Load prediction method and system based on user order demand

Abstract

The invention discloses a load prediction method and system based on user order demands, and relates to the technical field of power load prediction. Calculating a standard production load of a single product based on the historical production log and the historical power load data, and determining a characteristic operation power value corresponding to each product type according to the standard production load; generating a baseline load prediction curve based on planned starting and ending time and a characteristic operation power value in the production plan data; combining the baseline load prediction curve and the historical power load data to obtain an input feature, and inputting the input feature into the time sequence prediction model to obtain an initial load prediction curve; determining an efficiency correction factor and an environment correction factor based on the historical production log and the historical power load data; and correcting the initial load prediction curve according to the efficiency correction factor and the environment correction factor to obtain a final load prediction curve. By implementing the technical scheme provided by the invention, the accuracy of power load prediction of the industrial enterprise is improved.

Description

Technical Field

[0001] This application relates to the field of power load forecasting technology, specifically to a load forecasting method and system based on user order demand. Background Technology

[0002] Driven by both "dual carbon" targets and energy cost optimization, refined energy management is becoming increasingly important for industrial enterprises. Electricity load forecasting, as a core technology, directly impacts an enterprise's production cost control, electricity market trading strategies, and the stable operation of the power grid. Therefore, researching and applying high-precision industrial load forecasting methods has significant economic and social value.

[0003] In existing technologies, mainstream load forecasting methods typically employ time series models, such as Long Short-Term Memory (LSTM) networks. These methods learn the inherent periodicity and trends in historical electricity load data, and may combine external variables such as date type and weather to construct predictive models that infer future electricity load curves.

[0004] However, existing technologies essentially treat the object of prediction as a "black box," relying primarily on the statistical characteristics of historical load sequences for extrapolation. They cannot incorporate specific production planning information such as future production orders, product types, and planned start-up and shutdown times into the prediction model. When a company's production plan deviates significantly from historical patterns, models that rely solely on historical data will fail to capture these discontinuous, structured changes, leading to decreased prediction accuracy and making it difficult to meet the company's actual needs for refined production scheduling and cost control. Summary of the Invention

[0005] This application provides a load forecasting method and system based on user order demand. This method comprehensively considers information on production plans, equipment status, and environmental factors to improve the accuracy of power load forecasting for industrial enterprises.

[0006] Firstly, this application provides a load forecasting method based on user order demand. The method includes: acquiring production plan data, historical production logs, and historical power load data of a target enterprise; calculating the standard production load of a single production line producing a single product based on the historical production logs and historical power load data, and calculating the baseline non-production load of all production lines during non-production periods; calculating the difference between the standard production load and the baseline non-production load to obtain the net increase in average load corresponding to a single product, and averaging the net increase in average load of multiple historical production tasks belonging to the same product type to obtain the characteristic operating power value corresponding to each product type; and obtaining the planned start and end times of the target production tasks from the production plan data. Based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production tasks, a baseline load forecast curve for the target production tasks is generated. The baseline load forecast curve and historical power load data are merged to obtain input features, which are then input into the time series forecast model to obtain the initial load forecast curve for the forecast period. The baseline operating efficiency and cumulative operating time of the production equipment are extracted from the historical production logs, and the baseline operating efficiency is calculated based on the cumulative operating time to obtain the efficiency correction factor. Based on the historical operating temperature and historical power load data in the historical production logs, the environmental correction factor is determined. The initial load forecast curve is corrected according to the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

[0007] By adopting the above technical solution, the standard production load of a single product and the benchmark non-production load of the production line during non-production periods are calculated based on historical production logs and historical power load data. The difference between the standard production load and the benchmark non-production load is calculated to obtain the net increase average load corresponding to a single product. The net increase average load of multiple historical production tasks belonging to the same product type is calculated to obtain the characteristic operating power value corresponding to each product type. A baseline load prediction curve is generated based on the planned start and end times of the target production task and the characteristic operating power value. The baseline load prediction curve and historical power load data are merged and input into the time series prediction model to obtain the initial load prediction curve within the prediction period. This organic integration ensures that the prediction results meet the rigid constraints of the production plan and can capture the complex time series patterns contained in historical data. The calculation of efficiency correction factors fully considers the inevitable performance degradation of industrial equipment during long-term operation. The calculation of environmental correction factors makes up for the shortcomings of traditional load prediction methods that ignore environmental impact. The initial load prediction curve is corrected based on efficiency correction factors and environmental correction factors to obtain the final load prediction curve. This method can significantly improve the accuracy of power load prediction for industrial enterprises by comprehensively considering multi-dimensional information while maintaining computational efficiency.

[0008] Optionally, based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production task, a baseline load forecast curve for the target production task is generated. Specifically, this includes: determining the target product type from the target production task and retrieving the start-up transition time and shutdown transition time for the target product type; dividing the complete cycle of the target production task based on the planned start and end times, start-up transition time, and shutdown transition time to obtain the start-up transition period, running period, and shutdown transition period; obtaining the planned start time and planned end time from the planned start and end times, obtaining multiple start-up time points based on the planned start time and start-up transition time, and using the ratio of the characteristic operating power value to the start-up transition time as the start-up slope; multiplying the difference between each start-up time point and the planned start time by the start-up slope to obtain the start-up load value corresponding to each start-up time point; sorting the multiple start-up load values ​​in chronological order to obtain the load sequence of the start-up transition period; and adding the planned start time to the start-up transition time to obtain the start time of the running period. In this process, the difference between the planned termination time and the shutdown transition duration is used as the end time of the operating segment. Multiple steady-state time points are determined based on the start and end times, and the characteristic operating power value is used as the steady-state load value corresponding to each steady-state time point. All steady-state load values ​​are sorted in chronological order to obtain the load sequence of the operating segment. The difference between the planned termination time and the shutdown transition duration is used as the start time of the shutdown transition period. Multiple shutdown time points are generated based on the start time of the shutdown transition duration. The negative value of the characteristic operating power value is divided by the shutdown transition duration to obtain the shutdown slope. The difference between each shutdown time point and the start time of the shutdown transition period is calculated to obtain the time span. The time span is multiplied by the shutdown slope and added to the characteristic operating power value to obtain the shutdown load value corresponding to each shutdown time point. All shutdown load values ​​are sorted in chronological order to obtain the load sequence of the shutdown transition period. The load sequences of the startup transition period, the operating segment, and the shutdown transition period are spliced ​​together in chronological order to obtain the baseline load prediction curve.

[0009] By adopting the above technical solution, the target product type is determined from the target production task, and the start-up transition time and shutdown transition time are retrieved. Based on the planned start and end times, start-up transition time, and shutdown transition time, the cycle of the target production task is divided into start-up transition period, running period, and shutdown transition period. Load values ​​are calculated for different periods to obtain the load sequence of start-up transition period, running period, and shutdown transition period. The load sequences of start-up transition period, running period, and shutdown transition period are then spliced ​​together in chronological order to obtain the baseline load prediction curve. This ensures that the baseline load prediction curve can reflect the power gradual change characteristics during start-up and shutdown, as well as the constant power characteristics during steady-state operation, throughout the entire production task cycle, and can accurately predict load fluctuations during task handover periods.

[0010] Optionally, the baseline load forecast curve and historical power load data are merged to obtain input features. Specifically, this includes: determining a target time axis based on the time range of the historical power load data and the time range of the baseline load forecast curve; mapping the historical power load data and the baseline load forecast curve onto the target time axis; selecting a target time point from the target time axis; extracting multiple historical load values ​​within a preset historical window length from the historical power load data before the target time point; sorting the multiple historical load values ​​in chronological order to obtain a historical load feature sequence; the target time point being any time point on the target time axis; numerically encoding the time attribute information of the target time point to obtain time attribute features; extracting multiple baseline load forecast values ​​within a preset guiding window from the baseline load forecast curve based on the target time point; sorting the multiple baseline load forecast values ​​in chronological order to obtain a baseline guiding feature sequence; concatenating and combining the historical load feature sequence, time attribute features, and baseline guiding feature sequence corresponding to the target time point to obtain a high-dimensional feature vector; and combining the high-dimensional feature vectors corresponding to all target time points within the forecast period in chronological order to obtain the input features.

[0011] By adopting the above technical solution, the target time axis is determined based on the time range of historical power load data and the time range of baseline load prediction curves. The historical power load data and baseline load prediction curves are mapped onto the target time axis, realizing the unified alignment of heterogeneous time series data. The historical load values ​​of each target time point on the target time axis are calculated to obtain the historical load feature sequence. The time attribute information is encoded to obtain the time attribute features. Then, the historical load feature sequence, time attribute features, and baseline guiding feature sequence corresponding to the target time point are concatenated and combined to obtain a high-dimensional feature vector. This fusion strategy fully integrates information from three dimensions: historical operating status, time cycle pattern, and future production plan. This makes the prediction of each time point no longer rely on a single information source in isolation. The high-dimensional feature vectors corresponding to all target time points within the prediction period are combined in chronological order to obtain the input features.

[0012] Optionally, an efficiency correction factor is calculated based on the cumulative runtime to obtain the baseline operating efficiency. Specifically, this includes: using the characteristic operating power value as the baseline net power consumption of the product type under ideal equipment conditions; retrieving the equipment's most recent maintenance time from historical production logs; determining the cumulative runtime based on the current time and the most recent maintenance time; marking the current time as a historical efficiency anchor point when the cumulative runtime exceeds the preset calibration periodicity; subtracting the baseline non-production load from the actual average load of historical production tasks to obtain the actual net power consumption of the anchor point corresponding to the historical efficiency anchor point; calculating the difference between the actual net power consumption of the anchor points corresponding to two temporally adjacent historical efficiency anchor points to obtain the power consumption change; and dividing the power consumption change by the runtime between the two historical efficiency anchor points to obtain the target operating area. The segmented power consumption increment gradient within the time interval is used, with the target operating interval being the interval corresponding to two temporally adjacent historical efficiency anchor points. Multiple segmented power consumption increment gradients are sorted to obtain a historical power consumption increment gradient sequence. The most recent segmented power consumption increment gradient from this sequence is selected as the prediction baseline gradient. The average cumulative runtime of the target production task is calculated, and the difference between this average cumulative runtime and the target cumulative runtime at the most recent historical efficiency anchor point is calculated to obtain the prediction interval runtime. The prediction interval runtime is multiplied by the prediction baseline gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net increase in power consumption at the anchor point corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase in average load. The predicted net increase in average load is then divided by the baseline net increase in power consumption to obtain the efficiency correction factor.

[0013] By adopting the above technical solution, the characteristic operating power value is used as the baseline net increase in power consumption. The cumulative running time is determined based on the current time point and the time point of the most recent maintenance. When the cumulative running time exceeds the preset calibration period, the current time point is marked as the historical efficiency anchor point. The actual net increase in power consumption at the anchor point is obtained by subtracting the baseline non-production load from the actual average load of historical production tasks. This effectively isolates the basic energy consumption unrelated to production tasks. The difference between the actual net increase in power consumption at two adjacent historical efficiency anchor points is calculated to obtain the power consumption change. The power consumption change is divided by the running time between the two historical efficiency anchor points to obtain the segmented power consumption increment gradient within the target operating range. This segmented processing adapts to the equipment performance. The attenuation may exhibit nonlinear characteristics; multiple segmented power consumption increment gradients are sorted to obtain a historical power consumption increment gradient sequence. The most recent segmented power consumption increment gradient is selected from the historical power consumption increment gradient sequence as the prediction benchmark gradient. The average cumulative runtime of the target production task is calculated to determine the runtime of the prediction interval. The runtime of the prediction interval is multiplied by the prediction benchmark gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net increase power consumption of the anchor point corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase average load. The predicted net increase average load is divided by the benchmark net increase power consumption to obtain the efficiency correction factor, thereby compensating for and correcting the equipment aging effect and improving the continuous accuracy of load prediction during the long-term operation of the equipment.

