A device energy-saving strategy optimization method based on big data analysis

Abstract

The application discloses a kind of equipment energy-saving strategy optimization methods based on big data analysis, it is related to equipment energy-saving control technical field, including the following steps: step one: acquisition carries SIS equipment operating data of quality code, associated control instruction and equipment topology;Step two: screening and registration generation equipment state sequence and control action event;Step three: construct directed energy flow diagram and generate energy flow topology record;Step four: search reachable node generation action propagation cone;Step five: assign state change attribution generation action source exclusive propagation field;Step six: construct source field kernel matrix of improved HSIC-Lasso model and divide strategy cell;Step seven: form strategy feasible region and generate strategy cell screening result;Step eight: generate equipment energy-saving strategy and output energy-saving evaluation result.The application improves the accuracy, safety and explainability of energy-saving strategy generation.

Description

Technical Field

[0001] This invention relates to the field of equipment energy-saving control technology, and in particular to a method for optimizing equipment energy-saving strategies based on big data analysis. Background Technology

[0002] With the increasing digitalization of power plant production and operation and the growing demand for energy-saving control, the joint analysis of SIS (System-Independent System) equipment operation data, control commands, and equipment topology, as well as energy-saving strategy optimization technologies, have received widespread attention. Existing energy-saving management systems primarily rely on statistical analysis of operating indicators, manual experience rules, and adjustments to single-point control parameters for energy consumption optimization. However, these systems commonly suffer from the following problems in practical applications: The collected SIS equipment operation data often carries quality anomalies caused by communication interruptions, bad measurement points, and range out-of-range. If not filtered in conjunction with quality codes, it can easily lead to distortion of the equipment state sequence. There is a time deviation between control commands and sampled data, and the action time and sampling time are difficult to accurately match, making it difficult to distinguish the source of the influence of control actions on equipment state changes. The medium flow direction, propagation time window, and safety boundary in the equipment topology are not fully incorporated into the modeling process, and traditional methods are unable to characterize the influence range formed by the propagation of control actions along the energy flow path. When the node time overlaps in the propagation path of control actions, existing methods usually have difficulty assigning the state change to the appropriate entity, resulting in a lack of interpretability and consistency of safety constraints in the energy-saving strategy screening results.

[0003] Therefore, how to provide a method for optimizing equipment energy-saving strategies based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a method for optimizing equipment energy-saving strategies based on big data analysis. This invention fully utilizes SIS operating data quality screening, energy flow topology modeling, action propagation cone analysis, and an improved HSIC-Lasso model. It details the optimization process from identifying the attribution of control action propagation to screening the feasible domain of energy-saving strategies. This method can solve the problems of unclear sources of control action influence, lack of safety constraints on energy-saving strategies, and insufficient interpretability of energy consumption optimization results. It has the advantages of high data reliability, accurate strategy screening, controllable safety boundaries, and stable energy-saving effects.

[0005] A method for optimizing equipment energy-saving strategies based on big data analysis according to an embodiment of the present invention includes the following steps: Step 1: Collect the operating data of SIS equipment carrying quality codes, associate control commands with equipment topology, and generate an energy-saving basic dataset; Step 2: Filter the energy-saving basic dataset based on the quality code, and register it according to the sampling time and action time to generate equipment status sequence and control action events; Step 3: Construct a directed energy flow graph based on the medium flow direction of the device topology, mark the propagation time window and safety boundary for the energy flow edges, and generate an energy flow topology record; Step 4: Based on the energy flow topology record, search for reachable nodes using control action events as the propagation source, and generate an action propagation cone; Step 5: Based on the energy flow topology record and device state sequence, assign state change ownership when the time of the action propagation cone generation node overlaps, and generate an exclusive propagation field for the action source; Step 6: Construct the source field kernel matrix of the improved HSIC-Lasso model based on the exclusive propagation field of the action source, and divide the strategy cells according to the adjustment direction, adjustment amplitude and adjustment duration; Step 7: Based on the energy flow topology record and safety boundary, form the strategy feasible region, reconstruct the energy-saving target kernel with the source field kernel matrix and solve the strategy cell selection coefficient to generate the strategy cell screening result; Step 8: Generate equipment energy-saving strategies based on the strategy cell screening results, and output energy-saving evaluation results.

[0006] Optionally, step one specifically includes: Access the SIS data transmission server, read the SIS device operation data carrying the quality code through a dedicated read-only account, extract the sampling time, measurement point value and device ownership information according to the measurement point identifier, and generate operation measurement point records; Read the action time, controlled object, command direction and command amplitude from the control command to generate a control command record; read the device nodes, connection edges, media flow direction and safety boundary from the device topology to generate a device topology record. Based on the equipment ownership information, the operation measurement point records are mapped to the equipment nodes, and the control command records are mapped to the equipment nodes based on the control object. The operation measurement point records, control command records, and equipment topology records are associated to generate an energy-saving basic dataset.

[0007] Optionally, step two specifically involves: Read the SIS device operation data from the energy-saving basic dataset, calibrate the valid status of the sampled values ​​according to the quality code, retain the sampling time and measurement point value corresponding to the valid status, and remove the sampling time and measurement point value corresponding to the communication interruption status, the bad point status of the measurement point, and the range out of bounds status to generate a valid operation record. Collect valid operation records according to the measurement point identifier, sort the measurement point values ​​according to the sampling time, and register the measurement point values ​​to a unified sampling time scale to generate an equipment status sequence. The control commands are read from the energy-saving basic dataset, and the control commands are registered with the equipment state sequence according to the action time. The control object, command direction and command amplitude corresponding to the action time are extracted to generate control action events.

