A device energy-saving strategy generation method and system based on big data processing

Abstract

This invention discloses a method and system for generating equipment energy-saving strategies based on big data processing, relating to the field of equipment energy-saving control technology. The method includes the following steps: S1, forming a historical operating sequence of equipment and obtaining the correlation between equipment; S2, forming adjustment response segments; S3, matching the adjustment response segments with unadjusted response segments that have not undergone corresponding adjustment actions, and aligning the adjustment response segments and unadjusted response segments; S4, forming a preliminary adjustment response; S5, inputting the preliminary adjustment response into an improved HAVOK model to obtain a new adjustment response for the current adjustment action; S6, obtaining the system's net energy-saving response; S7, generating an equipment energy-saving strategy. This invention can separate the new adjustment response and generate a system's net energy-saving strategy.

Description

Technical Field

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

[0002] Industrial production equipment generates a large amount of operational data during long-term operation, including equipment status, energy consumption, control actions, and production tasks. Existing energy-saving management typically relies on historical energy consumption of individual equipment, empirical thresholds, or predictive models to determine control parameter adjustments and evaluates energy-saving effects based on the difference in energy consumption before and after adjustment. However, equipment energy consumption is simultaneously affected by factors such as production task progress, load changes, operating modes, and environmental conditions. Directly comparing data before and after adjustment makes it difficult to accurately distinguish between energy consumption changes caused by control actions and natural equipment operating changes. When multiple adjustment actions occur consecutively, the response of the previous action may extend into the stage of the next action, easily leading to confusion regarding the attribution of responses between different actions. Furthermore, equipment in a production system is often interconnected through electrical, steam, gas, or heat transfer pipelines. A reduction in the energy consumption of a single piece of equipment may increase the energy consumption of related equipment. Existing methods, which only evaluate energy-saving effects based on the energy consumption changes of the affected equipment, fail to reflect the actual net energy savings of the system. Current energy-saving strategies also often prioritize adjustment actions based solely on energy savings, without fully considering the connection between equipment statuses of adjacent adjustments, potentially resulting in a sequence of actions that may not be executed continuously.

[0003] Therefore, how to provide a method and system for generating energy-saving strategies for equipment based on big data processing 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 and system for generating equipment energy-saving strategies based on big data processing. This invention locates the position of control parameter changes in the historical operating sequence of equipment, extracts the equipment state and energy consumption changes before and after the adjustment action, and matches unadjusted response segments that did not undergo the same adjustment action. It aligns the operating process and deducts common changes according to the production task progress, reducing the interference of production progress, load fluctuations, and inherent equipment operating changes on the extraction results of the adjustment response. By setting a forced state recursively across sampling positions in the HAVOK model, and using continuation direction projection and complementary projection to separate the continuation correction component and the newly added forced component, the state vector evolution is driven by the continuation forced state and the complete forced state, respectively. The continuation response of the preceding adjustment action is extracted from the complete adjustment response, improving the accuracy of identifying the newly added response of the current adjustment action. Simultaneously, the energy consumption reduction and increase of the equipment affected by the adjustment action and related equipment are combined along the equipment energy correlation relationship to obtain the system's net energy-saving response. The execution sequence is determined based on the entry state, end state, and system net energy-saving response of the adjustment action, ensuring that the generated equipment energy-saving strategy takes into account both equipment state continuity and the overall system net energy-saving effect, improving the executability of the strategy and the actual energy-saving level.

[0005] A method for generating device energy-saving strategies based on big data processing according to an embodiment of the present invention includes the following steps: S1. Acquire equipment operation data, energy consumption data, control action data, and production task data to form a historical operation sequence of the equipment and obtain the correlation between the equipment; S2. Locate the position of control parameter change in the historical operation sequence of the equipment, extract the equipment status and energy consumption changes before and after the control parameter change, and form an adjustment response segment; S3. Match the adjustment response segment with an unadjusted response segment that has not undergone a corresponding adjustment action, and align the adjustment response segment and the unadjusted response segment according to the device operation process; S4. Subtract the common changes between the aligned and unadjusted response segments to form the initial adjusted response; S5. Input the preliminary adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the state vector evolution to strip away the response that continues from the previous adjustment action and obtain the new adjustment response of the current adjustment action. S6. By combining the energy consumption increase or decrease of the device currently performing the adjustment action and the associated devices along the correlation between the devices, the net energy-saving response of the system is obtained; S7. Connect multiple adjustment actions according to the equipment status before and after the adjustment action, and determine the execution order of the multiple adjustment actions according to the corresponding system net energy saving response to generate an equipment energy saving strategy.

[0006] Optionally, S1 includes: S11. Collect equipment operation data, energy consumption data, and control action data with equipment identification and occurrence time, and obtain the participating equipment and production task progress corresponding to the production task. S12. Match various types of data with equipment identifiers, arrange the data of the same equipment according to the time of occurrence, and classify the production tasks and production task progress into the corresponding time of occurrence to form the equipment historical operation sequence. S13. Pair the energy outputs and energy inputs that are interconnected between different devices, connect the energy output device to the corresponding energy input device, and obtain the relationship between the devices.

[0007] Optionally, S2 includes: S21. Compare adjacent values ​​of the same control parameter one by one along the historical operation sequence of the equipment, determine the position where the control parameter begins to deviate from the original value as the starting point of the change, and determine the position where the control parameter stops changing continuously as the ending point of the change; S22. Extract device status and energy consumption data from the starting point of the change along the decreasing time direction, and extract device status and energy consumption data from the ending point of the change along the increasing time direction, while retaining the control action data between the starting point and the ending point of the change. S23. Connect the equipment status and energy consumption data before the control parameter change, the control action data, and the equipment status and energy consumption data after the control parameter change in time sequence to form an adjustment response segment.

[0008] Optionally, S3 includes: S31. Extract the equipment status parameters, production task progress and load change of the adjustment response segment before the adjustment action occurs, and extract candidate operation segments from the equipment historical operation sequence of the same equipment that have not experienced corresponding adjustment actions. S32. Convert the equipment status parameters, production task progress and load change to the corresponding value range, calculate the difference between each candidate operation segment and the adjustment response segment, and select the candidate operation segment with the smallest total difference as the unadjusted response segment. S33. Convert the production task progress in the adjusted response segment and the unadjusted response segment into relative progress respectively, and pair the sampling positions with the same relative progress bit by bit to complete the alignment of the equipment operation process.

[0009] Optionally, S4 includes: S41. The last sampling position before the adjustment action occurs in the adjustment response segment is determined as the first reference position, and the position in the unadjusted response segment that is the same as the device operation process at the first reference position is determined as the second reference position. S42. Subtract the corresponding reference position's equipment status parameters and energy consumption data from the equipment status parameters and energy consumption data at each sampling location to form an adjusted change sequence and an unadjusted change sequence. S43. At the same position in the equipment operation process, subtract the corresponding changes in the unadjusted change sequence from the equipment state changes and energy consumption changes in the adjustment change sequence to form a preliminary adjustment response.

[0010] Optionally, the improved HAVOK model includes a delayed coordinate input layer, a coordinate decomposition layer, a forced state recursion layer, a state vector evolution layer, and a response output layer; The delayed coordinate input layer arranges the initial adjustment response in a staggered manner along the sampling position to form a Hankel matrix; The coordinate decomposition layer performs singular value decomposition on the Hankel matrix, and forms a state vector from the remaining multiple delayed coordinates excluding the last one, and determines the last delayed coordinate as the forced coordinate. The forced state recursion layer evolves the forced state of the previous sampling position through the forced recursion matrix, and transforms the forced coordinates of the current sampling position through the coordinate mapping matrix, forming the continued forced state and forced state of the current sampling position. The state vector evolution layer processes the state vector of the previous sampling position through the state evolution matrix, and combines the processing result with the result obtained by the forced coupling matrix in processing the continued forced state and the forced state to form the continued state vector and the total state vector. The response output layer converts the continued state vector and the total state vector into the response of the previous adjustment action and the total adjustment response respectively through the response reconstruction matrix. The response of the previous adjustment action is subtracted from the total adjustment response, and the new adjustment response of the current adjustment action is output.

[0011] Optionally, the continuing forced state and forced state that form the current sampling position include: Multiply the forced state of the previous sampling position by the forced recursion matrix to obtain the continued forced state of the current sampling position; Multiply the forced coordinates of the current sampling position with the coordinate mapping matrix to transform the forced coordinates to the state space where the continued forced state is located; Construct a projection matrix pointing to the subspace where the continued forced state is located and a corresponding complementary projection matrix. Extract the continued correction component from the transformed forced coordinates using the projection matrix and extract the new forced component using the complementary projection matrix. The continued correction component is incorporated into the continued forced state, and the corrected continued forced state and the newly added forced component are combined to form the forced state of the current sampling position. The corrected continuation forced state and the forced state at the current sampling position are passed to the state vector evolution layer, and the forced state at the current sampling position is used as the recursive input of the forced state at the next sampling position.

[0012] Optionally, S6 includes: S61. Starting from the device currently acting on the adjustment action, track the devices whose state changes or energy consumption changes within the sampling location range corresponding to the newly added adjustment response level by level along the correlation relationship between the devices, and determine the associated devices. S62. Extract the energy consumption changes of the current regulating action device and the associated device at each sampling position from the newly added regulating response, and classify the energy consumption increase part as energy consumption increase amount and the energy consumption decrease part as energy consumption reduction amount. S63. Combine the energy consumption increase and energy consumption reduction of the current regulating action device and related devices, and subtract the combined energy consumption increase from the combined energy consumption reduction to obtain the system net energy saving response corresponding to the current regulating action.