[0014] Optionally, based on historical operating temperature and historical power load data in historical production logs, an environmental correction factor is determined. Specifically, this includes: extracting ambient temperature data from all historical production tasks in the historical production logs; determining a temperature range covering all historical operating conditions based on multiple ambient temperature data points; dividing the temperature range to obtain multiple continuous, non-overlapping operating temperature intervals, and selecting the operating temperature interval containing the most historical production tasks as the benchmark temperature interval; using the difference between the standard production load and the benchmark non-production load corresponding to each historical production task as the actual net increase in power consumption, and dividing the actual net increase in power consumption by the characteristic operating power value of the historical production task to obtain the comprehensive power consumption skewness; and obtaining the cumulative operating time of the equipment for each historical production task. The process involves several steps: First, determining the historical equipment efficiency correction factor based on the cumulative operating time of the equipment. Second, using the ratio of the comprehensive power consumption skewness to the historical equipment efficiency correction factor as the environmental impact coefficient. Third, binding the environmental impact coefficient to each historical production task and summarizing the average operating temperature of all historical production tasks into each operating temperature range. Finally, calculating the interval average environmental impact coefficient for each operating temperature range. Fourth, performing load forecasting for the target production task, obtaining the predicted ambient temperature within the forecast period, comparing the predicted ambient temperature with each operating temperature range, determining the interval average environmental impact coefficient corresponding to the target historical production task based on the comparison results, and using the interval environmental impact coefficient as the environmental correction factor.

[0015] By adopting the above technical solution, ambient temperature data during historical production tasks is extracted from historical production logs, and a temperature range covering all historical operating conditions is determined based on multiple ambient temperature data. The temperature range is divided, and the operating temperature interval containing the most historical production tasks is selected as the benchmark temperature interval. The difference between the standard production load and the benchmark non-production load corresponding to each historical production task is taken as the actual net increase in power consumption. The actual net increase in power consumption is divided by the characteristic operating power value of the historical production task to obtain the comprehensive power consumption skewness. Based on the cumulative operating time of the equipment, the historical equipment efficiency correction factor is determined, and the ratio of the comprehensive power consumption skewness to the historical equipment efficiency correction factor is taken as the environmental impact factor. The environmental impact coefficient is linked to each historical production task and summarized into various operating temperature ranges based on the average operating temperature of all historical production tasks. The average environmental impact coefficient within each operating temperature range is calculated by averaging all environmental impact coefficients contained therein. The forecast environmental temperature within the prediction period is obtained, and the forecast environmental temperature is compared with each operating temperature range. Based on the comparison results, the average environmental impact coefficient corresponding to the target historical production task is determined, and the interval environmental impact coefficient is used as an environmental correction factor. This realizes the effective transfer of environmental factors from historical analysis to future prediction and can adaptively adjust the prediction results based on weather forecast information.

[0016] Optionally, the initial load forecast curve is corrected based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve. Specifically, this includes: obtaining the current forecast time point within the forecast period and obtaining the initial forecast load value corresponding to the current forecast time point from the initial load forecast curve; retrieving the baseline load value corresponding to the current forecast time point from the baseline load forecast curve; multiplying the efficiency correction factor and the environmental correction factor to obtain the comprehensive correction coefficient; multiplying the baseline load value and the comprehensive correction coefficient to obtain the corrected net increase in production load; subtracting the baseline load value from the initial forecast load value to obtain the non-production-related baseline load component, and adding the corrected net increase in production load and the non-production-related baseline load component to obtain the final forecast load value corresponding to the current forecast time point; and combining the final forecast load values ​​corresponding to all forecast time points in chronological order to obtain the final load forecast curve.

[0017] By adopting the above technical solution, the current forecast time point is obtained within the forecast period, and the initial forecast load value corresponding to the current forecast time point is obtained from the initial load forecast curve. The baseline load value corresponding to the current forecast time point is retrieved from the baseline load forecast curve, thus achieving precise positioning of production-related load components. The efficiency correction factor and the environmental correction factor are multiplied to obtain the comprehensive correction coefficient. The influence of the two types of correction factors is combined to avoid correction deviations that may be caused by simple superposition. The baseline load value is multiplied by the comprehensive correction coefficient to obtain the corrected net increase in production load. The baseline load value is subtracted from the initial forecast load value to obtain the non-production-related baseline load component. The corrected net increase in production load is added to the non-production-related baseline load component to obtain the final forecast load value. The final forecast load values ​​corresponding to all forecast time points are combined in chronological order to obtain the final load forecast curve. This completes the aggregation process from point-by-point correction to a complete curve, ensuring that the final load forecast curve has undergone refined correction considering the evolution of equipment status and the influence of environmental factors at each time point, thus improving the accuracy of the forecast curve for actual operating conditions.

[0018] Optionally, after correcting the initial load forecast curve based on efficiency and environmental correction factors to obtain the final load forecast curve, the method further includes: when a sudden failure occurs during the actual execution of a target production task, obtaining the type and scope of the sudden failure, and retrieving the planned delivery time of the target production task from the production plan data; calculating the interval between the planned delivery time and the current time, analyzing the type and scope of the sudden failure based on the interval, and obtaining the failure handling level; matching the failure handling level and the type of the sudden failure from a preset emergency plan library to obtain the target emergency handling measures; calculating the expected power load change resulting from the implementation of the target emergency handling measures, generating an emergency load curve starting from the time of the failure; and superimposing or replacing the emergency load curve with the final load forecast curve based on the time of the failure to obtain the real-time corrected emergency load forecast curve.

[0019] By adopting the above technical solution, when a sudden failure occurs during the actual execution of the target production task, the type and scope of the sudden failure are obtained, and the planned delivery time of the target production task is retrieved from the production plan data. The interval between the planned delivery time and the current time is calculated, and the type and scope of the sudden failure are analyzed based on the interval to obtain the failure handling level. Based on the failure handling level and the type of sudden failure, the target emergency handling measures are obtained by matching from the preset emergency plan library. The expected power load change generated by the implementation of the target emergency handling measures is calculated, and an emergency load curve is generated from the time of failure. This realizes the quantitative extrapolation from the emergency response plan to the impact on power demand. Different emergency measures will lead to drastically different load change patterns. Based on the time of failure, the emergency load curve is superimposed or replaced with the final load forecast curve to obtain the real-time corrected emergency load forecast curve. According to the nature of the emergency measures, an appropriate curve fusion method is selected to ensure that the emergency load forecast curve can accurately reflect the actual power demand trajectory of the enterprise after the failure occurs, avoiding the problem of prediction failure of traditional static forecasting methods when facing unplanned disturbances.

[0020] The second aspect of this application provides a load forecasting system based on user order demand. The system includes an acquisition unit, a calculation unit, a processing unit, and a correction unit. The acquisition unit acquires production plan data, historical production logs, and historical power load data of the target enterprise; calculates the standard production load for a single production line producing a single product based on the historical production logs and historical power load data, and calculates the baseline non-production load for all production lines during non-production periods. The calculation unit calculates the difference between the standard production load and the baseline non-production load to obtain the net increase average load corresponding to a single product, and averages the net increase average load of multiple historical production tasks belonging to the same product type to obtain the characteristic operating power value corresponding to each product type. The processing unit obtains data from the production plan data... The system extracts the planned start and end times of the target production task and generates a baseline load forecast curve based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production task. It then merges the baseline load forecast curve with historical power load data to obtain input features, which are input into a time series forecast model to obtain the initial load forecast curve for the forecast period. The system extracts the baseline operating efficiency and cumulative operating time of the production equipment from historical production logs and calculates the efficiency correction factor based on the cumulative operating time. It determines the environmental correction factor based on historical operating temperature and historical power load data from the historical production logs. Finally, a correction unit corrects the initial load forecast curve based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

[0021] In a third aspect, this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory, causing the electronic device to perform any of the methods described above in this application.

[0022] In a fourth aspect, this application provides a computer-readable storage medium storing instructions that, when executed, perform any of the methods described above in this application.

[0023] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Based on historical production logs and historical power load data, the standard production load of a single product and the baseline non-production load during off-peak hours of the production line are calculated. The difference between the standard production load and the baseline non-production load is calculated to obtain the net increase in average load for a single product. The net increase in average load for multiple historical production tasks belonging to the same product type is calculated to obtain the characteristic operating power value corresponding to each product type. A baseline load forecast curve is generated based on the planned start and end times of the target production task and the characteristic operating power value. The baseline load forecast curve and historical power load data are merged and input into the time series forecast model to obtain the initial load forecast curve within the forecast period. This organic integration ensures that the forecast results meet the rigid constraints of the production plan and can capture the complex time series patterns contained in historical data. By calculating the efficiency correction factor, the inevitable performance degradation of industrial equipment during long-term operation is fully considered. By calculating the environmental correction factor, the shortcomings of traditional load forecasting methods that ignore environmental impact are compensated for. The initial load forecast curve is corrected based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve. While maintaining computational efficiency, this method comprehensively considers multi-dimensional information and significantly improves the accuracy of power load forecasting for industrial enterprises. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a load forecasting method based on user order demand provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a load forecasting system based on user order demand provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0025] Explanation of reference numerals in the attached drawings: 201, acquisition unit; 202, calculation unit; 203, processing unit; 204, correction unit; 300, electronic device; 301, processor; 302, memory; 303, user interface; 304, network interface; 305, communication bus. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0027] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0028] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, 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 indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0029] Therefore, how to address the issue of decreased prediction accuracy caused by relying solely on historical data in existing technologies is a pressing problem. This application provides a load forecasting method based on user order demand, applied in a server. The server in this application is a platform providing power load forecasting services to enterprises. Figure 1 This is a flowchart illustrating a load forecasting method based on user order demand provided in an embodiment of this application. (Refer to...) Figure 1 The method includes the following steps S101-S108.

[0030] S101: Obtain the target company's production plan data, historical production logs, and historical power load data.

[0031] In S101 above, a standardized interface protocol is used to connect with the target enterprise's ERP (Enterprise Resource Planning) system and MES (Manufacturing Execution System). Production planning data is retrieved in real-time from the ERP system. This data includes details of product orders received by the enterprise in the future, specifically covering key fields such as order quantity, product type identifier, planned delivery time, and order priority for each order. It also includes the corresponding Bill of Materials (BOM) information, which details the types and quantities of raw materials required to produce the product, as well as the processing steps for each production process. Shop floor scheduling plans are extracted from the MES system. These plans are the concrete implementation of ERP orders at the shop floor execution level, containing detailed execution information such as the process arrangements for each order on different production lines, the specific equipment numbers used, the planned start and end times for each process, and the expected production period. This data clearly decomposes order requirements in both time and space dimensions, enabling the system to accurately know which equipment will be running and what production tasks will be performed at each future point in time.

[0032] While acquiring production plan data, historical power load data is collected from the target company's power monitoring system or energy management platform. This historical power load data records the company's actual electricity consumption over a period of time in time series format. The sampling frequency is usually 15 minutes or higher. The data includes total load power, sub-load power, and electrical parameters such as power factor at each time point. Although historical power load data cannot be directly used to predict future load changes driven by orders, it contains the basic patterns of the company's electricity consumption and the periodic characteristics of non-productive loads, providing necessary historical references for model learning. It is also necessary to extract the execution records of past production tasks from historical production logs. Historical production logs are automatically generated by the MES system during the production process and record in detail the actual execution of each historical production task, including task start time, end time, equipment list used, actual processing time of each process, equipment operating parameters during the production process such as speed setpoints and actual values, temperature control parameters, etc. It also includes equipment start and stop event records, fault downtime records, equipment cumulative running time and other equipment status information, as well as environmental data of the production site such as measured values ​​of workshop temperature and humidity. By establishing a time correspondence between production task records in historical production logs and historical power load data, it is possible to retrospectively analyze the actual energy consumption performance of equipment when performing production tasks for a certain type of product in the past.

[0033] During the data collection process, data quality verification is performed. For the collected production plan data, the integrity of the order fields is checked, and abnormal records that lack key information such as product type or delivery time are removed. For historical power load data, continuity is checked, missing data periods are identified and marked, and short-term missing data is filled in using methods such as linear interpolation or forward filling. For equipment operating parameters in historical production logs, the rationality is verified, and abnormal values ​​that obviously exceed the rated range of the equipment are excluded.

