Search

CN-122000874-A - Order-driven power load prediction method, equipment, medium and product

CN122000874ACN 122000874 ACN122000874 ACN 122000874ACN-122000874-A

Abstract

The application provides an order-driven power load prediction method, equipment, medium and product, wherein the method comprises the following steps of obtaining historical power load data and production business data corresponding to a target period, wherein the production business data comprises an order bill of materials and a workshop scheduling plan of a manufacturing execution system; the method comprises the steps of constructing a pre-scheduling production characteristic sequence aligned with historical load time according to a working procedure operation time window and equipment energy consumption standard, generating a time sequence input tensor through multidimensional splicing and inputting the time sequence input tensor into a cyclic neural network model to obtain an initial predicted value, monitoring the time deviation between an actual production progress and a scheduling plan in real time, correcting the characteristic sequence based on the actual progress when the deviation exceeds a threshold value, and inputting the model again to output a corrected predicted value. By implementing the technical scheme provided by the application, the dynamic coupling of the load prediction and the production rhythm is realized, and the prediction precision is obviously improved.

Inventors

  • PAN YINGCHAO
  • DUAN XIAOHAN

Assignees

  • 北京如实智慧电力科技有限公司

Dates

Publication Date
20260508
Application Date
20260115

Claims (10)

  1. 1. An order-driven power load prediction method, the method comprising: acquiring historical power load data before a target prediction period, and production business data corresponding to the target prediction period, wherein the production business data comprises an order bill of materials and a workshop scheduling plan from a manufacturing execution system; Constructing a pre-arrangement generation characteristic sequence aligned with the historical power load data in time resolution according to operation time windows of all working procedures in the workshop arrangement plan and equipment energy consumption references corresponding to the order bill of materials; performing multidimensional characteristic splicing on the historical power load data and the pre-arrangement generated characteristic sequence to generate a time sequence input tensor; Inputting the time sequence input tensor into a pre-trained cyclic neural network prediction model to obtain an initial power load predicted value of the target prediction period; Real-time monitoring the actual production progress of the manufacturing execution system, and calculating the time deviation value of the actual production progress and the workshop scheduling plan; When the time deviation value exceeds a preset adjustment threshold value, correcting the pre-arrangement production characteristic sequence based on the actual production progress, inputting the corrected pre-arrangement production characteristic sequence into the cyclic neural network prediction model again, and outputting a corrected power load predicted value.
  2. 2. The method according to claim 1, wherein the constructing a pre-arrangement production feature sequence aligned with the historical power load data in time resolution according to the operation time window of each process in the shop scheduling plan and the equipment energy consumption reference corresponding to the order bill of materials specifically comprises: analyzing material coding information contained in the order bill of materials, and inquiring a production equipment identifier associated with the material coding information and rated power parameters corresponding to the production equipment identifier in a preset equipment energy consumption database; Analyzing the workshop scheduling plan, and extracting a start-stop time point and an operation duration of the production equipment identifier corresponding to each procedure; generating a theoretical energy consumption curve according to the rated power parameter, the starting and stopping time point and the operation duration; and resampling and interpolating the theoretical energy consumption curve according to the time sampling frequency of the historical power load data to generate the pre-arrangement generation characteristic sequence.
  3. 3. The method according to claim 1, wherein said multi-dimensional feature stitching of said historical electrical load data and said pre-rank production feature sequences generates a time-sequential input tensor, in particular comprising: Detecting and eliminating abnormal values of the historical power load data, and filling the generated data gaps by using a Lagrange interpolation method; Mapping the filled historical power load data and the pre-arrangement generated characteristic sequence to a preset numerical interval by adopting a maximum and minimum normalization algorithm; Extracting calendar characteristic data corresponding to the target prediction period, wherein the calendar characteristic data at least comprises a working day identifier, a holiday identifier and a season index identifier; And carrying out parallel splicing on the normalized historical power load data, the normalized pre-arrangement generated characteristic sequence and the calendar characteristic data in a characteristic dimension to construct the time sequence input tensor forming a three-dimensional structure.
  4. 4. A method according to claim 3, wherein said inputting the time-series input tensor into a pre-trained recurrent neural network prediction model to obtain an initial power load prediction value for the target prediction period, comprises: Transmitting the time sequence input tensor into an input layer of the cyclic neural network prediction model, and converting the time sequence input tensor into an implicit layer feature vector through the input layer; processing the hidden layer feature vector through a long-period memory network unit in the cyclic neural network prediction model, and outputting a hidden state vector; And inputting the hidden state vector to a full-connection layer of the cyclic neural network prediction model to perform linear transformation to obtain the initial power load predicted value corresponding to the time step of the target prediction period.
  5. 5. A method according to claim 3, characterized in that before said inputting of said time series input tensor into a pre-trained recurrent neural network prediction model, the method further comprises a training step of constructing said recurrent neural network prediction model, said training step comprising in particular: acquiring historical actual measurement load data, historical actual scheduling records and historical calendar characteristic data in a preset historical time period; carrying out characteristic processing on the history actual production record, and constructing a history production characteristic sequence aligned with the history actual measurement load data in the time dimension; splicing the history actual measurement load data, the history production characteristic sequence and the history calendar characteristic data in characteristic dimension to generate a history training sample set; And transmitting the historical training sample set into a to-be-trained cyclic neural network prediction model, calculating a loss function value between model output and the historical actual measurement load data, and updating weight parameters of the to-be-trained cyclic neural network prediction model by using an optimization algorithm until the loss function value meets a preset convergence condition to obtain the pre-trained cyclic neural network prediction model.
  6. 6. The method according to claim 5, wherein the characterizing the historical actual production record constructs a historical production signature sequence aligned with the historical measured load data in a time dimension, specifically comprising: removing abnormal record entries which are not aligned with the time stamp of the history actual measurement load data in the history actual production record; analyzing production equipment operation data contained in the historical actual production record, and generating a historical energy consumption curve according to the production equipment operation data and preset rated power parameters; resampling and interpolating the historical energy consumption curve according to the time sampling frequency of the historical actual measurement load data to generate the historical production characteristic sequence; And mapping the historical production characteristic sequence to a preset numerical interval consistent with the historical actual measurement load data by using a maximum and minimum normalization algorithm.
  7. 7. The method according to claim 1, wherein when the time deviation value exceeds a preset adjustment threshold, the pre-ranking production feature sequence is modified based on the actual production schedule, specifically comprising: Determining the delay time length or the advance time length of the actual production progress relative to the workshop scheduling plan according to the positive and negative attributes of the time deviation value; positioning a characteristic data segment corresponding to an unexecuted procedure after the current moment in the pre-arrangement generating characteristic sequence; translating the characteristic data segment along the time axis direction according to the delay time length or the advance time length to generate a translated intermediate characteristic sequence; And carrying out zero-value filling on a vacant time window generated by translation of the intermediate characteristic sequence or carrying out average value filling according to characteristic values of adjacent time steps to obtain the corrected pre-arranged generated characteristic sequence.
  8. 8. An electronic device comprising a processor and a memory; the memory is for storing computer program code comprising computer instructions that the processor invokes to cause the electronic device to perform the method of any of claims 1-7.
  9. 9. A computer readable storage medium storing computer instructions which, when run on an electronic device, cause the electronic device to perform the method of any one of claims 1-7.
  10. 10. A computer program product, characterized in that the computer program product, when run on an electronic device, causes the electronic device to perform the method of any of claims 1-7.