[0013] Optionally, S7 includes: S71. Determine the state of the last device before the adjustment action occurs as the entry state, determine the state of the last device within the sampling position range corresponding to the newly added adjustment response as the end state, and encapsulate the adjustment action, entry state, end state and system net energy saving response into an adjustment action unit. S72. Pair the end state of the previous adjustment action unit with the entry state of the next adjustment action unit according to the device identifier and status parameters, and connect the adjustment action units with the same device identifier and the same or different status parameter values, or whose differences do not exceed the sampling resolution of the corresponding status parameters, to form a status connection relationship. S73. Determine the last device state in the historical operation sequence of the device as the starting device state. Pair the entry state of each adjustment action unit with the starting device state according to the device identifier and state parameters. Select the adjustment action unit with the largest net energy saving response of the system from the adjustment action units with the same device identifier, the same state parameter value or the difference does not exceed the sampling resolution of the corresponding state parameter and the system net energy saving response is greater than zero. S74. Remove the already arranged adjustment action units, and from the remaining adjustment action units that have a state connection relationship with the current adjustment action unit and whose net energy saving response of the system is greater than zero, successively select the adjustment action unit with the largest net energy saving response of the system and continue to arrange them until there are no adjustment action units that can be connected, thus forming the execution sequence of the adjustment action. S75. Arrange the adjustment actions in each adjustment action unit according to the execution sequence to generate an energy-saving strategy for the equipment.

[0014] A device energy-saving strategy generation system based on big data processing according to an embodiment of the present invention includes the following modules: The data generation module is used to acquire equipment operation data, energy consumption data, control action data, and production task data, form historical operation sequences of equipment, and obtain the correlation between equipment. The response interception module is used to locate the position of control parameter change in the historical operation sequence of the equipment, intercept the equipment status and energy consumption changes before and after the control parameter change, and form an adjustment response segment. The segment matching module is used to match the adjustment response segment with an unadjusted response segment that has not undergone a corresponding adjustment action, and to align the adjustment response segment and the unadjusted response segment according to the device operation process; The response extraction module is used to subtract the common changes between the aligned adjusted response fragments and the unadjusted response fragments to form a preliminary adjusted response; The response separation module is used to input the preliminary adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the state vector evolution to strip away the response that continues from the previous adjustment action and obtain the new adjustment response of the current adjustment action. The energy-saving calculation module is used to aggregate the energy consumption increase or decrease of the device currently performing the adjustment action and the associated devices along the correlation between the devices to obtain the net energy-saving response of the system. The strategy generation module is used to connect multiple adjustment actions based on the equipment status before and after the adjustment action, and determine the execution order of the multiple adjustment actions according to the corresponding system net energy saving response, so as to generate an equipment energy saving strategy.

[0015] The beneficial effects of this invention are: 1. This invention locates the position of control parameter change from the equipment's historical operation sequence, extracts the equipment state and energy consumption changes before and after the adjustment action, and matches the adjustment response segment with the unadjusted response segment that did not have the same adjustment action. It completes the alignment of the operation process and deduction of common changes according to the production task progress, which can reduce the interference of production task progress, natural load fluctuations and inherent equipment operation changes on the adjustment response extraction results and improve the accuracy of energy-saving response identification of adjustment actions.

[0016] 2. This invention improves the HAVOK model by writing the forced coordinates into the forced state recursively across sampling positions. It separates the continued correction component and the newly added forced component through the continued direction projection and complementary projection. The state vector evolution is driven by the continued forced state and the complete forced state, respectively. This can separate the continued response of the previous adjustment action from the complete adjustment response, reduce the confusion of response attribution between consecutive adjustment actions, and improve the distinguishability of the newly added response of the current adjustment action.

[0017] 3. This invention combines the energy reduction and energy increase of the regulating action on the device and related devices along the energy correlation between devices to obtain the net energy saving response of the system. Based on the entry state, end state of the regulating action and the net energy saving response of the system, multiple regulating actions are connected. This can avoid judging the energy saving effect based on the energy consumption change of a single device. The generated device energy saving strategy can simultaneously meet the device state connection requirements and the overall net energy saving requirements of the system, thereby improving the feasibility of the energy saving strategy and the actual energy saving effect. Attached Figure Description

[0018] 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 method for generating equipment energy-saving strategies based on big data processing, as proposed in this invention. Figure 2 This is a schematic diagram of the improved HAVOK model structure for a device energy-saving strategy generation method based on big data processing proposed in this invention. Figure 3 This is a structural diagram of a device energy-saving strategy generation system based on big data processing proposed in this invention. Detailed Implementation

[0019] 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.

[0020] refer to Figures 1-2 A method for generating energy-saving strategies for devices based on big data processing includes the following steps: S1. Acquire equipment operation data, energy consumption data, control action data, and production task data to form a historical operation sequence of the equipment and obtain the correlation between the equipment; S2. Locate the position of control parameter changes in the historical operating sequence of the positioning equipment, extract the equipment status and energy consumption changes before and after the control parameter changes, and form an adjustment response segment; S3. Match the unadjusted response segments that have not undergone corresponding adjustment actions to the adjustment response segments, and align the adjustment response segments and unadjusted response segments according to the device operation process; S4. Subtract the common changes between the aligned and unadjusted response segments to form the initial adjusted response; S5. Input the initial adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the evolution of the state vector. The response of the previous adjustment action is stripped away to obtain the new adjustment response of the current adjustment action. S6. By combining the energy consumption increase or decrease of the device currently performing the adjustment action and its associated devices along the relationships between the devices, the net energy-saving response of the system is obtained. S7. Connect multiple adjustment actions according to the equipment status before and after the adjustment action, and determine the execution sequence of multiple adjustment actions based on the corresponding system net energy saving response to generate equipment energy saving strategy.

[0021] In this embodiment, S1 includes: S11. Collect equipment operation data, energy consumption data, and control action data with equipment identification and occurrence time, and obtain the participating equipment and production task progress corresponding to the production task. S12. Match various types of data with equipment identifiers, arrange the data of the same equipment according to the time of occurrence, and classify the production tasks and production task progress into the corresponding time of occurrence to form the equipment historical operation sequence. S13. Pair the energy outputs and energy inputs that are interconnected between different devices, connect the energy output device to the corresponding energy input device, and obtain the relationship between the devices.

[0022] In the specific implementation process, equipment operation data is obtained from the historical records of equipment controllers, distributed control devices, or equipment monitoring platforms. Each equipment operation record includes at least the equipment identification, sampling time, start / stop status, operating mode, load, and operating parameters. Energy consumption data is obtained from electricity meters, steam flow meters, gas flow meters, or other energy metering devices bound to the equipment. Each energy consumption record includes the equipment identification, energy type, metered value, and metering time. Control action data is obtained from control command records. Each control action record includes the controlled equipment identification, control parameter identification, value before adjustment, value after adjustment, and time of action occurrence.

[0023] When using cumulative electrical energy values, the cumulative electrical energy values ​​at adjacent metering times are subtracted to obtain the electrical energy consumption for the corresponding time period. When using instantaneous electrical power, the average electrical power at adjacent sampling times is taken and multiplied by the sampling interval to obtain the electrical energy consumption for the corresponding time period. When using steam flow rate, the average steam flow rate at adjacent sampling times is taken and multiplied by the sampling interval to obtain the steam mass, which is then multiplied by the difference between the supply steam enthalpy and the return water enthalpy to obtain the steam energy consumption for the corresponding time period. When using gas flow rate, the average gas flow rate at adjacent sampling times is taken and multiplied by the sampling interval to obtain the gas volume, which is then multiplied by the lower heating value of the gas to obtain the gas energy consumption for the corresponding time period. The supply steam enthalpy, return water enthalpy, and lower heating value of the gas are read from the enterprise's energy metering configuration table. The enterprise's energy metering configuration table stores the supply steam enthalpy, return water enthalpy, lower heating value of the gas, parameter effective time, and parameter ineffective time according to the energy medium identifier. When calculating the energy consumption for each sampling interval, parameter values ​​within the valid time range are read according to the sampling time; the same set of parameter values ​​is used within the same sampling interval.

[0024] Production task data is obtained from production management records. Each production task record includes a production task identifier, task start time, task completion record time, planned completion quantity, cumulative completion quantity, and participating equipment identifier. The ratio of cumulative completion quantity to planned completion quantity is determined as the production task progress, and the production task identifier, participating equipment identifier, and production task progress are retained in the same production task record. When a production task spans multiple sampling times, the corresponding production task identifier is continuously retained from the task start time to the task completion record time, and the production task progress is updated as the cumulative completion quantity changes.

[0025] When forming the historical operation sequence of equipment, the timestamps of each data source are converted to a unified time format within the same time zone, and the sampling time of the equipment operation data is used as the sorting benchmark. In this embodiment, the unified sampling interval is set to 1 second. For data with an original sampling interval shorter than 1 second, the average value is calculated based on the valid records within each 1-second sampling interval; for data with an original sampling interval longer than 1 second, the most recent valid value is retained until the next valid record appears. Energy consumption records, control action records, and production task records are respectively assigned to the corresponding equipment according to the equipment identifier and assigned to the first equipment operation sampling position no earlier than the time the record occurred. Each sampling position stores the equipment identifier, sampling time, equipment operating status, operating parameters, energy consumption, control parameter values, control action identifier, production task identifier, and production task progress, which are arranged in ascending order along the sampling time to form the historical operation sequence of equipment.