[0034] S102: Calculate the standard production load for a single production line producing a single product based on historical production logs and historical power load data, and calculate the baseline non-production load for all production lines during non-production periods.

[0035] In S102 above, the standard production load for a single production line producing a single product is calculated based on historical production logs and historical power load data. The baseline non-production load for all production lines during non-production periods is also calculated. Specifically, this includes: extracting the actual start and end times of each historical production task from the historical production logs and extracting the load data sequence corresponding to the actual start and end times from the historical power load data; obtaining the target load data sequence corresponding to the target historical production task; removing the load data corresponding to the start-up and shutdown impact times from the target load data sequence to obtain a load subsequence, where the target historical production task is any one of multiple historical production tasks; performing data point density analysis on the load subsequence to obtain multiple power intervals and counting the number of data points within each power interval; selecting the highest power interval with the most data points from the multiple power intervals and calculating the arithmetic average of all data points within the highest power interval to obtain the standard production load; determining multiple non-production time periods where all production lines have not executed production tasks based on the historical production logs; selecting target non-production time periods with a value greater than a preset idle threshold from the multiple non-production time periods; and averaging the power load data within all target non-production time periods to obtain the baseline non-production load.

[0036] Specifically, the actual start and end times of each historical production task are extracted from historical production logs. This is because historical production logs record the complete timeline of each production task from start to completion, while historical power load data records the company's electricity consumption in a continuous time series format. Accurately aligning these two data points in the time dimension is crucial for accurately extracting the load performance corresponding to a specific production task's execution period. After extracting the actual start and end times, the load data sequence corresponding to these times is extracted from the historical power load data. This involves performing a range query in the time index of the historical power load data based on the start and end timestamps of each historical production task, locating the load power values ​​of all sampling points within that time period. Since power load data typically records power values ​​at fixed sampling intervals, such as every 15 minutes, all data from the first sampling point after the task's start time to the last sampling point before the task's end time is extracted, forming a load data sequence corresponding to the time span of that production task. During the extraction process, the timestamp of each sampling point in the load data sequence is extracted simultaneously to maintain the integrity of the time sequence information.

[0037] After obtaining the load data sequences corresponding to all historical production tasks, for each target historical production task to be analyzed, the load data corresponding to the start-up impact duration and shutdown impact duration are removed from the target load data sequence corresponding to the target historical production task. This is because the energy consumption characteristics of equipment during the start-up and shutdown phases differ significantly from those during steady-state operation. During the start-up phase, factors such as motor starting current surges and equipment preheating cause instantaneous power to be significantly higher than the steady-state value. Similarly, during the shutdown phase, factors such as equipment inertia and braking processes also generate load patterns different from normal production. Including data from these transitional phases in the calculation of standard production load would result in calculation results that do not accurately reflect the typical energy consumption level of the equipment under stable production conditions. By using a pre-set start-up impact duration parameter, which can be configured according to different equipment types (e.g., the start-up impact duration for large machining equipment might be set to 5 to 10 minutes, while for electric furnace equipment it might be as long as 30 minutes or even longer), the corresponding start-up impact duration parameter is automatically selected based on the equipment type used in the target historical production task, and data points corresponding to this duration are truncated from the beginning of the target load data sequence. Similarly, based on the preset downtime impact duration parameter, data points of a corresponding duration are truncated from the end of the target load data sequence. The setting of the downtime impact duration also takes into account the downtime characteristics of different equipment. For example, the downtime impact duration of an injection molding machine may be shorter, while the downtime impact duration of a continuous production line needs to consider the time consumed by material evacuation, equipment cooling, and other processes. The load subsequence obtained after the first and last elimination operation represents the stable operation phase of this production task.

[0038] Secondly, the maximum and minimum power values ​​of all data points in the load subsequence are calculated to determine the dynamic range of the power values. This range is then evenly divided into several power intervals. The number of intervals can be adaptively adjusted based on the total number of data points and the desired resolution. Generally, 20 to 50 intervals are suitable, ensuring that the width of each interval is neither too large to distinguish different operating modes nor too small to cause excessive data point dispersion. A histogram statistical method is used to traverse each power value data point in the load subsequence, determining which power interval each data point falls into, and incrementing the counter for the corresponding interval. After traversal, the distribution of data points within each power interval is obtained. For stably operating production equipment, the histogram typically presents one or more peaks. The power interval containing the peak corresponds to the equipment's main operating load level. For example, some equipment switching between full load and half load will form a bimodal distribution, while equipment continuously operating at full load will form a single significant peak in the high power interval.

[0039] The highest power range with the most data points from multiple power ranges is selected as the range representing the steady-state full-load operation characteristics of the production task. When selecting the highest power range, if the number of data points in multiple power ranges is similar, the range with the higher power value is preferred, because in the production load modeling scenario, the energy consumption level of the equipment during the main production phase is more important. After determining the highest power range, the arithmetic mean of all data points within the highest power range is calculated. That is, the sum of all load power values ​​falling within this power range is divided by the number of data points, and the average value obtained is the standard production load of the production task. At this point, the standard production load value represents the steady-state power demand of a single production line producing a single product under typical operating conditions after eliminating the impact of start-stop transitions and focusing on the main operating state. Referring to the above calculation process for the standard production load of a single production line producing a single product, the calculation is performed for products on other production lines to obtain the standard production load corresponding to each product.

[0040] While calculating the standard production load, the baseline non-production load for all production lines during non-production periods is calculated. This baseline energy consumption level represents the company's basic energy consumption when not conducting major production activities. This includes energy consumption from lighting systems, air conditioning and ventilation systems, office equipment, security monitoring systems, insulation and pressure maintenance of standby equipment, and line losses in the power distribution system. Although these energy consumptions are not directly related to specific production orders, they are still an important component of the company's total load. The historical production task timelines for all production lines are extracted. For each production line, the start and end times of all historical production tasks are marked on the timeline, resulting in an alternating pattern of production and idle periods. The overlapping idle periods of all production lines are identified. These overlapping periods represent the non-production periods during which no production tasks are performed, and the company's electricity load during these periods is primarily contributed by non-production facilities. From the identified non-production periods, target non-production periods exceeding a preset idle threshold are selected. The necessity of setting an idle threshold lies in the fact that excessively short non-production periods may still be affected by preceding and following production tasks; therefore, the preset idle threshold is typically set to 1 to 2 hours to ensure that the selected target non-production periods are sufficiently long. During the screening process, the duration of each non-production time period is checked one by one, and time periods whose duration exceeds the preset idle threshold are marked as target non-production time periods.

[0041] The baseline non-production load is obtained by averaging the power load data for all target non-production time periods. Based on the start and end time range of the target non-production time periods, the power values ​​of all sampling points within these time periods are extracted from historical power load data. The arithmetic mean of the load data for all target non-production time periods is then calculated, and the resulting mean is the baseline non-production load.

[0042] S103: Calculate the difference between the standard production load and the baseline non-production load to obtain the net increase average load corresponding to a single product, and average the net increase average load of multiple historical production tasks belonging to the same product type to obtain the characteristic operating power value corresponding to each product type.

[0043] In S103 above, the difference between the standard production load and the baseline non-production load is the net increase in average load. Physically, the net increase in average load represents the net increase in total load when the enterprise switches from a completely non-production state to performing the production task for that product. However, it is important to note the time-period matching issue of the baseline non-production load. Since multiple different baseline non-production load values ​​have been calculated based on time-period attributes, for example, the baseline non-production load may be higher during weekdays due to the full operation of office equipment and lighting systems, lower at night during weekdays due to the closure of office areas, and lowest on weekends and holidays due to the minimum number of personnel. When calculating the net increase in average load, it is necessary to select the corresponding baseline non-production load for difference calculation based on the actual time attribute of the historical production task. Only then will the net increase in average load represent the load within the same time period.

[0044] Based on the above differences, the net increase in average load value corresponding to each historical production task is calculated. Since the net increase in average load of the same product type may vary across different batches of historical production tasks due to factors such as different production quantities, equipment conditions, operator skill levels, and raw material batch differences, statistical analysis and averaging are performed on the net increase in average load of multiple historical production tasks belonging to the same product type. Here, "product type" refers to the same product on the same production line. That is, when performing statistical classification, not only must the product model and specifications be completely identical, but these historical production tasks must also be executed on the same production line. This is because different production lines may have different equipment configurations, process parameters, and levels of automation, even if they produce the same model of product. For example, all historical tasks of production line A producing product A are grouped together, and all historical tasks of production line A producing product B are grouped together. For each group, the net increase in average load value corresponding to all historical production tasks within the group is extracted, and the arithmetic average is calculated to obtain the average value of the group. This average value is defined as the characteristic operating power value corresponding to the product type. The characteristic operating power value represents the typical incremental water of the enterprise's total load relative to the non-production state when producing the product on the production line. It is the standardized energy consumption characteristic of the product type on the production line.

[0045] S104: Obtain the planned start and end times of the target production task from the production plan data, and generate the baseline load prediction curve of the target production task based on the planned start and end times and the characteristic operating power value corresponding to the target production task.

[0046] In step S104 above, detailed information about the target production task is obtained from the production plan data. The target production task refers to the production orders that the company plans to execute in the future. Production plan data is typically formulated by the company's production scheduling department based on factors such as customer orders, inventory status, equipment capacity, and raw material supply. It includes basic attribute information for each production task, the most crucial of which is the planned start and end time. The planned start and end times define the expected start and completion times of the production task, constituting the task's time interval on the timeline. The planned start and end times of the target production task are extracted from the production plan data.

[0047] Furthermore, based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production tasks, a baseline load forecast curve for the target production tasks is generated. Specifically, this includes: determining the target product type from the target production tasks and retrieving the start-up transition time and shutdown transition time for each target product type; dividing the complete cycle of the target production tasks based on the planned start and end times, start-up transition time, and shutdown transition time to obtain start-up transition periods, running periods, and shutdown transition periods; obtaining the planned start and end times from the planned start and end times, and obtaining multiple start-up time points based on the planned start and start-up transition time, using the ratio of the characteristic operating power value to the start-up transition time as the start-up slope; multiplying the difference between each start-up time point and the planned start time by the start-up slope to obtain the start-up load value corresponding to each start-up time point, sorting the multiple start-up load values ​​in chronological order to obtain the load sequence of the start-up transition period; and adding the planned start time to the start-up transition time to obtain the start time of the running period. The process involves several steps: First, the difference between the planned termination time and the shutdown transition duration is used as the end time of the operating period. Multiple steady-state time points are determined based on the start and end times. The characteristic operating power value is used as the steady-state load value corresponding to each steady-state time point. All steady-state load values ​​are sorted chronologically to obtain the load sequence of the operating period. Second, the difference between the planned termination time and the shutdown transition duration is used as the start time of the shutdown transition period. Multiple shutdown time points are generated based on the start time of the shutdown transition duration. The negative value of the characteristic operating power value is divided by the shutdown transition duration to obtain the shutdown slope. The difference between each shutdown time point and the start time of the shutdown transition period is calculated to obtain the time span. The time span is multiplied by the shutdown slope and added to the characteristic operating power value to obtain the shutdown load value corresponding to each shutdown time point. All shutdown load values ​​are sorted chronologically to obtain the load sequence of the shutdown transition period. Finally, the load sequences of the startup transition period, the operating period, and the shutdown transition period are concatenated chronologically to obtain the baseline load prediction curve.