[0008] Optionally, step three specifically includes: Read the device topology records from the energy-saving basic dataset, configure the device nodes as graph nodes, configure the connecting edges as graph edges, and determine the direction of the graph edges according to the medium flow direction to generate a directed energy flow graph; The measurement point values ​​of the device nodes at both ends of the connection edge are read based on the device state sequence. The propagation time window of the energy flow edge is determined according to the start time and arrival time of the change of the measurement point values, and the propagation time window is written into the directed energy flow graph. Read the security boundary from the device topology record, bind the security boundary to the graph node and power flow edge in the directed power flow graph, and combine the directed power flow graph, propagation window and security boundary to generate a power flow topology record.

[0009] Optionally, step four specifically includes: Read the directed energy flow graph, propagation window, and safety boundary from the energy flow topology record; determine the propagation source node based on the control object in the control action event; and determine the propagation starting point based on the action time in the control action event. Starting from the propagation source node, search for graph nodes along the energy flow edge direction in the directed energy flow graph, accumulate the propagation arrival time according to the propagation time window, and verify the running status of graph nodes according to the safety boundary to generate reachable node records; By combining the propagation source node, reachable node record, action time, propagation arrival time, and energy flow edge path, an action propagation cone with the control action event as the propagation source is generated.

[0010] Optionally, step five specifically includes: Based on the graph nodes, propagation arrival time, and energy flow edge path in the action propagation cone, the node time of the action propagation cone corresponding to the control action event is compared. The action propagation cones with consistent graph nodes and overlapping propagation arrival times are classified into the overlapping event set. The overlapping interval of propagation arrival times is determined as the node overlapping period, and a node time overlap record containing graph nodes, node overlapping period, and overlapping event set is generated. The continuity of the energy flow edge path and the energy flow intensity of each control action event in the overlapping event set are calculated based on the energy flow topology record. The change of measurement point values ​​of the graph nodes during the node overlap period is calculated based on the device state sequence. The attribution weight value is generated according to the order of propagation arrival time, the continuity of the energy flow edge path and the energy flow intensity. Normalize the attribution weight values ​​to generate the attribution share of each control action event in the overlapping event set. Allocate the change in measurement point values ​​to the control action events according to the attribution share, forming a state change attribution that includes control action events, graph nodes, node overlap time periods, energy flow edge paths, attribution shares, and state change after allocation. Generate an exclusive propagation field for the action source based on the state change attribution.

[0011] Optionally, step six specifically includes: Based on the exclusive propagation field of the action source, the control action event is used as the sample object. The graph nodes, node overlap time period, energy flow edge path, ownership share and state change after allocation corresponding to the control action event are extracted to generate source field samples. An improved HSIC-Lasso model is constructed. The improved HSIC-Lasso model uses source field samples instead of measurement point variables as feature objects, energy flow edge paths to represent propagation sources, ownership shares to represent the distribution ratio of state changes, and the amount of state change after distribution to represent the intensity of independent actions. The source field kernel matrix is ​​generated based on the similarity relationship between source field samples. The similarity of source field samples is calculated based on the consistency of energy flow edge paths, the proximity of node overlap periods, the proximity of ownership shares, and the proximity of state changes after allocation. This allows the source field kernel matrix to represent the similarity of control action events in propagation paths, propagation times, and state change allocation. The adjustment direction is determined based on the direction of the instruction in the control action event, the adjustment amplitude is determined based on the amplitude of the instruction in the control action event, and the adjustment duration is determined based on the overlapping period of nodes. Source field samples with the same adjustment direction, the same amplitude range, and the same duration range are divided into the same strategy cell.

[0012] Optionally, step seven specifically includes: Based on the directed energy flow graph, propagation window, and safety boundary in the energy flow topology record, and combined with the adjustment direction, adjustment amplitude, and adjustment duration of the strategy cell, a strategy feasible region is generated, which is limited by the direction mutual exclusion condition, the full propagation domain safety condition, the energy flow closure condition, and the propagation conflict condition. Based on the source field samples and source field kernel matrix corresponding to the strategy cell, the source field kernel matrix is ​​collected according to the strategy cell identifier to generate a kernel dictionary that represents the similarity relationship of independent propagation effects of strategy cells; Based on the equipment state sequence, the energy consumption change and safety margin corresponding to the strategy cell are extracted, the energy-saving target kernel is constructed, and the improved HSIC-Lasso model is used to reconstruct the energy-saving target kernel with the kernel dictionary within the policy feasible domain, and the policy cell selection coefficient is solved. Based on the strategy cell selection coefficient, strategy cells that fall into the policy feasible region are selected, and strategy cell selection results containing strategy cell identifiers and strategy cell selection coefficients are generated.

[0013] Optionally, the energy-saving target kernel is constructed from the energy consumption change and safety margin in the equipment state sequence. The energy consumption change is determined based on the equipment state items before and after the propagation of the control action event corresponding to the strategy cell. The safety margin is determined based on the distance between the equipment state item after propagation and the safety boundary. The energy-saving target kernel is used to characterize the target similarity relationship of the strategy cell in reducing energy consumption and maintaining the safety boundary, and serves as the target kernel object reconstructed by the source field kernel matrix in the improved HSIC-Lasso model.

[0014] Optionally, step eight specifically includes: Read the strategy cell identifier and strategy cell selection coefficient from the strategy cell screening results, locate the strategy cell based on the strategy cell identifier, extract the adjustment direction, adjustment amplitude and adjustment duration from the strategy cell, and associate the control action event and control object in the source field sample corresponding to the strategy cell to generate candidate energy-saving strategy units. Based on the directed energy flow graph and propagation time window in the energy flow topology record, the propagation arrival time of the candidate energy-saving strategy unit from the controlled object to the reachable node along the energy flow edge path is determined, and the strategy execution sequence is generated according to the propagation arrival time order and the size of the strategy cell selection coefficient. Based on the strategy execution sequence, the controlled object, adjustment direction, adjustment range, adjustment duration, propagation arrival time and strategy cell selection coefficient are written into the strategy record to generate the equipment energy-saving strategy. Based on the equipment state sequence, the equipment state items before and after the execution of the energy-saving strategy are read. The energy consumption change is calculated based on the equipment state items, and the safety margin change is calculated based on the safety boundary. The energy-saving strategy, energy consumption change, and safety margin change are combined to generate the energy-saving evaluation result.

