sequence, the data that change continuously in the same direction are marked with intervals, and the locations where the change direction is reversed or the change rate changes abruptly are marked with boundaries to form the first segment candidate point set;

[0039] Based on the first segmented candidate point set, a sliding window scan is performed on adjacent candidate intervals to extract peak segments, valley segments, and plateau segments within the intervals. The candidate intervals are then further subdivided according to morphological changes to form a second segmented candidate interval. Specifically, the sliding window scan involves setting a fixed-length scanning window along the time sequence, using adjacent candidate intervals as the range. Data within each window is traversed segment by segment, and local extreme points and inflection points are extracted within each window. The location distribution of temperature rise segments, fall segments, and relatively stable segments is recorded. Then, as the window moves forward along the time axis, the data entering new windows is processed in the same way, and data in overlapping window regions is continuously tracked to maintain the feature continuity between adjacent windows. During the scan, the identified change segments within each window are marked and concatenated according to their order of appearance to form a local morphological data set covering the entire candidate interval. The morphological distribution sequence is processed by scanning all data within the interval when the window slides to the end of the candidate interval, and outputs the corresponding morphological feature marking results. The refinement segmentation is as follows: based on the morphological feature marking results obtained by the sliding window scan, the rising segment, falling segment, and stationary segment are used as the basic division units, and the connection position of adjacent morphological segments is identified. The position from rising to stationary or from stationary to falling is marked as the refinement boundary point. The consecutive morphological segments of the same type are merged to form sub-intervals with consistent structure. When a local reverse change occurs within a single morphological segment, a new boundary point is inserted at the change position. Furthermore, the interval with frequent morphological changes is split into segments according to the change order, and different change stages are assigned to independent sub-intervals. After the division is completed, each sub-interval is renumbered in chronological order, and the corresponding data is collected to form a refined candidate interval set.

[0040] The candidate intervals for the second segment are filtered based on interval length and change density. Intervals with continuous and complete change processes are retained, while intervals with scattered changes are merged to form a primary segment sequence. Specifically, the time span covered by each candidate interval for the second segment is counted, and the number of inflection points and the number of changes in direction within the interval are simultaneously counted to form a change density identifier. Then, the candidate intervals are sorted from longest to shortest time span, and the intervals are classified based on the change density identifier. Intervals with shorter time spans and lower change densities are merged into adjacent, continuous candidate intervals. Intervals with longer time spans but uneven internal change density distribution are split at locations with dense change based on the distribution of inflection points. Intervals with high change density and continuous distribution are retained, maintaining their original boundaries. After the filtering is completed, the retained intervals and the adjusted intervals are rearranged in chronological order, and their boundary positions are uniformly marked to form a set of filtered candidate intervals.

[0041] Cross-segment trajectory backtracking is performed on the primary segment sequence. The change paths of adjacent segments are compared segment by segment in chronological order. Segments with continuously extending change paths are spliced ​​together, and segments with multi-stage changes are further segmented. Specifically, the cross-segment trajectory backtracking process involves: traversing each segment in the primary segment sequence chronologically, extracting the change trajectory features at the end of each segment, and comparing them with the change trajectories at the starting positions of subsequent adjacent segments. Then, the change direction, magnitude, and continuity relationship between adjacent segments are matched to identify the connection relationship between preceding and following segments. When adjacent segments exhibit continuous change at the boundary, the end trajectory of the previous segment is extended into the subsequent segment, and its change path is continuously tracked. For segment boundaries with interruptions or abrupt changes, the process backtracks to the corresponding position within the previous segment, relocates the change inflection point, and adjusts the segment division range accordingly. During the backtracking process, the changes in each segment are recorded. The trajectory connection relationship between segments is identified, and segments forming continuous changing paths are marked, ultimately forming a trajectory backtracking result sequence containing the connection relationship between segments. The specific process of splicing segments with continuous extension of changing paths is as follows: adjacent segments that are determined to be continuously extended after trajectory backtracking are sequentially combined, the end data sequence of the previous segment and the start data sequence of the next segment are extracted, and the two data segments are aligned at the boundary position. Then, according to the alignment result, the data sequence of the next segment is directly appended to the end of the previous segment in chronological order. During the splicing process, data with duplicate time markers at the boundary are deduplicated, and only one set of data records is retained. For time points with intervals at the boundary, the data of the interval interval is inserted into the splicing position in chronological order. After the splicing is completed, a unified time interval marker is regenerated for the merged new segment, and its corresponding segment number is updated. Finally, the extended segment sequence formed by merging multiple continuous segments is output.

[0042] After completing the segment splicing and re-segmentation, the data within each segment is reorganized in chronological order, and the segment boundary positions are aligned with the neighborhood. The data points at the boundary are adjusted to positions where the direction of change is consistent and the magnitude of change is significantly higher than that of the adjacent intervals, thus forming stable boundary points.

[0043] The stability boundary points are sequentially numbered, and the data within the stability boundary points are collected and packaged according to time intervals to form a multi-level thermal evolution fragment cluster with a hierarchical structure.

[0044] In this embodiment, local response projection is performed on the thermal behavior correlation chain structure to map the thermal change relationships in the correlation chain into a propagable local thermal response field, generating a regional thermal response mapping volume, as follows:

[0045] The thermal behavior association chain structure is divided into layers according to the time dimension, and a corresponding layer identifier is assigned to each time layer to form a hierarchical chain structure. Specifically, based on the time identifier corresponding to each node in the thermal behavior association chain structure, the entire association chain is continuously expanded in chronological order, and the overall time range is determined. Within this time range, the time axis is divided at equal intervals to form multiple continuous time intervals. Then, each node in the association chain is mapped to the corresponding time interval according to its time identifier. For nodes located at the boundary of the time interval, they are assigned to intervals with closer time spans based on their time position and the distance relationship between adjacent intervals. After completing the node division, the nodes within the same time interval are sorted in order, keeping the association order between nodes unchanged, thus forming a chain structure layered according to the time dimension.

[0046] Based on the hierarchical chain structure, node fragments with continuous temporal order within the same level are grouped together to form a hierarchical response fragment set.

[0047] Spatial mapping is performed on the hierarchical response fragment set to bind the spatial positions of nodes in the fragments to their corresponding regions, so that each response fragment corresponds to a preset region unit in space. Specifically, based on the hierarchical response fragment set, the spatial positions of each node are converted into a unified coordinate representation, and the corresponding spatial regions are determined. Then, for each response fragment, according to the spatial regions of the internal nodes, the fragment data is mapped one by one to the corresponding region unit position, and the nodes are arranged according to their spatial coordinates within the region unit. Furthermore, for multiple node data in the same spatial region, they are sorted according to the relative spatial position relationship between the nodes. For node data that crosses the region boundary, according to their spatial adjacency relationship, they are simultaneously mapped to the boundary position of the adjacent region, and the corresponding spatial identifier is retained at the boundary. After completing the mapping from node to region, the time identifier in the fragment is bound to the spatial position to form a mapping result indexed by the spatial region.

[0048] After completing the hierarchical mapping, the response distribution in the next level of the adjacent time level is connected to the corresponding area of ​​the previous level one by one according to the node position, and the node position and time marker are kept in a continuous arrangement during the splicing process to form a cross-level continuous response sequence.

