A method for identifying and classifying power consumer load characteristics

Abstract

The application provides a power user load feature recognition classification evaluation method, comprising: identifying start interval and power superposition time through analysis of start-stop sequence change identification, grouping the start interval and power superposition time according to equipment type and power level to obtain equipment start-stop time combination; evaluating start time difference and power superposition time difference of different users with the same production process in the same industry according to the equipment start-stop time combination, and identifying user electricity behavior similarity; identifying industrial user electricity load data through subdivision of user category label, evaluating load prediction accuracy and electricity mode stability according to the industrial user electricity load data to obtain electricity load feature; updating equipment start-stop time record database according to the electricity load feature, and obtaining final electricity load data recognition classification evaluation result through analysis of the electricity load feature.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, in particular to a power user load feature identification and classification evaluation method. BACKGROUND

[0002] In the field of power user load feature identification and classification evaluation, it is of great significance to study the regularity of user electricity consumption behavior, which not only relates to the stable operation of the power system, but also directly affects the efficiency of energy distribution and the accuracy of user service. By analyzing the user's electricity consumption mode, key basis can be provided for power dispatching and equipment management, thereby improving the intelligent level of the entire power grid. However, some current methods are more concerned with static power data when dealing with complex electricity consumption behavior, and fail to deeply mine the behavior information contained in the device start-stop time sequence, especially after the user adjusts the operation strategy, it is often difficult to capture the difference in the use habit and operation sequence of the device, resulting in inaccurate classification results. The device start-stop sequence reflects the user's specific operation habit, such as some industrial users starting the device according to a fixed process, while different users show significant differences in the starting time interval and power superposition opportunity. The dynamic change of this sequence directly affects the identification effect of electricity consumption behavior. If the change of this sequence cannot be accurately grasped, it is difficult to distinguish between user groups with similar devices but different use habits, thereby causing classification evaluation deviation. For example, in a certain industrial scene, the user originally operates according to the sequence of starting the main device first and then starting the auxiliary device, but due to the adjustment of the maintenance plan, the sequence changes to starting the auxiliary device first, and then starting the main device, and the time interval also changes. The change of this sequence and interval makes the original behavior pattern no longer applicable, and the identification system fails to adjust in time, resulting in classification errors. Therefore, how to re-evaluate the similarity between different users when the device start-stop sequence changes dynamically has become a key problem to be solved. SUMMARY

[0003] The present application provides a power user load feature identification and classification evaluation method, which can accurately capture the difference in time interval and power superposition when the device start-stop sequence changes dynamically, and re-evaluate the similarity between different users accordingly.

[0004] To solve the above technical problems, the method provided by the embodiments of the present application mainly includes:

[0005] The system collects power-on time, power-off time, operating power values, and start-up time differences between devices corresponding to industrial user equipment types and power levels. It retrieves historical start-up / stop sequence data from a device start-up / stop time record database, identifying the time deviation between the power-on time and the historical start-up / stop sequence data, and the power difference between the operating power value and the historical power record. Based on the time deviation and power difference values, it assesses the time difference between the start-up / stop sequence at the current power-on time and the historical start-up / stop sequence data. When the time difference exceeds a preset threshold, it determines that the start-up / stop sequence has been adjusted and generates a start-up / stop sequence change identifier. Based on the start-up / stop sequence change identifier, it identifies the time difference between device start-up times. The startup interval and power superposition time are grouped according to the equipment type and power level to generate equipment start-up and shutdown time combinations. Based on the equipment start-up and shutdown time combinations, the startup time difference and power superposition time difference of different users in the same industrial park are evaluated to generate user electricity consumption behavior similarity. Based on the user electricity consumption behavior similarity, high similarity user groups are identified, the electricity consumption time distribution of the high similarity user groups is statistically analyzed, the peak and valley time overlap rate and load fluctuation amplitude are calculated, and subdivided user categories are generated. Based on the subdivided user categories and electricity load data, electricity load characteristics are generated, and the equipment start-up and shutdown time record database is updated.

[0006] Furthermore, the process of collecting the start-up time, shutdown time, operating power value, and start-up time difference between equipment corresponding to the type and power level of industrial user equipment, obtaining historical start-up and shutdown sequence data from the equipment start-up and shutdown time record database, and identifying the time deviation value between the start-up time and the historical start-up and shutdown sequence data, and the power difference value between the operating power value and the historical power record, includes:

[0007] The system acquires real-time operating data of industrial user production equipment, records the start-up time, the shutdown time, and the operating power value, generates equipment start-up and shutdown sequence records according to the equipment type and power level, calculates the start-up time difference between adjacent equipment, and generates a start-up and shutdown dataset containing the equipment number and the start-up time difference; it extracts historical start-up and shutdown sequence data of the corresponding user from the equipment start-up and shutdown time record library, compares the start-up time of the start-up and shutdown dataset with the start-up time of the historical start-up and shutdown sequence data, calculates the time deviation value, and compares the operating power value with the historical power record to calculate the power difference value.

[0008] Furthermore, the step of evaluating the time difference between the start-up / stop sequence at the power-on time and the historical start-up / stop sequence data based on the time deviation value and the power difference value, and determining that the start-up / stop sequence is adjusted when the time difference exceeds a preset threshold, and generating a start-up / stop sequence change identifier, includes:

[0009] Based on the time deviation value and the power difference value, a time difference value is calculated; when the time difference value exceeds a preset threshold, the timing data of the power-on time is extracted, the average interval difference between the start-up time interval of adjacent devices and the historical start-up and shutdown sequence data is calculated, and a change amplitude marker is generated based on the difference ratio; based on the change amplitude marker, a start-up and shutdown sequence change identifier containing the adjustment type and device number is generated.

[0010] Furthermore, the step of identifying the device start-up interval and power superposition time based on the start-up / stop sequence change identifier, and grouping the start-up interval and power superposition time according to the device type and the power level to generate a device start-up / stop time combination includes:

[0011] The start-stop sequence change identifier is parsed to extract the adjustment type and device number. The device start-stop time record database is queried to obtain the power-on time and the power-off time, and the device start-up interval is calculated. The overlapping period of the power-on time and the power-off time is identified, and the power superposition time is recorded. The operating power value of the overlapping period is accumulated to generate a power superposition record. The device start-up interval and the power superposition record are grouped according to the device type and the power level to generate a device start-stop time combination containing the device number, the device start-up interval, and the power superposition time.