[0015] The beneficial effects of this invention are: This invention collects operational data from SIS (System-In-Service) devices carrying quality codes and associates control commands with the device topology. This allows for the removal of abnormal data corresponding to communication interruptions, faulty measurement points, and out-of-range measurement conditions before the data enters energy-saving analysis, thus improving the reliability of device state sequences from the source. By registering sampling time with action time, control action events can be correlated with changes in device operation, avoiding the problem of unclear sources of control influence caused by simply relying on statistical analysis of operational indicators. Furthermore, a directed energy flow graph is constructed based on the medium flow direction, and propagation windows and safety boundaries are marked for the energy flow edges. This ensures that the device's energy-saving strategy is no longer limited to single-point parameter adjustments but reflects the true impact range of control actions along the energy flow path, improving the consistency between energy-saving analysis results and the actual operation of the power plant.

[0016] This invention assigns state changes to nodes when their propagation cones overlap in time, generating an exclusive propagation field for the action source. This allows state changes of the same node within the overlapping period to be assigned to control action events based on propagation arrival time, energy flow edge path continuity, and energy flow intensity, enhancing the interpretability of control action impact identification. Based on the exclusive propagation field for the action source, an improved HSIC-Lasso model source field kernel matrix is ​​constructed, and the energy-saving target kernel is reconstructed within the policy feasible region. This simultaneously considers energy-saving benefits, safety margins, propagation conflicts, and directional exclusivity conditions, selecting policy cells that better meet equipment topology and safety boundary requirements. This generates equipment energy-saving strategies with execution order, adjustment direction, adjustment amplitude, and adjustment duration, improving the accuracy of strategy selection, reducing energy consumption fluctuation risks, and ensuring safe equipment operation. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a device energy-saving strategy optimization method based on big data analysis proposed in this invention; Figure 2 This invention presents a flowchart illustrating the exclusive propagation field generation process of an energy-saving strategy optimization method for equipment based on big data analysis. Figure 3 This is a flowchart of the strategy cell selection process for an equipment energy-saving strategy optimization method based on big data analysis proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A method for optimizing equipment energy-saving strategies based on big data analysis includes the following steps: Step 1: Collect the operating data of SIS equipment carrying quality codes, associate control commands with equipment topology, and generate an energy-saving basic dataset; Step 2: Filter the energy-saving basic dataset based on the quality code, and register it according to the sampling time and action time to generate equipment status sequence and control action events; Step 3: Construct a directed energy flow graph based on the medium flow direction of the device topology, mark the propagation time window and safety boundary for the energy flow edges, and generate an energy flow topology record; Step 4: Based on the energy flow topology record, search for reachable nodes using control action events as the propagation source, and generate an action propagation cone; Step 5: Based on the energy flow topology record and device state sequence, assign state change ownership when the time of the action propagation cone generation node overlaps, and generate an exclusive propagation field for the action source; Step 6: Construct the source field kernel matrix of the improved HSIC-Lasso model based on the exclusive propagation field of the action source, and divide the strategy cells according to the adjustment direction, adjustment amplitude and adjustment duration; Step 7: Based on the energy flow topology record and safety boundary, form the strategy feasible region, reconstruct the energy-saving target kernel with the source field kernel matrix and solve the strategy cell selection coefficient to generate the strategy cell screening result; Step 8: Generate equipment energy-saving strategies based on the strategy cell screening results, and output energy-saving evaluation results.

[0020] In this embodiment, step one specifically includes: After accessing the SIS data sending server, the system uses a dedicated read-only account to establish a data retrieval session. It reads the SIS device operation data carrying the quality code from the data packets returned by the data retrieval session, uses the measurement point identifier as the index field, writes the sampling time and measurement point value under the same measurement point identifier into the same operation record, and reads the device ownership information from the measurement point configuration table. It combines the measurement point identifier, sampling time, measurement point value and device ownership information to generate an operation measurement point record. When the system reads control commands, it uses the control command number as the index field to extract the action time, controlled object, command direction, and command amplitude from the control command. The extracted action time, controlled object, command direction, and command amplitude are combined to generate a control command record. When the system reads the device topology, it uses the device number as the index field to extract the device nodes, connection edges, media flow direction, and security boundaries from the device topology. The extracted device nodes, connection edges, media flow direction, and security boundaries are combined to generate a device topology record. The system searches for the corresponding device node in the device topology record based on the device ownership information, binds the running measurement point record to the corresponding device node, searches for the corresponding device node in the device topology record based on the controlled object, binds the control command record to the corresponding device node, and uses the device node as the association primary key to associate the running measurement point record, control command record and device topology record to generate an energy-saving basic dataset.

[0021] In this embodiment, step two specifically involves: The system reads the SIS equipment operation data from the energy-saving basic dataset. Using the quality code as a filtering field, the system compares the quality code corresponding to each sampling time with the preset valid status identifier. When the quality code matches the valid status identifier, the measurement point value under the same sampling time is retained. When the quality code matches the communication interruption status, the bad point status of the measurement point, or the range out of bounds status, the measurement point value under the same sampling time is deleted. The retained sampling time, measurement point identifier, and measurement point value are combined to generate a valid operation record. The system uses the measurement point identifier as the collection index to allocate the valid operation records to the data sequence under the corresponding measurement point identifier. Within each data sequence, the measurement point values ​​are arranged in ascending order of sampling time. Then, using a unified sampling time scale as the time base, the measurement point values ​​under different measurement point identifiers are aligned to the same time position, so that each unified sampling time scale forms a device status item containing the values ​​of each measurement point. The device status items are arranged in ascending order of unified sampling time scale to generate a device status sequence. The system reads control commands from the energy-saving basic dataset, uses the action time as the registration field between the control command and the equipment state sequence, determines the unified sampling time scale corresponding to the action time in the equipment state sequence, and reads the control object, command direction and command amplitude under the corresponding unified sampling time scale. The action time, control object, command direction and command amplitude are written into the same event record to generate a control action event.