[0048] Specifically, the target product type is determined from the target production task, and the start-up transition time and shutdown transition time for the target product type are retrieved. This is because different product types have significantly different production processes on different production lines. For example, reactors in the chemical industry require a series of processes during startup, including preheating, temperature increase, pressurization, and material feeding, which may take 2 to 3 hours. In contrast, chip mounters in the electronics assembly industry only need to complete system self-checks and positioning calibrations, and the startup process may be completed within 10 minutes. Ignoring these product and process differences and treating all task start-up and shutdown processes with uniform fixed durations would lead to distortion of the baseline load forecast curve in the time dimension. The product model and assigned production line number of the target production task are extracted from the production plan data. These two attributes are combined as a unique identifier for the product type. For example, if the product model of a target production task is "Type A motor" and the assigned production line is "Production Line 3," then the target product type is identified as the combination "Production Line 3 - Type A motor." The start-up transition time and shutdown transition time corresponding to the target product type are then retrieved. Start-up transition time is the time span required for quantitative production equipment to transition from a static state to rated operating conditions. Shutdown transition time is the time span required for quantitative production equipment to transition from rated operating conditions to complete shutdown.

[0049] After obtaining the start-up transition duration and shutdown transition duration corresponding to the target product type, the complete cycle of the target production task is divided into time periods based on the planned start-up and end times, start-up transition duration, and shutdown transition duration. This results in the start-up transition period, the running period, and the shutdown transition period. In other words, the time interval occupied by the target production task on the time axis is decomposed into three sub-intervals with different load characteristics. First, the planned start-up and end times of the target production task are extracted from the production plan data. The planned start-up and end times contain two timestamp parameters, namely the planned start time and the planned end time. For example, if the planned start time of a production task is "January 10, 202x, 08:00:00" and the planned end time is "January 10, 202x, 16:00:00", then the planned execution cycle of the task is 8 hours.

[0050] The time range of the start-up transition period is determined based on the planned start time and the start-up transition duration. The start time of the start-up transition period is the planned start time, and the end time of the start-up transition period is the planned start time plus the start-up transition duration. This end time marks the completion of the equipment start-up sequence and entry into steady-state operation. For example, if the start-up transition duration of the above task is 30 minutes, then the start-up transition period is from "January 10, 202x 08:00:00" to "January 10, 202x 08:30:00".

[0051] The runtime phase is the main stage of a production task, corresponding to the time window during which equipment operates continuously under steady-state conditions. During this period, the load should remain at a stable level near the characteristic operating power value. The start time of the runtime phase is the end time of the start-up transition period, which is the planned start time plus the start-up transition duration. The end time of the runtime phase needs to be pre-allocated a shutdown transition duration based on the planned end time, i.e., the planned end time minus the shutdown transition duration. This ensures that there is sufficient time window to execute the shutdown process after the runtime phase ends, so that the shutdown process is completed exactly at the planned end time. For example, if the shutdown transition duration for the above task is 20 minutes, then the end time of the runtime phase is "January 10, 202x, 15:40:00," meaning that the shutdown process begins 20 minutes before the planned end time "January 10, 202x, 16:00:00." The complete time range of the runtime phase is from "January 10, 202x, 08:30:00" to "January 10, 202x, 15:40:00."

[0052] The shutdown transition period is the final stage of a production task, corresponding to the time window during which equipment transitions from steady-state operation to complete shutdown. The start time of the shutdown transition period is the end time of the running segment, which is the planned termination time minus the shutdown transition duration. The end time of the shutdown transition period is the planned termination time, because the task should complete all shutdown operations by the planned termination time. For example, the shutdown transition period for the task mentioned above is from "January 10, 202x, 15:40:00" to "January 10, 202x, 16:00:00". After dividing the above three periods, the load sequences corresponding to each of the three periods are executed sequentially.

[0053] First, we introduce the specific calculation process of the load sequence during the start-up transition period. The planned start time is extracted from the planned start and end times and used as the time origin for the start-up transition period. Then, multiple start-up time points are generated based on the planned start time and the start-up transition duration. These start-up time points are discrete sampling points on the time axis of the start-up transition period. The start-up transition duration is divided equally according to a preset time resolution. For example, if the time resolution is set to 15 minutes and the start-up transition duration is 30 minutes, the start-up transition period is divided into two time periods, corresponding to three time points: the start time point "08:00:00", the intermediate time point "08:15:00", and the end time point "08:30:00". After generating the start-up time points, the characteristic operating power value is obtained, and the ratio of the characteristic operating power value to the start-up transition duration is used as the start-up slope, which is the rate of load growth per unit time. For each startup time point within the startup transition period, the time difference between that time point and the planned start time needs to be calculated. This difference represents the time span that has elapsed from the start of startup to the current time point. For example, the difference between the time point "08:15:00" and the planned start time "08:00:00" is 15 minutes, or 0.25 hours. Multiplying this time difference by the startup slope yields the cumulative load increase from the start of startup to the current time point. For example, 0.25 hours multiplied by 1200 kilowatts per hour equals 300 kilowatts, indicating that after a 15-minute startup process, the startup load value is 300 kilowatts. This process is repeated for all startup time points within the startup transition period, calculating the corresponding startup load value for each time point. These startup load values ​​are then sorted chronologically to obtain the load sequence for the startup transition period.

[0054] Next, we will introduce the specific calculation process of the runtime load sequence. First, we determine the accurate time range of the runtime segment. For example, if the planned start time is "08:00:00" and the start-up transition duration is 30 minutes, then the start time of the runtime segment is "08:30:00". The end time of the runtime segment is calculated by the difference between the planned end time and the stop-up transition duration. If the planned end time is "16:00:00" and the stop-up transition duration is 20 minutes, then the end time of the runtime segment is "15:40:00". Based on the start and end times of the runtime segment, multiple steady-state time points within this period are generated. The steady-state time points are discrete sampling points of the runtime segment on the time axis. The sampling strategy is the same as that for the start-up transition period, which will not be explained in detail here. For each steady-state time point within the operating period, the characteristic operating power value is taken as the steady-state load value corresponding to that time point. This is because the load should remain at a constant level of the characteristic operating power value during the steady-state operation phase and should not change with time. For example, if the characteristic operating power value is 300 kilowatts, then the steady-state load value at all steady-state time points is 300 kilowatts. All steady-state load values ​​are sorted in chronological order to obtain the load sequence of the operating period.

[0055] Finally, let's introduce the specific calculation process of the load sequence during the shutdown transition period. The start time of the shutdown transition period is determined by calculating the difference between the planned end time and the shutdown transition duration. This start time is the same as the end time of the running period, marking the end of steady-state operation and the beginning of the shutdown process. For example, if the planned end time is "16:00:00" and the shutdown transition duration is 20 minutes, then the start time of the shutdown transition period is "15:40:00", and the end time of the shutdown transition period is the planned end time "16:00:00". Therefore, the complete time range of the shutdown transition period is from "15:40:00" to "16:00:00", lasting 20 minutes.

[0056] Based on the start time and duration of the shutdown transition period, multiple shutdown time points are generated within this period. These shutdown time points are discrete sampling points on the time axis of the shutdown transition period. The sampling strategy still uses uniform sampling at fixed time intervals, similar to the sampling process for the start-up transition period described above, which will not be elaborated upon here. The shutdown slope is obtained by negatively dividing the characteristic operating power value by the shutdown transition duration. For example, if the characteristic operating power value is 300 kW and the shutdown transition duration is 20 minutes (1 / 3 hour), then the shutdown slope is -300 divided by 1 / 3 equals -1800 kW per hour. This means that during the shutdown process, the load decreases by an average of 1800 kW per hour, or 30 kW per minute. The physical meaning of the negative slope is the rate of load reduction over time. For each shutdown time point within the shutdown transition period, calculate the time span between the shutdown time point and the start time of the shutdown transition period. Multiply this time span by the shutdown slope to obtain the cumulative load decrease from the start of the shutdown to the current time point. Add this decrease to the characteristic operating power value to obtain the shutdown load value corresponding to the shutdown time point. Repeat this process for all shutdown time points within the shutdown transition period, calculating the corresponding shutdown load value for each time point. Then, sort these shutdown load values ​​in chronological order to obtain the load sequence for the shutdown transition period. After generating load sequences for the start-up transition period, the running period, and the shutdown transition period, the load sequences of the three periods are spliced ​​together in chronological order to obtain the complete baseline load forecast curve for the target production task. The core of the splicing is to merge three independent time series arrays into a continuous time series array, while ensuring the continuity of time and the smooth transition of load values. According to the chronological order of the start-up transition period, the running period, and the shutdown transition period, the load sequences of each period are appended to the result array in turn, forming a time series with a length equal to the sum of the total number of sampling points in the three periods.

[0057] S105: Merge the baseline load forecast curve and historical power load data to obtain input features, and input the input features into the time series forecast model to obtain the initial load forecast curve within the forecast period.

[0058] In step S105 above, the baseline load forecast curve and historical power load data are merged to obtain input features. Specifically, this includes: determining a target time axis based on the time range of the historical power load data and the time range of the baseline load forecast curve; mapping the historical power load data and the baseline load forecast curve onto the target time axis; selecting a target time point from the target time axis; extracting multiple historical load values ​​within a preset historical window length from the historical power load data before the target time point; sorting the multiple historical load values ​​in chronological order to obtain a historical load feature sequence; the target time point being any time point on the target time axis; numerically encoding the time attribute information of the target time point to obtain time attribute features; extracting multiple baseline load forecast values ​​within a preset guidance window from the baseline load forecast curve based on the target time point; sorting the multiple baseline load forecast values ​​in chronological order to obtain a baseline guidance feature sequence; concatenating and combining the historical load feature sequence, time attribute features, and baseline guidance feature sequence corresponding to the target time point to obtain a high-dimensional feature vector; and combining the high-dimensional feature vectors corresponding to all target time points within the forecast period in chronological order to obtain the input features.

[0059] Specifically, to address the time mismatch between the baseline load forecast curve and historical power load data, time alignment is required using a unified time coordinate system. This ensures that each time point is simultaneously associated with corresponding values ​​in both historical data and the forecast curve. Boundary information for the time range is extracted from both the historical power load data and the baseline load forecast curve. Based on these two time ranges, the boundaries of the target time axis are determined. The start time of the target time axis is the earlier of the earliest timestamp from the historical data and the start timestamp from the forecast curve, and the end time is the later of the latest timestamp from the historical data and the end timestamp from the forecast curve. After determining the start and end boundaries and sampling interval of the target time axis, a sequence of all time points on the target time axis is generated. Mapping the historical power load data onto the target time axis involves assigning a corresponding historical load value to each time point on the target time axis, a process that requires handling time alignment. Similarly, mapping the baseline load forecast curve onto the target time axis involves assigning a corresponding baseline forecast value to each time point on the target time axis, a process similar to that for historical power load data. After the mapping is completed, each point in time on the target time axis is associated with information from historical power load data and baseline load forecast curves.

[0060] A target time point is selected from the target time axis, and a high-dimensional feature vector is constructed for this target time point as model input. The target time point can be any point on the target time axis. For each target time point, multiple historical load values ​​within a preset historical window length are extracted from the historical power load data. The historical window length is a pre-defined hyperparameter that indicates how long back to extract historical load features. The historical window length is typically set to several hours to several days, covering the intraday periodic changes and short-term trends of the load. The extracted historical load values ​​are sorted in chronological order to obtain a historical load feature sequence. This sorting operation ensures that the elements in the sequence are arranged in chronological order from oldest to newest, reflecting the true trajectory of load evolution over time.

[0061] It is also necessary to extract the time attribute features of the target time point. These time attribute features are a multi-dimensional characterization of the target time point's position, containing rich calendar and event information. For example, weekday workloads are typically higher than weekend workloads, and daytime workloads are typically higher than nighttime workloads. The time attribute information of the target time point is numerically encoded, converting the non-numerical time representation of timestamps into numerical feature vectors, enabling machine learning models to process and learn from them. Time attribute features include two main categories: periodic time features and event-based time features. Periodic time features characterize the periodic structure of time, i.e., the position of a time point within various cycles. These cycles include natural time cycles such as daily, weekly, and yearly cycles. Multiple periodic time features are extracted from the timestamp of the target time point, including dimensions such as the current hour, day of the week, day of the year, and month. The current hour feature indicates which hour of the day the target time point falls within, with a value ranging from 0 to 23. Event-based time features characterize whether a point in time falls within a specific event or state. These features are typically expressed as Boolean symbols, with values ​​of 0 or 1, indicating whether an event has occurred or whether a state is established. The extracted event-based time features include Boolean symbols for whether it is a weekend or a public holiday. All extracted periodic and event-based time features are combined to form a time attribute feature vector for the target time point.