[0049] For cross-level continuous response sequences, diffusion connection paths between nodes are constructed along spatial adjacency relationships, and response changes are propagated from the source node to its adjacent nodes level by level to form a spatially extended response distribution. Specifically, based on the cross-level continuous response sequence, spatial adjacency relationships between nodes are extracted, and the set of adjacent nodes is identified for each node. An initial connection relationship is established with the node as the starting point. Then, according to the temporal order of the nodes in the continuous response sequence, the connection relationships between adjacent nodes are oriented and arranged. Local connection paths are established first for nodes in the same region, and cross-regional connection relationships are established for nodes across regional boundaries according to spatial adjacency relationships. During the path establishment process, each connection relationship is bound to the corresponding time identifier, and the connection relationships of all nodes are sequentially connected to form a set of diffusion connection paths covering multiple nodes and multiple regions.

[0050] The response data within the same spatial region in the spatially extended response distribution are summarized and organized in chronological order, and the response data from different time levels are sequentially spliced ​​together to form a continuous response sequence covering a single region.

[0051] The continuous response sequences are uniformly numbered and spatially arranged, and the relative positional relationships between the regions are reorganized according to the overall spatial topology to form a regional thermal response mapping body.

[0052] S2, based on the multi-node temperature control graph, performs cross-node cooling resource collaborative allocation to form a hierarchical cooling scheduling sequence;

[0053] In this embodiment, based on the multi-node temperature control graph, cross-node cooling resource collaborative allocation is performed to form a hierarchical cooling scheduling sequence, as follows:

[0054] Based on the temperature distribution and spatial topology of each node in the multi-node temperature control diagram, the temperature differences and distribution structure between nodes are extracted to form a node thermal state sequence. Specifically, the nodes are sorted by adjacency to form a node adjacency sequence. In the node adjacency sequence, the temperature values ​​of adjacent nodes are compared one by one to calculate the temperature difference between nodes and record the direction of the difference to form a node temperature difference sequence. Then, nodes in the same area are compared laterally to obtain the temperature differences between multiple nodes and arranged according to spatial location. Furthermore, for nodes that cross the boundary of an area, the temperature differences between boundary nodes are extracted based on their spatial adjacency relationship and included in the overall difference sequence. After the node difference extraction is completed, the temperature difference sequence is bound to the node spatial location and organized according to the spatial topology order to form structured data reflecting the temperature differences and spatial distribution relationship between nodes.

[0055] A thermal potential gradient expansion process is performed on the node thermal state sequence to form a node thermal potential gradient map. Specifically, based on the temperature values ​​and corresponding spatial positions of each node in the node thermal state sequence, the nodes are sorted by adjacency according to spatial topological relationships. In the adjacency sequence, the temperature difference between adjacent nodes is calculated pairwise. Then, the direction of difference change is marked according to the spatial connection direction. Starting from the node with higher temperature, adjacent nodes are connected step by step along the temperature decreasing direction. The gradient relationships between multiple nodes are connected in series to form a continuous gradient path. Furthermore, multiple gradient paths are expanded and arranged according to their spatial positions. Adjacent or intersecting paths are combined side by side. During the path expansion process, the temperature value and gradient direction of each node are bound to the path position and rearranged according to the spatial distribution order to finally form a node thermal potential gradient map.

[0056] Based on the node thermal gradient map, hotspot migration trajectory recognition processing is performed to form a set of thermal migration trajectories. Specifically, the nodes with higher temperatures and at the starting position of the gradient in the node thermal gradient map are extracted as hotspot starting nodes, and their spatial position and time identifier are recorded. Then, the adjacent nodes of the hotspot starting node are traversed step by step along the thermal gradient direction. Nodes with the same temperature change direction and continuous gradient are selected and connected. During the traversal, the subsequent adjacent nodes of each node are filtered, and nodes that satisfy the gradient continuity relationship are added to the current path in sequence and arranged in chronological order. When a node has multiple extension directions, independent path branches are established for each direction and the path branches are numbered and identified. The formed path branches are extended according to the time dimension, and the paths that maintain a continuous change relationship in adjacent time segments are spliced ​​to form a continuous migration path across time. After the path is generated, the spatial position of the nodes in each path is sorted in order, and the temporal order relationship of the nodes in the path is preserved. Finally, a set of thermal migration trajectories covering multiple nodes and multiple time segments is formed.

[0057] The set of thermal potential migration trajectories is divided into node affiliations, and nodes located on the same migration path are grouped to form path-related node clusters;

[0058] Perform region mapping and reorganization processing on the path-related node clusters, regroup the nodes that overlap across paths into several cooling response blocks, and generate a block association index structure;

[0059] The load mapping decomposition process is performed based on the cooling response block. The thermal load of the nodes in the block is expanded according to the spatial location to form a block cooling demand distribution map. Specifically, the temperature distribution and spatial location of each node in the cooling response block are obtained and expanded and arranged according to the spatial distribution order of the nodes in the block to form a block node sequence. Then, the node temperature change of each node in the block node sequence and the temperature difference of adjacent nodes are calculated to generate a load difference sequence between nodes. Further, based on the load difference sequence between nodes, nodes with similar temperature characteristics and spatial adjacency are grouped into sub-region sets, and the node load in each sub-region set is split node by node. The node load is mapped to the corresponding sub-region position according to its spatial coordinates to form a sub-region load distribution. After the sub-region division is completed, the load distribution of each sub-region is uniformly numbered and rearranged according to the spatial topology order. Finally, all sub-region load distributions are summarized according to the block boundary to generate a block cooling demand distribution map that is continuously expanded in space.

[0060] The block cooling demand distribution map is sorted chronologically, and the demand changes of nodes within each block are arranged in time to form a block demand evolution sequence.

[0061] Perform cross-block linkage orchestration processing on the block demand evolution sequence, and connect the demand changes between adjacent blocks according to spatial adjacency to form a block collaborative scheduling chain;

[0062] Based on the block-based collaborative scheduling chain, a hierarchical cooling resource mapping and sequence reorganization process is performed to form a hierarchical cooling scheduling sequence. Specifically, the cooling demand distribution map corresponding to each block is extracted according to the arrangement order of each block in the block-based collaborative scheduling chain, and then expanded sequentially according to the order of the blocks in the scheduling chain. Then, nodes at similar spatial levels are grouped into the same resource level to form a hierarchical node set within the block. Further, the hierarchical node set is arranged in chronological order and bound to the corresponding block position. After completing the hierarchical mapping within each block, the node sets of the same level in adjacent blocks are connected in series according to the scheduling chain order to form a cross-block hierarchical resource sequence. Finally, all hierarchical resource sequences are rearranged to output a hierarchical cooling scheduling sequence arranged in spatial hierarchy and chronological order.

[0063] In this embodiment, a region mapping and reorganization process is performed on the path-associated node clusters to regroup nodes that overlap across paths into several cooling response blocks and generate a block association index structure, as follows:

[0064] Extract the spatial coordinates, migration path, and time series of each node in the path-related node cluster, and mark the overlapping nodes that exist in multiple migration paths simultaneously;

[0065] Dynamic path stripping and assignment optimization are performed on overlapping nodes. Nodes are temporarily stripped from their original paths and reassigned to a defined set of target destination paths. Specifically, the position sequence and corresponding time order of the marked overlapping nodes in each migration path are extracted. The predecessor and successor nodes of each node in different paths are recorded. The adjacent nodes of the overlapping nodes in each path are compared to obtain their spatial adjacency relationships and thermal potential change directions, forming a multi-path adjacency set. Then, based on each adjacency set, the connection relationships of the overlapping nodes in different paths are split, and their... The path connection with the node that does not meet the spatial continuity or thermal potential continuity relationship is disconnected, forming multiple candidate attribution relationships. Further, each candidate attribution relationship is sorted according to the extension direction of the node in the path, and the path that best fits its spatial position and has a continuous connection relationship is selected as the target attribution path. After the attribution path is determined, the overlapping nodes are stripped from the other paths and re-inserted into the corresponding position of the target path. At the same time, the connection order of the nodes before and after is adjusted. After the insertion is completed, the time series of the node in the path is rearranged and its path identifier is updated to form a node path structure with clear attribution.