[0022] In this embodiment, step three specifically includes: The system reads device topology records from the energy-saving basic dataset, uses the device number of the device node as the graph index, writes each device node into the graph node set of the directed energy flow graph, writes the start and end identifiers of the connecting edges into the edge set of the directed energy flow graph, reads the medium flow direction carried by the connecting edges, determines the device node on the medium entry side as the graph edge start point, and determines the device node on the medium exit side as the graph edge end point. The graph edge direction is configured according to the direction from the graph edge start point to the graph edge end point. The graph edge after the direction configuration is completed is recorded as the energy flow edge, and a directed energy flow graph reflecting the medium transfer direction and energy transfer direction between device nodes is generated. Based on the device state sequence, the system reads the measurement values ​​of the device nodes at both ends of the connection edge. The system takes the measurement value of the device node at the starting point of the graph edge as the upstream state quantity and the measurement value of the device node at the ending point of the graph edge as the downstream state quantity. When the upstream state quantity continuously deviates from the reference value before the change, the system records the change start time. When the downstream state quantity changes in the same direction along the medium flow direction and reaches the response judgment condition, the system records the change arrival time. The time interval between the change start time and the change arrival time is used as the energy flow edge propagation time window, and the energy flow edge propagation time window is written into the corresponding energy flow edge in the directed energy flow graph. The system reads the security boundary from the device topology record, binds the security boundary to the graph node in the directed energy flow graph according to the device number of the device node, and binds the security boundary to the energy flow edge in the directed energy flow graph according to the start and end identifiers of the connecting edge, forming a directed energy flow graph that carries both the security boundary of the graph node and the security boundary of the energy flow edge. Then, the directed energy flow graph that has been written with the propagation window and security boundary is organized into an energy flow topology record.

[0023] In this embodiment, step four specifically includes: The system reads the directed energy flow graph, propagation window, and safety boundary from the energy flow topology record. Using the control object in the control action event as the matching field, the system searches for graph nodes in the directed energy flow graph whose device identifier matches the control object. The found graph node is identified as the propagation source node, and the action time in the control action event is written into the start time field of the propagation source node, forming a propagation starting point composed of the propagation source node and the action time. The system uses the propagation source node as the starting point for the search, reads the energy flow edges connected to the propagation source node and pointing from the propagation source node to downstream graph nodes, and takes the downstream graph nodes connected by the energy flow edges as candidate graph nodes. It reads the propagation time window corresponding to the energy flow edge, adds the action time to the propagation time window to obtain the propagation arrival time of the candidate graph node, and continues to search along the downstream energy flow edges of the candidate graph nodes. In each search, it accumulates the propagation time windows of the energy flow edges that have been passed to obtain the propagation arrival time of each graph node. Then, based on the propagation arrival time, it reads the graph node running status from the device state sequence, compares the graph node running status with the safety boundary, and writes the graph node identifier, propagation arrival time, and energy flow edge path into the reachable node record when the graph node running status is within the safety boundary. The system writes the propagation source node, reachable node record, action time, propagation arrival time, and energy flow edge path into the same propagation structure, arranges the reachable node record in the order of action time to propagation arrival time, and organizes the graph nodes, energy flow edge paths, and propagation arrival times that can be reached by the same control action event into node time range records, generating an action propagation cone with the control action event as the propagation source.

[0024] In this embodiment, step five specifically includes: Based on the graph nodes, propagation arrival time, and energy flow edge path in the action propagation cone, the system uses the graph node identifier as the first comparison field and the propagation arrival time interval as the second comparison field to perform pairwise comparisons on the action propagation cones formed by control action events. When two action propagation cones hit the same graph node and their propagation arrival time intervals overlap, the control action events to which the two action propagation cones belong are classified into the overlapping event set. The time from the start of the overlap to the end of the overlap is determined as the node overlap period. The graph node identifier, node overlap period, overlapping event set, control action event identifier, and energy flow edge path are combined to generate a node time overlap record. Based on the energy flow topology record, the system calculates the continuity of the energy flow edge path and the energy flow intensity of each control action event in the overlapping event set. The system checks whether the endpoint graph node of the adjacent energy flow edge is consistent with the starting graph node along the energy flow edge path. The ratio of the number of continuously connected energy flow edges to the length of the energy flow edge path is determined as the continuity of the energy flow edge path. The system accumulates the energy flow intensity of each energy flow edge in the energy flow edge path to obtain the path energy flow intensity. The system then reads the measurement point values ​​of the graph node at the beginning and end of the node overlap period based on the device state sequence. The difference between the measurement point values ​​at the two ends is determined as the change in measurement point value. The system generates the assigned weight value based on the propagation arrival time, the continuity of the energy flow edge path, and the path energy flow intensity. The system performs normalization processing on the attribution weight values. It divides the attribution weight value of each control action event by the sum of the attribution weight values ​​in the overlapping event set to obtain the attribution share of each control action event. The sum of the attribution shares in the overlapping event set is equal to one. Then, the change in the measured point value is multiplied by the attribution share to obtain the state change after allocation. The control action event, graph node, node overlap time period, energy flow edge path, attribution share and state change after allocation are combined to form the state change attribution. The state change attribution is collected according to the control action event to generate the action source exclusive propagation field.

