A device energy consumption anomaly identification method based on time sequence feature mining

Abstract

The application provides a device energy consumption anomaly recognition method based on time sequence feature mining, relates to the technical field of power system device operation state monitoring and data processing, and comprises the following steps: obtaining a time sequence energy consumption sequence of a power transformation device and an instruction issuing time and an expected regulation power contained in a scheduling instruction; obtaining an expected mechanical hysteresis time length and an expected load ramping slope according to a steady-state active load rate before the instruction issuing; obtaining an expected action starting time in combination with the instruction issuing time and the mechanical hysteresis time length; finding an actual action starting time in the sequence and performing time alignment; constructing a physical transition trajectory in combination with the expected regulation power and the ramping slope, splicing to a steady-state operation baseline to generate a dynamic baseline; and determining energy consumption anomaly when the aligned energy consumption sequence deviates from a follow-up interval defined by the dynamic baseline. The application contains an actual physical transition process of a device and improves the accuracy of anomaly determination.

Description

Technical Field

[0001] This invention relates to the field of power system equipment operation status monitoring and data processing technology, and in particular to a method for identifying equipment energy consumption anomalies based on time-series feature mining. Background Technology

[0002] Substations are typically equipped with electrical equipment capable of dynamic load regulation, and energy consumption is monitored by collecting power data from this equipment during operation and maintenance. Existing data monitoring logic largely relies on historical energy consumption performance of equipment under stable, resting conditions to establish a fixed status assessment benchmark. This data processing mechanism has the capability to evaluate substation equipment that has not received external dispatch instructions, supporting the routine basic operation and maintenance of the distribution network.

[0003] In actual operation, power substations frequently respond to load regulation commands issued by the power grid dispatching system. Upon receiving control messages, their internal mechanical mechanisms must overcome inherent damping to generate displacement. Due to the mechanical response characteristics of the internal components, there is a time delay between the issuance of the network command and the actual power transfer, and the gradual ramp-up rate during power conversion is also physically constrained by the pre-regulation base load level. This controlled action causes the active power of the power substation to exhibit a nonlinear temporal evolution accompanied by physical displacement characteristics.

[0004] Conventional monitoring and comparison models, due to their static and fixed comparison thresholds and time reference benchmarks, rarely incorporate the start-up lag and transient ramp-up patterns of power equipment during dynamic dispatch transitions into their evaluation calculations. When directly using a fixed baseline to verify transient transition data, power fluctuations generated by the equipment's normal execution of controlled actions can easily exceed the original tolerance limits, thus affecting the accuracy of the system's assessment of the equipment's operating status. The industry needs an energy consumption monitoring data analysis mechanism that can integrate the physical evolution characteristics of equipment. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying abnormal energy consumption in equipment based on time-series feature mining, in order to solve the technical problem that existing energy consumption monitoring methods, which use static fixed benchmarks, are difficult to adapt to the start-up lag and dynamic power ramp-up evolution when power equipment responds to dispatch instructions, and thus easily misjudge the physical transition response process as an abnormal energy consumption exceeding the limit.

[0006] This invention provides a method for identifying abnormal device energy consumption based on time-series feature mining, comprising:

[0007] Obtain the time-series energy consumption sequence of the target substation, and simultaneously extract the instruction issuance time and expected adjustment power contained in the load scheduling instruction for the target substation;

[0008] Obtain the steady-state active power load rate of the target substation prior to the time the instruction is issued;

[0009] Using the steady-state active power load rate, the corresponding expected mechanical hysteresis duration and expected load ramp-up slope are obtained.

[0010] Add the expected mechanical hysteresis duration to the time when the instruction is issued to obtain the expected start time of the action.

[0011] Find the power mutation point in the time-series energy consumption sequence as the start time of the actual action;

[0012] The expected start time of the action and the actual start time of the action are time-aligned to output an aligned energy consumption sequence.

[0013] Starting from the expected start time of the expected action, a physical transition trajectory is constructed by combining the expected adjustment power and the expected load ramp slope;

[0014] The physical transition trajectory is spliced ​​to a pre-established steady-state operating baseline to generate a dynamic baseline;

[0015] Compare the aligned energy consumption sequence with the dynamic baseline;

[0016] When the aligned energy consumption sequence deviates from the following interval defined by the dynamic baseline, it is determined that the target power equipment has an energy consumption anomaly.

[0017] Optionally, the steps for constructing the pre-established steady-state operating baseline include:

[0018] Collect stable energy consumption data of the target substation under historical no-dispatch conditions;

[0019] The median filter is applied to the steady-state energy consumption data using a sliding time window to remove transient spike noise contained in the steady-state energy consumption data;

[0020] The average value of the stable energy consumption data after removing transient peak noise is calculated according to a preset time step.

[0021] The average values ​​corresponding to each preset time step are connected to form the steady-state operation baseline.

[0022] Optionally, the step of using a sliding time window to perform median filtering on the stable energy consumption data to remove transient spike noise contained in the stable energy consumption data includes:

[0023] Extract all discrete power sample values ​​within the sliding time window;

[0024] Sort all discrete power sample values ​​within the sliding time window by amplitude;

[0025] Select the middle amplitude value after sorting as the representative power value at the center position of the current window;

[0026] Calculate the absolute difference between the original center power value and the representative power value;

[0027] When the absolute difference exceeds the preset electromagnetic interference tolerance limit, the original center power value is determined to be the transient spike noise and is removed, and the representative power value is used to fill and cover it.

[0028] Optionally, finding the power abrupt change point in the time-series energy consumption sequence as the starting time of the actual action includes:

[0029] Starting from the moment the instruction is issued, calculate the power change rate between adjacent energy consumption sampling points in the time-series energy consumption sequence;

[0030] Compare the power change rate with the inherent mechanical response start-up dead zone threshold of the target power equipment;

[0031] In the time-series energy consumption sequence, the energy consumption sampling point corresponding to the first crossing of the mechanical response start-up dead zone threshold is identified;

[0032] The energy consumption sampling point that first crosses the mechanical response start-up dead zone threshold is established as the starting moment of the actual action when the physical state begins to change.

[0033] Optionally, the step of time-aligning the expected start time of the action with the actual start time of the action and outputting an aligned energy consumption sequence includes:

[0034] Calculate the absolute value of the time difference between the actual start time of the action and the expected start time of the action;

[0035] Determine whether the absolute value of the time difference is less than the maximum transmission hysteresis limit allowed by the power grid dispatching network;

[0036] If the absolute value of the time difference is less than the maximum transmission hysteresis limit, the time-series energy consumption sequence is shifted as a whole on the time axis by the absolute value of the time difference, so that the start time of the actual action coincides with the start time of the expected action, and the aligned energy consumption sequence is obtained in a synchronized alignment state.