[0066] A set of spatially continuous regions with consistent thermal migration directions is divided into unit segments, and their path source and time series information are retained to form initial region segments. The criteria for spatial continuity and consistent thermal migration directions are as follows: the spatial coordinates of each node in the region set are determined by adjacency relationship. Nodes that are directly connected or adjacent in spatial topology are considered as continuous nodes, and a continuous node sequence is formed. In the continuous node sequence, the thermal migration direction corresponding to each node is extracted. The migration directions of adjacent nodes are compared one by one. Nodes with the same migration direction or on the same extension trend are determined as nodes with consistent direction. Nodes that simultaneously satisfy spatial adjacency relationship and consistent migration direction are merged and assigned to the same continuous region set. For nodes at the boundary position, based on their spatial distance from adjacent nodes and the matching of migration directions, they are assigned to a region set of nodes with small migration direction deviation and consistent with adjacent nodes. After the determination is completed, the nodes in each continuous region set are arranged in spatial order, and this is used as the basic criterion for dividing unit segments.

[0067] The initial region segments undergo cross-path merging to identify segments belonging to different migration paths but spatially highly overlapping or adjacent, and connect the segments sequentially to form a complete cooling response block. Specifically, the migration path identifiers and spatial location ranges are extracted for each segment in the initial region, and segments under different paths are compared pairwise to obtain the spatially overlapping areas and adjacent boundary nodes between segments. For segments with spatial overlap or adjacent boundaries, the thermal potential migration direction and time sequence of the corresponding nodes are extracted. Segments with continuous migration directions and sequential connection are marked as mergeable segments. For segments marked as mergeable, the merging order is determined based on their spatial position relationship. The segment in the preceding position is used as the reference segment, and the node sequence of subsequent segments is appended to the reference segment in spatial order. During the merging process, the nodes in the overlapping area are deduplicated, and the boundary nodes are aligned. After the segment merging is completed, the path identifiers of the merging result are updated, and the original segment source information is retained to form a cooling response block.

[0068] The cooling response blocks are subjected to hierarchical structure optimization, and the nodes in each block and its sub-levels are uniformly numbered to generate a block index table. Specifically, the spatial location of each node in each cooling response block and its corresponding thermal potential change sequence are extracted. The nodes are initially sorted according to their spatial distribution relationship within the block to form a block node sequence. Then, based on the block node sequence, nodes with similar change trends and spatial adjacency are divided into the same subset. According to the spatial location of each subset in the block, the set of nodes in the central region of the block is divided into a higher-level set, and the set of nodes distributed along the spatial boundary is divided into a lower-level set. Furthermore, for nodes at the boundary of different subsets, according to their spatial adjacency relationship with adjacent sets and the continuity of thermal potential change, they are assigned to the subset with a more obvious continuity relationship. After the subset division is completed, each subset is hierarchically sorted according to its spatial location from the inside to the outside or from high temperature to low temperature, and a corresponding hierarchical identifier is assigned to each level to form the hierarchical structure division result within the block.

[0069] The cooling response block is linked with adjacent blocks or blocks with continuous thermal gradients. The path connection order between blocks is recorded, and the final cooling response block and block association index structure is output. Specifically, the spatial coverage area and boundary node set of each cooling response block are extracted. Block pairs that are in contact or adjacent in space are identified, and the adjacency relationship between blocks is established. Then, in the adjacent block pairs, the thermal change sequence of the boundary nodes is extracted. The nodes at the boundary of adjacent blocks are compared point by point. Block pairs with consistent thermal change direction and continuous change order are marked as connectable blocks. Then, based on the connectable blocks, the blocks are sorted according to their spatial position. The block in the preceding position is taken as the starting block, and subsequent blocks are connected level by level along the spatial adjacency direction. During the connection process, the boundary nodes of each block are aligned according to the corresponding relationship, and the connection order between blocks is recorded. Finally, a block association chain structure is formed by connecting multiple blocks in series according to spatial adjacency and thermal continuity relationship.

[0070] In this embodiment, cross-block linkage orchestration processing is performed on the block demand evolution sequence, connecting the demand changes between adjacent blocks according to spatial adjacency to form a block collaborative scheduling chain, as detailed below:

[0071] The demand changes of each block in the block demand evolution sequence are unfolded in chronological order, and the magnitude and direction of demand changes of each block in each time segment are extracted to form the block demand change trajectory.

[0072] Based on the demand change trajectory of blocks, the demand fluctuation structure within each block is identified and identified in chronological order to form a block fluctuation structure sequence. Specifically, the demand values ​​arranged in chronological order in the block demand change trajectory are traversed point by point to extract the change amplitude and direction between adjacent time segments. Data points at the intersection of different change directions are marked as boundaries. The position where the change direction changes is switched is used as a segmentation node, and multiple continuous fluctuation intervals are divided accordingly. After the interval division is completed, each fluctuation interval is numbered according to its chronological order in the time series, and each interval is given a corresponding change direction identifier and time range identifier. Finally, all fluctuation intervals arranged in chronological order are connected in series to form a block fluctuation structure sequence.

[0073] The process involves pairwise comparison of the fluctuation structure sequences between different blocks, establishing connections between structural segments with similar change amplitudes and adjacent times to form cross-block fluctuation matching pairs. Specifically, based on the block fluctuation structure sequence, structural segments between adjacent blocks are paired segment by segment, and candidate comparison relationships are established according to chronological order. In the candidate comparison relationships, the change amplitudes of the two structural segments are compared point by point to obtain their amplitude difference sequence, and structural segment pairs with the same change direction are retained. Furthermore, the time intervals of the structural segments are compared, and structural segments with adjacent or partially overlapping time intervals are marked as time-connectable structural segments. Structural segment pairs that meet the conditions of similar change amplitudes and time interval connections are numbered, and their block and time sequence positions are recorded. Finally, all structural segment pairs that meet the conditions are organized to form cross-block fluctuation matching pairs.

[0074] Based on cross-block fluctuation matching pairs, extract the set of spatial boundary nodes between adjacent blocks, and align the demand change sequences of the boundary nodes within their respective blocks to generate boundary node change trajectories.

[0075] Node pairs with continuous change direction and sequential change in the boundary node change trajectory are marked as bridging nodes. Using bridging nodes as the connection benchmark, the demand sequences of the corresponding blocks are locally spliced ​​to form cross-block connection segments. Specifically, the corresponding position of the bridging node in the adjacent blocks is used as the benchmark point. The time position of the node in the demand sequences of the two blocks and its adjacent node sequences are extracted respectively. The time segments corresponding to the bridging nodes in the two blocks are aligned. Segments with adjacent or overlapping time positions are mapped to the same time sequence position. In the aligned sequence, the demand sequence in the previous block before the bridging node is retained, and the demand sequence in the next block after the bridging node is connected in chronological order, with the bridging node as the boundary. Nodes with duplicate time segments in the splicing process are deduplicated, and the node data at the same time position is retained in order. At the splicing boundary, the adjacent nodes on both sides of the bridging node are aligned, and finally, a cross-block connection segment with the bridging node as the connection point is formed.

