A method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production

Abstract

This invention discloses a method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production. The method includes: collecting time-series energy consumption data from the discrete manufacturing enterprise and constructing a dataset to obtain the energy consumption curve; rearranging the 1D time-series energy consumption data into a 2D matrix using a preset calculation method; processing the 2D matrix using a preset convolution module, performing a weighted summation calculation on the data after the convolution, stacking the convolution modules and using residual connections; and inputting the output of the convolution modules into a fully connected layer to generate a prediction result for the next time step, thereby achieving energy consumption prediction for the discrete manufacturing enterprise. This invention employs a time-series periodic information mining method to identify time-series changes in energy consumption caused by production and equipment dynamics in discrete manufacturing enterprises, thereby improving the accuracy of energy consumption prediction for discrete manufacturing enterprises and achieving accurate prediction of energy consumption for discrete manufacturing enterprises with large energy consumption fluctuations and insignificant consumption patterns.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of industrial big data analysis and industrial energy management, and in particular to a method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production. Background Technology

[0002] Energy consumption forecasting for discrete manufacturing enterprises is crucial for understanding future energy demands in production processes and achieving a balanced energy supply and demand. Especially in situations of energy shortages, accurate energy consumption forecasting helps enterprises formulate reasonable energy supply plans, prioritizing the energy needs of critical production activities, thereby reducing manufacturing energy costs while promoting green and sustainable development. However, due to the highly variable production tasks in discrete manufacturing processes, energy consumption in discrete manufacturing enterprises fluctuates significantly, making accurate forecasting difficult. How to accurately utilize the large amount of production energy consumption data generated during the production process to achieve precise energy consumption forecasting for the entire discrete manufacturing enterprise park has become a pressing issue in the field of industrial energy conservation and carbon reduction.

[0003] Early research on energy consumption prediction for discrete manufacturing enterprises focused on modeling energy consumption mechanisms across multiple levels, including equipment, process, workshop, and enterprise, leading to mechanism-driven prediction methods. Because these methods require constructing multi-variable correlations between production information and the operating mechanisms of multiple devices, they are primarily used in process or semi-process manufacturing scenarios with fixed production patterns and stable production conditions. However, they still struggle to meet the demands of discrete manufacturing enterprises with complex manufacturing processes and frequently changing production plans for rapid, low-cost prediction.

[0004] With the development of industrial artificial intelligence technology, enterprises can monitor and collect time-series energy consumption data during the production process by deploying smart devices and the Industrial Internet. Based on the collected time-series energy consumption data, researchers have proposed a data-driven method for modeling and predicting the energy consumption system of discrete manufacturing enterprises, using methods such as random forests and support vector machines for time-series prediction. Traditional machine learning-based methods have good predictive performance for stable time-series energy consumption data, but they cannot meet the requirements for predicting large variations in energy consumption data.

[0005] In summary, accurate prediction of energy consumption for discrete manufacturing enterprises, which have large fluctuations and indistinct consumption patterns, still faces the following technical challenges: the time-series energy consumption data of discrete manufacturing enterprises spans a long period of time and has a large data volume, requiring high stability of the prediction model; frequent changes in production tasks and plans cause the time-series energy consumption data to exhibit complex periodicity and random fluctuations, making feature extraction and trend identification difficult. Summary of the Invention

[0006] This invention provides an energy consumption prediction method for discrete manufacturing enterprises that adapts to dynamic production. It adopts a time series periodic information mining method to identify the time series changes in energy consumption caused by production and equipment changes in discrete manufacturing enterprises, so as to improve the accuracy of energy consumption prediction for discrete manufacturing enterprises and achieve accurate prediction of energy consumption for discrete manufacturing enterprises with large energy consumption fluctuations and insignificant consumption patterns.

[0007] This invention provides a method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production, including:

[0008] Collect time-series energy consumption data from discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curves of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data;

[0009] The 1D time-series energy consumption data is rearranged into a 2D matrix using a preset calculation method in order to capture the dynamic characteristics of short-term and long-term time-series energy consumption data.

[0010] The 2D matrix is ​​processed using a preset convolutional module, and the data after the data is weighted and summed. Then, the convolutional modules are stacked and residual connections are used to ensure that information is effectively transmitted and aggregated in the deep network.

[0011] The output of the convolution module is input into the fully connected layer to generate a prediction result for the next time step, thereby enabling the prediction of energy consumption of the discrete manufacturing enterprise.