[0012] Furthermore, the step of evaluating the start-up time difference and power superposition time difference of different users within the same industrial park based on the combination of equipment start-up and shutdown times, and generating user electricity consumption behavior similarity, includes:

[0013] Based on the equipment start-up and shutdown time combinations, extract the equipment configurations of users within the industrial park, filter users with the same equipment type and power level, obtain the start-up time sequence and power change curve, calculate the start-up time difference between users, and generate a start-up difference record; compare the power superposition times, calculate the time offset and peak difference, and generate a power curve difference metric; based on the start-up time difference and the power curve difference metric, generate an electricity consumption behavior feature vector, calculate the similarity between vectors, and generate user electricity consumption behavior similarity.

[0014] Furthermore, the step of identifying highly similar user groups based on the similarity of user electricity consumption behavior, statistically analyzing the electricity consumption time distribution of the highly similar user groups, calculating the peak-valley time overlap rate and load fluctuation amplitude, and generating subdivided user categories includes:

[0015] Based on the similarity of user electricity consumption behavior, high-similarity user groups are screened, the power curves of the high-similarity user groups are extracted, the operating power values ​​for each hourly period are statistically analyzed, and a time-period power matrix is ​​generated. The peak and valley periods of the time-period power matrix are identified, and the overlap rate of the peak and valley periods and the load fluctuation amplitude are calculated. Based on the overlap rate of the peak and valley periods and the load fluctuation amplitude, subdivided user categories including peak-consumption type, stable-consumption type, and intermittent-consumption type are generated.

[0016] Furthermore, the step of generating electricity load characteristics and updating the equipment start-up and shutdown time record database based on the subdivided user categories and electricity load data includes:

[0017] Based on the subdivided user categories, historical electricity load data of users is extracted to generate a load dataset containing time series and operating power values; based on the load dataset, prediction accuracy and pattern stability are calculated to generate electricity load features containing the prediction accuracy and pattern stability; based on the electricity load features, the feature fields of the equipment start-up and shutdown time record library are updated.

[0018] Furthermore, after generating the electrical load characteristics, the process includes:

[0019] Update the user records in the equipment start-up and shutdown time record library according to the power load characteristics; extract the subdivided user categories, prediction accuracy and user power consumption behavior similarity, and integrate them to generate an evaluation result that includes the subdivided user categories and prediction accuracy.

[0020] The technical solutions provided in this application embodiment may include the following beneficial effects:

[0021] This application discloses a method for identifying, classifying, and evaluating the load characteristics of electricity users. Addressing the challenges of complex electricity consumption behaviors among multiple users and devices in industrial parks, characterized by dynamic changes in start-up and shutdown sequences and insufficient load forecasting accuracy, the method collects equipment start-up and shutdown times, operating power, and historical data. It calculates time deviations and power differences to identify changes in start-up and shutdown sequences, and then analyzes the start-up intervals and power superposition times in groups to generate equipment start-up and shutdown time combinations. Based on this, the application evaluates the start-up time differences and power superposition time differences among users, identifies highly similar user groups, analyzes their electricity consumption period distribution and peak-valley overlap rate, classifies users into peak, stable, and intermittent electricity consumption types, further updates the equipment start-up and shutdown record database, optimizes load forecasting accuracy and electricity consumption pattern stability, and ultimately generates a classification and evaluation result of electricity load data that includes user category, predicted value, and similarity score. Through multi-dimensional data fusion and dynamic analysis, this application significantly improves the accuracy of industrial user electricity consumption behavior classification and load forecasting capabilities, providing efficient support for park energy management. Attached Figure Description

[0022] Figure 1 This is a flowchart of a power user load characteristic identification, classification and evaluation method according to this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; "and / or" in this text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0024] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0025] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0026] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner to facilitate understanding.

[0027] It should be understood that in this application, "at least one (item)" means one or more. "More than one" means two or more. "At least two (items)" means two or three or more. "And / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural.

[0028] The character " / " generally indicates that the preceding and following objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any single or multiple items. For example, "at least one of a, b, or c" can be expressed as: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0029] Both "...when" and "if" indicate that a corresponding action will be taken under certain objective circumstances. They are not time limits, nor do they require a judgment action to be taken when the action is taken, nor do they imply any other limitations.

[0030] The method provided in this application embodiment can be executed by a wind power system network security risk identification device. This wind power system network security risk identification device can be an electronic device or a device applied in an electronic device, such as a data analysis module. The electronic device can refer to a mobile phone, server, or other such device, and this application embodiment does not limit this.

[0031] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0032] like Figure 1 This embodiment of a method for identifying, classifying, and evaluating the load characteristics of electricity users may specifically include:

[0033] Step S101: Collect the start-up time, shutdown time, operating power value, and start-up time difference between different equipment types and power levels of industrial users; obtain historical start-up and shutdown sequence data from the pre-established equipment start-up and shutdown time record library; and identify the time deviation value and power difference value between the current collected data and the historical data.

[0034] As an example, real-time operating data of various production equipment of industrial users can be obtained through smart meters and data acquisition terminals. This operating data can include, for example, the start-up and shutdown times of equipment such as air compressors, conveyor belts, and machining centers. The real-time power values ​​of each device during operation are recorded. Equipment start-up and shutdown sequence records are established based on equipment number and power level. The start-up time interval between adjacent devices is calculated, forming a current equipment start-up and shutdown dataset containing equipment number, start-up and shutdown times, and start-up intervals. Historical start-up and shutdown sequence data for the corresponding industrial user are extracted from a pre-established equipment start-up and shutdown time record library. The start-up time of each device in the current dataset is compared one by one with the historical records. The time deviation value is obtained by calculating the difference between the current start-up time and the historical start-up time. The power difference value is obtained by calculating the difference between the current power value and the historical power record. It is then determined whether the time deviation value exceeds a preset 30-minute deviation threshold. If the time deviation value exceeds the 30-minute deviation threshold, the change in equipment operating status is assessed based on the ratio of the power difference value to the historical average power. The start-stop correlation characteristics of the equipment are identified by statistical analysis of the change trend of the start-up time interval, and the equipment operating deviation identification results containing the time deviation value, power difference value and correlation characteristic markers of each equipment are obtained.

[0035] Specifically, in one implementation, smart meters and data acquisition terminals are deployed in the power distribution cabinets and on various production equipment of industrial users to collect electrical parameters of equipment such as air compressors, conveyor belts, and machining centers in real time. The smart meters record the operating status of the equipment once per minute. When the current jumps from zero to more than 10% of the rated value, it is recorded as the start-up time; when the current drops below 5% of the rated value and remains below 30 seconds, it is recorded as the shutdown time. Real-time power values ​​are calculated based on the equipment nameplate power and the measured current value, forming a raw data stream containing timestamps, equipment numbers, and power values.

[0036] Specifically, the process of establishing equipment start-up and shutdown timing records involves processing and organizing the raw data stream. The collected data is grouped according to the equipment number, the changes in equipment status within the same time window are sorted, and the start-up time interval between adjacent devices is calculated.