[0026] When establishing relationships between devices, the energy medium identifier, energy output device identifier, output interface identifier, energy input device identifier, input interface identifier, and transmission direction are retrieved from the device ledger or energy pipeline connection table. Two devices with directly connected output and input interfaces, the same energy medium, and consistent transmission direction are paired. The device belonging to the energy output is connected to the corresponding energy input device, and the device belonging to the energy output is recorded as the preceding device, and the device belonging to the energy input is recorded as the succeeding device. When one energy output interface connects to multiple energy input interfaces, corresponding device connections are established separately. When multiple energy output interfaces converge to the same energy input interface, the connections from each energy output device to the energy input device are retained, and the relationships between all device connections constitute the relationships between the devices.

[0027] In this embodiment, S2 includes: S21. Compare adjacent values ​​of the same control parameter one by one along the historical operating sequence of the equipment, determine the position where the control parameter begins to deviate from the original value as the starting point of the change, and determine the position where the control parameter stops changing continuously as the ending point of the change; S22. Extract equipment status and energy consumption data from the starting point of the change along the decreasing time direction, and extract equipment status and energy consumption data from the ending point of the change along the increasing time direction, while retaining the control action data between the starting point and the ending point of the change. S23. Connect the equipment status and energy consumption data before the control parameter change, the control action data, and the equipment status and energy consumption data after the control parameter change in time sequence to form an adjustment response segment.

[0028] In the specific implementation process, the historical operating sequence of the equipment is scanned item by item according to the equipment identifier and control parameter identifier. The sampling resolution of the control parameters is read from the parameter configuration record of the equipment controller; if the parameter configuration record does not record the sampling resolution, the non-zero adjacent changes of the same control parameter in the historical operating sequence are counted, and the smallest non-zero adjacent change is taken as the sampling resolution. For discrete control parameters such as start / stop status, gear position, or operating mode, the difference between two adjacent valid gear positions is taken as the sampling resolution.

[0029] Compare the values ​​of the same control parameter at adjacent sampling positions. When the absolute value of the difference between adjacent values ​​reaches the corresponding sampling resolution, a parameter change marker is written at the next sampling position. When a parameter change marker appears for the first time after three consecutive sampling positions without being written, the position of the first appearance of the parameter change marker is determined as the change start point. When parameter change markers appear consecutively, the last parameter change marker in the consecutive change segment is determined as the change end point. When a single step adjustment generates only one parameter change marker, the change start point and change end point are taken from the same sampling position.

[0030] Equipment status and energy consumption data are extracted from the starting point of the change along a decreasing time direction. The extraction process continues until the control parameters remain unchanged for five consecutive sampling positions and the equipment operating mode remains unchanged; the extracted data is then used as the data before the control parameter change. If the historical data length is less than five sampling positions, all continuous data obtainable before the starting point of the change is extracted. Equipment status includes equipment start / stop status, operating mode, load, and operating parameters corresponding to the control parameters; energy consumption data uses the energy consumption within each sampling interval.

[0031] Equipment status and energy consumption data are extracted from the endpoint of the change along the time-increasing direction. Extraction in the time-increasing direction ends when adjacent changes in equipment status parameters are less than the sampling resolution of the corresponding measuring point, and adjacent changes in energy consumption per unit sampling interval are less than the metering resolution of the energy metering device, and this continues for five consecutive sampling positions. If another control action occurs during extraction, extraction ends before the location where the other control action occurs to avoid responses from different control actions entering the same adjustment response segment. The measuring point sampling resolution and metering resolution are read from the measuring point configuration record and the energy metering device configuration record, respectively.

[0032] The control action data between the start and end points of the change should retain at least the control parameter identifier, the value before adjustment, the value after adjustment, and the time of action occurrence. When the time of action occurrence of the control action record falls between the sampling times of two devices, the control action record should be assigned to the first sampling position no earlier than the time of action occurrence.

[0033] After the data capture is completed, with the starting point of the change as the relative zero point, the data before the control parameter change are arranged in negative order, the control action data between the starting point and the ending point of the change are arranged in occurrence order, and the equipment status and energy consumption data after the ending point of the change are arranged in positive order. Each sampling position retains the equipment identifier, relative position, sampling time, control parameter value, equipment status parameter, and energy consumption, which are then connected along the time sequence to form a regulation response segment.

[0034] In this embodiment, S3 includes: S31. Extract the equipment status parameters, production task progress and load change of the adjustment response segment before the adjustment action occurs, and extract candidate operation segments from the equipment historical operation sequence of the same equipment that have not experienced corresponding adjustment actions. S32. Convert the equipment status parameters, production task progress and load change to the corresponding value range respectively, calculate the difference between each candidate operation segment and the adjustment response segment, and select the candidate operation segment with the smallest total difference as the unadjusted response segment. S33. Convert the production task progress in the adjusted response segment and the unadjusted response segment into relative progress respectively, and pair the sampling positions with the same relative progress bit by bit to complete the alignment of the equipment operation process.

[0035] In the specific implementation process, all sampling positions before the occurrence of the adjustment action are extracted from the adjustment response segment, and the corresponding equipment status parameters, production task progress, and load are obtained for each sampling position. Equipment status parameters are selected as continuous operating parameters with valid sample values ​​within the adjustment response segment; start / stop status, operating mode, and control method are retained according to discrete states. The average value of each continuous equipment status parameter before the occurrence of the adjustment action is taken as the equipment status parameter before adjustment; the production task progress at the last sampling position before the occurrence of the adjustment action is taken as the production task progress before adjustment; and the load difference between the first and last sampling positions before the occurrence of the adjustment action is taken as the load change.

[0036] In the historical operation sequence of the same equipment, each historical sampling position is used as a candidate starting position to extract continuous operation data with the same production task progress span as the adjustment response segment. The starting and ending production task progress of the candidate operation segment correspond to the starting and ending production task progress of the adjustment response segment, respectively. If the historical sampling position does not cover the complete production task progress span, the corresponding operation data is deleted. The control action records within the candidate operation segments are read one by one. If there is a control action in the candidate operation segment with the same control parameter identifier as the current adjustment action, the corresponding candidate operation segment is deleted. If the equipment operation mode of the candidate operation segment is inconsistent with the equipment operation mode before the adjustment action occurred, the corresponding candidate operation segment is deleted.

[0037] For each candidate operation segment, the same value position as the adjustment response segment is used to obtain the average value of equipment state parameters, production task progress, and load change before the corresponding position of the adjustment action. For continuous equipment state parameters, production task progress, and load change, the effective minimum and maximum values ​​of the corresponding parameters are read from the equipment's historical operation sequence for the same equipment. The parameter value is subtracted from the effective minimum value, and then divided by the difference between the effective maximum and effective minimum values ​​to convert it to a value range of 0 to 1. When the effective maximum and effective minimum values ​​are the same, the conversion result of the corresponding parameter is recorded as 0 and is not included in the difference ranking among candidate operation segments. Discrete states do not undergo numerical conversion, and candidate operation segments with inconsistent discrete states are directly deleted.

[0038] The absolute differences in equipment state parameters, production task progress, and load change between candidate running segments and regulation response segments are calculated item by item. All absolute differences are summed to obtain the total difference for each candidate running segment. When multiple candidate running segments have the same total difference, the segment whose occurrence time is closest to the regulation response segment is selected first; if they still cannot be distinguished, the segment with the higher sampling data integrity rate is selected. The candidate running segment with the smallest total difference is determined as the unregulated response segment. When all candidate running segments are deleted, the corresponding regulation response segment is marked as an unmatched segment, and equipment running process alignment, common change deduction, and improved HAVOK model processing are not performed on the corresponding regulation response segment.

[0039] During equipment operation process alignment, the starting and ending production task progress of both the adjusted and unadjusted response segments are read. The production task progress at each sampling position is subtracted from the starting production task progress of its corresponding segment, and then divided by the difference between the ending and starting production task progress of that segment to obtain a relative progress between 0 and 1. When the starting and ending production task progresses are the same, the relative progress is calculated according to the sampling position's position within the segment.

[0040] Using the sampling positions in the regulated response segment as a reference, find sampling positions with the same relative progress in the unregulated response segment. If no two relative progresses are exactly the same, select the two sampling positions with the smallest relative progress difference, and calculate the corresponding equipment status parameters and energy consumption data according to the distance ratio between the two relative progresses. Pair the sampling positions with the same relative progress or those obtained by proportional calculation to form a correspondence between the equipment operation progress of the regulated response segment and the unregulated response segment.

[0041] After identifying the unadjusted response segment, starting from the device acting on the adjustment action, read the device identification of the energy output device or the device connection whose energy input device identification is the same as the current device identification, and obtain the device identification within the connection range level by level. For each obtained device identification, extract the device status and energy consumption data from the time range corresponding to the adjustment action and the time range corresponding to the unadjusted response segment, respectively.

[0042] For devices other than those performing the adjustment action, candidate operation segments are no longer sorted separately; instead, the start and end times of the unadjusted response segment selected by the device performing the adjustment action are directly used. Data for each device is paired position by position according to the relative progress of the same production task. For each relative progress position, a set of device status parameters and energy consumption data is saved according to the device identifier, ensuring that the adjustment response segment, unadjusted response segment, preliminary adjustment response, and newly added adjustment response all include the same device range.