[0062] Multiple baseline load forecast values ​​are extracted from the baseline load forecast curve within a preset guiding window based on the target time point. The guiding window length is a pre-set hyperparameter, typically set to several hours to a day, covering the production task cycle near the target time point. Based on practical experience, the guiding window length can be set to 12 hours or 24 hours. The preset guiding window length is extended forward from the target time point to determine the end time point of the guiding window. The baseline forecast values ​​corresponding to these time points within the guiding window are extracted from the target time axis. The extracted baseline load forecast values ​​are sorted chronologically to obtain the baseline guiding feature sequence. After extracting the historical load feature sequence, time attribute features, and baseline guiding feature sequence, these three features are concatenated to form a high-dimensional feature vector corresponding to the target time point. Referring to the above extraction and concatenation process for the target time point, extraction and concatenation operations are performed for other time points within the prediction period to generate a high-dimensional feature vector corresponding to each target time point. The high-dimensional feature vectors corresponding to all target time points within the prediction period are combined chronologically to obtain the input features.

[0063] Before inputting the input features into the time series forecasting model to obtain the initial load forecast curve within the forecast period, it is necessary to first construct the time series forecasting model. A Long Short-Term Memory (LSTM) network is used as the core architecture of the time series forecasting model. Choosing an LSTM-based network effectively solves the gradient vanishing and gradient exploding problems faced by traditional recurrent neural networks when processing long-sequence data, thus capturing short-term fluctuation features while preserving long-term historical information. For the application scenario of this application, the length of the input sequence to be processed by the model is set to 168 hours. This time window covers a complete one-week production cycle, fully encompassing the weekly cycle of enterprise production scheduling, equipment maintenance cycles, and load differences between weekends and weekdays. The memory unit structure of the LSTM network enables it to effectively transmit key information within this relatively long time span, identifying the load evolution patterns during the execution of production orders.

[0064] After preprocessing and feature engineering, baseline load forecast curves, historical power load data, equipment operating parameters, and environmental condition data are constructed into a standardized three-dimensional input tensor. The input layer uses embedding and fully connected layers to initially encode and transform these multi-dimensional features, mapping heterogeneous features to a unified feature space. The model's hidden layers employ a multi-layer stacked LSTM unit structure, with the first LSTM unit receiving the feature sequence from the input layer. To further enhance the model's feature extraction capability, a second LSTM unit is stacked on top of the first LSTM layer. A dropout regularization mechanism is introduced between each LSTM unit layer, with a dropout ratio set to 0.2, randomly discarding 20% ​​of neuron connections to effectively prevent the model from overfitting the training data during training. The model's output layer design converts the temporal features extracted by the LSTM hidden layers into the final load forecast values. A fully connected layer is used as the output layer, with the number of neurons set to the number of hours in the prediction time domain, i.e., 24 neurons corresponding to hourly load forecasts for the next 24 hours. The output layer uses a linear activation function to map the hidden state of the last time step of the LSTM into a continuous sequence of load forecast values.

[0065] The model's loss function uses mean squared error (MSE) as the primary loss function, calculated as the average of the squared difference between the predicted and actual load values ​​across all samples and time steps. The MSE loss function penalizes prediction bias with the square of the bias magnitude, making the model more focused on eliminating larger biases during training. This aligns with the high accuracy requirements for peak load prediction when enterprises participate in the electricity market.

[0066] The model training process employs a batch training strategy, dividing the preprocessed training dataset into batches of 32, with each batch containing 32 sample sequences. Each sample sequence includes 168 hours of input features and 24 hours of target load values. The model parameters are progressively optimized through multiple iterations. Training ends when the iteration conditions are met, and performance is evaluated using a test set. The test set contains samples from production data from different time periods than the training set, ensuring that the evaluation results accurately reflect the model's predictive ability in the face of new orders and new production schedules.

[0067] After converting the input features into a structure that meets the input requirements of the time series forecasting model, the input features are fed into the time series forecasting model. The time series forecasting model adopts a Long Short-Term Memory (LSTM) network architecture. Through recursive calculation of multiple LSTM units, it progressively extracts the coupling pattern between the baseline load curve and historical load data, identifying the correlation pattern between the load change trend driven by production orders and historical energy consumption inertia. The output layer of the time series forecasting model generates the load forecast value for each time node within the forecast period (set as the next 24 hours). These forecast values ​​together constitute the initial load forecast curve.

[0068] S106: Extract the baseline operating efficiency and cumulative operating time of the production equipment from the historical production logs, calculate the baseline operating efficiency based on the cumulative operating time, and obtain the efficiency correction factor.

[0069] In S106 above, the baseline operating efficiency and cumulative operating time of the production equipment are extracted from the historical production logs. The baseline operating efficiency is calculated based on the cumulative operating time to obtain an efficiency correction factor. Specifically, this includes: using the characteristic operating power value as the baseline net power consumption of the product type under ideal equipment conditions; retrieving the equipment's most recent maintenance time from the historical production logs; determining the cumulative operating time based on the current time and the most recent maintenance time; marking the current time as a historical efficiency anchor point when the cumulative operating time exceeds the preset calibration periodicity; subtracting the baseline non-production load from the actual average load of historical production tasks to obtain the actual net power consumption of the anchor point corresponding to the historical efficiency anchor point; calculating the difference between the actual net power consumption of the anchor points corresponding to two temporally adjacent historical efficiency anchor points to obtain the power consumption change; and dividing the power consumption change by the two historical efficiency values. The runtime between anchor points is used to obtain the segmented power consumption increment gradient within the target operating interval, which is the interval corresponding to two temporally adjacent historical efficiency anchor points. Multiple segmented power consumption increment gradients are sorted to obtain a historical power consumption increment gradient sequence. The most recent segmented power consumption increment gradient is selected from this sequence as the prediction baseline gradient. The average cumulative runtime of the target production task is calculated, and the difference between the average cumulative runtime and the target cumulative runtime at the most recent historical efficiency anchor point is calculated to obtain the prediction interval runtime. The prediction interval runtime is multiplied by the prediction baseline gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net increase in power consumption at the anchor point corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase in average load. The predicted net increase in average load is divided by the baseline net increase in power consumption to obtain the efficiency correction factor.

[0070] Specifically, since the characteristic operating power value represents the standard energy consumption level of the equipment when processing products under ideal and healthy conditions, it is used as the baseline net increase in power consumption. The most recent maintenance time for each piece of equipment is retrieved from the equipment maintenance record module of the historical production log. This time point may correspond to routine maintenance, fault repair, or component replacement. By calculating the time span between the current time point and the most recent maintenance time point, and combining this with the actual operating hours recorded in the equipment operation log, the cumulative runtime of the equipment is determined. It should be noted that the cumulative runtime is not simply a difference in calendar time, but rather the sum of the effective duration of the equipment's actual operation. The accurate cumulative runtime value is obtained by summing the durations of all operating segments.

[0071] The cumulative runtime is compared with the preset calibration period, which is configured differently based on the characteristics of different equipment types. For high-speed machining equipment, the preset calibration period can be set to 500 hours, while for equipment such as heat treatment furnaces with relatively stable operating loads, this period can be extended to 1000 hours. When the cumulative runtime of a piece of equipment exceeds the preset calibration period, the current time point is marked as a historical efficiency anchor point. For each marked historical efficiency anchor point, after the anchor point is marked, the first production task of the same type of product completed immediately after the anchor point, which is the same as the one used in the baseline net increase in power consumption calculation, is selected as the historical production task. The reason for selecting the same type of product is to ensure consistency in comparison. The load monitoring data of the equipment throughout the entire processing process is extracted from the execution log of the historical production task, and the actual average load of the task is calculated. Then, the baseline non-production load is subtracted from the actual average load of the historical production task, and the fixed base load unrelated to the production task is removed to obtain the actual net increase in power consumption of the anchor point corresponding to the historical efficiency anchor point.

[0072] After accumulating multiple historical efficiency anchor points and their corresponding actual net power increase, two historical efficiency anchor points that are temporally adjacent are paired, for example, the i-th anchor point and the (i+1)-th anchor point. Their respective actual net power increase is extracted and denoted as Pi and Pi+1. The power change between these two anchor points is obtained by subtracting Pi from Pi+1. The sign of the power change reflects the direction of the device's energy efficiency change; a positive value indicates increased energy consumption, i.e., decreased efficiency, while a negative value may correspond to maintenance or upkeep between the two anchor points, resulting in an efficiency recovery. The cumulative runtime recorded for each of these two adjacent anchor points is then obtained and denoted as Ti and Ti+1. The runtime span between the two anchor points is obtained by subtracting Ti from Ti+1. The power change is divided by the runtime span to obtain the segmented power increment gradient within the target operating interval. This process is repeated for all historical efficiency anchor points, calculating the segmented power increment gradient between all adjacent anchor point pairs. These gradient values ​​are then arranged in chronological order to form a historical power increment gradient sequence. The most recent segmented power increment gradient is selected from the historical power increment gradient sequence; that is, the gradient value in the sequence that is closest to the current time is used as the prediction baseline gradient. After determining the prediction baseline gradient, the planned execution information of the target production task is obtained. By analyzing the estimated execution time of all previously scheduled tasks in the production schedule, the average cumulative runtime of the equipment when the target production task begins is accumulated. The average cumulative runtime represents the time-weighted center position of the equipment's operating status during task execution.

[0073] The cumulative runtime recorded at the most recent historical efficiency anchor point is found from the historical efficiency anchor point sequence and recorded as the most recent anchor point cumulative runtime. The difference between the average cumulative runtime of the target production task and the most recent anchor point cumulative runtime is calculated to obtain the prediction interval runtime. The prediction interval runtime quantifies the extra time span of equipment operation from the most recent efficiency observation point to the execution time of the target task. The prediction interval runtime is multiplied by the previously determined prediction baseline gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net increase in power consumption corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase in average load. The predicted net increase in average load comprehensively reflects the current health status of the equipment and the efficiency evolution trend over a period of time. The predicted net increase in average load is divided by the baseline net increase in power consumption to obtain the efficiency correction factor. The efficiency correction factor is a dimensionless ratio parameter. The value of the efficiency correction factor directly reflects the degree of deviation of the actual energy consumption of the equipment from the ideal state. When the efficiency correction factor is equal to 1, it indicates that the equipment operating efficiency is maintained in the ideal state and the actual energy consumption is consistent with the benchmark. When the efficiency correction factor is greater than 1, it indicates that the equipment efficiency has declined and the actual energy consumption is higher than the benchmark. The larger the value, the more severe the decline. When the efficiency correction factor is less than 1, it may correspond to the situation where the efficiency of the equipment is better than the initial benchmark after maintenance or there is an abnormality in the measurement.

[0074] S107: Determine the environmental correction factor based on historical operating temperature and historical power load data in historical production logs.

[0075] In S107 above, based on historical operating temperature and historical power load data in the historical production log, an environmental correction factor is determined. Specifically, this includes: extracting ambient temperature data from all historical production tasks from the historical production log; determining a temperature range covering all historical operating conditions based on multiple ambient temperature data points; dividing the temperature range to obtain multiple continuous, non-overlapping operating temperature intervals; selecting the operating temperature interval containing the most historical production tasks as the benchmark temperature interval; using the difference between the standard production load and the benchmark non-production load corresponding to each historical production task as the actual net increase in power consumption; and dividing the actual net increase in power consumption by the characteristic operating power value of the historical production task to obtain the comprehensive power consumption skewness; and obtaining the cumulative operating power of the equipment for each historical production task. The system calculates the environmental impact coefficient by determining the historical equipment efficiency correction factor based on the cumulative operating time of the equipment. It then uses the ratio of the comprehensive power consumption skewness to the historical equipment efficiency correction factor as the environmental impact coefficient. This environmental impact coefficient is linked to each historical production task, and the average operating temperature of all historical production tasks is categorized into various operating temperature ranges. All environmental impact coefficients within each operating temperature range are calculated to obtain the range-average environmental impact coefficient. Finally, the system performs load forecasting for the target production task, obtains the predicted ambient temperature within the forecast period, compares the predicted ambient temperature with each operating temperature range, determines the range-average environmental impact coefficient corresponding to the target historical production task based on the comparison results, and uses this range-average environmental impact coefficient as the environmental correction factor.