[0037] For example, if the air compressor starts at 7:00:00 AM and the conveyor belt starts at 7:03:30 AM, the start-up time interval between the two is 210 seconds. This time interval reflects the process requirements of industrial production; the air compressor needs to establish air pressure before the conveyor belt can operate normally.

[0038] It should be noted that the time deviation value is calculated using the start-up time difference method for the corresponding equipment. Historical data from the same period is extracted from the equipment start-up and shutdown time record database, for example, start-up and shutdown records for the same time period in the previous 30 days are extracted as a reference benchmark. If the current start-up time of the air compressor is 7:00:00, while the historical average start-up time is 6:30:00, then the time deviation value is 30 minutes. The power difference value is obtained by the difference between the current measured power and the historical average power. If the current power of the air compressor is 55 kW, and the historical average power is 50 kW, then the power difference value is 5 kW.

[0039] In one possible implementation, the 30-minute deviation threshold is set based on a comprehensive consideration of industrial production shift changes and equipment warm-up time. When the time deviation exceeds this threshold, it is determined that a change in the start-up and shutdown sequence may have occurred due to production plan adjustments or equipment maintenance. The ratio of the power difference value to the historical average power is used to assess changes in equipment load; a ratio exceeding 15% is considered a significant change in the equipment's operating status.

[0040] Preferably, the correlation between device start-up and shutdown is identified by statistically analyzing the changing trends of start-up time intervals of multiple devices. A histogram of time interval distribution is constructed to identify the central tendency and dispersion of the intervals. When the standard deviation of the interval distribution exceeds 20% of the mean, it is marked as a weakening correlation; otherwise, it is marked as a stable correlation.

[0041] Step S102: Evaluate the time difference between the current start-stop sequence and the historical start-stop sequence based on the time deviation value and power difference value. If the time difference exceeds the preset time difference threshold, it is determined that the start-stop sequence has changed. Then, the start-stop sequence change indicator is obtained by analyzing the change range of the device start-stop time.

[0042] The system extracts start-up and shutdown records of all devices within a preset time period from the historical database. These records are grouped by device number and start-up time. The average start-up and shutdown interval and standard deviation between each group are calculated. The actual start-up and shutdown time of each device within the current acquisition cycle is obtained. The time interval between adjacent devices is calculated. The difference between this time interval and the historical average start-up and shutdown interval is calculated. The absolute values ​​of all differences are summed to obtain the time deviation value. A fixed-length time window is used to scan the power data of the most recent acquisition cycles. The ratio of the power difference to the time difference at adjacent time points for each device is calculated as the power change rate. This power change rate is compared with the historical average power change rate. If it exceeds a preset change rate threshold, it is marked as a power anomaly. The ratio of the number of all anomalies to the total number of acquisition points is counted to obtain the power difference value. The time deviation value and the power difference value are evaluated separately. If the time deviation value exceeds a preset time difference threshold, or the power difference value exceeds a preset power difference threshold, the start-up and shutdown sequence is determined to have changed. Simultaneously, the difference between the current start / stop time and the historical start / stop time is calculated. The difference between the current start / stop time and the historical start / stop time is divided by the historical start / stop time to obtain the relative change rate. Based on the preset range of the relative change rate, the corresponding start / stop sequence change indicator is determined.

[0043] Specifically, historical database data extraction employs a hierarchical indexing approach. First, a primary index is created based on the equipment number. Then, a secondary index is created under each equipment number based on its start-up time, forming a dual-grouping structure. Specifically, for key equipment such as compressor units, cooling towers, and circulating pumps within the industrial park, the system initially groups start-up and stop records within a preset time period by equipment number. Subsequently, within each equipment group, these records are arranged in chronological order of start-up time. The average start-up and stop interval is calculated by statistically analyzing the time difference of the same equipment over multiple start-up cycles. The mean is obtained by summing all time intervals and dividing by the total number of intervals. The standard deviation is obtained by calculating the sum of squared deviations of each interval from the mean, dividing by the sample size, and then taking the square root. For example, the actual start-up and stop times within the current acquisition cycle rely on real-time data acquisition from the equipment power monitoring system. When the equipment power jumps from zero power in a static state to more than 15% of the operating power, the system automatically records this moment as the equipment start-up time. When the power drops from normal operating conditions to below 8% of the rated power and remains below this level for more than three seconds, it is recorded as the equipment stop time. The time interval between adjacent devices is calculated by subtracting timestamps. The start time of the next device in the device sequence is subtracted from the start time of the previous device to form the start interval sequence between devices. The difference between the time interval and the historical average start-stop interval is calculated by comparing the current interval with the corresponding historical average interval one by one. After calculating the difference between each pair of intervals, the absolute value is taken, and all absolute values ​​are summed to obtain the time deviation value.

[0044] It should be noted that the fixed-length time window is determined based on the power data acquisition frequency and the device response characteristics. When the data acquisition frequency is twice per second, the time window length is typically set between 30 and 90 seconds. The power change rate is calculated by the ratio of the power difference between adjacent sampling points within the sliding window to the corresponding time difference; that is, the power at the current sampling point minus the power at the previous sampling point, divided by the time interval between the two sampling points. The historical average power change rate is calculated as a baseline value by collecting power change rate data of similar devices within the same time period and using an arithmetic mean method. When the real-time calculated power change rate exceeds the historical average plus a preset change rate threshold, the sampling point is marked as a power anomaly. The power difference index is formed by the ratio of the total number of anomalies to the total number of sampling points within the statistical time window.

[0045] In one possible implementation, the dual evaluation mechanism employs independent judgment and parallel processing. The time deviation evaluation module and the power difference evaluation module respectively compare their respective index values ​​against thresholds. When the time deviation value exceeds a preset time difference threshold, or the power difference value exceeds a preset power difference threshold, either condition being met triggers a change in the start-stop sequence. The relative change rate is calculated by dividing the absolute difference between the current start-stop time and the historical start-stop time by the historical start-stop time, with the historical start-stop time as the denominator. The resulting value reflects the degree of change of the current start-stop time relative to the historical baseline. Preferably, the determination of the start-stop sequence change indicator is based on the numerical range of the relative change rate. When the relative change rate is between 0% and 5%, the system assigns a "minor change" indicator, indicating that the start-stop sequence remains basically stable; when it is between 5% and 15%, it assigns a "slight change" indicator; when it is between 15% and 30%, it assigns a "moderate change" indicator; and when it exceeds 30%, it assigns a "severe change" indicator.

[0046] Step S103: By analyzing the start-stop sequence change identifiers, the start interval and power superposition time are identified. The start interval and power superposition time are grouped according to the equipment type and power level to obtain the equipment start-stop time combination.