[0076] Based on cross-block connection segments, blocks with spatial adjacency in the already connected blocks are sequentially included, and their demand sequences are appended to the end of the connection segment according to the adjacency order, forming a continuously expanding spatial linkage sequence. Specifically, based on the formed cross-block connection segments, the set of spatial boundary nodes of the end blocks is extracted, and candidate blocks adjacent to the boundary nodes are found. Then, blocks in the candidate blocks that are directly in contact with or adjacent to the current end block are retained, and their corresponding demand sequences and time identifiers are obtained. Further, the demand sequences of the candidate blocks are compared with the time sequences of the end of the current connection segment to determine blocks that are adjacent or connectable in time sequence, and the candidate blocks are sorted in time sequence. According to the sorting results, target blocks are selected in sequence, and the part of their demand sequences that connects with the time sequence of the end of the current connection segment is extracted and appended to the end of the connection segment in time sequence. Then, the end blocks of the connection segment are updated, and the above process is repeated until no new adjacent blocks can be included.

[0077] The spatial linkage sequence is organized according to spatial location and time order. The demand change trajectory of blocks within the path is sequentially spliced ​​together, and the fluctuation structure matching relationship and bridging node identifier are preserved, ultimately forming a block collaborative scheduling chain containing multi-path structure.

[0078] S3 performs spatiotemporal orchestration and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node, driving the cooling units to coordinate their actions and forming a set of state offsets.

[0079] In this embodiment, the hierarchical cooling scheduling sequence is spatiotemporally orchestrated and trajectory constrained at each edge node to drive the cooling units to coordinate their actions and form a set of state offsets, as detailed below:

[0080] The process involves obtaining the execution segments of the hierarchical cooling scheduling sequence at each edge node and expanding them according to node identifier and time sequence to construct a node-level cooling driver primitive set. Specifically, this involves extracting the node identifier, execution time segment, and cooling action identifier corresponding to each scheduling instruction in the hierarchical cooling scheduling sequence, and sorting all scheduling instructions in chronological order. In the sorted sequence, scheduling instructions belonging to the same edge node are grouped into the same node set. Then, in each node set, the scheduling instructions are expanded one by one in chronological order, retaining the execution start position and duration interval of each instruction. Furthermore, scheduling instructions that are temporally adjacent and have continuous execution actions are sequentially connected to form continuous action segments. For instructions with time intervals or execution interruptions, segments are marked at the corresponding positions to form independent action segment units. After completing the expansion within the node, the node identifier, time sequence, and action segments are bound together to finally form a node-level cooling driver primitive set composed of multiple node action segments.

[0081] Spatiotemporal embedding rearrangement is performed on the node-level cooling driving primitive set to map each driving primitive to a unified spatiotemporal grid. Conflicting primitives are relocated within the grid to generate a spatiotemporally aligned driving field. Specifically, the spatial position of each driving primitive and its execution time interval are extracted from the node-level cooling driving primitive set. Each driving primitive is mapped to its corresponding time segment position according to its execution time interval, and simultaneously mapped to its corresponding spatial coordinate point based on its node spatial position to form an initial spatiotemporal distribution structure. Then, driving primitives in the same time segment and adjacent in spatial position that overlap in execution time and have an adjacency relationship in space are identified as conflicting primitives. The identified conflicting primitives are adjusted according to the order of their time intervals by shifting them backward along the time axis to a position that does not overlap with the preceding primitives. For primitives that overlap in space or have intersecting paths, their positions are offset according to the spatial adjacency relationship between nodes. After completing the time and spatial position adjustment, all driving primitives are reordered in chronological order to form a rearranged spatiotemporal embedding structure.

[0082] Based on a spatiotemporally aligned driving field, the propagation path of cooling action is extracted, and the path is continuously extended and spatially folded to construct a cross-node cooling conduction trajectory field. Specifically, in the spatiotemporally aligned driving field, the spatial position and temporal sequence of each driving primitive are extracted. Taking the primitive that triggers the cooling action as the starting point of the path, adjacent driving primitives are searched step by step along the adjacency direction between nodes. Primitives that are temporally continuous and spatially adjacent are selected for connection. During the connection process, only the temporally consecutive connection relationship is retained. When there are multiple connectable directions, independent branch paths are established and extended according to the spatial extension direction. Finally, the driving primitives in each path are arranged in chronological order to form a cooling action field that covers multiple nodes and has temporal continuity. Using the propagation path, the extracted cooling effect propagation path is traversed segment by segment in chronological order. The driving primitive at the end of each path is extracted, and the temporal position of the candidate primitives is compared. Primitives that are temporally adjacent and spatially continuous are selected to be connected to the current path. Then, branches are established for paths that extend in multiple directions, and they are continuously extended in the spatial extension direction to form a multi-segment continuous path structure. After the path extension is completed, the path segments with intervals or cross-regional distribution are remapped according to their spatial adjacency relationship. The path segments that are in non-continuous spatial positions but have a connection relationship are mapped to the same continuous coordinate sequence. The mapped path nodes are arranged in a new spatial order while keeping the original temporal order unchanged, and finally a cross-node cooling conduction trajectory field with continuous cross-node distribution is formed.

[0083] Constraint domain embedding processing is performed on the cross-node cooling conduction trajectory field. Path blocking areas and priority passage areas are introduced into the trajectory, and the trajectory is trimmed and recombined to generate a constrained scheduling trajectory field.

[0084] Based on the constrained scheduling trajectory domain, trajectory nodes are mapped to corresponding cooling execution actions, and the cooling execution actions are hierarchically embedded and arranged to form a node action-driven graph. Specifically, the spatial location, time identifier, and trajectory connection relationship of each trajectory node in the constrained scheduling trajectory domain are extracted, and the corresponding cooling execution action identifier is matched for each trajectory node. During the matching process, the time sequence of the trajectory nodes is aligned with the execution sequence of the cooling actions, and the corresponding cooling execution actions of consecutive nodes on the same trajectory path are linked together in time sequence to form an action chain. Furthermore, the action chains in all trajectory paths are summarized according to their spatial distribution, and the actions in the core spatial region are divided into high-level actions, and the actions in the peripheral region are divided into low-level actions. After the hierarchical division is completed, the cooling execution actions in each level are arranged in time sequence, and the actions of different levels are nested and organized in order from the inside to the outside or from the top to the bottom, finally forming a cooling execution action arrangement structure with a spatial hierarchy.

[0085] The action paths of spatially adjacent nodes in the node action-driven graph are interwoven and combined, and temporally adjacent actions are synchronized and aligned to generate a collaborative driving structure. Specifically, the action paths corresponding to each node in the node action-driven graph are extracted, and the action paths of adjacent nodes are paired. In the paired paths, the action segments of each path are extracted in chronological order. Action segments that are temporally adjacent or have overlapping intervals are aligned. Within the aligned time interval, the action segments of different nodes are alternately inserted in chronological order to form a temporally interleaved action sequence. For action segments with continuous execution relationships, they are interspersed and connected according to their spatial position. During the interweaving process, duplicate actions of repeated time segments are removed, and the order of action segments at the boundary is adjusted. Finally, the action paths of multiple adjacent nodes are merged into a composite action path structure with interleaved connections.