[0076] Specifically, complete ambient temperature data is extracted from historical production logs. For each historical production task, corresponding ambient temperature monitoring data is stored during task execution. The ambient temperature data for each historical production task is acquired, and the minimum and maximum values ​​are calculated. Based on these minimum and maximum values, a temperature range covering all historical operating conditions is determined. After obtaining the historical operating temperature range, based on the climate characteristics of the company's location and the distribution patterns of historical temperature data, the temperature range is divided into multiple continuous, non-overlapping operating temperature intervals. For example, for a manufacturing company with a temperature range of -5℃ to 40℃, this is divided into five operating temperature intervals: -5℃ to 5℃, 5℃ to 15℃, 15℃ to 25℃, 25℃ to 35℃, and 35℃ to 40℃. After interval division, each historical production task is assigned to its corresponding operating temperature interval based on its average operating temperature. The number of historical production tasks included in each operating temperature interval is counted. The operating temperature interval containing the most historical production tasks is selected as the benchmark temperature interval to measure the degree of energy consumption deviation under other temperature conditions.

[0077] Next, the energy consumption deviation of each historical production task during actual execution is calculated. For any historical production task, the standard production load data during the task's execution period is extracted from the production log. The standard production load refers to the average power consumed by the equipment during the entire process from task start-up to completion. The actual net increase in power consumption is obtained by subtracting the baseline non-production load from the standard production load corresponding to the historical production task. This parameter, actual net increase in power consumption, purely reflects the incremental power consumption caused by executing the production task, eliminating the interference of base load fluctuations at different times. The characteristic operating power value of the product type corresponding to the historical production task under ideal equipment conditions is extracted from the historical database. The actual net increase in power consumption is divided by the characteristic operating power value to obtain the comprehensive power consumption skewness. The comprehensive power consumption skewness reflects the overall energy consumption deviation caused by the superposition of all factors such as equipment efficiency degradation, ambient temperature influence, and process parameter fluctuations.

[0078] To isolate the impact of equipment efficiency degradation from the overall power consumption skewness, the above-mentioned backtracking calculation process for the efficiency correction factor can be referenced to determine the historical equipment efficiency correction factor based on the cumulative duration of the equipment.

[0079] The environmental impact coefficient is obtained by dividing the overall power consumption skewness by the historical equipment efficiency correction factor. The environmental impact coefficient represents the energy consumption correction factor caused by environmental conditions after eliminating equipment efficiency factors. A coefficient of 1 indicates that the ambient temperature during the execution of the historical task had no significant impact on energy consumption or was within standard environmental conditions. A coefficient greater than 1 indicates increased energy consumption due to high or low temperatures, while a coefficient less than 1 may correspond to reduced energy consumption under suitable temperatures. The calculated environmental impact coefficient is then linked to the corresponding historical production tasks. An environmental impact coefficient field is added to each historical production task record in the database and populated with the corresponding value. Simultaneously, the average operating temperature information for that task is stored. Based on the average operating temperature of all historical production tasks, they are categorized into various operating temperature ranges. For example, historical tasks with an average operating temperature of 18℃ are categorized into the 15℃ to 25℃ range, and tasks with an average operating temperature of 32℃ are categorized into the 25℃ to 35℃ range.

[0080] For each operating temperature range, the environmental impact coefficient values ​​corresponding to all historical production tasks within that range are extracted. These coefficients are then calculated to determine the central trend value. The calculation of the central trend value can include, but is not limited to, arithmetic mean, weighted summation, or averaging, to obtain the interval average environmental impact coefficient for that operating temperature range. The forecast ambient temperature information corresponding to the planned execution period of the target production task is obtained. Weather forecasts typically provide temperature prediction data for the next few hours to days, including hourly temperature change curves. The forecast temperature sequence from the expected start time to the end time of the target production task is extracted, and the forecast average ambient temperature during the task execution period is calculated using a time-weighted average. The forecast ambient temperature is compared with the boundary values ​​of the previously established operating temperature ranges to determine which operating temperature range the forecast temperature falls into. For example, a forecast temperature of 22℃ matches the 15℃ to 25℃ range, and a forecast temperature of 38℃ matches the 35℃ to 40℃ range. Based on the matching results, the average environmental impact coefficient corresponding to the target operating temperature range is queried from the mapping table from temperature range to environmental impact coefficient, and this coefficient value is used as the environmental correction factor for the target production task.

[0081] S108: Correct the initial load forecast curve based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

[0082] In step S108 above, the initial load forecast curve is corrected based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve. Specifically, this includes: obtaining the current forecast time point from the forecast period and obtaining the initial forecast load value corresponding to the current forecast time point from the initial load forecast curve; retrieving the baseline load value corresponding to the current forecast time point from the baseline load forecast curve; multiplying the efficiency correction factor and the environmental correction factor to obtain the comprehensive correction coefficient; multiplying the baseline load value and the comprehensive correction coefficient to obtain the corrected net increase in production load; subtracting the baseline load value from the initial forecast load value to obtain the non-production-related baseline load component, and adding the corrected net increase in production load and the non-production-related baseline load component to obtain the final forecast load value corresponding to the current forecast time point; and combining the final forecast load values ​​corresponding to all forecast time points in chronological order to obtain the final load forecast curve.

[0083] Specifically, based on the start and end times and time resolution of the prediction period, a sequence of all time points requiring prediction within the prediction period is generated. For example, for a 24-hour hourly prediction scenario, a timestamp sequence of the next 24 hours from the current moment is generated. The timestamp sequence is traversed in chronological order, and independent correction calculations are performed for each time point. The currently processed time point is the current prediction time point. The initial predicted load value corresponding to the current prediction time point is extracted from the initial load prediction curve. Simultaneously, the baseline load value corresponding to the current prediction time point is retrieved from the baseline load prediction curve. After obtaining the initial predicted load value and the baseline load value, the efficiency correction factor and the environmental correction factor are multiplied to obtain the comprehensive correction coefficient. That is, the energy consumption increment caused by equipment efficiency degradation and the energy consumption offset caused by changes in ambient temperature can be approximately regarded as independent mechanisms, and the overall energy consumption correction factor is equal to the product of the two correction factors. For example, if an efficiency correction factor of 1.08 indicates that equipment wear and tear leads to an 8% increase in energy consumption, and an environmental correction factor of 1.05 indicates that ambient temperature leads to a 5% increase in energy consumption, then the comprehensive correction coefficient is 1.08 × 1.05 = 1.134, which means that under the combined effect of the two factors, the actual energy consumption increases by 13.4% compared to the ideal baseline.

[0084] The corrected net increase in production load is obtained by multiplying the baseline load value by the comprehensive correction factor. Since the initial forecast load value is actually the sum of production-related load and non-production baseline load, the estimated net increase in production load calculated based on ideal assumptions is subtracted from the initial forecast load value to obtain the non-production-related baseline load component. The corrected net increase in production load is then added to the non-production-related baseline load component to obtain the final forecast load value corresponding to the current forecast time point. The mathematical structure of the final forecast load value is: Final forecast load value = Non-production-related baseline load component + Corrected net increase in production load. For example, for a certain time point, the initial forecast load value is 500 kW, of which the baseline load is 200 kW and the ideal net increase in production load is 300 kW. After correction by the comprehensive correction factor of 1.134, the net increase in production load becomes 340.2 kW, and the final forecast load value is 540.2 kW, an increase of 40.2 kW compared to the initial forecast.

[0085] The above correction calculation process is repeated for all forecast time points within the forecast period in chronological order, generating a corresponding final forecast load value for each time point. After the correction calculations for all time points are completed, these discrete final forecast load values ​​are combined and concatenated in chronological order to construct a continuous time series data structure, thus obtaining the final load forecast curve.

[0086] In one possible implementation, in an industrial production environment, although enterprises can accurately predict the electricity demand for planned production tasks through refined load forecasting methods, various unforeseen failures are inevitable during actual production. These unplanned events can cause the original production arrangements and load forecasts to fail. If the load forecast curve is not dynamically corrected in a timely manner, it will lead to a serious mismatch between power supply and actual demand, resulting in emergency response measures being unable to be implemented due to power shortages or causing a huge waste of power resources. Therefore, this application can also dynamically correct the final load forecast curve based on the impact of emergency measures on the power load.

[0087] When a sudden failure occurs during the actual execution of a target production task, the type and scope of the failure are obtained, and the planned delivery time of the target production task is retrieved from the production plan data. The interval between the planned delivery time and the current time is calculated, and the type and scope of the failure are analyzed based on the interval to obtain the failure handling level. Based on the failure handling level and the type of the failure, a matching solution is obtained from a pre-set emergency plan library to obtain the target emergency handling measures. The expected change in power load resulting from the implementation of the target emergency handling measures is calculated, and an emergency load curve starting from the time of failure is generated. Based on the time of failure, the emergency load curve is superimposed or replaced with the final load forecast curve to obtain the real-time corrected emergency load forecast curve.

[0088] Specifically, during operation, the system continuously collects operational status data from each production line and key equipment through a real-time monitoring module. When a sudden failure is detected during the actual execution of a target production task, the monitoring system immediately triggers a fault identification and analysis process. The detection of sudden failures relies on a multi-layered anomaly identification mechanism, including various data sources such as equipment sensor alarm signals, process parameter limit violations, product quality inspection failure records, and equipment protection device action information. Key features are extracted from this raw fault information, and the specific type of sudden failure is determined through a fault type classification algorithm. The fault type classification system is established based on the actual situation of industrial production and typically includes mechanical equipment failures such as bearing damage, gear breakage, and transmission system failure; electrical failures such as motor burnout, control system malfunction, and power outage; process failures such as temperature runaway, pressure anomalies, and raw material quality problems; and human error. By matching the real-time monitored fault feature parameters with typical fault patterns in a fault knowledge base, the specific type of fault is identified. For example, bearing failures can be identified through vibration spectrum analysis, motor winding short circuits through current waveform analysis, and cooling system failures through temperature change curves.

[0089] After determining the type of failure, the scope of its impact is assessed. This assessment is conducted from multiple dimensions, analyzing the location and role of the faulty equipment in the production process; examining the product batches currently being processed on that equipment and subsequent planned product orders. Next, the time-level impact is considered. Based on the typical repair time and spare parts supply cycle of the faulty equipment, the time span required for failure repair is estimated, and the number of production tasks affected within that time span is determined. The planned delivery time information for the target production tasks is retrieved from the production planning management database. The planned delivery time is a key benchmark for assessing the urgency of the failure, because for the same type and scope of failure, if it occurs very close to the delivery deadline, the urgency and handling requirements will be much higher than if it occurs at the beginning of the production task. By calculating the interval between the planned delivery time and the current time, the time margin for completing the task is quantitatively assessed. The interval is divided into several segments, for example, less than 24 hours is extremely urgent, 24-72 hours is highly urgent, 72 hours-7 days is moderately urgent, and more than 7 days is for routine handling. The shorter the interval, the more expensive but faster emergency measures need to be taken, while when the interval is longer, a lower-cost conventional maintenance solution can be chosen.