[0047] By parsing the encoded information in the start-stop sequence change identifier, three dimensions of data are extracted: sequence type label, change magnitude level, and number of affected devices. Based on the encoded information, the device start-stop time record database is queried to obtain the detailed start-stop time sequence of the corresponding devices. The order in which devices start is identified, and the time interval between adjacent starting devices is calculated, forming a time-series dataset containing device numbers and start-up intervals. Based on the start-up and shutdown times in the time-series dataset, overlapping periods of multiple devices operating simultaneously within a preset time window are identified. When the operating periods of two or more devices overlap, the start time of power superposition is recorded. The power superposition value is obtained by accumulating the operating power of each device during the overlapping period, constructing a power superposition record containing the superposition time, a list of participating device numbers, and the superposition power value. The power superposition record is classified according to device type attributes, categorized into power equipment, heating equipment, and auxiliary equipment based on the device's process function. Within each device type, it is further divided into three subgroups based on rated power: high power level, medium power level, and low power level. The start-up intervals within each subgroup are statistically processed to obtain the interval duration distribution characteristics. Based on the interval duration distribution characteristics and power superposition records, a combined data structure for device start-stop time is constructed. Each combined record contains a set of participating device numbers, a sequence of start-stop time intervals between devices, the time when power superposition occurs, and the corresponding power superposition value. A complete combination of device start-stop time is formed by associating device numbers with time intervals.

[0048] Specifically, in one implementation, the parsing process of the start-stop sequence change identifier extracts multi-dimensional information through bitwise operations and masking. After receiving the eight-bit binary code, bit shifting and AND operations are used to separate the data of each dimension. The first bit is ANDed with 0x80 to extract the sequence type marker; the middle three bits are right-shifted by four bits and then ANDed with 0x07 to obtain the change magnitude level; the last four bits are directly ANDed with 0x0F to obtain the number of affected devices. The parsed data is used to query the device start-stop time record database, locate the specific historical start-stop record through the device number index, and extract complete time-series data including the start-up time, shutdown time, and operating power.

[0049] Specifically, constructing a time-series dataset requires sorting and associating the extracted start-up and shutdown times. All devices are sorted according to their start-up times, forming a device startup sequence. For each device in the sequence, the time interval between it and the previous started device is calculated; this interval reflects the dependencies and process requirements of device startup in industrial production.

[0050] For example, a company's blast furnace blower must be started within 10 minutes after the blast furnace is ignited, while the dust removal equipment needs to be turned on 5 minutes before the blower is started. This timing relationship is reflected by the starting interval value, forming a timing data record that includes equipment number pairs, interval duration, and starting sequence index.

[0051] It should be noted that the identification of power superposition moments uses a sliding time window method. A 30-second time window is set, sliding in 1-second increments throughout the entire operating cycle. At each time window position, the operating status of all devices is checked. When an overlap is detected between the operating segments of two or more devices, the start time of overlap is recorded. The power superposition value is obtained through real-time cumulative calculation. The system reads the actual operating power of each participating device at that moment and performs algebraic summation.

[0052] For example, if the air compressor has an operating power of 75 kW, the cooling water pump has an operating power of 15 kW, and the conveyor belt motor has an operating power of 22 kW, then the total power at that moment is 112 kW.

[0053] Preferably, equipment types are classified based on their process function characteristics. Power equipment includes motors, air compressors, fans, etc., which provide mechanical power; the start and stop of this type of equipment directly affects the operation of the production line. Heating equipment includes electric furnaces, heaters, drying equipment, etc., whose power consumption has phased characteristics. Auxiliary equipment includes lighting, ventilation, cooling systems, etc., which typically operate as supporting facilities for production. Within each equipment type, the power level is divided using a logarithmic scaling method, grouping rated power according to orders of magnitude to ensure that the power characteristics of equipment within each level are similar, facilitating subsequent statistical analysis and pattern recognition.

[0054] In one possible implementation, the statistical processing of the interval duration distribution characteristics involves the calculation of multiple statistics. Frequency statistics are performed on the start-up interval data within each subgroup to construct a frequency distribution histogram of the interval duration. By calculating central tendency indicators such as the mean, median, and mode of the intervals, typical start-up patterns of the equipment in that group are identified. Standard deviation and coefficient of variation are used to measure the dispersion of the start-up intervals; when the coefficient of variation exceeds 0.3, it indicates significant variability in the start-up timing of the equipment in that group. Skewness and kurtosis coefficients further describe the distribution pattern; a positively skewed distribution indicates the presence of occasional long time intervals, which may correspond to equipment maintenance or process adjustments.

[0055] For example, a chemical plant's reactor group comprises three reactors of different capacities, belonging to the high-power class of heating equipment. Statistical analysis revealed a bimodal distribution in the start-up intervals of the three reactors: one peak around 5 minutes, corresponding to sequential start-up during normal production; the other peak around 45 minutes, corresponding to interval start-up during batch switching. This distribution reflects the batch characteristics and process requirements of chemical production, providing a benchmark for identifying abnormal start-up and shutdown patterns. Furthermore, the design of the equipment start-up and shutdown time combination data structure employs a nested organization. The main structure includes metadata such as combination identifier, creation timestamp, and expiration date marker. The set of participating equipment numbers is stored in array form to maintain the equipment start-up order. The start-up and shutdown time interval sequence records the time difference between adjacent equipment, using relative time representation to reduce storage space. Power superposition information is organized in the form of time-value pairs, recording only key moments of power changes; the complete power curve can be reconstructed through interpolation.

[0056] Understandably, the complete combination formed by associating device numbers with time intervals not only contains static device information but also implies dynamic operating modes. A pattern matching algorithm compares the similarity between the current combination and historical typical combinations, considering three dimensions: device composition, startup order, and interval duration. When the similarity falls below a preset threshold, a new operating mode is identified, triggering a reassessment process for subsequent electricity consumption behavior.

[0057] For example, the equipment start-stop time combination display in a machining workshop shows that the original start-up sequence of machining center-conveyor line-cleaning machine has been adjusted to conveyor line-machining center-cleaning machine, with the start-up of the machining center delayed by 15 minutes. This adjustment reflects the optimization of the production process, reducing the waiting time of the machining center by starting the conveyor line in advance for material preparation. The corresponding power superposition pattern has also changed, with peak power shifting from a concentrated occurrence to a more dispersed distribution, which helps reduce the instantaneous load impact on the power grid.

[0058] Step S104: Based on the combination of equipment start-up and shutdown times, assess the time difference of start-up time and power superposition between different users with the same equipment in the same industrial park, and identify the similarity of users' electricity consumption behavior.