[0025] In this embodiment, step six specifically includes: Based on the exclusive propagation field of the action source, and using control action events as sample objects, the system first collects the state changes according to the control action event identifier. Then, under the same control action event identifier, it organizes the graph nodes, node overlap periods, energy flow edge paths, attribution shares, and the amount of state changes after allocation. The graph nodes are used to identify the equipment location reached by the control action event, the node overlap periods are used to identify the time range of the control action event, the energy flow edge paths are used to identify the transmission route of the control action event from the propagation source node to the graph node, the attribution shares are used to identify the proportion of state changes allocated to the control action event within the node overlap period, and the amount of state changes after allocation is used to identify the independent effect intensity of the control action event on the operating state of the graph node. In this way, the source field sample is generated. An improved HSIC-Lasso model is constructed. The improved HSIC-Lasso model retains the kernel matrix structure used in HSIC-Lasso to measure the nonlinear dependency between feature objects and energy-saving targets, and retains the sparse selection structure used to screen effective feature objects. The system replaces the traditional HSIC-Lasso method of directly using measurement point variables as feature objects with the method of using source field samples as feature objects. This changes the model input from a single measurement point value to independent propagation action records corresponding to control action events. The system uses energy flow edge paths to represent propagation sources, ownership shares to represent the distribution ratio of state changes, and the amount of state change after distribution to represent the intensity of independent actions. The system also generates a source field kernel matrix based on the similarity relationship between source field samples. The similarity of source field samples is calculated based on the consistency of energy flow edge paths, the proximity of node overlap periods, the proximity of ownership shares, and the proximity of state changes after allocation. The system increases the similarity of source field samples when two source field samples pass through the same energy flow edge path, have similar node overlap periods, similar ownership shares, and similar state changes after allocation. The system decreases the similarity of source field samples when the difference in energy flow edge paths between two source field samples increases, the distance between node overlap periods increases, the difference in ownership shares increases, and the difference in state changes after allocation increases. This allows the source field kernel matrix to characterize the similarity relationship of control action events in propagation paths, propagation times, and state change allocation. The adjustment direction is determined based on the direction of the command in the control action event, the adjustment amplitude is determined based on the magnitude of the command in the control action event, and the adjustment duration is determined based on the start and end time difference of the node overlap period. The system divides the amplitude interval by the boundary of the distribution of command amplitude in the historical control commands and divides the duration interval by the boundary of the duration distribution of the node overlap period. Source field samples with the same adjustment direction, the same amplitude interval, and the same duration interval are divided into the same strategy cell.

[0026] This invention improves upon the traditional HSIC-Lasso model by using source field samples as feature objects instead of directly using measured point variables. It incorporates energy flow edge paths, node overlap periods, attribution shares, and post-allocation state changes into the source field kernel matrix construction process. This allows the model to no longer rely solely on single-point measured values ​​to determine energy-saving relevance, but rather to characterize the independent roles of control action events in propagation source, propagation time, and state change allocation. Furthermore, by adjusting direction, amplitude, and duration to divide strategy cells, the energy-saving strategy selection object is transformed from discrete control actions into strategy units with execution meaning. This improves the accuracy of control action impact identification, reduces attribution interference caused by node time overlap, enhances the interpretability of the energy-saving target reconstruction process, and improves the reliability and safety constraint adaptation capability of equipment energy-saving strategy generation.

[0027] In this embodiment, step seven specifically includes: Based on the directed energy flow graph, propagation window, and safety boundary in the energy flow topology record, the system maps the adjustment direction, adjustment magnitude, and adjustment duration of the strategy cell to the graph nodes and energy flow edges in the directed energy flow graph. The direction mutual exclusion condition is used to restrict the same adjustment parameter to retain only one adjustment direction in the same execution period. The full propagation domain safety condition is used to verify that the adjustment action is within the safety boundary when it reaches each graph node along the propagation window. The energy flow closure condition is used to verify that the energy flow changes when the adjustment action enters the graph node and leaves the graph node maintain the corresponding relationship. The propagation conflict condition is used to restrict strategy cells with overlapping propagation arrival times and opposite adjustment directions from entering the screening range at the same time. The direction mutual exclusion condition, the full propagation domain safety condition, the energy flow closure condition, and the propagation conflict condition jointly limit the feasible domain of the strategy. Based on the source field samples and source field kernel matrices corresponding to the strategy cells, the system collects the source field kernel matrices according to the strategy cell identifiers, treats the source field kernel matrices under the same strategy cell identifier as the same kernel atom, establishes a correspondence between the kernel atom and the strategy cell identifier, and retains the similar information formed by the energy flow edge path, node overlap period, ownership share and state change after allocation in the source field samples, so that the kernel dictionary can characterize the similarity relationship of independent propagation between strategy cells; Based on the device state sequence, the system extracts the energy consumption change and safety margin corresponding to the strategy cell. The system takes the device state item before the strategy cell occurs as the baseline state and the device state item after the strategy cell has completed propagation as the response state. The energy consumption change is determined based on the difference in energy consumption index between the baseline state and the response state. The safety margin is determined based on the distance between the response state and the safety boundary. The energy-saving target kernel is constructed based on the similarity relationship between the energy consumption change and the safety margin. The energy-saving target kernel is reconstructed with the kernel dictionary using the improved HSIC-Lasso model within the policy feasible domain. The strategy cell selection coefficient corresponding to each strategy cell is obtained through sparse selection. The system filters strategy cells that fall within the policy feasible region based on the strategy cell selection coefficient. Strategy cells that do not meet the policy feasible region are excluded, while strategy cells that meet the policy feasible region and whose strategy cell selection coefficient meets the filtering criteria are retained. The system then combines the retained strategy cell identifier and strategy cell selection coefficient to generate the strategy cell filtering result.