[0037] If the absolute value of the time difference is greater than or equal to the maximum transmission hysteresis limit, a communication interruption fault is established, and the time alignment step is terminated directly.

[0038] Optionally, constructing the physical transition trajectory by combining the expected adjustment power with the expected load ramp slope includes:

[0039] The original power value corresponding to the expected start time of the action is taken as the starting point of the trajectory, based on the steady-state operation baseline.

[0040] By combining the expected adjustment power with the original power value, the target power endpoint that must be reached to complete the load scheduling task is calculated.

[0041] Starting from the trajectory origin, construct a linearly varying slope according to the expected load ramp rate;

[0042] The linear change line is extended and calculated until its value reaches the target power endpoint, and the line segment between the trajectory starting point and the time node corresponding to the value is taken as the physical transition trajectory.

[0043] Optionally, the step of stitching the physical transition trajectory to a pre-established steady-state operating baseline to generate a dynamic baseline includes:

[0044] Locate the entry node corresponding to the expected start time of the action in the steady-state operation baseline;

[0045] The target end time of the physical transition process is calculated using the expected adjustment power and the expected load ramp slope.

[0046] Locate the cut-out node corresponding to the target end time in the steady-state operation baseline;

[0047] Remove the original horizontal line segment of the steady-state operation baseline that is located between the cut-in node and the cut-out node;

[0048] The physical transition trajectory is filled between the cut-in node and the cut-out node, and the dynamic baseline containing the actual physical response slope is output.

[0049] Optionally, before comparing the aligned energy consumption sequence with the dynamic baseline, the step of setting the following interval is further included:

[0050] Retrieve the nameplate parameters of the target substation equipment to obtain the percentage of inherent measurement error calibrated at the factory.

[0051] Extract the reference power values ​​corresponding to each time node on the dynamic baseline;

[0052] Multiply the reference power value by the inherent measurement error percentage to obtain the dynamic floating tolerance specific to each time point;

[0053] The dynamic floating tolerance is superimposed and subtracted above and below the dynamic baseline to generate parallel upper and lower bound envelopes.

[0054] The following interval is formed by the upper limit envelope and the lower limit envelope together in space.

[0055] Optionally, determining that the target power equipment has an energy consumption anomaly when the aligned energy consumption sequence deviates from the following interval defined by the dynamic baseline includes:

[0056] Extract the actual power sample values ​​from the aligned energy consumption sequence one by one;

[0057] The actual power sampling value is compared with the upper limit envelope and the lower limit envelope corresponding to the same time node.

[0058] When an actual power value greater than the value corresponding to the upper limit envelope or less than the value corresponding to the lower limit envelope is found, it is confirmed that the aligned energy consumption sequence deviates from the following interval.

[0059] Abnormal data points that deviate from the following interval are continuously counted. When the total number of consecutive counts exceeds the fault tolerance threshold, the target power equipment is determined to have an energy consumption abnormality.

[0060] Optionally, the method further includes:

[0061] If the aligned energy consumption sequence is within the following interval, then the target power equipment is considered to have successfully responded to the scheduling task and the physical execution process is normal.

[0062] Intercept over-limit alarm actions that were originally expected to be triggered by load changes in the substation monitoring backend;

[0063] Extract the aligned energy consumption sequence that is within the following interval as the actual normal transition sample;

[0064] The actual normal transition samples are saved to the equipment feature archive for periodic correction of the mapping parameters of the expected mechanical hysteresis duration and the expected load ramp-up slope.

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

[0066] This invention obtains the steady-state active power load rate of the target substation and derives the corresponding expected mechanical hysteresis duration and expected load ramp-up slope. Combined with the expected regulation power, a dynamic baseline encompassing the actual physical response characteristics of the equipment is constructed, improving upon the limitations of fixed evaluation benchmarks that cannot adapt to the transient regulation evolution of the equipment. In time-series analysis, this method identifies the actual start time of the substation's action and aligns it with the expected start time, eliminating time coordinate reference deviations caused by network communication delays and start-up hysteresis of the equipment's actuators. This invention compares the aligned energy consumption sequence with the following interval derived from the dynamic baseline, incorporating the load step changes during substation response to scheduling tasks into the evaluation scope. This dynamic analysis mechanism, which integrates the actual physical transition laws of the substation, reduces the possibility of misjudgments in the monitoring system during controlled equipment regulation and improves the accuracy of energy consumption status identification when the target substation is under complex scheduling conditions.

[0067] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention can be realized and obtained by means of the technical solutions particularly pointed out in the written description and drawings.

[0068] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0069] 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:

[0070] Figure 1 This is a schematic diagram of the hardware architecture of the data processing device that performs the energy consumption anomaly identification method in an embodiment of the present invention;

[0071] Figure 2 This is an overall flowchart of a device energy consumption anomaly identification method based on time-series feature mining in an embodiment of the present invention;

[0072] Figure 3 This is a flowchart illustrating the steps involved in constructing a steady-state operating baseline in an embodiment of the present invention.

[0073] Figure 4 This is a flowchart illustrating the steps for locating the actual start time of an action in an embodiment of the present invention.

[0074] Figure 5 This is a flowchart illustrating the process of determining abnormal energy consumption in an embodiment of the present invention.

[0075] Figure 6This is a schematic diagram of the timing waveforms of the dynamic baseline and the following interval in an embodiment of the present invention. Detailed Implementation

[0076] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0077] This invention provides a method for identifying abnormal energy consumption in equipment based on time-series feature mining. The execution entity of this method can be a data processing device configured within the substation control layer or process layer nodes. For example... Figure 1 As shown, the data processing device includes a processor, random access memory, a non-volatile memory module, a network communication interface, and a system bus for connecting the various components in its hardware architecture. The memory module is pre-configured with a computer-readable instruction sequence. The processor extracts, decodes, and executes this instruction sequence to achieve multi-dimensional time-series data acquisition, network packet parsing, numerical matrix operations, and logical state determination.

[0078] This invention provides a method for identifying abnormal energy consumption in equipment based on time-series feature mining. The overall process is as follows: Figure 2 As shown, the specific steps include:

[0079] Step S10: Construct a pre-established steady-state operating baseline.

[0080] like Figure 3 As shown, the specific construction process of the steady-state operation baseline is as follows:

[0081] Collect stable energy consumption data of target substation equipment under historical, un-dispatched conditions. Target substation equipment includes individual electrical devices with dynamic load regulation capabilities, such as on-load tap-changing transformers and static var compensators (SVCs) deployed in the distribution network. During continuous time windows when the power grid dispatching system does not issue power regulation control commands to the target substation equipment, the target substation equipment operates under basic resting load conditions. The data processing equipment receives a sequence of discrete active power values ​​continuously sampled and output by the metering module through a network communication interface connected to the substation process layer bus. This sequence is stored in a one-dimensional buffer array in memory, establishing it as the stable energy consumption data. Each data node in the stable energy consumption data contains a discrete power value and a timestamp parameter associated with it.