[0059] Based on the combined data of equipment start-up and shutdown times, equipment configuration information of each user within the same industrial park is extracted. User groups with equipment of the same type and power level are selected, and the equipment start-up time sequence and power change curve for each user are obtained. The start-up time difference between corresponding devices of different users is calculated, forming a user start-up difference record containing user pair identifiers, equipment types, and time difference values. Based on this user start-up difference record, the time points of power superposition for each user are extracted. The power superposition patterns of different users within the same time window are compared, and the offset of superposition time and the difference of superposition peak value are calculated. A dynamic time warping algorithm is used to align the power curves of different users along the time axis, outputting the aligned power curve difference metric. Based on the power curve difference metric and start-up time difference, a user electricity consumption behavior feature vector is constructed. A sequence consistency score is obtained by statistically analyzing the consistency of the start-up sequence, an interval stability score is obtained by calculating the variance of the time interval, and a superposition overlap rate score is obtained based on the overlap ratio of the power superposition periods. These three scores are combined into a feature vector, and the cosine value between the vectors is calculated to obtain the behavioral similarity value between user pairs. A user similarity matrix is ​​constructed using the aforementioned behavioral similarity values. Users in the matrix are merged level by level using a hierarchical clustering algorithm. When the average similarity of users within a group exceeds a preset threshold, a user group is formed. The output includes user grouping results and average similarity within the group, which are used to identify the similarity of user electricity consumption behavior.

[0060] Specifically, in one implementation, the filtering of user equipment configuration information within the industrial park is achieved by establishing an equipment feature database. A list of all connected users' equipment is extracted from the park's power management platform, including basic information such as equipment model, rated power, manufacturer, and commissioning date. By matching equipment type codes and power ratings, user groups with the same or similar equipment are identified.

[0061] For example, in a certain industrial park, there are 12 companies, 8 of which are equipped with the same model of weaving machines and dyeing equipment. These companies are grouped into the same comparison group. An independent equipment startup time sequence is established for each user. The sequence records all start-up and stop events of each device within the monitoring period, forming a triplet data structure of timestamp-device number-action type. The startup time difference is calculated using a corresponding device matching method. First, corresponding devices between different users are identified. For example, user A's air compressor #1 and user B's air compressor #3 are of the same model, establishing a device mapping relationship. For each pair of mapped devices, their startup times within the same time period are extracted, and the absolute value of the time difference is calculated. When a device starts at one user but does not run at another user, it is recorded as a missing value and specially marked. By traversing all device pairs, a user-to-user startup difference record is formed, containing a matrix of time difference values. The rows and columns of the matrix correspond to the devices of different users, and the element values ​​are the startup time differences.

[0062] It should be noted that the identification and comparison of power overlay patterns involves a complex time-series analysis process. Using a 5-minute time window, a sliding scan is performed within a 24-hour period to identify power overlay events for each user. Power overlay occurs when multiple devices are operating simultaneously; the system records the start time of overlay, the list of participating devices, the peak overlay power, and the duration. Overlay patterns from different users are compared using two dimensions: time offset and power difference.

[0063] For example, user A has three devices running simultaneously at 9:00 AM, with a total power of 180 kilowatts; user B has a similar simultaneous operation at 9:15 AM, with a total power of 165 kilowatts. The time offset is 15 minutes, and the power difference is 15 kilowatts.

[0064] Preferably, the dynamic time warping algorithm is used to handle the time axis differences of power curves from different users. This algorithm finds the optimal alignment path between two curves by constructing a cumulative distance matrix. The algorithm first discretizes the power time series of the two users into equally spaced sampling points, each containing a time and power value. A distance matrix is ​​constructed, where each element is the square of the power difference between corresponding points in the two series. Using dynamic programming, the algorithm finds the path with the minimum cumulative distance from the top left corner to the bottom right corner of the matrix; this path defines the alignment relationship between the two curves. The aligned curves can then be compared point-by-point to calculate an overall difference metric, which reflects the degree of difference in power usage patterns between the two users.

[0065] For example, constructing a user electricity consumption behavior feature vector requires integrating information from multiple dimensions. The sequence consistency score is calculated using the Kendall rank correlation coefficient, which measures the consistency of the startup sequence of two user devices, ranging from -1 to 1, where 1 indicates complete consistency and -1 indicates complete reversal. The time interval stability score is represented by the reciprocal of the coefficient of variation; first, the ratio of the standard deviation to the mean of each device's startup interval is calculated, and then the reciprocal is taken, resulting in a higher score for higher stability. The power overlap rate is obtained by calculating the proportion of the intersection duration to the union duration of the overlapping periods of two users. These three scores are then normalized to form a three-dimensional feature vector.

[0066] In one possible implementation, cosine similarity reflects the overall similarity of user behavior patterns. The cosine value of two feature vectors is equal to the product of their dot product and their magnitudes, with the result ranging from 0 to 1, where a value closer to 1 indicates greater similarity.

[0067] For example, the feature vectors of two companies in a certain industrial park are [0.85, 0.72, 0.68] and [0.82, 0.75, 0.71], respectively. The calculated cosine similarity is 0.996, indicating that the two companies' electricity consumption behaviors are highly similar. Furthermore, the user similarity matrix is ​​a symmetric matrix, with each element recording the behavioral similarity values ​​between any two users. The hierarchical clustering algorithm groups users based on this matrix. Initially, each user is treated as an independent cluster. In each iteration, the two clusters with the highest similarity are merged. The distance between clusters is updated using the average connectivity method, meaning the distance between the new cluster and other clusters is equal to the average distance between the original two clusters and other clusters. The clustering process continues until all users are merged into one cluster or a preset similarity threshold is reached. When the average similarity of all user pairs within a cluster exceeds 0.75, the users within that cluster are considered to have the same electricity consumption behavior pattern. The clustering results form a dendrogram structure. By cutting the dendrogram at different heights, user groups of different granularities can be obtained.

[0068] For example, cluster analysis of 15 companies in an electronics manufacturing park resulted in three main user groups: the first group, comprising 6 companies, primarily engaged in surface mount and soldering operations, with equipment startup concentrated at the start of the morning shift; the second group, consisting of 4 companies, focused on assembly and testing, with relatively stable equipment operation; and the third group, comprising 5 companies, engaged in injection molding, exhibiting clear batch production characteristics. The average similarity within each group was 0.82, 0.79, and 0.86, respectively, while the inter-group similarity was all below 0.45, indicating that the grouping results had good discriminative power.

[0069] Step S105: Identify high-similarity user groups based on the similarity of electricity consumption behavior, identify the electricity consumption time distribution of high-similarity user groups, evaluate the overlap rate of peak and valley periods and the load fluctuation amplitude of high-similarity user groups based on the electricity consumption time distribution, and obtain subdivided user categories.