[0086] The collaborative driving structure drives the cooling units of each edge node to perform actions, and records the node state changes in real time during the action execution process to generate a node state evolution diagram.

[0087] Spatial reclamation and temporal folding are performed on the node state evolution diagram to form a state offset set. Specifically, the state data of each node in the node state evolution diagram at different time segments are extracted and unfolded in chronological order to form a node state sequence. In the node state sequence, all state data are classified according to the spatial location of the nodes, and the states of nodes belonging to the same spatial region or adjacent regions are grouped together. Then, the grouped node state sequence is rearranged according to spatial topological order, and the scattered node states are remapped to a continuous spatial location sequence. After spatial aggregation, state segments that are continuous in time and have consistent change trends are merged, and time segments with repetition or small changes are removed, retaining only key time nodes with significant changes. During the folding process, the temporal sequence information of each node is rebound with its spatial location. Finally, the node states after spatial reclamation and temporal folding are organized in a unified manner according to spatial topological order and time sequence to form a state offset set.

[0088] In this embodiment, constraint domain embedding processing is performed on the cross-node cooling conduction trajectory field. Path blocking regions and priority passage regions are introduced into the trajectory, and the trajectory is pruned and recombined to generate a constrained scheduling trajectory domain, as detailed below:

[0089] Structural analysis is performed on each trajectory path in the cross-node cooling conduction trajectory field to extract the spatial position, temporal sequence and path connection relationship of each node in the path, and the complete path is decomposed into multiple continuous trajectory segments according to the node connection order.

[0090] The system identifies trajectory segments that have spatial overlap, path intersection, or repeated coverage among multiple continuous trajectory segments, and marks these segments as a set of conflict segments. At the same time, trajectory segments that are spatially continuous and have a single connection relationship are marked as a set of candidate passable segments.

[0091] The trajectory segments in the conflict segment set are removed from the path and their connection positions in the original path are recorded. At the same time, the segment combinations in the candidate passable segment set that can form a continuous path structure are retained.

[0092] Based on the combination of segments with continuous path structure, each trajectory segment is used as a connection unit in the graph, and connection edges are established according to the spatial adjacency relationship between segments to form a trajectory segment topology graph.

[0093] Based on the topology diagram of the trajectory segment, the connecting edges are traversed one by one to identify the connecting edges that have cross connections, loop structures, or multiple path overlaps. The corresponding connecting areas are defined as path blocking areas, while the connecting path areas that maintain unidirectional extension and have no intersections are defined as priority passage areas.

[0094] Based on path blocking zones and priority passage zones, path filtering is performed on the trajectory segment topology graph, retaining connecting paths located within priority passage zones to form a pre-filtered path set. Specifically, based on the trajectory segment topology graph, firstly, all path segments corresponding to connecting edges and their associated node sequences are extracted. Then, based on the labeling results of path blocking zones and priority passage zones, the region affiliation of each connecting edge is determined. Subsequently, a removal operation is performed on connecting edges located within path blocking zones, disconnecting their corresponding path segments from the topology, and simultaneously recording the location of the disconnection. After removing the obstruction zone path, the break points in the path caused by the removal operation are identified, and the unbroken segments before and after the break points are retained as candidate connection units. Furthermore, the connection edges located in the priority passage area are traversed segment by segment according to the spatial adjacency order, and the path segments that can be connected end to end are sequentially spliced ​​together, while maintaining the original time order of the nodes during the splicing process. After the path splicing is completed, the paths that still contain obstruction zone nodes or have duplicate nodes are removed, and finally, the set of trajectory paths that are continuous and composed of priority passage area connection edges is retained.

[0095] Based on the initial set of paths, multiple candidate paths with the same start node or the same end node are compared in parallel. Path branches with interruptions or duplicate coverage are eliminated, and paths with continuous structure are retained as the preferred set of paths.

[0096] Based on the preferred path set, the paths are sequentially reorganized, and the temporal order of the path nodes is aligned during the connection process. Specifically, for each path in the preferred path set, its contained trajectory segments are extracted, and each trajectory segment is initially sorted according to time identifiers. A path sequence is established based on the starting node of the path. In the path sequence, the connecting nodes of adjacent trajectory segments are matched to determine whether the ending node of the previous segment and the starting node of the next segment satisfy the spatial topological adjacency relationship. Segments that meet the conditions are connected end to end in sequence. When multiple candidate subsequent segments appear, they are filtered according to the order of the segment's starting time and spatial connection relationship. Segments that are temporally continuous and have consistent spatial adjacency relationships are selected as the continuation path. For segments with temporal overlap or inconsistent order, their arrangement position in the path is adjusted. After completing the internal reorganization of a single path, segments with the same node or adjacent spatial positions in different paths are aligned. These segments are then synchronously arranged according to a unified time benchmark. Finally, each path segment is rearranged according to the spatial connection order and temporal order to form a path sequence with a continuous structure and consistent order.

[0097] After path reassembly, compensation and connection processing are performed on segments that still have local breaks or incomplete connections in the reassembled path. The spatial position and temporal order of each node in the path are bound together to finally generate a constrained scheduling trajectory domain that meets the constraints. Specifically, each reassembled path is scanned segment by segment to extract the locations where there are connection breaks between adjacent trajectory segments. The identifiers of the nodes before and after the break point and their spatial coordinates are recorded. Then, for each break point, a set of candidate segments that are spatially adjacent to the nodes before and after the break point is retrieved in the trajectory segment topology diagram. The start node, end node and time identifier of the candidate segments are extracted. Subsequently, the candidate segments are screened, and segments whose start nodes are spatially adjacent to the nodes before the break point and whose temporal order is close are selected as connection candidates. After the candidate segments are determined, they are inserted into the break point to bridge and connect the trajectory segments before and after the break point. The temporal order of the inserted segments is adjusted. For break points that cannot be connected by a single segment, multiple candidate segments are selected and spliced ​​step by step until a continuous path structure is formed, and finally a complete and continuous trajectory path structure is formed.

[0098] S4, perform edge-side temporal correlation backtracking and local strategy reconstruction processing on the state offset set, generate feedback control commands, and write the feedback control commands back to the multi-node temperature control map;

[0099] In this embodiment, edge-side temporal correlation backtracking and local strategy reconstruction are performed on the state offset set to generate feedback control commands, which are then written back to the multi-node temperature control map, as detailed below:

[0100] The state offset set is processed by edge-side data expansion to extract the state change records of each node under different time segments, and arranged according to node identifier and time order to form a node state evolution sequence. Specifically, the node identifier, time identifier and state change corresponding to the state data of each node in the state offset set are extracted, and all state records are sorted in chronological order. In the sorted data sequence, state records belonging to the same node are divided into the same node data set, and each node data set is expanded point by point in chronological order to form a continuous node state sequence. Then, temporally adjacent state records are connected segment by segment, and records with continuous time and consistent state change direction are combined into continuous state segments. The positions with time intervals or state abrupt changes are segmented and marked. After the expansion within the node is completed, the states of different nodes in the same time segment are arranged synchronously, and the relative change order of each node in the time segment is preserved. Furthermore, the state records of spatially adjacent nodes are horizontally organized, and finally a node state evolution sequence is formed in an orderly expansion according to node identifier, time order and spatial position.