[0043] In this embodiment, S4 includes: S41. The last sampling position before the adjustment action occurs in the adjustment response segment is determined as the first reference position, and the position in the unadjusted response segment that is the same as the device operation process of the first reference position is determined as the second reference position. S42. Subtract the corresponding reference position's equipment status parameters and energy consumption data from the equipment status parameters and energy consumption data at each sampling location to form an adjusted change sequence and an unadjusted change sequence. S43. At the same position in the equipment operation process, subtract the corresponding changes in the unadjusted change sequence from the equipment state changes and energy consumption changes in the adjustment change sequence to form a preliminary adjustment response.

[0044] In the specific implementation process, the equipment historical operation sequence adopts a 1-second sampling interval. Each sampling position saves the equipment identifier, sampling time, relative progress of production task, load, temperature, pressure, speed, flow rate, valve opening, and energy consumption per unit sampling interval. After the regulated response segment and the unregulated response segment are aligned with the equipment operation progress, a sampling position pairing table is generated. Each row of the sampling position pairing table saves a relative progress, the sampling position of the regulated response segment, and the sampling position of the unregulated response segment; when there is no corresponding relative progress for the unregulated response segment, the pairing table simultaneously saves the two sampling positions before and after the relative progress and the interpolation ratio.

[0045] Taking a control action occurring at 120 seconds as an example, if the sampling position at 119 seconds simultaneously has complete equipment status parameters and energy consumption data, then the 119-second sampling position is determined as the first reference position. If the control action occurrence time coincides with the equipment operation data sampling time, the sampling data from the control action occurrence position is not used; the previous sampling position is used instead. If there are no values ​​at the 119-second sampling position, the system searches bit by bit along the decreasing time direction for the nearest sampling position with complete equipment status parameters and energy consumption data from 120 seconds, and this found sampling position is determined as the first reference position. The relative progress of the production task at the first reference position is recorded as the first reference progress.

[0046] The first reference progress is found in the sampling position pairing table. When unadjusted response segments have the same relative progress, the sampling position with the same relative progress is recorded as the second reference position. When unadjusted response segments do not have the same relative progress, the two sampling positions with the closest distance to each other on both sides of the first reference progress are selected.

[0047] Equipment status parameters are arranged in a fixed channel order: load, temperature, pressure, speed, flow rate, and valve opening. Start-up / shutdown status and operating mode are not included in the numerical subtraction but are only retained in the sampling location. Regulated and unregulated response segments use the same channel order, unit of measurement, and decimal precision. Load is uniformly expressed in kW, temperature in ℃, pressure in MPa, speed in r / min, flow rate in t / h, and valve opening as a percentage. Energy consumption is retained in units of measurement according to energy type, and the same unit of measurement is used in both regulated and unregulated response segments for the same energy metering channel. This step only performs baseline subtraction and common change subtraction for the same energy metering channel; it does not perform conversion for different energy types.

[0048] For each sampling position in the adjustment response segment, subtract the value of the same channel at the first reference position from the value of each channel at the current sampling position. Arrange the changes in each channel and the changes in energy consumption at all sampling positions in ascending order of relative progress to form an adjustment change sequence.

[0049] For each paired position in the unadjusted response segment, the channel values ​​at the current paired position are subtracted from the values ​​of the same channel at the second reference position. When a paired position is obtained by interpolation, the interpolation ratio stored in the sampling position pairing table is first used to calculate the channel values ​​and energy consumption data, and then the reference subtraction is performed. The subtraction results are arranged according to the same relative progress and channel order as the adjusted change sequence to form the unadjusted change sequence.

[0050] Read the sampling location pairing table line by line. For each relative progress and each equipment status parameter channel, subtract the change in the unadjusted change sequence from the change in the adjusted change sequence; for the energy consumption channel, subtract the energy consumption change in the unadjusted change sequence from the energy consumption change in the adjusted change sequence. For example, if the load change in the adjusted change sequence is -40kW and the load change in the unadjusted change sequence is -15kW under a certain relative progress, then the load change in the initial adjustment response is written as -25kW; if the energy consumption change in the adjusted change sequence is -0.012kWh and the energy consumption change in the unadjusted change sequence is -0.004kWh, then the energy consumption change in the initial adjustment response is written as -0.008kWh.

[0051] For each preliminary regulation response sampling location, the load difference, temperature difference, pressure difference, speed difference, flow difference, valve opening difference, and energy consumption difference are stored separately according to the equipment identification. All sampling locations are arranged in ascending order of relative progress to form the preliminary regulation response. Negative values ​​indicate a decrease in regulation action relative to the unregulated operation, while positive values ​​indicate an increase in regulation action relative to the unregulated operation.

[0052] In this embodiment, the improved HAVOK model includes a delayed coordinate input layer, a coordinate decomposition layer, a forced state recursion layer, a state vector evolution layer, and a response output layer. The delayed coordinate input layer will arrange the initial adjustment response in a staggered manner along the sampling position to form a Hankel matrix; The coordinate decomposition layer performs singular value decomposition on the Hankel matrix, and forms a state vector from the remaining multiple delayed coordinates except for the last one, and determines the last delayed coordinate as the forced coordinate. The forced state recursion layer evolves the forced state of the previous sampling position through the forced recursion matrix, and transforms the forced coordinates of the current sampling position through the coordinate mapping matrix, forming the continued forced state and the forced state of the current sampling position. The state vector evolution layer processes the state vector of the previous sampling position through the state evolution matrix, and combines the processing results with the results obtained from the forced state and the forced state processing by the forced coupling matrix to form the continued state vector and the total state vector. The response output layer converts the continued state vector and the total state vector into the response of the preceding adjustment action and the total adjustment response respectively through the response reconstruction matrix. The response of the preceding adjustment action is subtracted from the total adjustment response, and the new adjustment response of the current adjustment action is output.

[0053] In this embodiment, the formation of the continued forced state and forced state of the current sampling position includes: Multiply the forced state of the previous sampling position by the forced recursion matrix to obtain the continued forced state of the current sampling position; Multiply the forced coordinates of the current sampling position with the coordinate mapping matrix to transform the forced coordinates to the state space where the forced state continues; Construct a projection matrix pointing to the subspace where the continued forced state is located and the corresponding complementary projection matrix. Extract the continued correction component from the transformed forced coordinates through the projection matrix and extract the new forced component through the complementary projection matrix. The continued correction component is incorporated into the continued forced state, and the corrected continued forced state and the newly added forced component are combined to form the forced state of the current sampling position. The corrected continuation forced state and the forced state at the current sampling position are passed to the state vector evolution layer, and the forced state at the current sampling position is used as the recursive input of the forced state at the next sampling position.

[0054] In practice, the original HAVOK model misplaces the time series data into the Hankel matrix, performs singular value decomposition on the Hankel matrix, treats the first-order retained coordinates as linear states, and the last-order retained coordinates as external forcing coordinates, directly incorporating the forcing coordinates of the current sampling position into the state evolution process. When multiple adjustment actions occur consecutively, the dynamic response formed by the preceding adjustment action may still extend to the current sampling position. The original HAVOK model does not separately save the propagation state of the forcing action between adjacent sampling positions. The continuation and new parts contained in the current forcing coordinates will both enter the state vector, making it difficult to directly separate the new response added by the current adjustment action.

[0055] The improved HAVOK model retains the Hankel matrix, singular value decomposition, and linear state evolution structure, adding a forced state recursion layer between the forced coordinates and the state vector. This layer first propagates the forced state from the previous sampling position to the current sampling position, then decomposes the current forced coordinates into a continuation correction component pointing to the continuing forced state and a new forced component located in the complementary space. The corrected continuing forced state and the current forced state containing the new forced component drive the state vector evolution, forming the continuing state vector and the total state vector, respectively. These two state vectors are transformed into the continuation response and the total regulation response of the preceding regulation action through the same response reconstruction matrix. The two responses are subtracted from each sampling position and each parameter channel to obtain the new regulation response of the current regulation action. The improved model preserves the propagation relationship of the forced action between adjacent sampling positions and separates the continuation response of the preceding regulation action from the current total response, reducing confusion in response attribution between consecutive regulation actions.

[0056] Multiple preliminary adjustment responses are arranged according to the time of occurrence of the adjustment action. Within each preliminary adjustment response, the equipment is first arranged in ascending order of equipment identification, and then the parameter channels of the same equipment are arranged in the order of load difference, temperature difference, pressure difference, speed difference, flow difference, valve opening difference, and energy consumption difference. The same equipment identification order and parameter channel order are used for the same equipment range. The parameter channel values ​​of all equipment are connected at each sampling position in the aforementioned order to form the input vector of the corresponding sampling position. The same input vector length and channel position are maintained in training, verification, testing, and actual operation.

[0057] Before model identification, the preliminary adjustment responses are grouped according to the device identifier of the adjusting action and the set of device identifiers contained in the preliminary adjustment response. Preliminary adjustment responses with the same device identifier of the adjusting action and the same set of device identifiers are grouped into the same model data group. The same set of device identifiers means that the number of device identifiers and each device identifier is the same, regardless of the order of the device identifiers in the original record.