[0070] Based on the electricity consumption behavior similarity identification results, user pairs with similarity values ​​exceeding a preset threshold are selected and clustered to form high-similarity user groups. The 24-hour electricity consumption power curves of all members within each user group are extracted. Power values ​​for each user are statistically analyzed by hourly time period, constructing a time-period power matrix. Rows in the matrix represent users within the group, and columns represent 24-hour time periods, resulting in the electricity consumption time period distribution of the high-similarity user groups. Based on this distribution, peak power periods and valley power periods for each user are identified. Peak periods are defined as times when power exceeds a preset high multiple of the daily average power, and valley periods are defined as times when power is lower than a preset low multiple of the daily average power. The overlap rate of peak and valley periods among different users within the group is calculated as the proportion of the total number of peak and valley periods. The load fluctuation amplitude is calculated as the percentage of the difference between the maximum and minimum daily power values ​​for each user relative to the average power. Based on the peak and valley overlap rate and load fluctuation amplitude, the K-means clustering algorithm is used to classify the user groups. The overlap rate and fluctuation amplitude are used as two-dimensional feature inputs, and three cluster centers are determined through iterative calculation. Each center represents a typical electricity consumption pattern. If the peak-valley overlap rate of the cluster center exceeds a preset high threshold and the fluctuation amplitude exceeds a preset fluctuation threshold, the cluster is determined to be a peak power consumption type; if the fluctuation amplitude is lower than a preset low threshold, it is determined to be a stable power consumption type; if the fluctuation amplitude is between the high and low thresholds, it is determined to be an intermittent power consumption type, and the output is a subdivided user category including user group identifier and category label.

[0071] Specifically, in one implementation, the identification of highly similar user groups is based on user behavior similarity scores obtained in previous steps. A similarity threshold of 0.75 is set, and all user pairs with similarity exceeding this threshold are selected. An agglomerative hierarchical clustering method is then used to merge users whose similarity to each other all exceed the threshold into one group.

[0072] For example, if user A has a similarity of 0.82 with user B, user B has a similarity of 0.79 with user C, and user A has a similarity of 0.85 with user C, then all three are grouped into the same high-similarity user group. Each user group must contain at least 3 members to ensure the validity of the statistical analysis.

[0073] Specifically, constructing the time-period power matrix involves organizing and normalizing the 24-hour power data of all users within the group. Hourly power data for each user within the monitoring period is extracted from the database. For each hourly time, the average power for that hour is calculated as the representative value for that time period. Each row of the matrix corresponds to one user, containing the power values ​​for that user across 24 time periods from 0:00 to 23:00. To eliminate the impact of differences in installed capacity among different users, each row of data is standardized by dividing it by the user's daily average power, ensuring comparability of electricity usage patterns for businesses of different sizes.

[0074] It should be noted that a dynamic threshold method is used to identify peak and valley periods. First, the daily average power consumption for each user is calculated as the criterion. The threshold for peak periods is set by multiplying the daily average power consumption by a preset high multiplier, typically between 1.3 and 1.8; the threshold for valley periods is set by multiplying the daily average power consumption by a preset low multiplier, typically between 0.3 and 0.6. For each hourly period, if the power consumption exceeds the peak threshold, it is marked as a peak period; if it falls below the valley threshold, it is marked as a valley period; and periods in between are considered normal periods. This dynamic threshold method can adapt to the differences in industries and production scales.

[0075] Preferably, the calculation of the peak-valley overlap rate reflects the synchronicity of electricity consumption behavior among users within the group. The peak and valley time sets for each user are statistically analyzed. For any two users within the group, the peak time overlap rate is obtained by dividing the number of hours of their peak time overlap by the number of hours of their peak time union; similarly, the valley time overlap rate is calculated. The weighted average of the two overlap rates is taken as the overall peak-valley overlap rate.

[0076] For example, a user group in an industrial park includes 5 companies. The peak electricity consumption of 4 of these companies is concentrated between 8-11 am and 2-5 pm, with a peak overlap rate of 0.85, indicating that the production arrangements of these companies are highly synchronized.

[0077] For example, the peak-valley difference rate is used to calculate the load fluctuation amplitude. For each user, the maximum and minimum values ​​of their 24-hour power curve are extracted, and the percentage of the difference between the two to the daily average power is calculated. The fluctuation amplitude reflects the stability of the user's electricity consumption; the larger the value, the more uneven the electricity consumption. Some continuous production enterprises, such as those in the chemical and metallurgical industries, typically have load fluctuation amplitudes below 30%; while batch production enterprises, such as those in the machinery and food processing industries, may have fluctuation amplitudes exceeding 80%. By statistically analyzing the fluctuation amplitude distribution of all users within a group, the overall electricity consumption characteristics of that group can be identified.

[0078] In one possible implementation, the application of the K-means clustering algorithm requires pre-determining the number of clusters and initial centroids. Based on practical experience in power load management, the electricity consumption patterns of industrial users are mainly divided into three categories, therefore, K is set to 3. The algorithm uses the peak-valley overlap rate and load fluctuation amplitude as a two-dimensional feature space, where each user group is represented as a point. The initial centroids are selected using the K-means++ method; the first centroid is randomly selected, and the probability of selecting subsequent centroids is proportional to the distance to the previously selected centroids, ensuring a uniform distribution of initial centroids. Furthermore, the clustering iteration process includes two stages: assignment and update. In the assignment stage, the Euclidean distance from each user group to the three centroids is calculated, and the user group is assigned to the category corresponding to the nearest centroid. In the update stage, the feature mean of all user groups within each category is recalculated as the new centroid. Iteration continues until the change in centroid position is less than a preset threshold or the maximum number of iterations is reached. After clustering, the three categories exhibit significant feature differences.

[0079] Understandably, the determination of electricity consumption categories is based on feature analysis of clustering results. Peak-consumption types are characterized by a high overlap between peak and off-peak hours and large load fluctuations. These users are typically single-shift production enterprises that concentrate their electricity consumption during working hours and essentially cease production during off-peak hours. Stable-consumption types are characterized by small load fluctuations, regardless of the overlap between peak and off-peak hours. These users are mostly continuous production enterprises where equipment operates 24 hours a day, maintaining relatively stable power output. Intermittent-consumption types fall between the two, with moderate fluctuations, and may be enterprises with multi-shift production or batch production characteristics.