[0082] A median filter is applied to the stable energy consumption data using a sliding time window to remove transient spikes and noise. All discrete power sample values ​​within the sliding time window are extracted. The processor allocates a contiguous space in the memory data area to instantiate the sliding time window and configures the number of sampling nodes included in the sliding time window to a predetermined odd number. The specific value of this odd number parameter is derived based on the maximum duration of high-frequency transient bounce generated by the switch contacts of the power equipment during state switching. The system extracts the maximum contact bounce oscillation time parameter recorded in the type test report of similar electrical equipment and divides it by the discrete sampling time interval period at the hardware level of the analog-to-digital conversion module.

[0083] The quotient of the division operation is rounded up. If the result is even, it is incremented by one; otherwise, it remains unchanged. The resulting integer is then assigned to the number of sampling nodes. On the timeline sequence corresponding to the stationary energy consumption data, the processor controls the sliding time window to recursively advance from the beginning to the end of the time data column, with a time step equal to the time interval of a single sampling period. In each single-step sliding execution cycle, the processor uses pointer offset addressing to read all discrete power sampling numerical data elements within the current sliding time window boundary and loads them into a temporary processing queue.

[0084] All discrete power sample values ​​within the sliding time window are sorted by amplitude. The processor executes an internal sorting algorithm call instruction for the data elements in the temporary processing queue. The specific sorting logic expands based on the amplitude of the data elements, reorganizing the discrete power sample values ​​originally distributed in chronological order according to timestamps into a new data column that monotonically increases in amplitude.

[0085] The sorted median amplitude value is selected as the representative power value at the center of the current window. Since the total number of data nodes covered by the sliding time window is odd, the processor reads the total number of windows, divides it by two, and rounds down to generate a specific integer index offset. Using the starting address of the reorganized new data column as a reference, the processor adds this integer index offset value and reads the power value stored at the corresponding physical memory address. The amplitude value at this extracted center index position is established as the sorted median amplitude value and assigned to the representative power value variable in the program stack.

[0086] Calculate the absolute difference between the original center power value and the representative power value. The processor reads the discrete power sample value from the stationary energy consumption data sequence, where the time coordinate corresponds to the physical center position of the current sliding time window, and defines it as the original center power value. Perform an arithmetic subtraction calculation to subtract the representative power value from the original center power value, and call the absolute value function to calculate the absolute difference parameter between the two.

[0087] When the absolute difference exceeds the preset electromagnetic interference tolerance limit, the original center power value is determined to be a transient spike noise point and is removed, and the representative power value is used to fill and cover it. The preset electromagnetic interference tolerance limit parameter is pre-read from the system configuration file and resides in memory. The value of this preset electromagnetic interference tolerance limit parameter is calibrated based on the quantization accuracy of the analog-to-digital conversion at the data acquisition terminal and the actual background noise of the field environment. The underlying initialization program extracts the stable background noise power sequence when the device is in a no-service load state, calculates the peak-to-peak value of the sequence fluctuation, and superimposes the peak-to-peak value onto twice the power resolution value corresponding to the least significant bit of the analog-to-digital converter. The equivalent sum of the two is directly established as the preset electromagnetic interference tolerance limit parameter. The processor executes a numerical comparison instruction between the acquired absolute difference parameter and the preset electromagnetic interference tolerance limit parameter. If the absolute difference parameter is greater than the preset electromagnetic interference tolerance limit parameter, the system logic confirms that the original center power value corresponding to the current time node has deviated from the data. The memory overwrite control flow is triggered, locating the data address of the original center power value in the stable energy consumption data matrix. The binary data representing the power value is directly written to this address to perform an overwrite operation. If the absolute difference parameter is less than or equal to the preset electromagnetic interference tolerance upper limit parameter, the decision branch jumps to the skip overwrite operation code block, preserving the original state of the original center power value data element. After completing the above traversal loop, the stable energy consumption data sequence after removing transient spike noise is output.

[0088] The average value of the stable energy consumption data after removing transient spike noise is calculated according to a preset time step. The processor reads the global time segmentation parameters and divides the time axis of the filtered stable energy consumption data into multiple non-overlapping continuous time domain blocks. The time measure spanned by each continuous time domain block is defined as the preset time step. The value of this preset time step is subject to the physical refresh cycle of the telemetry data sent from the local substation process layer network to the remote dispatch master station. The underlying communication daemon thread reads the fixed frame interval time parameter for the general call or cyclic reporting of active power telemetry points in the internal communication protocol of the station-level remote control gateway, and directly assigns this fixed frame interval time parameter as the preset time step. The sequence is partitioned and traversed to extract all discrete power sample values ​​belonging to the same continuous time domain block. The accumulation instruction is executed to add these values ​​together, and then the sum generated by the accumulation calculation is divided by the total number of sampling nodes contained in the block. The result of the calculation generates the corresponding arithmetic mean. Bind the arithmetic mean to the data structure to the arithmetic midpoint timestamp between the start and end times of the corresponding block.

[0089] The average values ​​corresponding to each preset time step are connected to form the steady-state operating baseline. The processor maps and transforms the calculated timestamps and their corresponding average values ​​into a two-dimensional coordinate matrix structure containing time and power parameters. The system calls a linear interpolation calculation program to calculate and construct a first-order linear equation between two adjacent time-sequential average data points. Based on this linear equation, a continuous set of discrete coordinate points is generated within the two node intervals. By concatenating the interpolated data sequences of all adjacent nodes, a continuous piecewise linear data array is generated. This array sequence is established as the steady-state operating baseline characterizing the inherent operating base morphology of the target substation equipment.

[0090] Step S20: Obtain timing data and scheduling instruction parameters.

[0091] The time-series energy consumption sequence of the target substation equipment is obtained, and the load dispatch instructions for the target substation equipment, including the instruction issuance time and expected adjustment power, are extracted simultaneously. The local synchronous measurement system deployed in the substation continuously performs analog-to-digital conversion and calculation processing on the analog electrical signals of the secondary side of the instrument transformers, outputting a set of discrete active power sampling points carrying a synchronous clock tag. The data processing equipment continuously receives the above data stream through a network port and organizes it into a structured two-dimensional array column according to the timestamp sequence, establishing it as the time-series energy consumption sequence.