[0101] The node state evolution sequence is reconstructed by temporal segment folding, and segments with abrupt changes or shifts are independently marked to form state folding units with different structures. Specifically, the node state evolution sequence is traversed point by point in chronological order, the magnitude and direction of change between adjacent state records are extracted, and a continuous change trajectory is formed. In this trajectory, state records with consistent change direction and continuous change magnitude are divided into intervals and marked as the same continuous segment. Boundary markings are made at positions where the change direction is reversed or the change magnitude jumps. Subsequently, the divided continuous segments are scanned segment by segment, and state records with consistent change trends within the segments are compressed and integrated in chronological order to form a single representation unit. Furthermore, for segments that are short or have scattered changes, they are merged into adjacent segments. After the segment compression is completed, all segments are rearranged in the original chronological order, and the state records at the segment boundaries are aligned to form a state folding unit composed of multiple folded segments arranged in chronological order.

[0102] Reverse trajectory backtracking is performed based on state folding units, and the spatial position and evolution order of each state node in the path are aligned and organized during the backtracking process to form a state backtracking path set;

[0103] Path aggregation is performed on the state backtracking path set to merge backtracking paths with the same evolutionary starting point or similar change trends to form a state evolution association structure. Specifically, this involves: extracting the starting node, ending node, path length, and corresponding time interval for each path in the state backtracking path set, and recording the change order of each node in the path. Then, all backtracking paths are grouped according to the starting node of the path, and paths with the same starting node are grouped into the same candidate set. Within each candidate set, the change trajectories between paths are compared pairwise, and the node change sequences in the same or adjacent time segments of the path are extracted. It is then determined whether their change direction and change order are consistent. For paths with consistent change trajectories or continuous connection relationships, they are spliced ​​and connected in chronological order to form a longer composite path structure. For paths with partial overlap, the paths are truncated and merged according to the overlap interval, retaining the overlapping part once and connecting the non-overlapping part. After the path merging is completed, the node order in the aggregated path is reorganized. Finally, the paths in each candidate set are organized into several aggregated path structures with a unified evolutionary starting point and continuous change trajectories.

[0104] Based on the state evolution association structure, local behavior mapping is performed, which converts the state change relationship in the path into corresponding regulatory behavior segments, and maps the regulatory behavior segments segment by segment to form a set of behavior mapping segments.

[0105] The behavior segments in the behavior mapping segment set that are adjacent and have continuous control relationships are combined and rearranged. Conflicting or repetitive behavior segments are screened out and replaced to form a strategy combination unit with consistent structure. Specifically, the node identifier, time interval, and control action type corresponding to each segment in the behavior mapping segment set are extracted. All behavior segments are initially arranged in chronological order. In the arranged sequence, adjacent behavior segments are compared segment by segment to identify behavior segments that are continuous in time and act on the same node or spatially adjacent nodes. It is also determined whether their control directions are consistent. Behavior segments that meet the continuous control relationship are merged and connected in chronological order to form a continuous behavior chain. Behavior segments that have slight overlap in time are adjusted to arrange them in chronological order. Behavior segments with control conflicts or repetitive actions are screened out according to their position in the time series, and only effective segments that are connected in the continuous chain are retained. After the segment merging and screening are completed, each continuous behavior chain is grouped and organized according to the spatial position of the nodes, and rearranged in chronological order within the group to finally form a strategy combination unit with continuous structure and consistent order.

[0106] Multi-path fusion is performed on the strategy combination unit, and the temporal order of each strategy is uniformly aligned during the fusion process to generate a cross-node collaborative strategy structure. Specifically, the strategy sequences from different state backtracking paths in the strategy combination unit are extracted one by one and initially arranged according to their corresponding node identifiers and temporal order. Then, based on the starting node and time position of each strategy combination unit, strategy sequences with the same starting point or overlapping temporal relationships are aligned. During the alignment process, temporally overlapping strategy segments are segmented and further, spatially adjacent strategy segments with consistent control behavior directions are further... The process involves connecting multiple policy sequences along spatial adjacency. During this process, paths with branching structures are grouped and organized, and paths with the same expansion direction are merged into the same fusion branch. Each branch is then numbered according to its chronological order. Subsequently, cross-integration is performed between different fusion branches, connecting spatially intersecting branch paths and uniformly adjusting the policy execution order at the intersection nodes. After the integration between paths is completed, the policy segments in the overall fusion path are reordered according to chronological order, ultimately forming a multi-path fusion result that is spatially continuous and temporally consistent.

[0107] Arrange the control behaviors in the cross-node collaborative strategy structure according to the node and time sequence, and attach a corresponding execution identifier to each behavior to form a feedback control instruction;

[0108] The feedback control commands are mapped back to the corresponding node positions according to the node spatial location, and written to the corresponding node data structure in the multi-node temperature control diagram in chronological order, thereby realizing the write-back update of the feedback control commands.

[0109] In this embodiment, reverse trajectory backtracking is performed based on the state folding unit, and the spatial positions and evolution order of each state node in the path are aligned and organized during the backtracking process to form a state backtracking path set, as follows:

[0110] The process involves: acquiring each folded segment from the state folding unit, constructing a time inversion index, and arranging the folded segments in reverse time order to form a set of inverted time series segments. Specifically, this involves: extracting the start time identifier, end time identifier, and time order of the state nodes within each folded segment from the state folding unit; using the end time of the folded segment as the inversion benchmark, establishing a time mapping relationship for each folded segment; renumbering the times in the time order according to their relative distance from the end time to form inverted time identifiers; traversing the state nodes within each folded segment in time order and replacing the time identifiers of each node with the corresponding inverted time identifiers while preserving the node order; further sorting all folded segments according to their inverted time identifiers; assigning a unique index identifier to each inverted time layer; and establishing a multi-level mapping relationship consisting of inverted time index, folded segment, and node identifier; binding the position of each node in the inverted time axis to its corresponding folded segment and original node information; and finally forming a time inversion index structure covering all folded segments.

[0111] Based on the time inversion index, a segment-by-segment backtracking traversal is performed on the inverted time series segments. During the traversal, the end state of a segment is used as the starting anchor point, and its predecessor state records in the node state sequence are retrieved sequentially. These records are then connected level by level according to the time inversion relationship to form an initial inverted trajectory chain. Specifically, according to the sorting results of each inverted time layer in the time inversion index, starting from the inverted time layer corresponding to the end time, the set of folded segments corresponding to the current inverted time layer is selected sequentially. Each state node in this set is used as the backtracking starting point, and the time identifier and node identifier information bound to it in the time inversion index are read. Subsequently, the corresponding node is located in the node state sequence based on the node identifier. The predecessor state is recorded, and candidate nodes that are spatially adjacent to the current node and temporally connected are extracted from the predecessor state record. After obtaining the candidate predecessor nodes, the nodes that meet the conditions are connected to the current backtracking chain one by one in reverse time order. For each connection operation, the starting node, the ending node and the corresponding reverse time identifier are recorded. After completing the predecessor retrieval and connection of all nodes in the current reverse time layer, the obtained backtracking fragment result is used as the input to continue to advance to the previous reverse time layer and repeat the retrieval and connection process until the earliest reverse time layer is traversed, thus forming a complete backtracking path fragment sequence that extends layer by layer in reverse time.