[0012] Furthermore, the step of rearranging the 1D time-series energy consumption data into a 2D matrix using a preset calculation method to capture the dynamic characteristics of short-term and long-term time-series energy consumption data includes:

[0013] Fourier transform analysis is used to analyze the frequency domain of the time-series energy consumption data to identify the main periods in the energy consumption data. Then, by calculating the spectrum, the first k frequencies with amplitudes exceeding a set value are selected to determine the length of the corresponding period.

[0014] Based on the identified periodic features, the selected k original 1D time-series energy consumption data are rearranged into k 2D matrices to capture the dynamic features of short-term and long-term time-series energy consumption data respectively; where the columns of the matrix represent the energy consumption changes within a period, and the rows of the matrix represent the energy consumption changes at the same time point during different periods.

[0015] Furthermore, the step of using Fourier transform to analyze the frequency domain of the time-series energy consumption data to identify the main periods in the energy consumption data, and then selecting the first k frequencies with amplitudes exceeding a set value through spectrum calculation to determine the length of the corresponding period includes:

[0016] The formula for calculating the period of the time-series energy consumption data is as follows:

[0017] S = F ave (F amp (F FFT (X 1D )))

[0018]

[0019]

[0020] in, T×C represents enterprise time-series energy consumption data of length T and variable C; F FFT (·) indicates performing a Fourier transform on the time-series energy consumption data, converting the time-domain features into frequency-domain features; F amp (·) indicates the amplitude value obtained from the time-series energy consumption frequency domain curve; F ave (·) indicates averaging the amplitude values ​​obtained at different frequencies; argFirstk(·) indicates selecting the first k frequencies {f1,…f1,…f2} whose amplitude values ​​exceed the set value. k}, whose corresponding period length is {p1,…p k}, where k is a hyperparameter.

[0021] Furthermore, the step of rearranging the selected k original 1D time-series energy consumption data into k 2D matrices based on the identified periodic features, in order to capture the dynamic characteristics of short-term and long-term time-series energy consumption data respectively, includes:

[0022] Based on the selected frequencies {f1,…f k} and the corresponding period lengths {p1,…p k The 1D time-series energy consumption data is rearranged into a 2D matrix; where columns represent energy consumption changes within a period, and rows represent energy consumption changes at the same time point during different periods. The 2D matrix representation is shown in the following formula:

[0023]

[0024] Among them, F pad (·) indicates that the original time-series energy consumption data is filled with 0; F res (·) represents reconstructing 1D time-series energy consumption data into a 2D matrix based on periodic information, where f i p represents the frequency of the i-th frequency in the time-series energy consumption data. i Indicates the length of the period.

[0025] Further, the steps of processing the 2D matrix using a preset convolutional module, performing weighted summation on the data, stacking the convolutional modules, and using residual connections to ensure effective information transmission and aggregation in the deep network include:

[0026] The Inception module is used to convert the processed 2D matrix back to 1D form; the Inception module is a module that uses multi-scale 2D convolution kernels.

[0027] The calculated period amplitude is used as the weight, and the Softmax function is used to perform weighted summation on different 1D time series energy consumption data, which is then fused into the output of the current layer.

[0028] Convolutional modules are stacked and residual connections are used to ensure that information is effectively transmitted and aggregated in deep networks.

[0029] Furthermore, the step of using the Inception module to convert the processed 2D matrix back to 1D form includes:

[0030] The Inception module is used to perform convolution on the transformed 2D matrix to extract features, aggregating the changes in enterprise time-series energy consumption data at the same time point within one period and across different weeks:

[0031]

[0032] Among them, F inception (·) is a deep learning module that uses convolutional kernels of different sizes to process 2D matrices in parallel, capturing the short-term and long-term variation characteristics of time-series energy consumption data. It includes multi-scale periodicity and time-dependent features extracted from time-series energy consumption data.

[0033] Convert the 2D matrix after convolution to 1D and truncate the previously padded portion:

[0034]

[0035] Among them, F tru (·) indicates truncation, used to remove the time point used for filling, i.e., F pad Adding 0s in (·) restores the time-series energy consumption data to its original length; This indicates that the processed 2D matrix will be converted back into 1D time-series energy consumption data.