[0028] In this embodiment, the energy-saving target kernel is constructed from the energy consumption change and safety margin in the equipment state sequence. The system locates the strategy cell with the strategy cell identifier, and determines the action time and propagation completion time based on the control action event corresponding to the strategy cell. The system reads the equipment state items before the action time and the equipment state items after the propagation completion time from the equipment state sequence, determines the change in energy consumption index in the equipment state items as the energy consumption change, and determines the distance between the equipment state items after the propagation completion time and the safety boundary as the safety margin. The energy consumption change represents the energy-saving benefit of the strategy cell, and the safety margin represents the safety maintenance capability of the strategy cell. Strategy cells with similar energy consumption changes and similar safety margins are judged as having similar energy-saving targets. Strategy cells with increasing differences in energy consumption changes and increasing differences in safety margins are judged as having increasing differences in energy-saving targets. Thus, an energy-saving target kernel is formed, which represents the similarity relationship between strategy cells in terms of reducing energy consumption and maintaining the safety boundary. The energy-saving target kernel is used as the target kernel object reconstructed by the source field kernel matrix in the improved HSIC-Lasso model.

[0029] In this embodiment, step eight specifically includes: The system reads the strategy cell identifier and strategy cell selection coefficient from the strategy cell screening results. Based on the strategy cell identifier, the system locates the corresponding strategy cell in the strategy cell record. It reads the adjustment direction, adjustment amplitude, and adjustment duration from the strategy cell record. Then, based on the correspondence between the strategy cell and the source field sample, it traces back the source field sample, reads the control action event from the source field sample, and reads the control object from the control action event. The system combines the control object, adjustment direction, adjustment amplitude, adjustment duration, strategy cell identifier, and strategy cell selection coefficient to generate candidate energy-saving strategy units. Based on the directed energy flow graph and propagation time window in the energy flow topology record, the system takes the control object in the candidate energy-saving strategy unit as the propagation starting point, reads the reachable nodes along the energy flow edge path in the directed energy flow graph, and determines the propagation arrival time corresponding to each reachable node by accumulating according to the propagation time window corresponding to the energy flow edge path. Then, the propagation arrival time is written into the candidate energy-saving strategy unit, the candidate energy-saving strategy units are arranged in order of propagation arrival time, and when the propagation arrival time is consistent, the candidate energy-saving strategy units are arranged according to the size of the strategy cell selection coefficient to generate the strategy execution sequence. Based on the strategy execution sequence, the system reads the control object, adjustment direction, adjustment amplitude, adjustment duration, propagation arrival time and strategy cell selection coefficient in the candidate energy-saving strategy unit one by one. The control object is used as the device to which the strategy is applied, the adjustment direction is used as the adjustment direction, the adjustment amplitude is used as the adjustment amount, the adjustment duration is used as the duration of the strategy, the propagation arrival time is used as the influence time of the strategy, and the strategy cell selection coefficient is used as the contribution identifier of the strategy. These are written into the strategy record to generate the device energy-saving strategy. Based on the device state sequence, the system reads the device state items before and after the execution of the energy-saving strategy. The system uses the propagation arrival time in the strategy execution sequence as the time positioning basis, extracts the device state items before and after execution from the device state sequence, calculates the energy consumption change based on the difference in energy consumption indicators between the device state items before and after execution, and calculates the safety margin change based on the distance change between the device state items before and after execution relative to the safety boundary. The system combines the device energy-saving strategy, the energy consumption change, and the safety margin change to generate the energy-saving evaluation result.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to an energy-saving optimization scenario for thermal power unit equipment. The equipment analyzed included forced draft fans, induced draft fans, feedwater pumps, condensate pumps, circulating water pumps, condensers, and main steam-related equipment. The comparison method used was a commonly used manual experience threshold adjustment scheme. This scheme relies on the SIS trend curve, daily energy consumption reports, and fixed operating thresholds. Operators manually judge based on changes in fan current, pump frequency, main steam pressure, feedwater flow rate, and condenser vacuum, and then adjust the damper opening, pump frequency, and feedwater regulation. This manual experience threshold adjustment scheme does not establish a directed energy flow graph, does not calibrate the energy flow side propagation time window, does not assign state changes under overlapping node time conditions, and does not use a source-field kernel matrix to reconstruct the energy-saving target kernel. Therefore, when control actions occur intensively, it is easy to mix the state changes caused by multiple control actions into the same equipment node, resulting in unclear sources of action influence, conflicting energy-saving strategies, and insufficient safety boundary verification.

[0031] Under the same batch of SIS device operation data, control commands, and device topology conditions, this invention first accesses the SIS data transmission server through a dedicated read-only account, reads the SIS device operation data carrying quality codes, and extracts sampling time, measurement point values, and device affiliation information. For sampling data corresponding to communication interruption, measurement point failure, and range out-of-bounds states, this invention directly discards them based on the quality code, ensuring that all data entering the analysis process are valid operation records. The system then registers the measurement point values ​​to a unified sampling timescale based on the sampling time, extracts the controlled object, command direction, and command amplitude based on the action time in the control commands, and generates control action events. Device nodes, connection edges, medium flow direction, and safety boundaries in the device topology are converted into directed energy flow graphs. The system then determines the energy flow edge propagation time window based on the start time and arrival time of the measurement point value changes at both ends of the connection edge. For each control action event, the system uses the controlled object as the propagation source node and searches for reachable nodes along the energy flow edge direction to obtain the action propagation cone. When the propagation arrival time of the action propagation cones overlaps at the same graph node, the system calculates the attribution weight value based on the order of propagation arrival time, the continuity of the energy flow edge path, and the energy flow intensity, and allocates the change in the measurement point value during the node overlap period to the corresponding control action event according to the attribution share, generating an action source exclusive propagation field.