[0112] The time stamps of each node in the initial reverse trajectory chain are uniformly mapped to the reverse time axis. Under the same time anchor point, the state records belonging to different nodes are horizontally aligned to form a hierarchically aligned backtracking segment structure. Specifically, the time stamps corresponding to each node in the initial reverse trajectory chain are extracted, and the latest time stamp in the trajectory chain is used as the starting reference point for the reverse time axis. All time stamps are re-encoded. First, the node time stamps are sorted in ascending order of time, and the relative position index of each time stamp in the sequence is recorded. Then, the time stamps corresponding to the termination time are re-encoded. Using the reference time as a baseline, the original time identifier of each node is replaced with the reverse time identifier corresponding to the relative time difference between it and the reference time. The reverse time identifier is assigned a continuously decreasing time number. After the time identifier replacement is completed, nodes with the same reverse time number are merged and organized. Nodes belonging to the same reverse time layer are grouped into the same time group and sorted within the group according to spatial location or node identifier. Furthermore, a unified reverse time layer index is attached to each group, finally forming a reverse time axis structure with the termination time as the starting point and the time numbers arranged in reverse order.

[0113] Based on the hierarchical aligned backtracking fragment structure, the node state in the current time layer is matched and connected with the node state in the previous time layer that is spatially adjacent and has a continuous state evolution. In the case of multiple candidate predecessor nodes, the node is selected according to the spatial adjacency relationship, and multi-branch backtracking paths are generated layer by layer.

[0114] Backtracking paths with the same starting anchor point or the same starting time in multi-branch backtracking paths are grouped into the same set, and the node sequences are uniformly arranged in reverse chronological order within the set;

[0115] The multi-branch backtracking path, after being arranged, locally corrects the node sequences that are offset or misaligned in adjacent time layers to obtain backtracking path units with consistent structure. Specifically, based on the multi-branch backtracking path, the node sequences that are offset or misaligned between adjacent time layers are selected one by one, and the time identifier, spatial location information and preceding and following connection relationship of the corresponding nodes are extracted. A local alignment window is established with the current time layer as the benchmark. Within this window, the nodes of the previous time layer and the current time layer are compared one by one to identify node pairs that correspond in spatial adjacency but are intersected or misaligned in arrangement. For cases where the corresponding predecessor node is missing in the current time layer, candidate nodes that are spatially adjacent and temporally continuous are retrieved from adjacent time layers and inserted into the corresponding positions to complete the connection relationship. After completing the node rearrangement and connection correction, all node sequences in the local window are reorganized in chronological order, and the corrected node sequences are written back to the corresponding positions in the original backtracking path, thereby forming a path structure with consistent arrangement and continuous connection relationship between adjacent time layers.

[0116] The backtracking path units are summarized and arranged according to node identifiers and time order, and the spatial position and time level between paths are uniformly labeled to form a set of state backtracking paths with temporal reverse order consistency and spatial alignment relationship.

[0117] Example 2, Figure 2 The present invention provides a system for an intelligent operation and maintenance decision-making method based on the status of power facility assets, comprising a data mapping module, a collaborative allocation module, a scheduling control module, and a feedback optimization module, wherein the modules are interconnected.

[0118] The data mapping module is used to acquire real-time thermal state data streams, perform distributed feature segmentation and local thermal response mapping of edge nodes, and generate multi-node temperature control diagrams.

[0119] The collaborative allocation module is used to perform cross-node collaborative allocation of cooling resources based on a multi-node temperature control map, forming a hierarchical cooling scheduling sequence;

[0120] The scheduling and control module is used to perform spatiotemporal orchestration and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node, drive the cooling units to coordinate their actions, and form a set of state offsets.

[0121] The feedback optimization module is used to perform edge-side temporal correlation backtracking and local strategy reconstruction on the state offset set, generate feedback control commands, and write the feedback control commands back to the multi-node temperature control map.

[0122] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0123] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0125] The above description is merely a specific technical solution of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0126] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An edge-computing-based energy storage cooling system collaborative control method, characterized in that, include: Acquire real-time thermal status data streams, perform distributed feature segmentation and edge node local thermal response mapping, and generate multi-node temperature control maps; Based on the multi-node temperature control graph, cross-node cooling resource collaborative allocation is performed to form a hierarchical cooling scheduling sequence; The hierarchical cooling scheduling sequence is spatiotemporally orchestrated and trajectory constraint controlled at each edge node to drive the cooling units to coordinate their actions and form a set of state offsets. The state offset set is processed by edge-side temporal correlation backtracking and local strategy reconstruction to generate feedback control commands, and the feedback control commands are written back to the multi-node temperature control map.

2. The edge computing based cooperative control method of an energy storage cooling system according to claim 1, characterized in that, The process of acquiring real-time thermal state data streams and performing distributed feature segmentation and edge node local thermal response mapping to generate a multi-node temperature control map is as follows: Collect energy storage unit operation data, mark the collection identifier, and aggregate it to form a real-time thermal status data stream; Perform time series rearrangement and spatial aggregation on the real-time thermal state data stream to form a unified spatiotemporal calibration data sequence; Edge-side self-evolution segmentation is performed on the unified spatiotemporal calibration data sequence to form multi-level thermal evolution fragment clusters; Multidimensional feature decoupling extraction is performed on each thermal evolution fragment cluster to construct a combined feature set and form a fragment feature expression unit; Based on the fragment feature representation unit, cross-node thermal behavior association weaving is performed to form a thermal behavior association chain structure; Perform local response projection on the thermal behavior correlation chain structure to generate a regional thermal response mapping volume; Boundary-guided fusion is performed on the thermal response mapping volumes of each region to form an initial multi-node temperature control map; Perform edge-side scrolling refresh on the initial multi-node temperature control map and output the multi-node temperature control map.

3. The edge computing based cooperative control method of an energy storage cooling system according to claim 2, characterized in that, The process of performing edge-side self-evolutionary segmentation on the unified spatiotemporal calibration data sequence to form multi-level thermal evolution fragment clusters is as follows: The unified spatiotemporal calibration data sequence is expanded point by point in chronological order to form a continuously changing trajectory sequence; Based on the change trajectory sequence, the data that change continuously in the same direction are marked with intervals, and the locations where the change direction reverses or the change rate changes abruptly are marked with boundaries to form the first segment candidate point set; Based on the first segmented candidate point set, a sliding window scan is performed on adjacent candidate intervals to form the second segmented candidate interval; The candidate intervals for the second segment are filtered by interval length and change density to form a primary segment sequence; Perform cross-segment trajectory backtracking on the primary segment sequence and further segment the segments with multi-stage changes; The data within each segment after segmentation are reorganized in chronological order, and the segment boundary positions are aligned with the neighborhood to form stable dividing points; The stable boundary points are sequentially numbered, and the data is collected and packaged according to time intervals to form multi-level thermal evolution fragment clusters.

4. The edge computing based cooperative control method of an energy storage cooling system according to claim 1, characterized in that, Based on the multi-node temperature control graph, cross-node cooling resource collaborative allocation is performed to form a hierarchical cooling scheduling sequence, as detailed below: Based on the temperature distribution and spatial topology of each node in the multi-node temperature control diagram, a node thermal state sequence is formed; Perform thermal potential gradient expansion on the node thermal state sequence to form a node thermal potential gradient map; Based on the node thermal potential gradient map, hotspot migration trajectory identification processing is performed to form a set of thermal potential migration trajectories; The set of thermal potential migration trajectories is divided into node affiliations to form path-related node clusters; Perform region mapping and reorganization processing on the path-related node clusters and generate a block-related index structure; Based on the cooling response blocks, load mapping decomposition processing is performed to form a block cooling demand distribution map; The block cooling demand distribution map is sorted out in chronological order to form a block demand evolution sequence; Perform cross-block collaborative orchestration processing on the block demand evolution sequence to form a block collaborative scheduling chain; Based on the block-based collaborative scheduling chain, the cooling resource hierarchical mapping and sequence reorganization process is performed to form a hierarchical cooling scheduling sequence.