[0036] Furthermore, the step of using the calculated period amplitude as weights, employing the Softmax function to perform weighted summation on different 1D time-series energy consumption data, and fusing them into the output of the current layer includes:

[0037] The Softmax function is used to perform weighted fusion of different 1D time-series energy consumption data to achieve adaptive aggregation of the hierarchical convolution module:

[0038]

[0039]

[0040] Among them, F softmax (·) Convert the amplitude of each frequency into a weight, representing the importance of that frequency in the overall features; This means that multiple 1D features corresponding to different frequencies are aggregated by weighted summation to obtain the output.

[0041] Furthermore, in the step of stacking convolutional modules and using residual connections to ensure effective information transmission and aggregation in deep networks, the calculation formula is as follows:

[0042]

[0043] in, Φ(·) represents the output of the convolution module; This represents the output of the (l-1)th node.

[0044] The present invention also provides an energy consumption prediction device for discrete manufacturing enterprises adapted to dynamic production, comprising:

[0045] The data acquisition module is used to collect time-series energy consumption data of discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curve of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data;

[0046] The arrangement module is used to rearrange the 1D time-series energy consumption data into a 2D matrix using a preset calculation method, so as to capture the dynamic characteristics of short-term and long-term time-series energy consumption data.

[0047] The calculation module processes the 2D matrix using a preset convolution module, performs weighted summation on the data, and then stacks the convolution modules, using residual connections to ensure effective transmission and aggregation of information in the deep network.

[0048] The generation module is used to input the output of the convolution module into the fully connected layer to generate a prediction result for the next time step, thereby enabling the prediction of energy consumption of the discrete manufacturing enterprise.

[0049] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0050] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0051] The beneficial effects of this invention are as follows:

[0052] 1. This invention converts one-dimensional time-series energy consumption data into a two-dimensional matrix containing periodic information of multiple time periods. It can effectively identify the periodic characteristics in enterprise time-series consumption data, separate local fluctuations from global trends, thereby overcoming the characteristic of large fluctuations in the overall energy consumption of discrete manufacturing enterprises and achieving accurate prediction of energy consumption.

[0053] 2. This invention fully considers the dynamic nature of production tasks and plans of discrete manufacturing enterprises, and adopts a two-dimensional convolution method, which enables the model to simultaneously handle complex relationships within and between periods, improves feature extraction capabilities, and is suitable for achieving accurate prediction of the energy consumption of discrete manufacturing enterprises at the next moment. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the energy consumption prediction method for discrete manufacturing enterprises adapted to dynamic production, as described in this invention.

[0055] Figure 2 This is a schematic diagram of the energy consumption prediction device for discrete manufacturing enterprises adapted to dynamic production, as per the present invention.

[0056] Figure 3 This is a schematic diagram of the internal structure of the computer device in this invention.

[0057] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0058] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] like Figure 1 As shown, this invention provides a method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production, including:

[0060] S1. Collect time-series energy consumption data of discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curve of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data;

[0061] S2. The 1D time-series energy consumption data is rearranged into a 2D matrix using a preset calculation method to capture the dynamic characteristics of short-term and long-term time-series energy consumption data; specifically including:

[0062] S21. In order to process the time-series energy consumption data of discrete manufacturing enterprises, Fourier transform is used to analyze the frequency domain of the time-series energy consumption data to identify the main period in the energy consumption data. By analyzing the spectrum, the first k frequencies with amplitudes exceeding the set value are selected to determine the length of the corresponding period.

[0063] The formula for calculating the period of the time-series energy consumption data is as follows:

[0064] S = F ave (F amp (F FFT (X 1D )))

[0065]

[0066]

[0067] in, T×C represents enterprise time-series energy consumption data of length T and variable C; F FFT (·) indicates performing a Fourier transform on the time-series energy consumption data, converting the time-domain features into frequency-domain features; F amp (·) indicates the amplitude value obtained from the time-series energy consumption frequency domain curve; F ave (·) indicates averaging the amplitude values ​​obtained at different frequencies; argFirstk(·) indicates selecting the first k frequencies {f1,…f1,…f2} whose amplitude values ​​exceed the set value. k The period lengths corresponding to these frequencies are {p1,…p}. k}, where k is a hyperparameter. The three formulas mentioned above can be simplified as follows:

[0068] S,{f1,…f k},{p1,…p k}=F(X 1D )

[0069] S22. Based on the identified periodic features, the selected k original 1D time-series energy consumption data are rearranged into k 2D matrices to capture the dynamic features of short-term and long-term time-series energy consumption data respectively; wherein, the columns of the matrix represent the energy consumption changes within a period, and the rows of the matrix represent the energy consumption changes at the same time point during different periods.