[0058] Each model data set determines the input device order according to the ascending order of device identifiers, and separately determines the parameter channel order, channel average, and channel standard deviation. Training, validation, and testing Hankel matrices are then constructed for each set. When processing the initial adjustment response, the device identifiers and device identifier sets for the adjustment action are read, and the improved HAVOK model, where both the device identifiers and device identifier sets are identical, is invoked. Different model data sets do not connect Hankel sub-matrices and do not share state evolution matrices, forced coupling matrices, or response reconstruction matrices.

[0059] Before model identification, the mean and standard deviation of each parameter channel in the training data are calculated separately. Each original channel value is subtracted from its corresponding mean and then divided by its corresponding standard deviation to form a dimensionless input value. Parameter channels with a standard deviation of 0 are not written into the model input; during response output, the new change in the corresponding parameter channel is recorded as 0, and the data is backfilled according to the original device identification order and parameter channel order. After training is complete, the results of each channel generated by the response output layer are multiplied by their corresponding standard deviation and added back to their corresponding mean to restore the original units of measurement.

[0060] For each model data set, the initial conditioning responses are arranged in ascending order of the time of the conditioning action. The first 70% are assigned to the training set, the middle 15% to the validation set, and the remainder to the test set. The same initial conditioning response is not split across different data sets. The training set generates a Hankel matrix and calculates the model matrix. The validation set selects the delay length and regularization coefficient. The test set checks for state evolution error and response reconstruction error. If the number of adjacent sampling positions generated in the training set is less than the sum of the state vector dimension and the forced state dimension, model identification for the corresponding model data set is not performed.

[0061] The delay coordinate input layer constructs a Hankel matrix using consecutive sampling positions. In one embodiment, the sampling interval for device data is 1 second, with 10, 20, and 30 sampling positions used as candidate delay lengths, respectively. When the delay length is 20, all parameter channel values ​​are obtained from the 1st to the 20th sampling positions of the initial adjustment response, and arranged sequentially according to the parameter channel order and sampling position order as the first column of the Hankel matrix; the 2nd to the 21st sampling positions are arranged in the same order as the second column; this process is repeated, moving one sampling position backward until the end of the initial adjustment response. If the number of sampling positions in an initial adjustment response is less than the candidate delay length, it does not participate in the model identification for the corresponding delay length.

[0062] In the training set, each initial conditioning response forms a Hankel submatrix, and these Hankel submatrixes are concatenated using the same row order to form the training Hankel matrix. No cross-segment delay columns are constructed between adjacent initial conditioning responses. In the validation set, three candidate delay lengths are examined, and the root mean square error (RMSE) between the model output and the initial conditioning response is calculated for each delay length. The delay length with the smallest RMSE is selected; if the RMSEs are the same, the delay length with the smaller value is selected.

[0063] The coordinate decomposition layer performs singular value decomposition on the training Hankel matrix, obtaining the left singular vector, singular values, and right singular vector. The squares of the singular values ​​are accumulated in descending order of value. When the accumulated value reaches 99% of the sum of the squares of all singular values, the minimum number of coordinates required to reach 99% is retained. If the number of retained coordinates is less than 4, 4 is selected; if it is greater than 12, 12 is selected. The last coordinate in the retained coordinates is determined as the forced coordinate, and all coordinates before the last coordinate are used to form the state vector.

[0064] The training Hankel matrix is ​​projected onto the retained left singular vector to obtain the state vector and forced coordinates at each sampling position. The state vectors are arranged in ascending order according to the sampling position to form a state vector sequence; the forced coordinates are arranged according to the same sampling position to form a forced coordinate sequence. In the initial adjustment response, the state is not passed between different adjustment actions. When entering the first valid sampling position of each initial adjustment response, the forced state is initialized to an all-zero vector.

[0065] Four state channels are set up for the forced state, each storing the continuation results of the forced action at different recursion rates. The forced recursion matrix is ​​a 4x4 diagonal matrix, with the four diagonal elements set to 0.25, 0.50, 0.75, and 0.90 respectively, and the off-diagonal elements set to 0. After multiplying the forced state at the previous sampling position with the forced recursion matrix, the four forced state channels retain 25%, 50%, 75%, and 90% of the previous forced state, respectively, forming the continuation forced state at the current sampling position.

[0066] The coordinate mapping matrix adopts a 4x1 matrix structure, with all four matrix elements set to 0.5. The forced coordinates of the current sampling position are multiplied by the coordinate mapping matrix, and then simultaneously written into all four forced state channels to form the current forced coordinate mapping result. The sum of the squares of all elements in the coordinate mapping matrix is ​​1 to prevent the mapping process from altering the overall amplitude of the forced coordinates.

[0067] The continued forced state and the current forced coordinate mapping result occupy the same four state channels. The current forced coordinate mapping result can produce non-zero projections in the direction of the continued forced state and the corresponding complementary direction.

[0068] When the continued forced state is not a vector of all zeros, multiply the continued forced state by its transpose to form a 4x4 direction matrix. Divide the direction matrix by the sum of the squares of each channel of the continued forced state and 10 to the power of -8 to form a projection matrix pointing to the one-dimensional subspace of the continued forced state. Subtract the projection matrix from the 4x4 identity matrix to obtain the complementary projection matrix. Adding 10 to the power of -8 avoids division by zero when the continued forced state is close to all zeros.

[0069] Multiplying the projection matrix by the current forced coordinate mapping result yields the continuation correction component; multiplying the complementary projection matrix by the current forced coordinate mapping result yields the new forced component. When the continuation forced state is an all-zero vector, the continuation correction component is directly written as an all-zero vector, and the entire current forced coordinate mapping result is written into the new forced component.

[0070] The continuation correction component is added channel by channel to the continuation forced state to form the corrected continuation forced state. Then, the newly added forced component is added channel by channel to the corrected continuation forced state to form the forced state at the current sampling position. The corrected continuation forced state is passed to the continuation response branch of the state vector evolution layer, and the forced state at the current sampling position is passed to the total response branch of the state vector evolution layer. The forced state at the current sampling position is simultaneously written into the recursive input of the forced state at the next sampling position.

[0071] The state evolution matrix and the forced coupling matrix are identified using training group data. Adjacent sampling positions within the training group are combined to form training samples. Each training sample includes the state vector of the previous sampling position, the forced state of the current sampling position, and the state vector of the current sampling position. The state vector of the previous sampling position and the forced state of the current sampling position are concatenated column-wise to form the input sample; the state vector of the current sampling position is used as the target sample. After merging all training samples, a least-squares solution with regularization is performed. The matrix portion of the input sample corresponding to the previous state vector forms the state evolution matrix, and the matrix portion corresponding to the forced state forms the forced coupling matrix.

[0072] The regularization coefficients were set to 10 to the power of -8, 10 to the power of -6, and 10 to the power of -4, respectively. Three sets of state evolution matrices and forced coupling matrices were obtained using each regularization coefficient. The state vector prediction results were calculated at each sampling position in the validation set, and the root mean square error (RMSE) between the predicted state vector and the state vector obtained from singular value decomposition was calculated. The matrix with the smallest RMSE was selected; when the errors were the same, the larger regularization coefficient was selected to limit the excessive increase of matrix elements.

[0073] The response reconstruction matrix is ​​also obtained using the training group data. The state vectors at each sampling location in the training group are used as input, and the standardized preliminary adjusted response at the corresponding sampling location is used as the objective. A least-squares solution with a regularization term is performed to obtain the response reconstruction matrix. The regularization coefficients of the response reconstruction matrix use the same candidate values ​​as the state evolution matrix, and are selected using the root mean square error of the response reconstruction from the validation group.

[0074] When the model processes each initial adjustment response, it reads the state vector output by the coordinate decomposition layer at the last valid sampling position before the adjustment action occurs, and writes the corresponding state vector into the continuation state branch and the total state branch at the same time, as the initial state vector of the two branches.

[0075] Upon entering the next sampling position, the state evolution matrix processes the continuation state vector of the previous sampling position, and the forced coupling matrix processes the corrected continuation forced state of the current sampling position. The results of the two processes are added channel by channel to form the continuation state vector of the current sampling position.

[0076] The state evolution matrix processes the total state vector of the previous sampling position, while the forced coupling matrix processes the forced state of the current sampling position. The results of the two processes are added channel by channel to form the total state vector of the current sampling position.

[0077] The response reconstruction matrix processes the continuing state vector and the total state vector at the current sampling position to obtain the standardized preceding conditioning action continuing response and the total conditioning response. The preceding conditioning action continuing response and the total conditioning response use the same device identification order and parameter channel order as the model input. Each parameter channel value in both responses is multiplied by the corresponding standard deviation recorded before model identification, and then the corresponding average value is added to restore the original units of measurement.

[0078] By subtracting the values ​​of the same devices and the same parameter channels in the continuation response of the preceding regulation action from the values ​​of each parameter channel of each device in the overall regulation response, we can obtain the increase in the change of state parameters and the increase in the change of energy consumption at the current sampling location.

[0079] The continuation state vector at the current sampling position is used as the previous state vector of the continuation state branch at the next sampling position, the total state vector at the current sampling position is used as the previous state vector of the total state branch at the next sampling position, and the forced state at the current sampling position is used as the previous forced state at the next sampling position. The continuation state branch, the total state branch, and the forced state are recursively processed along the sampling position sequence until the last position of the initial adjustment response.

[0080] Each output sampling position, according to the device identifier order and parameter channel order used in the model input, saves the new changes in the state parameters, the new changes in energy consumption, and the corresponding relative progress of each device. All output sampling positions are arranged in ascending order along the relative progress, forming the new adjustment response of the current adjustment action.