[0092] Simultaneously, the network monitoring task thread of the data processing equipment continuously monitors the remote control network messages sent from the power grid dispatch center to the substation process layer communication bus. When an Ethernet message corresponding to the target substation equipment is intercepted, the data processing equipment performs unpacking and frame splitting processing on the application layer protocol stack based on the preset substation communication protocol (such as IEC 60870-5-104 or IEC 61850 MMS protocol) and extracts the corresponding field data. It reads the time synchronization field data in the message header control area, extracts the network precision time parameter of the control message generated or sent by the dispatch master station, formats it into a standard system timestamp data type, and establishes it as the instruction issuance time. It further analyzes the data load segment of the network message, locates the operational quantity control parameter field, and extracts the specific value of the target active power required by the target substation after responding to this dispatch action or the relative change in power required for adjustment. These values ​​are then uniformly converted into standard power dimension parameters and established as the expected adjustment power. The system-level clock synchronization protocol ensures that the timestamps on the timing power consumption sequence and the instruction issuance time parsed from the message are in the same absolute time reference system.

[0093] Step S30: Calculate the hysteresis duration and ramp slope based on the equipment's historical operating status.

[0094] The steady-state active power load rate of the target substation prior to the instruction issuance time is obtained. The processor locates the corresponding time index coordinate in the time-series energy consumption sequence array in memory, based on the instruction issuance time. A segment of sequence data of a set duration is extracted by working backwards along the earlier historical time direction on the time axis. The span boundary of this set duration is determined by the physical relaxation time constant of the large-mass mechanical transmission components or electromagnetic energy storage components inside the target substation. The processor reads the tail stabilization time parameter recorded in the target substation maintenance manual, which is the time required for the mechanical mechanism to complete its maximum displacement stroke until the internal mechanical damping fully attenuates the residual oscillation kinetic energy. To ensure that the extracted energy consumption sequence avoids the physical aftershocks of preceding actions, the system multiplies this tail stabilization time parameter by a safety margin constant defined between 1.2 and 1.5 (preferably 1.3 in this embodiment), and the resulting product value is established as the set duration. All active power sampling values ​​stored in this segment of sequence data are read, and an arithmetic mean summation is performed to calculate the power arithmetic mean. The system's underlying database interface performs a read operation, retrieving the set rated active power parameters from the target substation equipment ledger parameter table. Using the arithmetic mean of the power as the dividend and the rated active power parameters as the divisor, a division operation is performed, and the resulting quotient is established as the steady-state active load rate.

[0095] Using the steady-state active power load rate, the corresponding expected mechanical hysteresis duration and expected load ramp-up slope are derived. A multi-dimensional operating state mapping matrix structure data file is pre-programmed into the non-volatile storage module. This matrix structure data is divided into horizontal indices based on the discrete intervals of the steady-state active power load rate, corresponding to the historical actual hysteresis duration statistics and actual power ramp-up slope statistics of the target substation under the same load background. The processor uses the real-time calculated steady-state active power load rate parameter as the query key and inputs it into this mapping matrix structure for retrieval. If the currently calculated steady-state active power load rate lies between two known discrete load rate nodes recorded in the matrix, the system calls the algebraic proportional interpolation function to calculate two output values ​​that accurately match the current steady-state active power load rate. These two values ​​are then assigned to the expected mechanical hysteresis duration parameter and the expected load ramp-up slope parameter, respectively. The expected mechanical hysteresis duration characterizes the expected time interval required from the issuance of a control command to the commencement of substantial physical displacement of the equipment's mechanical mechanism; the expected load ramp slope characterizes the expected physical rate of change of active power over time after the equipment initiates a transient action. The initial data for this multi-dimensional operating state mapping matrix structure is obtained through physical bench step response calibration of the target substation before it leaves the factory or after maintenance. The calibration program divides the rated active load rate range from 0% to 100% into multiple discrete base load nodes using fixed percentage steps. The specific value of the fixed percentage step is determined by the load control resolution of the target substation, and its extraction steps are as follows: read the minimum adjustable active power step value set on the equipment's bottom-level control panel, divide it by the rated active power, and the resulting percentage value is established as the fixed percentage step; if the system cannot extract this minimum step value parameter from the bottom level, a default rule is triggered, forcibly setting the fixed percentage step to 5%. Under the static torque constraints corresponding to each base load node, a full-range adjustment command is issued to the equipment. The arithmetic mean of the measured time taken for the transmission mechanism to overcome the initial static friction and generate a substantial physical displacement from receiving the control electrical signal is recorded and established as the initial value of the expected mechanical hysteresis time for the corresponding node. Simultaneously, the average value of the first derivative of the power output waveform within the range from the response start point to the target steady-state value is extracted and established as the initial value of the expected load ramp-up slope. When the real-time steady-state active power load rate obtained by the system is between two known base load nodes, the processor locks the parameters of the adjacent preceding and following nodes and calls the first-order linear interpolation equation to calculate the corresponding expected mechanical hysteresis time and expected load ramp-up slope.

[0096] Step S40: Calculate the theoretical time point and locate the actual start time of the action.

[0097] like Figure 4 As shown, the specific process is as follows:

[0098] Adding the expected mechanical hysteresis duration to the instruction issuance time yields the expected action start time. Extracting the time value corresponding to the instruction issuance time parameter, and sending it to the processor's arithmetic execution unit, algebraically adds it to the calculated expected mechanical hysteresis duration parameter. The output time and value data are established as the expected action start time, representing the time reference point at which the physical state of the action response should change under ideal transmission conditions.

[0099] The processor identifies power abrupt change points in the time-series energy consumption sequence as the starting point for actual actions. Starting from the instruction issuance time, the power change rate between adjacent energy consumption sampling points in the time-series energy consumption sequence is calculated. The processor establishes an extraction index within the time-series energy consumption sequence array, initially setting the position to the index of the first energy consumption sampling point equal to or later than the instruction issuance time. In a time-incrementing traversal loop, data corresponding to two adjacent time coordinate nodes within the sequence is extracted. The power value of the energy consumption sampling point located at the later time coordinate is subtracted from the power value of the energy consumption sampling point located at the earlier time coordinate to calculate the power difference. Subsequently, this power difference is divided by the system's fixed sampling time interval parameter, and the quotient is used to generate the first-order power change rate corresponding to that specific interval.

[0100] The power change rate is compared with the inherent mechanical response start-up dead zone threshold of the target substation equipment. The system extracts the pre-written mechanical response start-up dead zone threshold of the target substation equipment from the configuration space. The setting of the mechanical response start-up dead zone threshold is intended to effectively isolate the inherent physical meshing clearance of the substation drive gears and the natural fluctuations of the grid fundamental voltage.