[0070] Based on the selected frequencies {f1,…f k} and the corresponding period lengths {p1,…p k The 1D time-series energy consumption data is rearranged into a 2D matrix; where columns represent energy consumption changes within a period, and rows represent energy consumption changes at the same time point during different periods. This 2D matrix representation better captures complex time-varying patterns, as shown in the following equation:

[0071]

[0072] Among them, F pad (·) indicates padding the original time-series energy consumption data with 0s to make its length adaptable to the dimensional requirements in the next reconstruction step, avoiding data loss or dimensional mismatch; F res (·) represents reconstructing 1D time-series energy consumption data into a 2D matrix based on periodic information, where f i p represents the frequency of the i-th frequency in the time-series energy consumption data. i Indicates the length of the period.

[0073] S3. The 2D matrix is ​​processed using a preset convolutional module, and the data after the data is weighted and summed. Then, the convolutional modules are stacked, and residual connections are used to ensure effective information transmission and aggregation in the deep network; specifically including:

[0074] The S31 Inception module processes time-series energy consumption data using multi-scale 2D convolutional kernels to capture changes across different periods and prevent short-term fluctuations from being mistaken for long-term trends. The Inception module then converts the processed 2D matrix back to 1D form for subsequent calculations.

[0075] The Inception module is used to perform convolution on the transformed 2D matrix to extract features, aggregating the changes in enterprise time-series energy consumption data at the same time point within one period and across different weeks:

[0076]

[0077] Among them, F inception (·) is a deep learning module that uses convolutional kernels of different sizes to process 2D matrices in parallel, capturing the short-term and long-term variation characteristics of time-series energy consumption data. It includes multi-scale periodicity and time-dependent features extracted from time-series energy consumption data.

[0078] Convert the 2D matrix after convolution to 1D and truncate the previously padded portion:

[0079]

[0080] Among them, F tru (·) indicates truncation, used to remove the time point used for filling, i.e., F pad Adding 0s in (·) restores the time-series energy consumption data to its original length; This indicates that the processed 2D matrix will be converted back into 1D time-series energy consumption data.

[0081] S32. Using the calculated period amplitude as weights, the Softmax function is used to perform weighted summation on different 1D time-series energy consumption data and merge them into the output of the current layer.

[0082] The Softmax function is used to perform weighted fusion of different 1D time-series energy consumption data to achieve adaptive aggregation of the hierarchical convolution module:

[0083]

[0084]

[0085] Among them, F softmax (·) Convert the amplitude of each frequency into a weight, representing the importance of that frequency in the overall features; This means that multiple 1D features corresponding to different frequencies are aggregated by weighted summation to obtain the output.

[0086] S33. Stack the convolutional modules and use residual connections to ensure effective information transfer and aggregation in deep networks. The calculation formula is as follows:

[0087]

[0088] in, Φ(·) represents the output of the convolution module; This represents the output of the (l-1)th node.

[0089] S4. Input the output of the convolution module into the fully connected layer to generate a prediction result for the next time step, thereby realizing the prediction of energy consumption of the discrete manufacturing enterprise.

[0090] The data used in this invention comes from a total of 31,395 power consumption values ​​from a discrete manufacturing enterprise between May 2023 and June 2024, with a data collection interval of 15 minutes. The experiment divides the dataset into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0091] The hyperparameters involved in the training and optimization process used in the experiment are shown in Table 1, the evaluation metrics used are shown in Table 2, and the experimental results are shown in Table 3.

[0092] Table 1. Training and optimization hyperparameters used in the experiment.

[0093]

[0094] Table 2 Evaluation indicators used in the experiment

[0095]

[0096] Table 3 Experimental Results

[0097]

[0098] The experimental results show that the method of this invention is superior to the Transformer model. The method of this invention considers overall dynamic variability better and can superimpose long-term information, effectively overcoming the influence of short-term dynamics.