[0081] In this embodiment, S6 includes: S61. Starting from the device currently acting on the regulation action, trace the devices that have changed their state or energy consumption within the sampling location range corresponding to the newly added regulation response level by level along the correlation between devices, and determine the associated devices. S62. Extract the energy consumption changes of the current regulating action device and associated device at each sampling location from the newly added regulation response, classify the energy consumption increase as energy consumption increase, and classify the energy consumption decrease as energy consumption reduction. S63. Combine the energy consumption increase and energy consumption reduction of the current regulating action device and related devices, and subtract the combined energy consumption increase from the combined energy consumption reduction to obtain the system net energy saving response corresponding to the current regulating action.

[0082] In the specific implementation process, each newly added regulation response saves the regulation action identifier, the identifier of the device involved in the regulation action, the starting sampling position, the ending sampling position, and the new changes in device status and energy consumption of each device at the corresponding sampling position. The starting sampling position is the location where the current regulation action occurred.

[0083] Starting from the initial sampling position, compare each energy metering channel of the currently regulating device and all installed devices position by position. If, within five consecutive sampling positions, the absolute difference between adjacent increases in energy consumption for all energy metering channels of all devices does not exceed the metering resolution of the corresponding energy metering device, the last sampling position among the five consecutive sampling positions is determined as the end sampling position. If any energy metering channel of any device has not yet met the aforementioned condition, continue comparing. If the newly added regulating response reaches the end of the segment but still does not meet the aforementioned condition, the end of the segment is determined as the end sampling position.

[0084] The relationships between devices are stored using a connection table. Each connection record includes the energy output device identifier, energy input device identifier, energy medium identifier, and transmission direction. A list of devices to be tracked and a set of visited devices are established, and the device currently performing the adjustment action is simultaneously written to both the list of devices to be tracked and the set of visited devices.

[0085] Each time, a device is retrieved from the first position of the list of devices to be tracked. Connection records where the energy output device identifier matches the current device identifier are read, and connection records where the energy input device identifier matches the current device identifier are also read. For each connection record, the device identifier of the other end of the connection is obtained. If the other end device has not yet been written to the visited device set, it is simultaneously written to both the list of devices to be tracked and the visited device set.

[0086] For each device removed from the list of devices to be tracked, the changes in device status and energy consumption from the start to the end of the new adjustment response sampling position are read. If the absolute value of the change in any device status parameter reaches the sampling resolution of the corresponding measurement point, or the absolute value of the change in energy consumption at any sampling position reaches the metering resolution of the energy metering device, the corresponding device is added to the associated device set. If neither the device status parameter nor the change in energy consumption reaches the corresponding resolution, the corresponding device is not added to the associated device set, but all connection records for the corresponding device are still read.

[0087] Repeat the device retrieval, connection read, device determination, and write operations to the other end device until the list of devices to be tracked is empty. The visited device set prevents the same device from repeatedly entering the list of devices to be tracked and avoids cyclic tracking caused by energy loops. The device currently performing the adjustment action is retained separately and is not written to the associated device set.

[0088] The increase in energy consumption is calculated as the difference in energy consumption per unit sampling interval relative to the unadjusted operation process. When the increase in energy consumption at each sampling point is greater than 0, the corresponding value is written as the increase in energy consumption, and the decrease in energy consumption is written as 0; when the increase in energy consumption is less than 0, the absolute value is written as the decrease in energy consumption, and the increase in energy consumption is written as 0; when the increase in energy consumption is equal to 0, both the increase in energy consumption and the decrease in energy consumption are written as 0.

[0089] When equipment is configured with multiple energy metering channels, the new energy consumption changes for electricity, steam, natural gas, compressed air, or other energy types are retained separately. Each energy type is first converted to a unified standard energy quantity before being merged. Electricity is converted according to the standard energy conversion factor; steam is converted according to steam energy consumption and the corresponding standard energy conversion factor; natural gas is converted according to gas volume, lower heating value, and the corresponding standard energy conversion factor. Each conversion factor is read from the same version of the enterprise energy management ledger or applicable energy conversion standard and remains unchanged during the generation of energy-saving strategies for the same batch of equipment. When all energy channels already use the same energy metering unit, the original new energy consumption changes are used directly.

[0090] For the device currently undergoing adjustment and all associated devices, the energy consumption increase is accumulated bit by bit from the starting sampling position to the ending sampling position to obtain the cumulative energy consumption increase for the corresponding device; the energy consumption reduction is also accumulated bit by bit to obtain the cumulative energy consumption reduction for the corresponding device. If the energy consumption per unit sampling interval is already stored at the sampling position, it is directly accumulated.

[0091] The total increase in system energy consumption is obtained by combining the cumulative energy consumption increase of the device affected by the current adjustment action and all associated devices. The total energy consumption reduction is then obtained by combining the cumulative energy consumption reduction of the device affected by the current adjustment action and all associated devices. The total system energy consumption reduction is then subtracted from the total system energy consumption increase to obtain the net energy-saving response of the current adjustment action. A net energy-saving response greater than 0 indicates that the current adjustment action generates net energy savings within the associated device range; a net energy-saving response less than 0 indicates that the increased energy consumption of the associated devices exceeds the reduced energy consumption.

[0092] In this embodiment, S7 includes: S71. Determine the state of the last device before the adjustment action occurs as the entry state, determine the state of the last device within the sampling position range corresponding to the newly added adjustment response as the end state, and encapsulate the adjustment action, entry state, end state and system net energy saving response into an adjustment action unit. S72. Pair the end state of the previous adjustment action unit with the entry state of the next adjustment action unit according to the device identifier and status parameters, and connect the adjustment action units with the same device identifier and the same or different status parameter values, or whose differences do not exceed the sampling resolution of the corresponding status parameters, to form a status connection relationship. S73. Determine the last device state in the historical operation sequence of the device as the starting device state. Pair the entry state of each adjustment action unit with the starting device state according to the device identifier and status parameters. Select the adjustment action unit with the largest system net energy saving response from the adjustment action units with the same device identifier, the same status parameter value or the difference does not exceed the corresponding status parameter sampling resolution and the system net energy saving response is greater than zero. S74. Remove the already arranged adjustment action units, and from the remaining adjustment action units that have a state connection relationship with the current adjustment action unit and whose net energy saving response of the system is greater than zero, successively select the adjustment action unit with the largest net energy saving response of the system and continue to arrange them until there are no adjustment action units that can be connected, thus forming the execution sequence of the adjustment action. S75. Arrange the adjustment actions in each adjustment action unit according to the execution sequence to generate an energy-saving strategy for the equipment.

[0093] In the specific implementation process, a regulation action unit record is established for each regulation action. The regulation action unit record includes regulation action identifier, acting equipment identifier, control parameter identifier, value before adjustment, value after adjustment, time of action, entry status, exit status, and system net energy saving response. The entry status and exit status only save the equipment status of the acting equipment, and the fields are in the following order: equipment identifier, start / stop status, operating mode, load, temperature, pressure, speed, flow rate, and valve opening.

[0094] When the adjustment action occurs between two sampling positions, the nearest sampling position before the action occurrence time is taken as the entry state position. When the adjustment action occurs at the same time as the sampling time, the sampling position preceding the action occurrence position is used. When parameters are missing at the entry state position, the sampling position closest to the action occurrence time with complete state parameters is searched sequentially along the decreasing time direction. The device status of the device acting on the adjustment action is then read to form the entry state.

[0095] The end state position is taken as the last sampling position within the sampling position range corresponding to the newly added adjustment response. If the last sampling position has missing state parameters, the sampling position with complete state parameters that is closest to the last sampling position is searched bit by bit along the decreasing time direction. The equipment status of the device that performed the adjustment action is read to form the end state.

[0096] After completing the entry and exit status writing, the system net energy-saving response is written to the same adjustment action unit. A set of adjustment action units to be arranged is established. All adjustment action units are read one by one. Adjustment action units with a system net energy-saving response greater than 0 are written to the set of adjustment action units to be arranged. Adjustment action units with a system net energy-saving response equal to or less than 0 are retained in the adjustment action unit record, but are not written to the set of adjustment action units to be arranged.

[0097] When establishing state connection relationships, one adjustment action unit is selected sequentially from the set of adjustment action units to be arranged as the previous adjustment action unit. Then, all adjustment action units in the set except the previous adjustment action unit are read one by one as the next adjustment action unit. State pairing is performed for each group of previous and next adjustment action units. First, the device identifiers in the ending state of the previous adjustment action unit and the entering state of the next adjustment action unit are compared. If the device identifiers in the two sets of states are inconsistent, no connection is established; if the device identifiers are consistent, they are compared item by item according to the device identifier and fixed state parameters.

[0098] Start-stop status and operating mode are discrete states; the corresponding discrete state in the ending state of the previous regulating action unit must be exactly the same as the corresponding discrete state in the entering state of the next regulating action unit. Load, temperature, pressure, speed, flow rate, and valve opening are continuous state parameters; calculate the absolute difference between the parameter values ​​in the ending state of the previous regulating action unit and the parameter values ​​in the entering state of the next regulating action unit.

[0099] The sampling resolution for each continuous status parameter is read from the device measurement point configuration table. The measurement point configuration table must at least store the device identifier, status parameter identifier, unit of measurement, and sampling resolution.