[0101] The system retrieves the first-order power change rate dataset of the target substation equipment over several consecutive resting periods without dispatch intervention and calculates the statistical standard deviation of this dataset. Based on the high confidence interval criterion of normal distribution (i.e., the 3σ principle), three times this standard deviation is taken as the rate tolerance base (in W / s) to cover conventional electrical background noise. Simultaneously, the mechanical transmission idle displacement parameters specified on the nameplate in the equipment ledger are extracted, and the corresponding theoretical transient power drift is calculated using the inherent transmission ratio. The specific calculation process is as follows: the system calls the preset electromechanical energy conversion equation. The solution is then performed. Among them, This represents the theoretical transient power drift (in W). The rated active power of the target substation equipment. To extract the mechanical transmission idle displacement parameters, This is the inherent transmission ratio. This refers to the total full-scale stroke parameter of the device's actuator. This represents the friction loss coefficient of the transmission mechanism. To ensure the dimensional homogeneity of the decision logic, the system uses the theoretical transient power drift. Divide by the system's fixed sampling time interval period This is converted into a drift rate component. Finally, the rate tolerance base is algebraically added to the drift rate component, and the sum is established as the dimensionlessly uniform mechanical response start-up rate dead zone threshold.

[0102] After calculating the power change rate in each iterative loop, its absolute value is calculated. The absolute value is then compared with the mechanical response start-up dead zone threshold value using an inequality check to determine their magnitude.

[0103] In the time-series energy consumption sequence, the energy consumption sampling point corresponding to the first crossing of the mechanical response start-up dead zone threshold is locked. During the forward time traversal, when the judgment statement captures the first power change rate data with an absolute value greater than or equal to the mechanical response start-up dead zone threshold, the system generates an internal interrupt signal to exit the current differential calculation loop. The processor locates two connected energy consumption sampling points participating in this over-limit power change rate calculation. The energy consumption sampling point located at the later timestamp position is extracted and a specific identifier bit is written into the memory data structure to complete the locking action. The energy consumption sampling point corresponding to the first crossing of the mechanical response start-up dead zone threshold is established as the actual action start time when the physical state begins to change. The timestamp attribute variable bound to the identified and locked energy consumption sampling point is extracted, persisted, and established as the actual action start time, used to characterize the starting point when the target substation body overcomes internal and external damping and the power begins to show a step-up characteristic.

[0104] Step S50: Perform time-domain alignment of the time series.

[0105] The expected start time of the action and the actual start time of the action are time-aligned, and an aligned energy consumption sequence is output. The absolute value of the time difference between the actual start time and the expected start time of the action is calculated. The processor retrieves the recorded values ​​of the actual start time and the expected start time from the memory variable space. A subtraction operation is performed, and the absolute value of the difference is taken. The final result is generated and established as the absolute value parameter of the time difference.

[0106] The system determines whether the absolute value of the time difference is less than the maximum allowable transmission hysteresis limit of the power grid dispatch network. The system configuration dictionary stores the maximum allowable transmission hysteresis limit parameter. The processor reads this parameter and performs a numerical comparison with the absolute value of the time difference parameter. The value of the maximum allowable transmission hysteresis limit parameter is anchored to the upper limit of the physical delay of the hardware topology routing in the substation process layer control network. The communication control process reads the total number of industrial switch hops required for the power grid dispatch command message to be sent to the target substation equipment. This total number of hops is multiplied by the worst-case backplane queuing forwarding time parameter measured under full-load concurrent operation of a single switch, and algebraically superimposed with the inherent hardware delay parameter of the electro-optic signal conversion in the transmission medium. The calculated sum of physical times is directly fixed as the maximum transmission hysteresis limit parameter. If the absolute value of the time difference is less than the maximum transmission hysteresis limit, the time-series energy consumption sequence is shifted along the time axis by the absolute value of the time difference, so that the start time of the actual action coincides with the start time of the expected action, resulting in the aligned energy consumption sequence in a synchronized state. Within the logical branch confirming that the network latency meets the preset boundaries, the processor executes a full batch processing instruction on the time-series energy consumption sequence array in memory. It iterates through each discrete sampled element in the array, performing a uniform algebraic addition or subtraction correction operation on its timestamp parameter and the absolute value of the time difference. The correction sign is determined based on the comparison between the actual action start time and the expected action start time. After this batch shift calculation, the timestamp value corresponding to the original actual action start time node in the array is updated and corrected to be equal to the expected action start time value. The entire data sequence array, after the overall shift correction of the time domain coordinates, is output and saved as an aligned energy consumption sequence.

[0107] If the absolute value of the time difference is greater than or equal to the maximum transmission hysteresis limit, a communication interruption fault is confirmed, and the time alignment step is terminated directly. If the comparison result falls into this conditional branch, the system determines that the data packet of the target substation has experienced an extended network congestion wait, or that the device's drive execution component has experienced a hardware deadlock fault. The system then outputs an abnormal interrupt status code to the control bus, stopping all subsequent sequence analysis and comparison processes for this instruction in the current round.

[0108] Step S60: Construct physical transition trajectories and stitch them together to generate dynamic baselines.

[0109] Starting from the expected start time of the action, a physical transition trajectory is constructed by combining the expected adjustment power and the expected load ramp slope. The original power value corresponding to the steady-state operating baseline at the expected start time of the action is used as the trajectory starting point. The processor reads the timestamp parameter of the expected start time of the action and uses this parameter as an addressing coordinate to query the two-dimensional data column structure corresponding to the steady-state operating baseline. The expected vertical active power value of the steady-state operating baseline at this time coordinate node is extracted, established, and assigned to the original power value parameter. The system establishes the coordinate point formed by the expected start time of the action and the original power value in the two-dimensional time-power space as the trajectory starting point.

[0110] Combining the expected adjustment power with the existing power value, the target power endpoint expected to be achieved after completing the load scheduling task is calculated. The processor retrieves the expected adjustment power parameter and executes addition or subtraction algebraic instructions to apply it to the existing power value. The scheduling route for the addition and subtraction signs of the algebraic instructions is determined by the bitmask parsing result of the scheduling instruction network message control word at the system's underlying layer. The message decomposition module extracts the first Boolean control flag bit representing the power adjustment direction in the Ethernet data payload area. If the parsed flag bit state is true, it indicates that the master station has issued a load increase action command, and the processor directly performs the addition operation based on the parsing result; if the parsed flag bit state is false, it indicates that a load decrease unloading command has been issued, and the route is routed to the subtraction algebraic operation channel. The sum or difference result obtained from the calculation represents the theoretical steady-state output level that the equipment should be at after the adjustment action is completed, and it is established as the target power endpoint value.