[0080] For example, after classification, an industrial park identified three typical user groups: the first group, comprising eight machinery manufacturing companies, is a peak-consumption type, with production peaking from 8:00 AM to 6:00 PM and essentially ceasing operations at night, resulting in a peak-valley overlap rate of 0.88 and load fluctuation of 92%; the second group, comprising five chemical companies, is a stable-consumption type, operating continuously in three shifts, with load fluctuation of only 18%; and the third group, comprising seven food processing companies, is an intermittent-consumption type, adjusting production according to orders, with load fluctuation of 55%. This segmentation of user categories provides a scientific basis for formulating differentiated electricity pricing policies and load dispatching strategies, helping to improve grid operation efficiency and the economic efficiency of user electricity consumption.

[0081] Step S106: Identify industrial user electricity load data by subdividing user category labels, evaluate load forecast accuracy and electricity consumption pattern stability based on industrial user electricity load data, and obtain electricity load characteristics.

[0082] Industrial users are grouped and labeled by subdividing user category tags. Historical electricity load data for corresponding users is extracted based on three tags: peak consumption, stable consumption, and intermittent consumption. Hourly power values ​​and equipment operation records for each user within the most recent preset period are obtained, constructing an electricity load dataset containing time series, power values, and category tags. Based on this electricity load dataset, an autoregressive moving average model is used to predict the load for each user. The load prediction accuracy index is obtained by calculating the percentage of the absolute value of the difference between the predicted and actual values ​​relative to the actual values ​​and averaging the results. Simultaneously, the coefficient of variation is obtained by calculating the ratio of the standard deviation to the mean of the daily load curve, and the autocorrelation coefficient is obtained by calculating the correlation coefficient between adjacent daily load curves. The stability of the electricity consumption pattern is assessed based on the coefficient of variation and the autocorrelation coefficient. Based on the load prediction accuracy index and the stability of the electricity consumption pattern, typical daily load curves are extracted from the electricity load dataset. The peak-to-valley difference rate is obtained by calculating the percentage of the difference between the daily maximum and minimum load relative to the maximum load, and the load factor is obtained by calculating the ratio of the average load to the maximum load. Combined with user category tags, an electricity load feature including prediction accuracy, stability, and load characteristic parameters is constructed.

[0083] Specifically, in one implementation, the application of detailed user category tags is achieved by establishing a category-load mapping relationship. Based on the obtained tags of peak power consumption, stable power consumption, and intermittent power consumption, all industrial users within the park are labeled. Each category corresponds to different load characteristics: peak power consumption users typically consume electricity concentratedly during daytime working hours, stable power consumption users maintain a continuous and stable load level, and intermittent power consumption users exhibit periodic load fluctuations.

[0084] Specifically, constructing an electricity load dataset requires integrating multi-source data. Historical data from the past 90 days for each user is extracted from the power monitoring platform, including hourly active power, reactive power, current, voltage, and other electrical parameters. Simultaneously, equipment operation logs are obtained from the equipment management system, recording the start-up and shutdown times, operating status, and fault information for each device. Through timestamp alignment, the electrical data is correlated with equipment status, forming a four-dimensional data structure of time series, power values, equipment status, and category labels, providing complete input information for load forecasting. The application of the autoregressive moving average model in load forecasting is based on the autocorrelation of time series data. The model establishes a linear relationship between load and its lag values ​​by analyzing the autoregressive and moving average characteristics of historical load data. For each user, a data stationarity test is first performed; if the data is non-stationary, differencing is performed. The model order is determined using the autocorrelation function and partial autocorrelation function, and the model parameters are estimated using the least squares method. During forecasting, the model uses a linear combination of previous load values ​​and random error terms to predict future load.

[0085] Preferably, the load forecasting accuracy is assessed using a combination of multiple indicators. The mean absolute percentage error is obtained by calculating the absolute value of the relative error for each forecast point and averaging it; this indicator eliminates the influence of load level differences.

[0086] For example, a company's daily load forecast was 850 kilowatts, while the actual load was 800 kilowatts, resulting in a single-point error rate of 6.25%. By statistically analyzing the error rates of all forecast points over a month and averaging them, the user's forecast accuracy was found to be 8.3%, indicating good forecasting performance.

[0087] For example, the stability of electricity consumption patterns is comprehensively assessed using the coefficient of variation (CVA) and autocorrelation coefficient (AC). The CVA reflects the relative magnitude of load fluctuations. For instance, a textile company has a daily average load of 500 kW and a standard deviation of 75 kW, with a CVA of 0.15, indicating relatively stable load. The AC measures the similarity between loads in adjacent periods. By calculating the Pearson correlation coefficient between today's load curve and yesterday's load curve, a high correlation of 0.92 is obtained, indicating that the user's electricity consumption pattern has strong regularity.

[0088] In one possible implementation, the construction of electricity load characteristics integrates two dimensions: predictive performance and load characteristics. The peak-valley difference rate is calculated by dividing the difference between the daily maximum load of 1200 kW and the minimum load of 300 kW (900 kW) by the maximum load, resulting in a peak-valley difference rate of 0.75. The load factor is the ratio of the daily average load of 650 kW to the maximum load of 1200 kW, which is 0.54. These characteristic parameters, along with the prediction accuracy of 8.3%, the stability score of 0.85, and the user category label, constitute a complete electricity load feature vector.

[0089] Step S107: Update the equipment start-up and shutdown time record database according to the power load characteristics, and obtain the final power load data identification, classification and evaluation results by analyzing the power load characteristics.

[0090] Based on the prediction accuracy, stability, and load characteristic parameters in the electricity load characteristics, the feature data is written into the historical records of the corresponding user in the equipment start-up and shutdown time record database. This replaces the original feature field values ​​and records the update timestamp, completing the data update of the record database. By querying the updated equipment start-up and shutdown time record database, the category label field, the latest load prediction value, and the calculated electricity behavior similarity for each user are extracted. These three types of data are then linked and integrated according to user identifiers, outputting an electricity load data identification, classification, and evaluation result that includes user category labels, load prediction values, and electricity behavior similarity.

[0091] Specifically, in one implementation, the update of the equipment start-up and shutdown time record library is achieved through database write operations. Parameters such as prediction accuracy, stability, peak-valley difference rate, and load factor from the electricity load characteristics are written into the record library according to a predefined data structure. Each record contains a unique user identifier, feature data fields, and a timestamp field. The timestamp uses a year-month-day-hour-minute-second format to record the update time, facilitating the tracking of data changes. The data integration process is completed through database queries and association operations. User category labels are extracted from the updated record library. These labels come from the previous K-means clustering classification results, including peak-consumption, stable-consumption, or intermittent-consumption types. Simultaneously, the latest load prediction value is extracted, based on the prediction results of the autoregressive moving average model. The electricity behavior similarity score is derived from the calculation results of the user similarity matrix. The electricity load data identification, classification, and evaluation results are output in a structured format. Each user corresponds to one evaluation record, containing a user ID, a category label string, a predicted power value, and a decimal similarity score.