5. The edge computing based cooperative control method of an energy storage cooling system according to claim 4, characterized in that, The process of performing region mapping and reorganization on the path-associated node clusters and generating a block association index structure is as follows: Extract the spatial coordinates, migration path, and time series of each node in the path-related node cluster, and mark the overlapping nodes that exist in multiple migration paths simultaneously; Dynamic path stripping and attribution optimization are performed on overlapping nodes, and they are then reassigned to the region set; A set of spatially continuous regions with consistent thermal potential migration directions is divided into unit segments to form initial region segments; The initial region segments are merged across paths to form a cooling response block; Perform hierarchical structure optimization on the cooling response blocks to generate a block index table; Establish an association chain for adjacent blocks or blocks with continuous thermal gradients in the cooling response block, and output the cooling response block and block association index structure.

6. The edge computing based cooperative control method of an energy storage cooling system according to claim 1, characterized in that, The hierarchical cooling scheduling sequence is spatiotemporally orchestrated and trajectory constraint controlled at each edge node to drive the cooling units to coordinate their actions and form a set of state offsets, as detailed below: Obtain the scheduling execution fragments of the hierarchical cooling scheduling sequence at each edge node, and construct a node-level cooling driver primitive set; Spatiotemporal embedding rearrangement is performed on the node-level cooling driving primitive set to generate a spatiotemporal aligned driving field; Based on the spatiotemporal alignment driving field, the cooling effect propagation path is extracted, and the path is continuously extended and spatially folded to construct a cross-node cooling conduction trajectory field. Constraint domain embedding processing is performed on the cross-node cooling conduction trajectory field to generate a constrained scheduling trajectory field; Based on the constrained scheduling trajectory domain, trajectory nodes are mapped to corresponding cooling execution actions, and the cooling execution actions are hierarchically embedded and arranged to form a node action-driven graph; Interweave and combine the action paths of spatially adjacent nodes in the node action-driven graph to generate a collaborative driving structure. The collaborative driving structure drives the cooling units of each edge node to perform actions, generating a node state evolution diagram; Spatial reclamation and temporal folding are performed on the node state evolution graph to form a set of state offsets.

7. The edge computing based cooperative control method of an energy storage cooling system according to claim 6, characterized in that, The constraint domain embedding process is performed on the cross-node cooling conduction trajectory field to generate a constrained scheduling trajectory field, as detailed below: Structural analysis is performed on each trajectory path in the cross-node cooling conduction trajectory field, and the complete path is decomposed into multiple continuous trajectory segments according to the node connection order. The system identifies overlapping trajectory segments among multiple continuous trajectory segments and marks these segments as a set of conflicting segments. At the same time, it marks spatially continuous trajectory segments with a single connection relationship as a set of candidate passable segments. The trajectory segments in the conflict segment set are removed from the path and their connection positions in the original path are recorded. At the same time, the segment combinations in the candidate passable segment set that can form a continuous path structure are retained. Based on the combination of segments, each trajectory segment is used as a connection unit in the graph, and connection edges are established according to the spatial adjacency relationship between segments to form a trajectory segment topology graph. Based on the topology diagram of the trajectory segment, the connecting edges are traversed one by one, and the corresponding connecting areas are defined as path blocking areas. At the same time, the connecting path areas that maintain unidirectional extension and have no intersections are defined as priority passage areas. Based on path blocking areas and priority passage areas, path filtering processing is performed on the trajectory segment topology map, and connecting paths located in the priority passage area are retained to form a preliminary filtered path set. Based on the initial set of paths, path branches with interruptions and duplicate coverage are removed, and paths with continuous structure are selected as the preferred set of paths. Based on the preferred path set, the paths are reordered. Compensation and connection processing are performed on segments that still have local breaks and incomplete connections in the recombined path after reorganization to generate a restricted scheduling trajectory domain.

8. The edge computing based cooperative control method of an energy storage cooling system according to claim 1, characterized in that, The process of performing edge-side temporal correlation backtracking and local strategy reconstruction on the state offset set to generate feedback control commands, and then writing the feedback control commands back to the multi-node temperature control map, is as follows: The state offset set is processed by edge-side data expansion to form a node state evolution sequence; Perform temporal segment folding reconstruction on the node state evolution sequence, and independently mark segments with abrupt changes and shifts to form state folding units; Reverse trajectory backtracking is performed based on state folding units to form a set of state backtracking paths; Perform path aggregation on the set of state backtracking paths to form a state evolution association structure; Local behavior mapping is performed based on the state evolution association structure to form a set of behavior mapping fragments; The behavior segments that are adjacent and have a continuous regulatory relationship in the behavior mapping segment set are combined and rearranged, and conflicting and repetitive behavior segments are screened out and replaced to form a strategy combination unit. Perform multi-path fusion on the strategy combination unit to generate a cross-node collaborative strategy structure; Arrange the various control behaviors in the cross-node collaborative strategy structure according to the node and time sequence to form feedback control instructions; The feedback control commands are mapped back to the corresponding node positions according to the node spatial location, and written into the corresponding node data structure in the multi-node temperature control diagram in chronological order.

9. The edge computing based cooperative control method of an energy storage cooling system according to claim 8, characterized in that, The reverse trajectory backtracking based on the state folding unit is performed to form a set of state backtracking paths, as follows: Obtain each fold segment in the state folding unit, construct a time inversion index, and arrange each fold segment in reverse time order to form a set of reverse time series segments; Based on the time inversion index, a segment-by-segment backtracking traversal is performed on the inverted time series segments to form an initial inverted trajectory chain; The time markers of each node in the initial reverse trajectory chain are uniformly mapped to the reverse time axis to form a backtracking segment structure. Based on the backtracking fragment structure, the node state in the current time layer is matched and connected with the node state in the previous time layer that is spatially adjacent and has a continuous state evolution. In the case of multiple candidate predecessor nodes, the node is selected according to the spatial adjacency relationship, and multi-branch backtracking paths are generated layer by layer. Backtracking paths with the same starting anchor point and the same starting time in multi-branch backtracking paths are grouped into the same set, and the node sequences are uniformly arranged in reverse chronological order within the set; After the arrangement is completed, the node sequences that have shifted and misaligned in adjacent time layers of the multi-branch backtracking path are locally corrected to obtain the backtracking path unit; The backtracking path units are summarized and arranged according to node identifiers and time order to form a state backtracking path set.

10. A system for using the collaborative control method of the edge computing based energy storage cooling system according to any one of claims 1-9, characterized in that, It includes a data mapping module, a collaborative allocation module, a scheduling control module, and a feedback optimization module, and the modules are interconnected. The data mapping module is used to acquire real-time thermal state data streams, perform distributed feature segmentation and local thermal response mapping of edge nodes, and generate multi-node temperature control diagrams. The collaborative allocation module is used to perform cross-node collaborative allocation of cooling resources based on a multi-node temperature control map, forming a hierarchical cooling scheduling sequence; The scheduling and control module is used to perform spatiotemporal orchestration and trajectory constraint control on the hierarchical cooling scheduling sequence at each edge node, drive the cooling units to coordinate their actions, and form a set of state offsets. The feedback optimization module is used to perform edge-side temporal correlation backtracking and local strategy reconstruction on the state offset set, generate feedback control commands, and write the feedback control commands back to the multi-node temperature control map.

Images