[0111] From the starting point of the trajectory Initially, this is combined with the expected load ramp-up slope. Construct the physical transition trajectory. In a two-dimensional time-power coordinate system, establish the system's time-varying trajectory. The first-order linear evolution equation of the change: The processor iteratively calculates discrete coordinate points using this equation, moving along the increasing direction of the time axis. The system monitors the calculation output in real time. numerical value, when The absolute value first reaches or exceeds the target power endpoint The calculation stops when the time is right. The system then captures the data from the trajectory starting point. To satisfy End point of condition The discrete data segments between these points are used to establish the set of line segments as the physical transition trajectory. For time The corresponding power values ​​at discrete coordinate points; The expected start time of the action calculated in the preceding steps; This represents the original power value corresponding to the expected start time of the action. The target end time is defined as the time corresponding to the achievement of the target power endpoint. The processor substitutes the target power endpoint value as the ordinate of the known equation function into the linear function and solves in reverse to obtain the corresponding horizontal time axis coordinate parameters. The solved time coordinate parameters are established as the intersection point. In the generated oblique line dataset, the discrete data combination sequence from the trajectory start coordinate to the intersection point coordinate is extracted, and it is established and assigned as an independent physical transition trajectory array.

[0112] The physical transition trajectory is stitched to a pre-established steady-state operating baseline to generate a dynamic baseline. The entry node corresponding to the expected start time of the action is located within the steady-state operating baseline. The target end time of the physical transition process is calculated using the expected adjustment power and the expected load ramp slope. The processor executes a division instruction, dividing the absolute value of the expected adjustment power by the expected load ramp slope parameter to obtain the time length variable required for the action state to maintain the transition. This time length variable is summed with the expected start time of the action, and the resulting timestamp value is established as the target end time. The exit node corresponding to the target end time is located within the steady-state operating baseline. Based on the calculated target end time, the corresponding time index node is searched in the steady-state operating baseline array and set as the exit node.

[0113] The original horizontal line segment of the steady-state operating baseline between the cut-in node and the cut-out node is removed. In the operating memory, all data elements in the steady-state operating baseline sequence whose timestamps are greater than or equal to the timestamp of the cut-in node and less than or equal to the timestamp of the cut-out node are removed. The physical transition trajectory is filled between the cut-in node and the cut-out node, outputting the dynamic baseline containing the actual physical response slope. The processor sequentially overwrites all data elements of the previously generated physical transition trajectory array into the address gaps left by the removal operation. Simultaneously, for the cut-out node and all steady-state operating baseline data nodes in the subsequent time series, the system performs a translation overlay calculation on their original vertical power coordinate values, uniformly adjusting them to the target power endpoint values. By executing a data reconstruction program including data pruning, data insertion, and subsequent longitudinal translation, a dynamic baseline sequence array is finally generated and output, covering the continuous changes in the initial steady state before load adjustment, the linear ramp transient during load adjustment, and the final steady state after load adjustment.

[0114] Step S70: Set the following interval based on the measurement error characteristics of the equipment.

[0115] Before comparing the aligned energy consumption sequence with the dynamic baseline, the process further includes setting the following interval. The nameplate parameters of the target substation are retrieved to obtain the inherent measurement error percentage calibrated at the factory. The system communication process sends a data query message to the equipment parameter ledger database via the local substation process layer network. A retrieval is performed based on the equipment code of the target substation, reading the comprehensive system accuracy level parameter noted in the instrument transformer accuracy verification report. The processor extracts the numerical part representing the ratio difference limit from the accuracy level parameter string (e.g., extracting the floating-point number 0.2 from 0.2S), divides it by 100, and assigns it to the inherent measurement error percentage parameter with a data type.

[0116] Extract the reference power values ​​corresponding to each time node on the dynamic baseline. Configure a read traversal process for the dynamic baseline sequence array generated by the preceding operation. Step by step along the time stamp increment order, extracting the expected power value in the storage structure of each time node. The extracted expected power value is used as the reference power value variable under the current loop step. Multiply the reference power value by the inherent measurement error percentage to obtain the dynamic floating tolerance specific to each time node. Perform multiplication calculation within the arithmetic unit. The calculated product value is scaled proportionally according to the different reference power value variables. Establish the calculated product value as the dynamic floating tolerance parameter at the current node.

[0117] The system allocates a starting address in the memory region to establish two independent array sequences, which are respectively identified as upper bound arrays. and lower bound envelope array At each discrete time node The processor makes ,and The resulting upper and lower bound envelopes define a continuous confidence threshold pipeline in the time domain, and the following interval is the one that satisfies the inequality condition. The set of numerical regions. Among them, Time node The corresponding upper limit envelope power value, Time node The corresponding lower limit envelope power value, Time node The corresponding dynamic baseline power value, Time node Dedicated dynamic floating tolerance, To be at the time node The acquired actual power sample value to be detected. The constructed following interval is a dynamically tolerant data corridor region whose bandwidth adaptively changes with the increase or decrease of the power reference value.

[0118] like Figure 6 As shown, a time-series waveform diagram of the constructed dynamic baseline, upper and lower bound envelopes, and following interval is presented.

[0119] Step S80: Verify the measured energy consumption waveform and determine if the energy consumption is abnormal.

[0120] like Figure 5 As shown, the specific logic for determining the anomaly is as follows:

[0121] The aligned energy consumption sequence is compared with the dynamic baseline; when the aligned energy consumption sequence deviates from the following interval defined by the dynamic baseline, it is determined that the target substation has an energy consumption anomaly. Actual power sample values ​​are extracted one by one from the aligned energy consumption sequence. A time extraction pointer is established starting at the beginning of the aligned energy consumption sequence array. As the instruction cycle progresses, the pointer iteratively steps forward at fixed data point intervals, reading the actual power sample value recorded inside the structure currently pointed to by the pointer. The actual power sample value is compared with the upper and lower bound envelopes corresponding to the same time node. The processor parses the timestamp identifier data associated with the current actual power sample value, uses this timestamp identifier data as the target key, and performs a search and read operation in the previously generated upper and lower bound envelope arrays to extract the upper and lower bound envelope power values ​​that are completely at the same time coordinate point.

[0122] When an actual power value greater than the value corresponding to the upper envelope or less than the value corresponding to the lower envelope is detected, it is confirmed that the aligned energy consumption sequence deviates from the following interval. The extracted actual power sample value, along with the upper and lower envelope power values, is fed into a value comparator. Branch decisions are executed to verify whether the actual power sample value is greater than the upper envelope power value and whether the actual power sample value is less than the lower envelope power value. If either branch returns a true Boolean logic signal, the processor performs a write operation in its recorded data status log, setting the timestamp to indicate an anomaly flag indicating deviation from the limit.