[0100] When all discrete states are identical, and the absolute difference of each continuous state parameter does not exceed the corresponding sampling resolution, a directed connection is established between the previous and subsequent adjustment units. The direction of the directed connection is from the previous adjustment unit to the next adjustment unit. If any discrete state is different, or if the absolute difference of any continuous state parameter exceeds the corresponding sampling resolution, no connection is established.

[0101] For each adjustment action unit, the last complete device state of the corresponding device is read from the device's historical operation sequence according to the device identifier. This last complete device state is used as the starting device state when the corresponding adjustment action unit participates in the first and second action matching. If there are missing state parameters at the last sampling position, the sampling position of the corresponding device that is closest to the last sampling position and has complete state parameters is searched bit by bit along the time decreasing direction. The starting device state of different devices is read separately, and the states of multiple devices are not merged into the same state record.

[0102] The entry state of each regulating action unit is compared item by item with the initial state of the equipment with the same function. The equipment identification must be the same, the start / stop state and the operating mode must be the same, and the absolute differences in load, temperature, pressure, speed, flow rate and valve opening must not exceed the sampling resolution of the corresponding state parameter. The regulating action unit that meets all the conditions and whose net energy-saving response of the system is greater than 0 is added to the first candidate set.

[0103] The candidate set is arranged from largest to smallest according to the system's net energy-saving response, and the first regulating action unit is selected as the first regulating action unit. If the system's net energy-saving responses are the same, the selection is based on the character order of the regulating action identifier. If the first candidate set is empty, no equipment energy-saving strategy is generated.

[0104] Establish an execution bit sequence table, write the first and first adjustment action unit into the first position of the execution bit sequence table, delete the first and first adjustment action unit from the set of adjustment action units to be arranged, and determine the first and first adjustment action unit as the current adjustment action unit.

[0105] Read all directed connections originating from the current adjustment action unit and pointing to the adjustment action units to be arranged. If no corresponding directed connection exists, end the arrangement; if only one directed connection exists, select the adjustment action unit pointed to by the directed connection; if multiple directed connections exist, arrange the adjustment action units pointed to by the directed connections in descending order of system net energy-saving response, and select the first adjustment action unit in the arrangement. If the system net energy-saving responses are the same, select according to the character order of the adjustment action identifier.

[0106] The selected adjustment action unit is written to the last position of the execution bit sequence list and deleted from the set of adjustment action units to be arranged. The newly written adjustment action unit is then designated as the current adjustment action unit. This process of reading directed connections and selecting adjustment action units is repeated until there are no more directed connections pointing to the adjustment action units to be arranged. Adjustment action units are immediately deleted from the set of adjustment action units to be arranged after being written to the execution bit sequence list to avoid repeated arrangement of the same adjustment action unit due to circular connections.

[0107] After the arrangement is completed, the device identifier, control parameter identifier, pre-adjustment value, and post-adjustment value of the adjustment action unit are read bit by bit according to the execution position sequence table to form the equipment energy-saving strategy. Each strategy position stores the execution sequence number, adjustment action identifier, device identifier, control parameter identifier, target value, entry state, exit state, and system net energy-saving response. The execution sequence number starts from 1 and increments continuously.

[0108] refer to Figure 3 A device energy-saving strategy generation system based on big data processing includes the following modules: The data generation module is used to acquire equipment operation data, energy consumption data, control action data, and production task data, form historical operation sequences of equipment, and obtain the correlation between equipment. The response interception module is used to locate the position of control parameter change in the historical operation sequence of the equipment, and to intercept the changes in equipment status and energy consumption before and after the control parameter change to form an adjustment response segment. The segment matching module is used to match the unadjusted response segments that have not undergone corresponding adjustment actions with the adjustment response segments, and to align the adjustment response segments and unadjusted response segments according to the device operation process; The response extraction module is used to subtract the common changes between the aligned adjusted response fragments and the unadjusted response fragments to form a preliminary adjusted response; The response separation module is used to input the initial adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the evolution of the state vector. It separates the response that continues from the previous adjustment action to obtain the new adjustment response of the current adjustment action. The energy-saving calculation module is used to aggregate the energy consumption increase or decrease of the device currently performing the adjustment action and its associated devices along the correlation between devices, and obtain the system's net energy-saving response. The strategy generation module is used to connect multiple adjustment actions based on the equipment status before and after the adjustment action, and determine the execution order of multiple adjustment actions based on the corresponding system net energy saving response to generate equipment energy saving strategies.

[0109] Example 1: To verify the feasibility of this invention in practice, it was applied to a coating and laminating production line of a flexible packaging material manufacturing company. This production line mainly produces food packaging composite films, and the production process includes unwinding, coating, drying, laminating, and rewinding. The drying section is equipped with a six-zone hot air oven, circulating fans, exhaust fans, a natural gas burner, and a heat transfer fluid circulation pump. The natural gas consumption and fan power consumption in the drying section account for approximately 64% of the total energy consumption of the entire production line.

[0110] The company's original energy-saving control mainly relied on operators adjusting oven temperature, fan frequency, and exhaust ratio based on product specifications and production experience. The original control method typically only counted the energy consumption changes of the controlled equipment, without simultaneously calculating the energy consumption increase of related equipment. Furthermore, during continuous adjustment, it was easy to include the response of the previous adjustment action that had not yet been completed in the subsequent adjustment action, leading to an overestimation of energy savings.

[0111] During implementation, data was continuously collected for 30 days from the production line control system, energy management system, and production management system, with a uniform sampling interval of 1 second. The collected data included equipment start / stop status, operating mode, production speed, oven temperature, fan speed, valve opening, gas flow, electrical power, control parameter adjustment records, and production task progress. A total of 178 production batches were collected, yielding approximately 2.59 million valid operating records, and 1764 changes in control parameters were identified.

[0112] Based on the location of control parameter changes, equipment status and energy consumption data before and after the adjustment action were extracted, resulting in 1496 adjustment response segments. Each adjustment response segment was matched with unadjusted response segments that had similar production task progress, pre-adjustment equipment status, and load changes, but did not undergo the same adjustment action. A total of 1268 adjustment response segments were successfully matched, while the remaining 228 adjustment response segments were not included in the energy-saving strategy generation because no matching unadjusted response segments were found.

[0113] The adjusted response segments and unadjusted response segments are aligned position by position according to the relative progress of the production task. The baseline state before the adjustment action is taken is deducted from each segment, and the common changes between the two segments are also deducted to form the preliminary adjustment response. The preliminary adjustment response is grouped according to the equipment affected by the adjustment action and the scope of related equipment, and then input into the improved HAVOK model.

[0114] The improved HAVOK model arranges the initial regulation response into a Hankel matrix along consecutive sampling positions, and obtains the state vector and forced coordinates through singular value decomposition. The forced coordinates are written into the forced state recursively across sampling positions and decomposed into a continuation part of the previous regulation action and a newly added part at the current sampling position. The continuation forced state and the complete forced state drive two state evolution branches respectively, and then, through response reconstruction and channel-by-channel subtraction, the newly added regulation response corresponding to the current regulation action is obtained.

[0115] Based on the energy connection relationship of the equipment, the associated equipment that has undergone changes in status or energy consumption is tracked step by step. The energy consumption decrease of each equipment is classified as energy consumption reduction, and the energy consumption increase is classified as energy consumption increase. The total energy consumption reduction of the system is subtracted from the total energy consumption increase of the system to obtain the net energy saving response of each adjustment action.

[0116] Taking the adjustment of the temperature setpoint of the third drying zone from 92℃ to 88℃ as an example, the total energy consumption of the oven and exhaust fan decreases by 3.36 kWh, while the total energy consumption of the circulating fan, heat medium circulation pump, and related processing equipment increases by 1.08 kWh, resulting in a net energy saving response of 2.28 kWh. Based on the entry and exit states of the adjustment action and the net energy saving response of the system, the adjustment actions that can be continuously connected are arranged sequentially to generate an equipment energy-saving strategy for the corresponding production batch.

[0117] To compare the effectiveness of different methods, 90 production batches with the same product specifications, similar production speeds, and similar ambient temperatures were selected for verification. The methods employed included the company's existing experience-based control methods, the original HAVOK model method, and the method of this invention, with 30 production batches for each method. The statistical results are shown in Table 1.

[0118] Table 1. Comparison of the implementation effects of different energy-saving strategy generation methods for different equipment.

[0119] Among them, unit product energy consumption is the ratio of total energy consumption of production batch to length of qualified product; system net energy saving rate is the ratio of the difference between benchmark energy consumption and actual energy consumption to benchmark energy consumption; strategy completion rate is the ratio of the number of production batches that complete all arrangement adjustment actions to the total number of production batches of the corresponding method.

[0120] As shown in Table 1, the original empirical control method did not fully consider the energy consumption increase of associated equipment, and the system's net energy saving rate was 5.21%. The original HAVOK model could extract the dynamic response of the adjustment action, but it was insufficient in separating the continuous response between continuous adjustment actions, and the system's net energy saving rate was 8.56%. The method of this invention, through unadjusted response segment matching, common change deduction, forced state recursion, and associated equipment energy consumption aggregation, reduced the unit product energy consumption to 172.3 kWh / km, achieved a system net energy saving rate of 12.37%, and a strategy completion rate of 96.67%.