[0090] A comprehensive analysis of the type and impact range of sudden failures based on the interval duration is used to determine the failure handling level. The severity of the failure type, the breadth of its impact range, and the urgency of the interval duration are quantified, scored, and weighted. For the failure type dimension, a base score is set based on the repair difficulty and time cost of different failures; for example, a simple sensor failure has a base score of 1 point, a complex spindle damage has a base score of 8 points, and a core control system failure has a base score of 10 points. For the impact range dimension, weighting coefficients are set based on the scale of output value affected and the number of production lines; for example, the coefficient for affecting a single piece of equipment is 1.0, for affecting an entire production line is 2.0, and for affecting multiple production lines is 3.0. For the interval duration dimension, a multiplier factor is set based on the urgency of time; for example, an interval greater than 7 days has a factor of 1.0, 3-7 days has a factor of 1.5, 1-3 days has a factor of 2.0, and less than 24 hours has a factor of 3.0. The scores of the three dimensions are multiplied or added together to obtain a comprehensive score. The comprehensive score is mapped to the fault handling level based on the preset level threshold, which is usually divided into several levels such as Level 1 (highest), Level 2 (high), Level 3 (medium), and Level 4 (low).

[0091] Based on the determined fault handling level and identified sudden fault type, intelligent matching is performed from a pre-set emergency solution library to obtain target emergency handling measures. The emergency solution library is a knowledge base pre-established by the enterprise based on historical fault handling experience and production resource allocation. This library structurally stores the mapping relationship between various fault scenarios and corresponding handling solutions. Target emergency handling measures include, but are not limited to, activating backup production lines, adjusting the operating parameters of existing production lines, or outsourcing processing. Multiple strategy options can also be set according to different actual production scenarios. Activating backup production lines is one of the common emergency measures. When the main production line cannot continue operating due to a fault and the enterprise has backup or redundant production lines, production tasks are quickly transferred to the backup production line for execution. Adjusting the operating parameters of existing production lines is another important emergency measure, applicable to scenarios where the fault affects equipment performance but does not completely lose production capacity. By reducing processing speed, increasing the number of process cycles, and optimizing toolpaths, the damaged equipment can continue to operate in degraded mode, or the capacity loss of the faulty equipment can be compensated by increasing the operating speed of other process equipment. Outsourcing processing is an external resource mobilization measure to address capacity gaps. When the internal production line cannot complete the task within the specified time, some or all processing processes are outsourced to external partners.

[0092] After determining the target emergency response measures, immediately initiate the calculation process for the impact of these measures on the power load. For measures involving the activation of backup production lines, query the equipment configuration list and rated power of each piece of equipment on the backup production line, and calculate the total installed power after the backup line is put into operation. Analyze the product types and production cycle times that the backup line needs to process, and generate time-by-time load curves for the backup line based on energy consumption curve templates in the product process characteristic database. For measures involving adjustments to the operating parameters of existing production lines, calculate the load changes based on the parameter adjustment scheme. For measures involving outsourced processing, the calculated load changes mainly reflect a reduction in internal production load. After completing the calculation of the expected power load changes, generate an emergency load curve starting from the time of the fault occurrence. The starting timestamp of the emergency load curve is the exact moment the fault was detected, and the time span of the curve is determined according to the expected duration of the emergency response measures.

[0093] Based on the time of the failure, the emergency load curve is superimposed or replaced with the final load forecast curve to obtain a real-time corrected emergency load forecast curve. The specific correction method depends on the degree of impact of the emergency measures on the original production plan, and the most appropriate correction method is selected. When the emergency measures add extra load to the original production plan, the superposition correction method is used, accumulating the load value of the emergency load curve to the corresponding position of the final load forecast curve point by point in time, realizing incremental load superposition. When the emergency measures replace the original production plan, the replacement correction method is used. First, the load component contributed by the failed production line in the final load forecast curve is identified. The total load curve is split into the failed equipment load component and other load components through a load decomposition algorithm. Then, the failed equipment load component is replaced with the emergency load curve. The replacement result is recombine with other load components to generate the corrected emergency load forecast curve.

[0094] This application also provides a load forecasting system based on user order demand. Figure 2 This is a schematic diagram of a load forecasting system based on user order demand provided in an embodiment of this application. (Refer to...) Figure 2 The system includes an acquisition unit 201, a calculation unit 202, a processing unit 203, and a correction unit 204. The acquisition unit 201 acquires the target enterprise's production plan data, historical production logs, and historical power load data; calculates the standard production load of a single production line producing a single product based on the historical production logs and historical power load data, and calculates the baseline non-production load of all production lines during non-production periods; The calculation unit 202 calculates the difference between the standard production load and the baseline non-production load to obtain the net increase average load corresponding to a single product, and averages the net increase average load of multiple historical production tasks belonging to the same product type to obtain the characteristic operating power value corresponding to each product type. Processing unit 203 obtains the planned start and end times of the target production task from the production plan data, and generates a baseline load forecast curve for the target production task based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production task; merges the baseline load forecast curve and historical power load data to obtain input features, and inputs the input features into the time series forecast model to obtain the initial load forecast curve within the forecast period; extracts the baseline operating efficiency and cumulative operating time of the production equipment from the historical production log, calculates the baseline operating efficiency based on the cumulative operating time, and obtains the efficiency correction factor; and determines the environmental correction factor based on the historical operating temperature and historical power load data in the historical production log. The correction unit 204 corrects the initial load forecast curve based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

[0095] In one possible implementation, the acquisition unit 201 is used to determine the target product type from the target production task and retrieve the start-up transition time and shutdown transition time of the target product type; the calculation unit 202 is used to divide the complete cycle of the target production task based on the planned start-up and end times, start-up transition time, and shutdown transition time to obtain the start-up transition period, the running period, and the shutdown transition period; the acquisition unit 201 is used to obtain the planned start time and planned end time from the planned start-up and end times, obtain multiple start-up time points based on the planned start time and start-up transition time, and use the ratio of the characteristic operating power value to the start-up transition time as the start-up slope; the calculation unit 202 is used to multiply the difference between each start-up time point and the planned start time by the start-up slope to obtain the start-up load value corresponding to each start-up time point, sort the multiple start-up load values ​​in chronological order to obtain the load sequence of the start-up transition period; add the planned start time and start-up transition time to obtain the start time of the running period, and terminate the planned... The difference between the start time and the shutdown transition duration is used as the end time of the running segment; multiple steady-state time points are determined based on the start and end times, and the characteristic operating power value is used as the steady-state load value corresponding to each steady-state time point. All steady-state load values ​​are sorted in chronological order to obtain the load sequence of the running segment; the difference between the planned termination time and the shutdown transition duration is used as the start time of the shutdown transition period, and multiple shutdown time points are generated based on the start time of the shutdown transition duration; the negative value of the characteristic operating power value is divided by the shutdown transition duration to obtain the shutdown slope, and the difference between each shutdown time point and the start time of the shutdown transition period is calculated to obtain the time span. The time span is multiplied by the shutdown slope and added to the characteristic operating power value to obtain the shutdown load value corresponding to each shutdown time point. All shutdown load values ​​are sorted in chronological order to obtain the load sequence of the shutdown transition period; the processing unit 203 is used to splice the load sequences of the start transition period, the running segment, and the shutdown transition period in chronological order to obtain the baseline load prediction curve.

[0096] In one possible implementation, processing unit 203 is used to determine a target time axis based on the time range of historical power load data and the time range of baseline load prediction curve, and to map the historical power load data and baseline load prediction curve onto the target time axis; acquisition unit 201 is used to select a target time point from the target time axis, extract multiple historical load values ​​within a preset historical window length before the target time point from the historical power load data, sort the multiple historical load values ​​in chronological order to obtain a historical load feature sequence, and the target time point is any time point on the target time axis; the time attribute information of the target time point is numerically encoded to obtain time attribute features; processing unit 203 is used to extract multiple baseline load prediction values ​​within a preset guidance window from the baseline load prediction curve based on the target time point, sort the multiple baseline load prediction values ​​in chronological order to obtain a baseline guidance feature sequence; the historical load feature sequence, time attribute features, and baseline guidance feature sequence corresponding to the target time point are concatenated and combined to obtain a high-dimensional feature vector; the high-dimensional feature vectors corresponding to all target time points within the prediction period are combined in chronological order to obtain input features.

[0097] In one possible implementation, the acquisition unit 201 is used to use the characteristic operating power value as the baseline net increase in power consumption of the product type under ideal equipment conditions, retrieve the most recent maintenance time point of the equipment from the historical production log, and determine the cumulative running time based on the current time point and the most recent maintenance time point; the processing unit 203 is used to mark the current time point as a historical efficiency anchor point when the cumulative running time is greater than the preset calibration periodicity; subtract the baseline non-production load from the actual average load of historical production tasks to obtain the actual net increase in power consumption of the anchor point corresponding to the historical efficiency anchor point; calculate the difference between the actual net increase in power consumption of the anchor points corresponding to two adjacent historical efficiency anchor points to obtain the power consumption change, and divide the power consumption change by the running time between the two historical efficiency anchor points to obtain the breakdown within the target operating range. The segmented power consumption increment gradient is used to determine the target operating interval, which is the interval corresponding to two adjacent historical efficiency anchor points in time. Multiple segmented power consumption increment gradients are sorted to obtain a historical power consumption increment gradient sequence. The most recent segmented power consumption increment gradient is selected from this sequence as the prediction baseline gradient. The average cumulative runtime of the target production task is calculated, and the difference between the average cumulative runtime and the target cumulative runtime at the most recent historical efficiency anchor point is calculated to obtain the prediction interval runtime. The prediction interval runtime is multiplied by the prediction baseline gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net increase in power consumption at the anchor point corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase in average load. The predicted net increase in average load is divided by the baseline net increase in power consumption to obtain the efficiency correction factor.

[0098] In one possible implementation, the acquisition unit 201 is used to extract ambient temperature data during all historical production tasks from historical production logs, and determine a temperature range covering all historical operating conditions based on multiple ambient temperature data; the calculation unit 202 is used to divide the temperature range to obtain multiple continuous non-overlapping operating temperature intervals, and select the operating temperature interval containing the most historical production tasks as the benchmark temperature interval; the difference between the standard production load corresponding to each historical production task and the benchmark non-production load is used as the actual net increase in power consumption, and the actual net increase in power consumption is divided by the characteristic operating power value of the historical production task to obtain the comprehensive power consumption skewness; the acquisition unit 201 is used to acquire the cumulative equipment runtime of each historical production task, and calculate the cumulative equipment runtime... The following steps are taken: First, a historical equipment efficiency correction factor is determined. Second, a calculation unit 202 uses the ratio of the overall power consumption skewness to the historical equipment efficiency correction factor as the environmental impact coefficient. Third, the environmental impact coefficient is linked to each historical production task, and the average operating temperature of all historical production tasks is summarized into each operating temperature range. Finally, all environmental impact coefficients within each operating temperature range are calculated to obtain the range-average environmental impact coefficient. Fourth, a processing unit 203 performs load forecasting for the target production task, obtains the predicted ambient temperature within the forecast period, compares the predicted ambient temperature with each operating temperature range, determines the range-average environmental impact coefficient corresponding to the target historical production task based on the comparison results, and uses the range-average environmental impact coefficient as the environmental correction factor.

[0099] In one possible implementation, the acquisition unit 201 is used to acquire the current forecast time point within the forecast period and acquire the initial forecast load value corresponding to the current forecast time point from the initial load forecast curve; retrieve the baseline load value corresponding to the current forecast time point from the baseline load forecast curve; the calculation unit 202 is used to multiply the efficiency correction factor and the environmental correction factor to obtain the comprehensive correction coefficient; multiply the baseline load value and the comprehensive correction coefficient to obtain the corrected net increase in production load; subtract the baseline load value from the initial forecast load value to obtain the non-production-related baseline load component, and add the corrected net increase in production load and the non-production-related baseline load component to obtain the final forecast load value corresponding to the current forecast time point; the correction unit 204 is used to combine the final forecast load values ​​corresponding to all forecast time points in chronological order to obtain the final load forecast curve.