[0123] Abnormal data points deviating from the following interval are continuously counted. When the total number of consecutive counts exceeds the fault tolerance threshold, the target substation is determined to have an energy consumption anomaly. The system allocates a separate register address to deploy an internal variable as a continuous anomaly counter. During the point-by-point verification loop of the sequence array, if an out-of-limit flag is generated by reading the current judgment, the continuous anomaly counter receives a control command and increments arithmetically by one. If, at a subsequent timing position, the actual power sample value extracted during verification falls within the range between the upper and lower limit envelope power values ​​without generating an out-of-limit flag, the processing system issues a clear signal to reset the continuous anomaly counter to zero. At the end of each counting cycle, the total number of consecutive counts stored in the continuous anomaly counter is extracted. The fault tolerance threshold parameter stored in the configuration library is retrieved. The fault tolerance threshold parameter characterizes the physical tolerance of the target substation for short-term transient power outages that do not constitute substantial insulation damage or electromechanical failure. The arc extinction and mechanical bounce effect of moving and stationary contacts during state transitions of internal components often induce high-frequency, short-lived out-of-bounds fluctuations in the monitored waveform. The system extracts the hardware architecture classification identifier of the target substation equipment: if the equipment is a mechanical electrical device with physical switch contacts (such as an on-load tap-changing transformer), the maximum arc extinction oscillation duration specified in the equipment datasheet is used as the divisor; if the equipment is a purely power electronic solid-state device (such as a static var compensator), the allowable transient error recovery time is used as the divisor. The fixed sampling time interval used to collect the time-series energy consumption sequence is used as the divisor. After performing the division, the quotient is rounded up, and the resulting integer is designated as the fault tolerance threshold parameter. A two-sided numerical comparison is performed. When the total number of consecutive counts exceeds the fault tolerance threshold parameter, the determination of continuous out-of-bounds fluctuations in the time series is established, and the underlying logic state machine outputs the characteristic parameters of the target substation's energy consumption anomaly and the execution interrupt result.

[0124] Step S90: State establishment and baseline parameter closed-loop feedback update.

[0125] If the aligned energy consumption sequence is within the following interval, the target substation is considered to have successfully responded to the scheduling task and the physical execution process is normal. When the last discrete data in the aligned energy consumption sequence has been traversed and the total number of consecutive counts recorded in the continuous anomaly counter has not reached the fault tolerance threshold throughout the entire cycle, the system's master control judgment stream outputs a valid status result. In the device management control tree, the session instance for this response action is marked with a normal physical execution status code.

[0126] The system intercepts over-limit alarm actions that were originally expected to be triggered by load changes in the substation monitoring backend. Within the scope of the established normal status code of the physical execution process, the network service process of the data processing equipment extracts the network access address parameters of the target substation equipment and the start and end time window span parameters affected by this adjustment action according to the internal protocol specifications. Combining these parameters, a network frame sequence for configuring intervention logic is generated and pushed to the substation monitoring backend host via the switch. The parsing application process of the monitoring backend host establishes a logical shielding strategy object for a specific active power telemetry attribute based on the instructions of this network frame sequence. Upon receiving the logical shielding strategy object and the time window span parameters, the system suspends the behavior of the backend setpoint judgment logic module in calculating over-limit alarms (specifically, by calling the API interface reserved by the backend system, the over-limit alarm enable flag position of the active power telemetry point within the corresponding time window is set to false), thereby suppressing the dispatch of alarm events for load step changes at the source.

[0127] The aligned energy consumption sequence within the current following interval is extracted as the actual normal transition sample. This aligned energy consumption sequence data column, verified by all validation logic, is separated and exported from memory. The steady-state active power load rate parameter calculated at the time of executing the scheduling task, as well as environmental condition data such as the expected adjustment power included in the instruction, are extracted. These environmental condition data are then combined with the exported sequence data using bitstream concatenation and manipulation to construct a storage file object with a system data structure, which is established as the actual normal transition sample. This actual normal transition sample is saved to the device feature archive for periodically correcting the mapping parameters of the expected mechanical hysteresis duration and the expected load ramp-up slope. The storage write application interface is called to serialize the storage file object and persistently save it to the corresponding interval subdirectory of the device feature archive on non-volatile storage media.

[0128] The system internally configures a timed daemon process associated with the absolute time generator. It triggers the wake-up of a parameter maintenance processing thread on a system-level daily cycle. This thread reads newly added actual normal transition sample files mounted in the feature archive in batches via a read / write interface. For each read actual normal transition sample file, it extracts its stored time sequence and control parameters. It performs an algebraic difference operation to calculate the time difference between the mutation start time and the instruction issuance time recorded in the sample, outputting the actual hysteresis time value for the current running cycle. The processor applies double-ended delimitation logic to the measured time-series data stream to define the truncation interval for the fitting algorithm. The system uses the actual action start time locked in the actual normal transition sample as the starting truncation cursor for the data extraction interval; it monitors point by point along the time axis, locking the specific timestamp node where the sampled power value first enters the following interval of the target power endpoint and continuously resides for a duration reaching the fault tolerance threshold. The system subtracts the physical time corresponding to the fault tolerance threshold from the timestamp of this node in reverse, and the resulting coordinate point is established as the termination truncation cursor. The system accurately extracts the purely monotonic ramp data set within the closed interval between the initial and final truncation cursors and stores it in the computation cache. For this finely trimmed dataset, a first-order least squares linear fitting algorithm is performed to obtain the linear equation. The slope coefficient of this equation is directly extracted as the actual work ramp slope parameter characterizing the current operating cycle. Multiple actual hysteresis time values ​​belonging to the same steady-state active power load rate interval are combined and summed arithmetically to obtain the average actual hysteresis duration; multiple actual work ramp slope parameters are combined and summed arithmetically to obtain the average actual ramp slope.

[0129] The processor addresses and loads the corresponding parameter positions in the multi-dimensional operating state mapping matrix structure within the storage system. It calls a preset configuration table to extract the decay influence factor parameters for the original historical data of the matrix and the current influence factor parameters for the newly entered sample data. The system applies asymmetric first-order exponential smoothing algebra constraint allocation logic in the underlying configuration table. The current influence factor parameter for the newly entered sample is set to a constant between 0.05 and 0.15 (preferably 0.1 in this embodiment), while the decay influence factor parameter for the original historical data is numerically constrained to the difference between a constant and the current influence factor parameter. The stored old expected mechanical hysteresis duration parameter and old expected load ramp slope parameter are multiplied by the decay influence factor parameter. The calculated average actual hysteresis duration and average actual ramp slope are multiplied by the current influence factor parameter. The corresponding matrix addition and update operation is performed, outputting the final expected value of the updated parameters combined with the smoothing factor. By utilizing the expected value of the updated parameter to overwrite and refresh the original data stored at the same location in the multidimensional operating state mapping matrix structure, a dynamic mapping association closed-loop system based on the physical characteristics of device evolution is established.