[0099] This invention discloses a method for predicting energy consumption in discrete manufacturing enterprises that considers production dynamics. It involves acquiring historical time-series energy consumption data from discrete manufacturing enterprises, constructing a dataset, and fitting an overall energy consumption curve for the enterprise. Fourier transform analysis is used to analyze the frequency domain characteristics of the time-series energy consumption data, selecting multiple frequencies with large amplitudes to determine the period length. A convolution module is constructed to convert the one-dimensional time-series energy consumption data into a two-dimensional matrix, where columns represent the energy consumption changes within a period, and rows represent the energy consumption changes during the same time period in different periods. Convolution processing is applied to the two-dimensional matrix to aggregate features within a period and during the same time period within the same period. The two-dimensional matrix is ​​then converted back into a one-dimensional sequence for weighted fusion. Residual connections are performed on the outputs of multiple convolution modules to output the final result, thereby achieving energy consumption prediction for discrete manufacturing enterprises. This invention can achieve accurate prediction of energy consumption in discrete manufacturing enterprises, thereby reducing manufacturing energy costs while promoting green and sustainable development.

[0100] like Figure 2 As shown, the present invention also provides an energy consumption prediction device for discrete manufacturing enterprises adapted to dynamic production, comprising:

[0101] Acquisition module 1 is used to collect time-series energy consumption data of discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curve of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data;

[0102] Arrangement module 2 is used to rearrange the 1D time-series energy consumption data into a 2D matrix using a preset calculation method, so as to capture the dynamic characteristics of short-term and long-term time-series energy consumption data.

[0103] The calculation module 3 processes the 2D matrix using a preset convolution module, performs weighted summation on the data, and then stacks the convolution modules, using residual connections to ensure that information is effectively transmitted and aggregated in the deep network.

[0104] The generation module 4 is used to input the output of the convolution module into the fully connected layer to generate a prediction result for the next time step, thereby realizing the prediction of energy consumption of the discrete manufacturing enterprise.

[0105] In one embodiment, the arrangement module 2 includes:

[0106] The identification unit is used to analyze the frequency domain of the time-series energy consumption data using Fourier transform to identify the main period in the energy consumption data, and then, by calculating the spectrum, select the first k frequencies whose amplitudes exceed a set value to determine the length of the corresponding period.

[0107] The permutation unit is used to rearrange the selected k original 1D time-series energy consumption data into k 2D matrices based on the identified periodic features, so as to capture the dynamic features of short-term and long-term time-series energy consumption data respectively; wherein, the columns of the matrix represent the energy consumption change within a period, and the rows of the matrix represent the energy consumption change at the same time point during different periods.

[0108] In one embodiment, the identification unit includes:

[0109] The formula for calculating the period of the time-series energy consumption data is as follows:

[0110] S = F ave (F amp (F FFT (X 1D )))

[0111]

[0112]

[0113] in, T×C represents enterprise time-series energy consumption data of length T and variable C; F FFT (·) indicates performing a Fourier transform on the time-series energy consumption data, converting the time-domain features into frequency-domain features; F amp (·) indicates the amplitude value obtained from the time-series energy consumption frequency domain curve; F ave (·) indicates averaging the amplitude values ​​obtained at different frequencies; argFirstk(·) indicates selecting the first k frequencies {f1,…f1,…f2} whose amplitude values ​​exceed the set value. k}, whose corresponding period length is {p1,…p k}, where k is a hyperparameter.

[0114] In one embodiment, the arrangement unit includes:

[0115] Based on the selected frequencies {f1,…f k} and the corresponding period lengths {p1,…p k The 1D time-series energy consumption data is rearranged into a 2D matrix; where columns represent energy consumption changes within a period, and rows represent energy consumption changes at the same time point during different periods. The 2D matrix representation is shown in the following formula:

[0116]

[0117] Among them, F pad (·) indicates that the original time-series energy consumption data is filled with 0; F res (·) represents reconstructing 1D time-series energy consumption data into a 2D matrix based on periodic information, where f i p represents the frequency of the i-th frequency in the time-series energy consumption data. i Indicates the length of the period.

[0118] In one embodiment, the computing module 3 includes:

[0119] The transformation unit is used to convert the processed 2D matrix back to 1D form using the Inception module; wherein the Inception module is a module that uses multi-scale 2D convolution kernels;

[0120] The weighting unit is used to use the calculated period amplitude as weights, and the Softmax function is used to perform weighted summation on different 1D time series energy consumption data, which are then fused into the output of the current layer.

[0121] Stacked units are used to stack convolutional modules and employ residual connections to ensure that information is effectively transmitted and aggregated in deep networks.