[0121] This embodiment demonstrates that the present invention can extract new responses to adjustment actions from historical equipment operating data, calculate actual net energy savings from a system-wide perspective, and generate an execution sequence of adjustment actions that is consistent with the equipment status. It is applicable to the generation of energy-saving strategies for production lines with multi-equipment energy correlation and continuous adjustment processes.

[0122] 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 generating equipment energy-saving strategies based on big data processing, characterized in that, Includes the following steps: S1. Acquire equipment operation data, energy consumption data, control action data, and production task data to form a historical operation sequence of the equipment and obtain the correlation between the equipment; S2. Locate the position of control parameter change in the historical operation sequence of the equipment, extract the equipment status and energy consumption changes before and after the control parameter change, and form an adjustment response segment; S3. Match the adjustment response segment with an unadjusted response segment that has not undergone a corresponding adjustment action, and align the adjustment response segment and the unadjusted response segment according to the device operation process; S4. Subtract the common changes between the aligned and unadjusted response segments to form the initial adjusted response; S5. Input the preliminary adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the state vector evolution to strip away the response that continues from the previous adjustment action and obtain the new adjustment response of the current adjustment action. S6. By combining the energy consumption increase or decrease of the device currently performing the adjustment action and the associated devices along the correlation between the devices, the net energy-saving response of the system is obtained; S7. Connect multiple adjustment actions according to the equipment status before and after the adjustment action, and determine the execution order of the multiple adjustment actions according to the corresponding system net energy saving response to generate an equipment energy saving strategy.

2. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S1 includes: S11. Collect equipment operation data, energy consumption data, and control action data with equipment identification and occurrence time, and obtain the participating equipment and production task progress corresponding to the production task. S12. Match various types of data with equipment identifiers, arrange the data of the same equipment according to the time of occurrence, and classify the production tasks and production task progress into the corresponding time of occurrence to form the equipment historical operation sequence. S13. Pair the energy outputs and energy inputs that are interconnected between different devices, connect the energy output device to the corresponding energy input device, and obtain the relationship between the devices.

3. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S2 includes: S21. Compare adjacent values ​​of the same control parameter one by one along the historical operation sequence of the equipment, determine the position where the control parameter begins to deviate from the original value as the starting point of the change, and determine the position where the control parameter stops changing continuously as the ending point of the change; S22. Extract device status and energy consumption data from the starting point of the change along the decreasing time direction, and extract device status and energy consumption data from the ending point of the change along the increasing time direction, while retaining the control action data between the starting point and the ending point of the change. S23. Connect the equipment status and energy consumption data before the control parameter change, the control action data, and the equipment status and energy consumption data after the control parameter change in time sequence to form an adjustment response segment.

4. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S3 includes: S31. Extract the equipment status parameters, production task progress and load change of the adjustment response segment before the adjustment action occurs, and extract candidate operation segments from the equipment historical operation sequence of the same equipment that have not experienced corresponding adjustment actions. S32. Convert the equipment status parameters, production task progress and load change to the corresponding value range, calculate the difference between each candidate operation segment and the adjustment response segment, and select the candidate operation segment with the smallest total difference as the unadjusted response segment. S33. Convert the production task progress in the adjusted response segment and the unadjusted response segment into relative progress respectively, and pair the sampling positions with the same relative progress bit by bit to complete the alignment of the equipment operation process.

5. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S4 includes: S41. The last sampling position before the adjustment action occurs in the adjustment response segment is determined as the first reference position, and the position in the unadjusted response segment that is the same as the device operation process at the first reference position is determined as the second reference position. S42. Subtract the corresponding reference position's equipment status parameters and energy consumption data from the equipment status parameters and energy consumption data at each sampling location to form an adjusted change sequence and an unadjusted change sequence. S43. At the same position in the equipment operation process, subtract the corresponding changes in the unadjusted change sequence from the equipment state changes and energy consumption changes in the adjustment change sequence to form a preliminary adjustment response.

6. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, The improved HAVOK model includes a delayed coordinate input layer, a coordinate decomposition layer, a forced state recursion layer, a state vector evolution layer, and a response output layer. The delayed coordinate input layer arranges the initial adjustment response in a staggered manner along the sampling position to form a Hankel matrix; The coordinate decomposition layer performs singular value decomposition on the Hankel matrix, and forms a state vector from the remaining multiple delayed coordinates excluding the last one, and determines the last delayed coordinate as the forced coordinate. The forced state recursion layer evolves the forced state of the previous sampling position through the forced recursion matrix, and transforms the forced coordinates of the current sampling position through the coordinate mapping matrix, forming the continued forced state and forced state of the current sampling position. The state vector evolution layer processes the state vector of the previous sampling position through the state evolution matrix, and combines the processing result with the result obtained by the forced coupling matrix in processing the continued forced state and the forced state to form the continued state vector and the total state vector. The response output layer converts the continued state vector and the total state vector into the response of the previous adjustment action and the total adjustment response respectively through the response reconstruction matrix. The response of the previous adjustment action is subtracted from the total adjustment response, and the new adjustment response of the current adjustment action is output.

7. The method for generating equipment energy-saving strategies based on big data processing according to claim 6, characterized in that, The continuing forced state and forced state that form the current sampling position include: Multiply the forced state of the previous sampling position by the forced recursion matrix to obtain the continued forced state of the current sampling position; Multiply the forced coordinates of the current sampling position by the coordinate mapping matrix to transform the forced coordinates into the state space where the continued forced state is located; Construct a projection matrix pointing to the subspace where the continued forced state is located and a corresponding complementary projection matrix. Extract the continued correction component from the transformed forced coordinates using the projection matrix and extract the new forced component using the complementary projection matrix. The continued correction component is incorporated into the continued forced state, and the corrected continued forced state and the newly added forced component are combined to form the forced state of the current sampling position. The corrected continuation forced state and the forced state at the current sampling position are passed to the state vector evolution layer, and the forced state at the current sampling position is used as the recursive input of the forced state at the next sampling position.

8. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S6 includes: S61. Starting from the device currently acting on the adjustment action, track the devices whose state changes or energy consumption changes within the sampling location range corresponding to the newly added adjustment response level by level along the correlation relationship between the devices, and determine the associated devices. S62. Extract the energy consumption changes of the current regulating action device and the associated device at each sampling position from the newly added regulating response, and classify the energy consumption increase part as energy consumption increase amount and the energy consumption decrease part as energy consumption reduction amount. S63. Combine the energy consumption increase and energy consumption reduction of the current regulating action device and related devices, and subtract the combined energy consumption increase from the combined energy consumption reduction to obtain the system net energy saving response corresponding to the current regulating action.

9. The method for generating equipment energy-saving strategies based on big data processing according to claim 1, characterized in that, S7 includes: S71. Determine the state of the last device before the adjustment action occurs as the entry state, determine the state of the last device within the sampling position range corresponding to the newly added adjustment response as the end state, and encapsulate the adjustment action, entry state, end state and system net energy saving response into an adjustment action unit. S72. Pair the end state of the previous adjustment action unit with the entry state of the next adjustment action unit according to the device identifier and status parameters, and connect the adjustment action units with the same device identifier and the same or different status parameter values, or whose differences do not exceed the sampling resolution of the corresponding status parameters, to form a status connection relationship. S73. Determine the last device state in the historical operation sequence of the device as the starting device state. Pair the entry state of each adjustment action unit with the starting device state according to the device identifier and state parameters. Select the adjustment action unit with the largest net energy saving response of the system from the adjustment action units with the same device identifier, the same state parameter value or the difference does not exceed the sampling resolution of the corresponding state parameter and the system net energy saving response is greater than zero. S74. Remove the already arranged adjustment action units, and from the remaining adjustment action units that have a state connection relationship with the current adjustment action unit and whose net energy saving response of the system is greater than zero, successively select the adjustment action unit with the largest net energy saving response of the system and continue to arrange them until there are no adjustment action units that can be connected, thus forming the execution sequence of the adjustment action. S75. Arrange the adjustment actions in each adjustment action unit according to the execution sequence to generate an energy-saving strategy for the equipment.

10. A system for generating energy-saving strategies for equipment based on big data processing, comprising executing the method for generating energy-saving strategies for equipment based on big data processing as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data generation module is used to acquire equipment operation data, energy consumption data, control action data, and production task data, form historical operation sequences of equipment, and obtain the correlation between equipment. The response interception module is used to locate the position of control parameter change in the historical operation sequence of the equipment, intercept the equipment status and energy consumption changes before and after the control parameter change, and form an adjustment response segment. The segment matching module is used to match the adjustment response segment with an unadjusted response segment that has not undergone a corresponding adjustment action, and to align the adjustment response segment and the unadjusted response segment according to the device operation process; The response extraction module is used to subtract the common changes between the aligned adjusted response fragments and the unadjusted response fragments to form a preliminary adjusted response; The response separation module is used to input the preliminary adjustment response into the improved HAVOK model. The improved HAVOK model writes the forced coordinates into the forced state recursively across the sampling positions, and the forced state drives the state vector evolution to strip away the response that continues from the previous adjustment action and obtain the new adjustment response of the current adjustment action. The energy-saving calculation module is used to aggregate the energy consumption increase or decrease of the device currently performing the adjustment action and the associated devices along the correlation between the devices to obtain the net energy-saving response of the system. The strategy generation module is used to connect multiple adjustment actions based on the equipment status before and after the adjustment action, and determine the execution order of the multiple adjustment actions according to the corresponding system net energy saving response, so as to generate an equipment energy saving strategy.

Images