Description

Order-driven power load prediction method, equipment, medium and product Technical Field The application relates to the technical field of industrial energy management, in particular to an order-driven power load prediction method, equipment, medium and product. Background How to predict the power load in the industrial production process with high precision, thereby accurately guiding the production scheduling and electricity purchasing decision, has become the core requirement for guaranteeing the power supply and demand balance of the power grid and reducing the operation cost of enterprises. In the prior art, the prediction function of industrial power load is generally realized by a pure time series analysis method based on historical power consumption data. Future electricity usage trends are deduced, for example, using historical contemporaneous average load values or statistical models trained based on historical data. However, in the prior art, deduction is mainly performed by relying on the statistical rule of historical load data, and real-time order details and production schedules in enterprises are not considered, so that a load prediction model and an actual production energy rhythm are in a splitting state. When facing to a production plan frequently changed by a discrete manufacturing enterprise, such a model cannot correct a prediction result in real time by sensing dynamic changes of a service side, so that a prediction curve is difficult to accurately follow actual load fluctuation driven by a specific order, and further, when the enterprise performs fine energy scheduling or participates in electric power spot market transaction, the risk of high-volume deviation assessment and economic loss is caused due to insufficient prediction accuracy. Disclosure of Invention In view of the above, the present application provides an order-driven power load prediction method, apparatus, medium and product to solve the above-mentioned problems. In a first aspect, there is provided an order-driven based power load prediction method, the method comprising: Acquiring historical power load data before a target prediction period and production service data corresponding to the target prediction period, wherein the production service data comprises an order bill of materials and a workshop scheduling plan from a manufacturing execution system; According to the operation time window of each procedure in the workshop scheduling plan and the equipment energy consumption standard corresponding to the order bill of materials, constructing a pre-scheduling production characteristic sequence aligned with the historical power load data in time resolution; performing multidimensional characteristic splicing on the historical power load data and the pre-arrangement generated characteristic sequence to generate a time sequence input tensor; Inputting the time sequence input tensor into a pre-trained cyclic neural network prediction model to obtain an initial power load predicted value of a target predicted period; real-time monitoring the actual production progress of the manufacturing execution system, and calculating the time deviation value of the actual production progress and the workshop scheduling plan; when the time deviation value exceeds a preset adjustment threshold value, correcting the pre-arrangement generated characteristic sequence based on the actual production progress, inputting the corrected pre-arrangement generated characteristic sequence into the cyclic neural network prediction model again, and outputting a corrected power load predicted value. According to the technical scheme, the order bill of materials and the workshop scheduling plan are combined with the historical power load data, so that the planeness and the instantaneity of the production business can be fully considered in the load prediction, and the actual influence of the production activity on the energy consumption can be dynamically reflected by the prediction result. When the actual production progress deviates from the plan, the load is predicted again by correcting the production characteristic sequence generated by the pre-arrangement, so that the prediction error caused by the production deviation can be effectively eliminated, and the real-time accuracy and the adaptability of the power load prediction are improved. Optionally, constructing a pre-scheduling production feature sequence aligned with the historical power load data in time resolution according to the operation time window of each procedure in the workshop scheduling plan and the equipment energy consumption standard corresponding to the order bill of materials, which specifically includes: analyzing material coding information contained in an order bill of materials, and inquiring a production equipment identifier associated with the material coding information and rated power parameters corresponding to the production equipment identifier in a preset equip