[0130] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying abnormal energy consumption in equipment based on time-series feature mining, characterized in that, include: Obtain the time-series energy consumption sequence of the target substation, and simultaneously extract the instruction issuance time and expected adjustment power contained in the load scheduling instruction for the target substation; Obtain the steady-state active power load rate of the target substation prior to the time the instruction is issued; Using the steady-state active power load rate, the corresponding expected mechanical hysteresis duration and expected load ramp-up slope are obtained. Add the expected mechanical hysteresis duration to the time when the instruction is issued to obtain the expected start time of the action. Find the power mutation point in the time-series energy consumption sequence as the start time of the actual action; The expected start time of the action and the actual start time of the action are time-aligned to output an aligned energy consumption sequence. Starting from the expected start time of the expected action, a physical transition trajectory is constructed by combining the expected adjustment power and the expected load ramp slope; The physical transition trajectory is spliced ​​to a pre-established steady-state operating baseline to generate a dynamic baseline; Compare the aligned energy consumption sequence with the dynamic baseline; When the aligned energy consumption sequence deviates from the following interval defined by the dynamic baseline, it is determined that the target power equipment has an energy consumption anomaly.

2. The method according to claim 1, characterized in that, The steps for constructing the pre-established steady-state operating baseline include: Collect stable energy consumption data of the target substation under historical no-dispatch conditions; The median filter is applied to the steady-state energy consumption data using a sliding time window to remove transient spike noise contained in the steady-state energy consumption data; The average value of the stable energy consumption data after removing transient peak noise is calculated according to a preset time step. The average values ​​corresponding to each preset time step are connected to form the steady-state operation baseline.

3. The method according to claim 2, characterized in that, The step of using a sliding time window to perform median filtering on the stable energy consumption data to remove transient spike noise points contained in the stable energy consumption data includes: Extract all discrete power sample values ​​within the sliding time window; Sort all discrete power sample values ​​within the sliding time window by amplitude; Select the middle amplitude value after sorting as the representative power value at the center position of the current window; Calculate the absolute difference between the original center power value and the representative power value; When the absolute difference exceeds the preset electromagnetic interference tolerance limit, the original center power value is determined to be the transient spike noise and is removed, and the representative power value is used to fill and cover it.

4. The method according to claim 1, characterized in that, The step of finding the power abrupt change point in the time-series energy consumption sequence as the starting time of the actual action includes: Starting from the moment the instruction is issued, calculate the power change rate between adjacent energy consumption sampling points in the time-series energy consumption sequence; Compare the power change rate with the inherent mechanical response start-up dead zone threshold of the target power equipment; In the time-series energy consumption sequence, the energy consumption sampling point corresponding to the first crossing of the mechanical response start-up dead zone threshold is identified; The energy consumption sampling point that first crosses the mechanical response start-up dead zone threshold is established as the starting moment of the actual action when the physical state begins to change.

5. The method according to claim 1, characterized in that, The step of aligning the expected start time of the action with the actual start time of the action and outputting an aligned energy consumption sequence includes: Calculate the absolute value of the time difference between the actual start time of the action and the expected start time of the action; Determine whether the absolute value of the time difference is less than the maximum transmission hysteresis limit allowed by the power grid dispatching network; If the absolute value of the time difference is less than the maximum transmission hysteresis limit, the time-series energy consumption sequence is shifted as a whole on the time axis by the absolute value of the time difference, so that the actual start time of the action coincides with the expected start time of the action, and the aligned energy consumption sequence in a synchronized alignment state is obtained. If the absolute value of the time difference is greater than or equal to the maximum transmission hysteresis limit, a communication interruption fault is established, and the time alignment step is terminated directly.

6. The method according to claim 1, characterized in that, The construction of the physical transition trajectory by combining the expected adjustment power with the expected load ramp slope includes: The original power value corresponding to the expected start time of the action is taken as the starting point of the trajectory, based on the steady-state operation baseline. By combining the expected adjustment power with the original power value, the target power endpoint that must be reached to complete the load scheduling task is calculated. Starting from the trajectory origin, construct a linearly varying slope according to the expected load ramp rate; The linear change line is extended and calculated until its value reaches the target power endpoint, and the line segment between the trajectory starting point and the time node corresponding to the value is taken as the physical transition trajectory.

7. The method according to claim 1, characterized in that, The step of stitching the physical transition trajectory to a pre-established steady-state operating baseline to generate a dynamic baseline includes: Locate the entry node corresponding to the expected start time of the action in the steady-state operation baseline; The target end time of the physical transition process is calculated using the expected adjustment power and the expected load ramp slope. Locate the cut-out node corresponding to the target end time in the steady-state operation baseline; Remove the original horizontal line segment of the steady-state operation baseline that is located between the cut-in node and the cut-out node; The physical transition trajectory is filled between the cut-in node and the cut-out node, and the dynamic baseline containing the actual physical response slope is output.

8. The method according to claim 1, characterized in that, Before comparing the aligned energy consumption sequence with the dynamic baseline, the method further includes the step of setting the following interval: Retrieve the nameplate parameters of the target substation equipment to obtain the percentage of inherent measurement error calibrated at the factory. Extract the reference power values ​​corresponding to each time node on the dynamic baseline; Multiply the reference power value by the inherent measurement error percentage to obtain the dynamic floating tolerance specific to each time point; The dynamic floating tolerance is superimposed and subtracted above and below the dynamic baseline to generate parallel upper and lower bound envelopes. The following interval is formed by the upper limit envelope and the lower limit envelope together in space.

9. The method according to claim 8, characterized in that, The step of determining that the target power equipment has an energy consumption anomaly when the aligned energy consumption sequence deviates from the following interval defined by the dynamic baseline includes: Extract the actual power sample values ​​from the aligned energy consumption sequence one by one; The actual power sampling value is compared with the upper limit envelope and the lower limit envelope corresponding to the same time node. When an actual power value greater than the value corresponding to the upper limit envelope or less than the value corresponding to the lower limit envelope is found, it is confirmed that the aligned energy consumption sequence deviates from the following interval. Abnormal data points that deviate from the following interval are continuously counted. When the total number of consecutive counts exceeds the fault tolerance threshold, the target power equipment is determined to have an energy consumption abnormality.

10. The method according to claim 9, characterized in that, Also includes: If the aligned energy consumption sequence is within the following interval, then the target power equipment is considered to have successfully responded to the scheduling task and the physical execution process is normal. Intercept over-limit alarm actions that were originally expected to be triggered by load changes in the substation monitoring backend; Extract the aligned energy consumption sequence that is within the following interval as the actual normal transition sample; The actual normal transition samples are saved to the equipment feature archive for periodic correction of the mapping parameters of the expected mechanical hysteresis duration and the expected load ramp-up slope.

Images