[0122] In one embodiment, the conversion unit includes:

[0123] Extracting sub-units, which are used to perform convolution on the transformed 2D matrix using the Inception module to extract features, and aggregating the changes in enterprise time-series energy consumption data at the same time point within a period and across different weeks:

[0124]

[0125] Among them, F inception (·) is a deep learning module that uses convolutional kernels of different sizes to process 2D matrices in parallel, capturing the short-term and long-term variation characteristics of time-series energy consumption data. It includes multi-scale periodicity and time-dependent features extracted from time-series energy consumption data.

[0126] Transformation subunits are used to convert the 2D matrix after convolution into 1D and truncate the previously padded portion:

[0127]

[0128] Among them, F tru (·) indicates truncation, used to remove the time point used for filling, i.e., F pad Adding 0s in (·) restores the time-series energy consumption data to its original length; This indicates that the processed 2D matrix will be converted back into 1D time-series energy consumption data.

[0129] In one embodiment, the weighting unit includes:

[0130] The Softmax function is used to perform weighted fusion of different 1D time-series energy consumption data to achieve adaptive aggregation of the hierarchical convolution module:

[0131]

[0132]

[0133] Among them, F softmax (·) Convert the amplitude of each frequency into a weight, representing the importance of that frequency in the overall features; This means that multiple 1D features corresponding to different frequencies are aggregated by weighted summation to obtain the output.

[0134] In one embodiment, the calculation formula for the stacked unit is:

[0135]

[0136] in, Φ(·) represents the output of the convolution module; This represents the output of the (l-1)th node.

[0137] Each of the above modules, units, and sub-units is used to perform the respective steps in the above-mentioned method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production. The specific implementation methods are as described in the above-mentioned method embodiments, and will not be repeated here.

[0138] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores all data required for the process of a discrete manufacturing enterprise energy consumption prediction method adapted to dynamic production. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the discrete manufacturing enterprise energy consumption prediction method adapted to dynamic production.

[0139] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0140] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described methods for predicting energy consumption in discrete manufacturing enterprises that adapt to dynamic production.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media provided in this application and in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0142] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0143] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting energy consumption in discrete manufacturing enterprises that adapts to dynamic production, characterized in that, include: Collect time-series energy consumption data from discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curves of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data; The 1D time-series energy consumption data is rearranged into a 2D matrix using a preset calculation method to capture the dynamic characteristics of short-term and long-term time-series energy consumption data; specifically including: Fourier transform analysis is used to analyze the frequency domain of the time-series energy consumption data to identify the main periods in the energy consumption data. Then, by calculating the spectrum, the first k frequencies with amplitudes exceeding a set value are selected to determine the length of the corresponding period. Based on the identified periodic features, the selected k original 1D time-series energy consumption data are rearranged into k 2D matrices to capture the dynamic features of short-term and long-term time-series energy consumption data respectively; where the columns of the matrix represent the energy consumption changes within a period, and the rows of the matrix represent the energy consumption changes at the same time point during different periods. The 2D matrix is ​​processed using a pre-defined convolutional module, and the processed data is then weighted and summed. The convolutional modules are then stacked, and residual connections are used to ensure effective information transfer and aggregation within the deep network. Specifically, this includes: The Inception module is used to convert the processed 2D matrix back to 1D form; the Inception module is a module that uses multi-scale 2D convolution kernels. The calculated period amplitude is used as the weight, and the Softmax function is used to perform weighted summation on different 1D time series energy consumption data, which is then fused into the output of the current layer. Convolutional modules are stacked and residual connections are used to ensure that information is effectively transmitted and aggregated in deep networks. The output of the convolution module is input into the fully connected layer to generate a prediction result for the next time step, thereby enabling the prediction of energy consumption of the discrete manufacturing enterprise.

2. The energy consumption prediction method for discrete manufacturing enterprises adapting to dynamic production as described in claim 1, characterized in that, The step of using Fourier transform to analyze the frequency domain of the time-series energy consumption data to identify the main periods in the energy consumption data, and then selecting the first k frequencies with amplitudes exceeding a set value through spectrum calculation to determine the length of the corresponding period, includes: The formula for calculating the period of the time-series energy consumption data is as follows: S=F ave (F amp (F FFT (X 1D ))) in, T×C represents enterprise time-series energy consumption data of length T and variable C; F FFT (·) indicates performing a Fourier transform on the time-series energy consumption data, converting the time-domain features into frequency-domain features; F amp (·) indicates the amplitude value obtained from the time-series energy consumption frequency domain curve; F ave (·) indicates averaging the amplitude values ​​obtained at different frequencies; argFirstk(·) indicates selecting the first k frequencies {f1,…f1,…f2} whose amplitude values ​​exceed the set value. k }, whose corresponding period length is {p1,…p k }, where k is a hyperparameter.