[0092] For example, the assessment results of a machinery processing enterprise show that it is classified as a peak-consumption type, with a predicted load of 850 kilowatts the following day, and a similarity score of 0.83 with similar users. This comprehensive assessment provides data support for power grid dispatching departments to formulate differentiated management strategies and optimize power resource allocation.

[0093] The above are only some preferred embodiments of this application, but this application is not limited thereto, and many improvements and modifications can be made. Any improvements and modifications made based on the basic principles of this application should be considered to fall within the protection scope of this application.

Claims

1. A method for identifying, classifying, and evaluating the load characteristics of electricity users, characterized in that, include: Collect relevant data of industrial user equipment, including at least: the start-up time, shutdown time, operating power value and start-up time difference between equipment corresponding to the equipment type and power level of the industrial user equipment; Based on the relevant data, assess the time difference between the start-up and shutdown sequence at the power-on time and the historical start-up and shutdown sequence data recorded in the equipment start-up and shutdown time record database; When the time difference exceeds a preset threshold, it is determined to adjust the start-up and shutdown sequence, and the device start-up interval and power superposition time are grouped according to the device type and the power level to generate a device start-up and shutdown time combination; Based on the equipment start-up and stop time combinations, assess the start-up time difference and power superposition time difference of different users in the same industrial park to determine the similarity of user electricity consumption behavior; Based on the similarity of user electricity consumption behavior, the electricity consumption time distribution of user groups with similarity of electricity consumption behavior greater than or equal to a preset similarity is statistically analyzed, the peak and valley time overlap rate and load fluctuation amplitude are calculated, and subdivided user categories are generated. Based on the subdivided user categories and power load data, power load characteristics are generated; the relevant data of the industrial user equipment includes: equipment start-up and stop sequence records generated according to the equipment type and the power level, and start-up and stop datasets containing equipment numbers and start-up time differences between adjacent equipment. Based on the relevant data, evaluate the time difference between the start-up and shutdown sequence at the power-on time and the historical start-up and shutdown sequence data recorded in the equipment start-up and shutdown time record database, including: Identify the time deviation between the power-on time and the historical start-stop sequence data recorded in the equipment start-stop time record library, as well as the power difference between the operating power value and the historical power records recorded in the equipment start-stop time record library; Based on the time deviation value and the power difference value, evaluate the time difference between the start-up and shutdown sequence at the power-on time and the historical start-up and shutdown sequence data; identify the time deviation value between the power-on time and the historical start-up and shutdown sequence data recorded in the equipment start-up and shutdown time record library, and the power difference value between the operating power value and the historical power records recorded in the equipment start-up and shutdown time record library, including: The method further includes: extracting historical start-stop sequence data for the corresponding user from the device start-stop time record database; comparing the start-up time of the start-stop dataset with the start-up time of the historical start-stop sequence data to calculate the time deviation value; and comparing the operating power value with the historical power record to calculate the power difference value. Calculate the time difference value based on the time deviation value and the power difference value; When the time difference value exceeds a preset threshold, extract the timing data of the power-on time and calculate the average interval difference between the start-up time interval of adjacent devices and the historical start-up and shutdown sequence data. Generate a change range marker based on the difference ratio; Based on the change magnitude marker, a start-stop sequence change identifier containing adjustment type and device number is generated; the method further includes: identifying device start-up intervals and power superposition times based on the start-stop sequence change identifier; grouping the start-up intervals and power superposition times according to the device type and the power level to generate device start-stop time combinations, including: The start-stop sequence change identifier is parsed to extract the adjustment type and device number. The device start-stop time record database is queried to obtain the power-on time and the power-off time, and the device start-up interval is calculated. The overlapping period of the power-on time and the power-off time is identified, and the power superposition time is recorded. The operating power value of the overlapping period is accumulated to generate a power superposition record. The device start-up interval and the power superposition record are grouped according to the device type and the power level to generate a device start-stop time combination containing the device number, the device start-up interval, and the power superposition time.

2. The method for identifying, classifying, and evaluating the load characteristics of electricity users according to claim 1, characterized in that, The step of evaluating the start-up time difference and power superposition time difference of different users in the same industrial park based on the combination of equipment start-up and shutdown times, and generating user electricity consumption behavior similarity includes: Based on the equipment start-up and shutdown time combinations, extract the equipment configurations of users within the industrial park, and filter users with the same equipment type and power level; Obtain the power-on time sequence and power change curve, calculate the startup time difference between users, and generate a startup difference record; By comparing the power superposition moments, the time offset and peak difference are calculated to generate a power curve difference metric. Based on the start-up time difference and the power curve difference metric, a power consumption behavior feature vector is generated, the similarity between the vectors is calculated, and the user power consumption behavior similarity is generated.

3. The method for identifying, classifying, and evaluating the load characteristics of electricity users according to claim 1, characterized in that, The process of identifying highly similar user groups based on the similarity of user electricity consumption behavior, statistically analyzing the electricity consumption time distribution of the highly similar user groups, calculating the peak-valley time overlap rate and load fluctuation amplitude, and generating subdivided user categories includes: Based on the similarity of user electricity consumption behavior, high-similarity user groups are screened, the power curves of the high-similarity user groups are extracted, the operating power values ​​for each hourly period are statistically analyzed, and a time-period power matrix is ​​generated. The peak and valley periods of the time-period power matrix are identified, and the overlap rate of the peak and valley periods and the load fluctuation amplitude are calculated. Based on the overlap rate of the peak and valley periods and the load fluctuation amplitude, subdivided user categories including peak-consumption type, stable-consumption type, and intermittent-consumption type are generated.

4. The method for identifying, classifying, and evaluating the load characteristics of electricity users according to any one of claims 1 to 3, characterized in that, The step of generating electricity load characteristics based on the subdivided user categories and electricity load data includes: Based on the subdivided user categories, historical electricity load data of users is extracted to generate a load dataset containing time series and operating power values; based on the load dataset, prediction accuracy and pattern stability are calculated to generate electricity load features containing the prediction accuracy and pattern stability; based on the electricity load features, the feature fields of the equipment start-up and shutdown time record library are updated.

5. The method for identifying, classifying, and evaluating the load characteristics of electricity users according to any one of claims 1 to 3, characterized in that, After generating the electrical load characteristics, the method further includes: Update the user records in the equipment start-up and shutdown time record library according to the electrical load characteristics; Extract the segmented user categories, prediction accuracy, and user electricity behavior similarity, and integrate them to generate an evaluation result that includes the segmented user categories and prediction accuracy.

Images