[0032] Based on the exclusive propagation field of the action source, this invention extracts the graph nodes, node overlap periods, energy flow edge paths, ownership shares, and post-allocation state changes corresponding to control action events to generate source field samples, which are then used as feature objects of the improved HSIC-Lasso model. The model generates a source field kernel matrix based on the consistency of energy flow edge paths, the proximity of node overlap periods, the proximity of ownership shares, and the proximity of post-allocation state changes. Then, it divides the strategy cells according to the adjustment direction, adjustment amplitude, and adjustment duration. Subsequently, the system combines the directed energy flow graph, propagation window, and safety boundary to form a feasible strategy domain, excluding strategy cells corresponding to mutually exclusive directions, full propagation domain out-of-bounds, non-closed energy flow, and propagation conflicts. The energy-saving target kernel is constructed from the energy consumption changes and safety margins in the equipment state sequence. The improved HSIC-Lasso model reconstructs the energy-saving target kernel within the feasible strategy domain using a kernel dictionary and solves for the strategy cell selection coefficients. Finally, the system generates an equipment energy-saving strategy containing the controlled object, adjustment direction, adjustment amplitude, adjustment duration, propagation arrival time, and strategy cell selection coefficients based on the strategy cell selection results. During the implementation process, 186,240 valid sampling points, 412 control commands, 37 device nodes, 52 energy flow edges, and 68 safety boundary items were used for verification. This resulted in 412 motion propagation cones, 126 node time overlap records, 94 strategy cells, and 31 executable energy-saving strategy units. The comparison results are shown in Table 1. Table 1. Energy Saving Optimization Comparison Table

[0033] As shown in Table 1, while the manual experience threshold adjustment scheme can provide adjustment suggestions based on changes in operating indicators, it lacks quality code screening, action time registration, energy flow propagation analysis, and overlapping state change attribution, resulting in low utilization rate of effective sampling points, success rate of control action registration, and accuracy of node overlap attribution. This invention eliminates abnormal sampling through quality codes, establishes control action events through sampling time and action time registration, and distinguishes the source of state changes through action propagation cones and exclusive propagation fields of action sources, thereby improving the accuracy of node overlap attribution from 58.9% to 91.3%. Simultaneously, this invention introduces directional mutual exclusion conditions, full propagation domain safety conditions, energy flow closure conditions, and propagation conflict conditions into the policy feasible domain, increasing the number of policy conflict detections from 9 to 27 and reducing the number of safety boundary violations from 6 to 1. The average unit energy consumption decreased from 313.4 g / kWh to 307.1 g / kWh, and the average safety margin increased from 8.5% to 12.4%, indicating that this invention can improve the efficiency of energy-saving strategy generation and the stability of energy-saving effects while maintaining safe equipment operation.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing equipment energy-saving strategies based on big data analysis, characterized in that, Includes the following steps: Step 1: Collect the operating data of SIS equipment carrying quality codes, associate control commands with equipment topology, and generate an energy-saving basic dataset; Step 2: Filter the energy-saving basic dataset based on the quality code, and register it according to the sampling time and action time to generate equipment status sequence and control action events; Step 3: Construct a directed energy flow graph based on the medium flow direction of the device topology, mark the propagation time window and safety boundary for the energy flow edges, and generate an energy flow topology record; Step 4: Based on the energy flow topology record, search for reachable nodes using control action events as the propagation source, and generate an action propagation cone; Step 5: Based on the energy flow topology record and device state sequence, assign state change ownership when the time of the action propagation cone generation node overlaps, and generate an exclusive propagation field for the action source; Step 6: Construct the source field kernel matrix of the improved HSIC-Lasso model based on the exclusive propagation field of the action source, and divide the strategy cells according to the adjustment direction, adjustment amplitude and adjustment duration; Step 7: Based on the energy flow topology record and safety boundary, form the feasible region of the strategy, reconstruct the energy-saving target kernel with the source field kernel matrix and solve the strategy cell selection coefficient to generate the strategy cell screening result; Step 8: Generate equipment energy-saving strategies based on the strategy cell screening results, and output energy-saving evaluation results.

2. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step one specifically involves: Access the SIS data transmission server, read the SIS device operation data carrying the quality code through a dedicated read-only account, extract the sampling time, measurement point value and device ownership information according to the measurement point identifier, and generate operation measurement point records; Read the action time, controlled object, command direction and command amplitude from the control command to generate a control command record; read the device nodes, connection edges, media flow direction and safety boundary from the device topology to generate a device topology record. Based on the equipment ownership information, the operation measurement point records are mapped to the equipment nodes, and the control command records are mapped to the equipment nodes based on the control object. The operation measurement point records, control command records, and equipment topology records are associated to generate an energy-saving basic dataset.

3. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step two specifically involves: Read the SIS device operation data from the energy-saving basic dataset, calibrate the valid status of the sampled value according to the quality code, retain the sampling time and measurement point value corresponding to the valid status, and remove the sampling time and measurement point value corresponding to the communication interruption status, the bad point status of the measurement point, and the range out of bounds status to generate a valid operation record. Collect valid operation records according to the measurement point identifier, sort the measurement point values ​​according to the sampling time, and register the measurement point values ​​to a unified sampling time scale to generate an equipment status sequence. The control commands are read from the energy-saving basic dataset, and the control commands are registered with the equipment state sequence according to the action time. The control object, command direction and command amplitude corresponding to the action time are extracted to generate control action events.

4. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step three specifically involves: Read the device topology records from the energy-saving basic dataset, configure the device nodes as graph nodes, configure the connecting edges as graph edges, and determine the direction of the graph edges according to the medium flow direction to generate a directed energy flow graph; The measurement point values ​​of the device nodes at both ends of the connection edge are read based on the device state sequence. The propagation time window of the energy flow edge is determined according to the start time and arrival time of the change of the measurement point values, and the propagation time window is written into the directed energy flow graph. Read the security boundary from the device topology record, bind the security boundary to the graph node and power flow edge in the directed power flow graph, and combine the directed power flow graph, propagation window and security boundary to generate a power flow topology record.

5. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step four specifically involves: Read the directed energy flow graph, propagation window, and safety boundary from the energy flow topology record; determine the propagation source node based on the control object in the control action event; and determine the propagation starting point based on the action time in the control action event. Starting from the propagation source node, search for graph nodes along the energy flow edge direction in the directed energy flow graph, accumulate the propagation arrival time according to the propagation time window, and verify the running status of graph nodes according to the safety boundary to generate reachable node records; By combining the propagation source node, reachable node record, action time, propagation arrival time, and energy flow edge path, an action propagation cone with the control action event as the propagation source is generated.

6. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step five specifically involves: Based on the graph nodes, propagation arrival time, and energy flow edge path in the action propagation cone, the node time of the action propagation cone corresponding to the control action event is compared. The action propagation cones with consistent graph nodes and overlapping propagation arrival times are classified into the overlapping event set. The overlapping interval of propagation arrival times is determined as the node overlapping period, and a node time overlap record containing graph nodes, node overlapping period, and overlapping event set is generated. The continuity of the energy flow edge path and the energy flow intensity of each control action event in the overlapping event set are calculated based on the energy flow topology record. The change of measurement point values ​​of the graph nodes during the node overlap period is calculated based on the device state sequence. The attribution weight value is generated according to the order of propagation arrival time, the continuity of the energy flow edge path and the energy flow intensity. Normalize the attribution weight values ​​to generate the attribution share of each control action event in the overlapping event set. Allocate the change in measurement point values ​​to the control action events according to the attribution share, forming a state change attribution that includes control action events, graph nodes, node overlap time periods, energy flow edge paths, attribution shares, and state change after allocation. Generate an exclusive propagation field for the action source based on the state change attribution.

7. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step six specifically involves: Based on the exclusive propagation field of the action source, the control action event is used as the sample object. The graph nodes, node overlap time period, energy flow edge path, ownership share and state change after allocation corresponding to the control action event are extracted to generate source field samples. An improved HSIC-Lasso model is constructed. The improved HSIC-Lasso model uses source field samples instead of measurement point variables as feature objects, energy flow edge paths to represent propagation sources, ownership shares to represent the distribution ratio of state changes, and the amount of state change after distribution to represent the intensity of independent actions. The source field kernel matrix is ​​generated based on the similarity relationship between source field samples. The similarity of source field samples is calculated based on the consistency of energy flow edge paths, the proximity of node overlap periods, the proximity of ownership shares, and the proximity of state changes after allocation. This allows the source field kernel matrix to represent the similarity of control action events in propagation paths, propagation times, and state change allocation. The adjustment direction is determined based on the direction of the instruction in the control action event, the adjustment amplitude is determined based on the amplitude of the instruction in the control action event, and the adjustment duration is determined based on the overlapping period of nodes. Source field samples with the same adjustment direction, the same amplitude range, and the same duration range are divided into the same strategy cell.

8. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step seven specifically involves: Based on the directed energy flow graph, propagation window, and safety boundary in the energy flow topology record, and combined with the adjustment direction, adjustment amplitude, and adjustment duration of the strategy cell, a strategy feasible region is generated, which is limited by the direction mutual exclusion condition, the full propagation domain safety condition, the energy flow closure condition, and the propagation conflict condition. Based on the source field samples and source field kernel matrix corresponding to the strategy cell, the source field kernel matrix is ​​collected according to the strategy cell identifier to generate a kernel dictionary that represents the similarity relationship of independent propagation effects of strategy cells; Based on the equipment state sequence, the energy consumption change and safety margin corresponding to the strategy cell are extracted, the energy-saving target kernel is constructed, and the improved HSIC-Lasso model is used to reconstruct the energy-saving target kernel with the kernel dictionary within the policy feasible domain, and the policy cell selection coefficient is solved. Based on the strategy cell selection coefficient, strategy cells that fall into the policy feasible region are selected, and strategy cell selection results containing strategy cell identifiers and strategy cell selection coefficients are generated.

9. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, The energy-saving target kernel is constructed from the energy consumption change and safety margin in the equipment state sequence. The energy consumption change is determined based on the equipment state items before and after the propagation of the control action event corresponding to the strategy cell. The safety margin is determined based on the distance between the equipment state item after propagation and the safety boundary. The energy-saving target kernel is used to characterize the target similarity relationship of the strategy cell in reducing energy consumption and maintaining the safety boundary, and serves as the target kernel object reconstructed by the source field kernel matrix in the improved HSIC-Lasso model.

10. The equipment energy-saving strategy optimization method based on big data analysis according to claim 1, characterized in that, Step eight specifically involves: Read the strategy cell identifier and strategy cell selection coefficient from the strategy cell screening results, locate the strategy cell based on the strategy cell identifier, extract the adjustment direction, adjustment amplitude and adjustment duration from the strategy cell, and associate the control action event and control object in the source field sample corresponding to the strategy cell to generate candidate energy-saving strategy units. Based on the directed energy flow graph and propagation time window in the energy flow topology record, the propagation arrival time of the candidate energy-saving strategy unit from the controlled object to the reachable node along the energy flow edge path is determined, and the strategy execution sequence is generated according to the propagation arrival time order and the size of the strategy cell selection coefficient. Based on the strategy execution sequence, the controlled object, adjustment direction, adjustment range, adjustment duration, propagation arrival time and strategy cell selection coefficient are written into the strategy record to generate the equipment energy-saving strategy. Based on the equipment state sequence, the equipment state items before and after the execution of the energy-saving strategy are read. The energy consumption change is calculated based on the equipment state items, and the safety margin change is calculated based on the safety boundary. The energy-saving strategy, energy consumption change, and safety margin change are combined to generate the energy-saving evaluation result.

Images