3. The energy consumption prediction method for discrete manufacturing enterprises adapting to dynamic production as described in claim 2, characterized in that, The step of rearranging k selected original 1D time-series energy consumption data into k 2D matrices based on the identified periodic features, in order to capture the dynamic characteristics of short-term and long-term time-series energy consumption data respectively, includes: Based on the selected frequencies {f1,…f k } and the corresponding period lengths {p1,…p k The 1D time-series energy consumption data is rearranged into a 2D matrix; where columns represent energy consumption changes within a period, and rows represent energy consumption changes at the same time point during different periods. The 2D matrix representation is shown in the following formula: Among them, F pad (·) indicates that the original time-series energy consumption data is filled with 0; F res (·) represents reconstructing 1D time-series energy consumption data into a 2D matrix based on periodic information, where f i p represents the frequency of the i-th frequency in the time-series energy consumption data. i Indicates the length of the period.

4. The energy consumption prediction method for discrete manufacturing enterprises adapting to dynamic production as described in claim 3, characterized in that, The step of using the Inception module to convert the processed 2D matrix back to 1D form includes: The Inception module is used to perform convolution on the transformed 2D matrix to extract features, aggregating the changes in enterprise time-series energy consumption data at the same time point within one period and across different weeks: Among them, F inception (·) is a deep learning module that uses convolutional kernels of different sizes to process 2D matrices in parallel, capturing the short-term and long-term variation characteristics of time-series energy consumption data. It includes multi-scale periodicity and time-dependent features extracted from time-series energy consumption data; Convert the 2D matrix after convolution to 1D and truncate the previously padded portion: Among them, F tru (·) indicates truncation, used to remove the time point used for filling, i.e., F pad Adding 0s in (·) restores the time-series energy consumption data to its original length; This indicates that the processed 2D matrix will be converted back into 1D time-series energy consumption data.

5. The energy consumption prediction method for discrete manufacturing enterprises adapting to dynamic production as described in claim 4, characterized in that, The step of using the calculated period amplitude as weights, employing the Softmax function to perform weighted summation on different 1D time-series energy consumption data, and fusing them into the output of the current layer includes: The Softmax function is used to perform weighted fusion of different 1D time-series energy consumption data to achieve adaptive aggregation of the hierarchical convolution module: Among them, F softmax (·) Convert the amplitude of each frequency into a weight, representing the importance of that frequency in the overall features; This means that multiple 1D features corresponding to different frequencies are aggregated by weighted summation to obtain the output.

6. The energy consumption prediction method for discrete manufacturing enterprises adapting to dynamic production as described in claim 5, characterized in that, In the step of stacking convolutional modules and using residual connections to ensure effective information transmission and aggregation in deep networks, the calculation formula is as follows: in, Φ(·) represents the output of the convolution module; This represents the output of the (l-1)th node.

7. An energy consumption prediction device for discrete manufacturing enterprises adapted to dynamic production, characterized in that, The energy consumption prediction method for discrete manufacturing enterprises adapted to dynamic production, based on any one of claims 1-6, specifically includes the following apparatus: The data acquisition module is used to collect time-series energy consumption data of discrete manufacturing enterprises and construct a dataset to obtain the energy consumption curve of the discrete manufacturing enterprises; wherein, the time-series energy consumption data is 1D time-series energy consumption data; The arrangement module is used to rearrange the 1D time-series energy consumption data into a 2D matrix using a preset calculation method, so as to capture the dynamic characteristics of short-term and long-term time-series energy consumption data. The calculation module processes the 2D matrix using a preset convolution module, performs weighted summation on the processed data, and then stacks the convolution modules, using residual connections to ensure that information is effectively transmitted and aggregated in the deep network. The generation module is used to input the output of the convolution module into the fully connected layer to generate a prediction result for the next time step, thereby enabling the prediction of energy consumption of the discrete manufacturing enterprise.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

Images