[0100] In one possible implementation, the acquisition unit 201 is used to acquire the type and scope of the sudden failure when a sudden failure occurs during the actual execution of the target production task, and retrieve the planned delivery time of the target production task from the production plan data; the calculation unit 202 is used to calculate the interval between the planned delivery time and the current time, analyze the type and scope of the sudden failure based on the interval, and obtain the failure handling level; the processing unit 203 is used to match the failure handling level and the type of the sudden failure from the preset emergency plan library to obtain the target emergency handling measures; calculate the expected power load change generated by the execution of the target emergency handling measures, and generate an emergency load curve starting from the time of the failure; the correction unit 204 is used to superimpose or replace the emergency load curve with the final load prediction curve based on the time of the failure to obtain the real-time corrected emergency load prediction curve.

[0101] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0102] This application also discloses an electronic device. (See reference...) Figure 3 , Figure 3 This application provides a schematic diagram of the structure of an electronic device. The electronic device 300 may include: at least one processor 301, at least one network interface 304, a user interface 303, a memory 302, and at least one communication bus 305.

[0103] The communication bus 305 is used to enable communication between these components.

[0104] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0105] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0106] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 302, and by calling data stored in memory 302. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and application requests; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0107] The memory 302 may include random access memory (RAM) or read-only memory. Optionally, the memory 302 may include a non-transitory computer-readable storage medium. The memory 302 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 302 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), instructions for implementing the various method embodiments described above, etc. The data storage area may store data involved in the various method embodiments described above. Optionally, the memory 302 may also be at least one storage device located remotely from the aforementioned processor 301.

[0108] like Figure 3 As shown, the memory 302, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application for load forecasting based on user order demand.

[0109] exist Figure 3In the electronic device 300 shown, the user interface 303 is mainly used to provide an interface for users to input data and obtain user input data; while the processor 301 can be used to call the application stored in the memory 302 based on the load forecast of user order demand. When executed by one or more processors, the electronic device performs one or more of the methods described in the above embodiments.

[0110] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0111] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some service interfaces; indirect couplings or communication connections between devices or units may be electrical or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0115] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0116] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and practical application disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure.

Claims

1. A load forecasting method based on user order demand, characterized in that, The method includes: Obtain the target company's production plan data, historical production logs, and historical power load data; calculate the standard production load for a single production line producing a single product based on the historical production logs and historical power load data, and calculate the baseline non-production load for all production lines during non-production periods; The difference between the standard production load and the baseline non-production load is calculated to obtain the net increase average load corresponding to the single product. The net increase average load of multiple historical production tasks belonging to the same product type is averaged to obtain the characteristic operating power value corresponding to each product type. The planned start and end times of the target production task are obtained from the production plan data. Based on the planned start and end times and the characteristic operating power value corresponding to the target production task, a baseline load prediction curve for the target production task is generated. The baseline load forecast curve and the historical power load data are merged to obtain input features, which are then input into the time series forecast model to obtain the initial load forecast curve for the forecast period. The baseline operating efficiency and cumulative operating time of the production equipment are extracted from the historical production logs. The baseline operating efficiency is calculated based on the cumulative operating time to obtain the efficiency correction factor. Based on the historical operating temperature and historical power load data in the historical production logs, an environmental correction factor is determined. The initial load forecast curve is corrected based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

2. The method according to claim 1, characterized in that, The step of generating a baseline load forecast curve for the target production task based on the planned start and end times and the characteristic operating power values ​​corresponding to the target production task specifically includes: Determine the target product type from the target production task, and retrieve the start-up transition time and shutdown transition time of the target product type; The complete cycle of the target production task is divided based on the planned start and end times, the start transition duration, and the shutdown transition duration to obtain the start transition period, the running period, and the shutdown transition period; The plan start time and plan end time are obtained from the plan start and end times. Multiple start time points are obtained based on the plan start time and the start transition duration. The ratio of the characteristic operating power value to the start transition duration is used as the start slope. The difference between each of the start time points and the planned start time is multiplied by the start slope to obtain the start load value corresponding to each of the start time points. The multiple start load values ​​are sorted in chronological order to obtain the load sequence of the start transition period. The start time of the running segment is obtained by adding the planned start time to the startup transition duration, and the difference between the planned end time and the shutdown transition duration is taken as the end time of the running segment. Based on the start time and the end time, multiple steady-state time points are determined, and the characteristic operating power value is used as the steady-state load value corresponding to each steady-state time point. All the steady-state load values ​​are sorted in chronological order to obtain the load sequence of the operating period. The difference between the planned termination time and the downtime transition duration is used as the start time of the downtime transition period, and multiple downtime points are generated based on the start time of the downtime transition duration. Divide the negative value of the characteristic operating power value by the shutdown transition duration to obtain the shutdown slope. Calculate the difference between the start time of each shutdown time point and the shutdown transition period to obtain the time span. Multiply the time span by the shutdown slope and add it to the characteristic operating power value to obtain the shutdown load value corresponding to each shutdown time point. Sort each shutdown load value in chronological order to obtain the load sequence of the shutdown transition period. The load sequences of the startup transition period, the running period, and the shutdown transition period are spliced ​​together in chronological order to obtain the baseline load prediction curve.

3. The method according to claim 2, characterized in that, The process of merging the baseline load forecast curve and the historical power load data to obtain input features specifically includes: A target time axis is determined based on the time range of the historical power load data and the time range of the baseline load prediction curve, and the historical power load data and the baseline load prediction curve are mapped onto the target time axis. Select a target time point from the target time axis, extract multiple historical load values ​​within a preset historical window length before the target time point from the historical power load data, sort the multiple historical load values ​​in chronological order to obtain a historical load feature sequence, and the target time point is any time point on the target time axis; The time attribute information of the target time point is numerically encoded to obtain time attribute features; Based on the target time point, multiple baseline load prediction values ​​are extracted from the baseline load prediction curve within a preset guidance window. The multiple baseline load prediction values ​​are then sorted in chronological order to obtain a baseline guidance feature sequence. The historical load feature sequence, the time attribute feature, and the baseline guidance feature sequence corresponding to the target time point are concatenated and combined to obtain a high-dimensional feature vector. The high-dimensional feature vectors corresponding to all target time points within the prediction period are combined in chronological order to obtain the input features.

4. The method according to claim 1, characterized in that, The calculation of the baseline operating efficiency based on the cumulative running time to obtain the efficiency correction factor specifically includes: The characteristic operating power value is used as the baseline net increase in power consumption of the product type under ideal equipment conditions. The most recent maintenance time of the equipment is retrieved from the historical production log. The cumulative running time is determined based on the current time and the most recent maintenance time. When the cumulative runtime exceeds the preset calibration period, the current time point is marked as a historical efficiency anchor point; Subtracting the baseline non-production load from the actual average load of the historical production task yields the actual net increase in power consumption at the anchor point corresponding to the historical efficiency anchor point. The difference between the actual net power consumption of the anchor points corresponding to the two time-adjacent historical efficiency anchor points is calculated to obtain the power consumption change. The power consumption change is divided by the running time between the two historical efficiency anchor points to obtain the segmented power consumption increment gradient in the target running interval. The target running interval is the interval corresponding to the two time-adjacent historical efficiency anchor points. The multiple segmented power increment gradients are sorted to obtain a historical power increment gradient sequence, and the most recent segmented power increment gradient is selected from the historical power increment gradient sequence as the prediction reference gradient. Calculate the average cumulative runtime of the target production task, and calculate the difference between the average cumulative runtime and the target cumulative runtime when it was most recently marked as the historical efficiency anchor point to obtain the predicted interval runtime. Multiply the predicted interval runtime by the predicted baseline gradient to obtain the predicted power consumption increment. The predicted power consumption increment is added to the actual net power consumption increase at the anchor point corresponding to the most recent historical efficiency anchor point to obtain the predicted net increase average load. The predicted net increase average load is then divided by the baseline net increase power consumption to obtain the efficiency correction factor.

5. The method according to claim 1, characterized in that, The determination of the environmental correction factor based on the historical operating temperature and historical power load data in the historical production logs specifically includes: Extract ambient temperature data from all historical production tasks from the historical production logs, and determine a temperature range covering all historical operating conditions based on multiple ambient temperature data. The temperature range is divided into multiple continuous and non-overlapping operating temperature intervals, and the operating temperature interval containing the most historical production tasks is selected as the benchmark temperature interval. The difference between the standard production load and the baseline non-production load corresponding to each historical production task is taken as the actual net increase in power consumption, and the actual net increase in power consumption is divided by the characteristic operating power value of the historical production task to obtain the comprehensive power consumption skewness. Obtain the cumulative equipment runtime for each of the historical production tasks, and determine the historical equipment efficiency correction factor based on the cumulative equipment runtime. The ratio of the overall power consumption skewness to the historical equipment efficiency correction factor is used as the environmental impact coefficient; The environmental impact coefficient is bound to each of the historical production tasks, and the average operating temperature of all the historical production tasks is summarized into each operating temperature range. The environmental impact coefficients contained in each operating temperature range are calculated to obtain the range average environmental impact coefficient. Load forecasting is performed on the target production task, the predicted ambient temperature within the forecast period is obtained, the predicted ambient temperature is compared with each of the operating temperature ranges, the interval average environmental impact coefficient corresponding to the target historical production task is determined based on the comparison results, and the interval environmental impact coefficient is used as the environmental correction factor.

6. The method according to claim 1, characterized in that, The step of correcting the initial load forecast curve based on the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve specifically includes: Obtain the current forecast time point within the forecast period, and obtain the initial forecast load value corresponding to the current forecast time point from the initial load forecast curve; Retrieve the baseline load value corresponding to the current forecast time point from the baseline load forecast curve; Multiply the efficiency correction factor by the environmental correction factor to obtain the comprehensive correction coefficient; Multiply the baseline load value by the comprehensive correction factor to obtain the corrected net increase in production load; Subtract the baseline load value from the initial predicted load value to obtain the non-production-related baseline load component, and add the corrected net increase in production load to the non-production-related baseline load component to obtain the final predicted load value corresponding to the current prediction time point. The final predicted load values ​​corresponding to all predicted time points are combined in chronological order to obtain the final load prediction curve.

7. The method according to claim 6, characterized in that, After correcting the initial load forecast curve according to the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve, the method further includes: When a sudden failure occurs during the actual execution of the target production task, the type and scope of the sudden failure are obtained, and the planned delivery time of the target production task is retrieved from the production plan data. Calculate the interval between the planned delivery time and the current time, analyze the type and impact of the sudden failure based on the interval, and obtain the failure handling level; Based on the fault handling level and the type of the sudden fault, a target emergency handling measure is obtained by matching from a preset emergency plan library. Calculate the expected power load change resulting from the implementation of the target emergency response measures, and generate an emergency load curve starting from the time of the fault occurrence; Based on the time point of the fault occurrence, the emergency load curve is superimposed or replaced with the final load prediction curve to obtain the real-time corrected emergency load prediction curve.

8. A load forecasting system based on user order demand, characterized in that, The system includes an acquisition unit, a calculation unit, a processing unit, and a correction unit. The acquisition unit acquires the target enterprise's production plan data, historical production logs, and historical power load data; calculates the standard production load for a single production line producing a single product based on the historical production logs and the historical power load data, and calculates the baseline non-production load for all production lines during non-production periods. The calculation unit calculates the difference between the standard production load and the benchmark non-production load to obtain the net increase average load corresponding to the single product, and averages the net increase average load of multiple historical production tasks belonging to the same product type to obtain the characteristic operating power value corresponding to each product type. The processing unit obtains the planned start and end times of the target production task from the production plan data, generates a baseline load forecast curve for the target production task based on the planned start and end times and the characteristic operating power value corresponding to the target production task, merges the baseline load forecast curve with the historical power load data to obtain input features, and inputs the input features into the time series forecast model to obtain the initial load forecast curve within the forecast period. The baseline operating efficiency and cumulative operating time of the production equipment are extracted from the historical production logs. The baseline operating efficiency is calculated based on the cumulative operating time to obtain an efficiency correction factor. An environmental correction factor is determined based on the historical operating temperature and historical power load data in the historical production logs. The correction unit corrects the initial load forecast curve according to the efficiency correction factor and the environmental correction factor to obtain the final load forecast curve.

9. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images