Energy consumption prediction method and device, storage medium and program product

Abstract

The embodiment of the invention provides an energy consumption prediction method and device, a storage medium and a program product. The method comprises the following steps: performing feature extraction and screening on a time sequence of an industrial system by using a machine learning model to obtain a feature sequence; through a long-term feature learning model and a short-term feature learning model, respectively extracting a long-term dependency relationship and a global feature between the features and a short-term dependency relationship and a local feature between the features; performing nonlinear fitting on the feature sequence to obtain a third energy consumption value by utilizing a multi-layer perceptron model and combining a long-term dependency relationship, a short-term dependency relationship, a global feature and a local feature; and integrating the energy consumption values to obtain a target energy consumption value. Through the mode, machine learning and multiple neural network models are combined to perform feature extraction and processing by using the advantages of each model, and the target energy consumption value corresponding to the time sequence of the industrial system can be more accurately obtained based on the long-term and short-term dependency relationships and the global and local features.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to an energy consumption prediction method, device, storage medium and program product. Background Art

[0002] A large amount of energy is consumed in the industrial production process, and this energy consumption is closely related to the production cost. Therefore, machine learning models are usually used to predict energy consumption, and the energy consumption prediction results are used as the basis to drive production behavior, thereby controlling energy consumption and production costs. For example, in traditional energy consumption prediction solutions based on machine learning models, machine learning models such as support vector machines (SVM), decision trees, and random forests are usually used to predict energy consumption and obtain prediction results. Usually these methods rely on manually extracted features and use simple regression models to predict future energy consumption. However, the accuracy of these existing energy consumption prediction methods based on machine learning models is low. Summary of the invention

[0003] Multiple aspects of the present application provide an energy consumption prediction method, device, storage medium and program product for more accurately predicting a target energy consumption value corresponding to a time series of an industrial system.

[0004] The embodiment of the present application provides an energy consumption prediction method, comprising: obtaining a target time series formed by at least one energy state information and at least one production line state information of a target industrial system within a target time period, wherein the target time series is a multi-dimensional time series; inputting the target time series into a machine learning model based on a decision tree, performing feature extraction on the target time series, and obtaining a first feature sequence; using the decision tree, screening out a plurality of second features whose contribution to energy consumption prediction meets preset conditions from a plurality of first features of the first feature sequence, to obtain a second feature sequence; inputting the second feature sequence into a long-term feature learning model and a short-term feature learning model, respectively, to identify the global features of the second feature sequence and the plurality of second features through the long-term feature learning model; The long-term dependency relationship between the local features of the second feature sequence and the multiple second features is identified through the short-term feature learning model, and the second feature sequence is nonlinearly fitted according to the short-term dependency and the local features to obtain a second energy consumption value; the multi-layer perceptron model is used to perform nonlinear fitting on the second feature sequence according to the long-term dependency, the short-term dependency, the global features and the local features to obtain a third energy consumption value; and the target energy consumption value of the target industrial system in the target time period is calculated according to the first energy consumption value, the second energy consumption value and the third energy consumption value.

[0005] Further optionally, the decision tree-based machine learning model includes: a random forest model and a gradient boosting model; using the decision tree, multiple second features whose contribution to energy consumption prediction meets preset conditions are screened out from multiple first features of the first feature sequence to obtain a second feature sequence, including: using the first decision tree in the random forest model to screen out multiple candidate features whose contribution to energy consumption prediction meets the first preset condition from multiple first features of the first feature sequence; using the second decision tree in the gradient boosting model to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from the multiple candidate features to obtain the second feature sequence.

[0006] Further optionally, the second decision tree in the gradient boosting model is used to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from the multiple candidate features to obtain the second feature sequence, including: using the second decision tree in the gradient boosting model to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from the multiple candidate features; identifying feature association relationships between the multiple second features based on the node positions of the multiple second features in the second decision tree; generating the second feature sequence based on the feature association relationships between the multiple second features and the multiple second features.

[0007] Further optionally, the long-term feature learning model includes: a first convolutional layer and a first relationship recognition layer based on an attention mechanism; identifying the long-term dependency relationship between the global features of the second feature sequence and the multiple second features through the long-term feature learning model, including: using the first convolutional layer to convolve the second feature sequence to obtain the global features of the second feature sequence; inputting the second feature sequence into the first relationship recognition layer, and based on the attention mechanism, identifying the influence weight of each second feature in the second feature sequence on other second features; determining the long-term dependency relationship according to the influence weight of each second feature on other second features.

[0008] Further optionally, the short-time feature learning model includes: a second convolutional layer and a second relationship recognition layer; identifying the short-term dependency relationship between the local features of the second feature sequence and the multiple second features through the short-time feature learning model, including: using the second convolutional layer to convolve the second feature sequence to obtain the local features of the second feature sequence; inputting the second feature sequence into the second relationship recognition layer, and performing short-term dependency recognition on any adjacent second features in turn to obtain the short-term dependency relationship between the multiple second features.

[0009] Further optionally, the target energy consumption value of the target industrial system within the target time period is calculated based on the first energy consumption value, the second energy consumption value and the third energy consumption value, including: performing weighted summation of the first energy consumption value, the second energy consumption value and the third energy consumption value according to weight information corresponding to each energy consumption value to obtain the target energy consumption value; or, calculating the average value of the first energy consumption value, the second energy consumption value and the third energy consumption value as the target energy consumption value.

[0010] Further optionally, it also includes obtaining the actual energy consumption value corresponding to the target time series; generating an error measurement index based on the actual energy consumption value and the target energy consumption value; optimizing the parameters of the full-link model based on the error measurement index, wherein the full-link model includes: the machine learning model, the long-time feature learning model, the short-time feature learning model and the multi-layer perceptron model; the machine learning model is respectively connected to the long-time feature learning model and the short-time feature learning model; the long-time feature learning model and the short-time feature learning model are respectively connected to the multi-layer perceptron model.

[0011] An embodiment of the present application also provides an electronic device, comprising: a memory and a processor; wherein the memory is used to: store one or more computer instructions; the processor is used to execute the one or more computer instructions to: execute the steps in the energy consumption prediction method.

[0012] An embodiment of the present application also provides a computer-readable storage medium, which, when the computer program is executed by a processor, enables the processor to implement the steps in the energy consumption prediction method.

[0013] An embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is enabled to implement the steps in the energy consumption prediction method.

[0014] In this embodiment, a machine learning model is used to extract and screen the features of the time series of the industrial system to obtain a feature sequence; the long-term feature learning model and the short-term feature learning model are used to extract the long-term dependency and global features between features, as well as the short-term dependency and local features between features; the multi-layer perceptron model is used to combine the long-term dependency, short-term dependency, global features and local features to perform nonlinear fitting on the feature sequence to obtain a third energy consumption value; and the target energy consumption value is obtained by combining various energy consumption values. In this way, machine learning and multiple neural network models are combined to utilize the advantages of each model for feature extraction and processing, and based on long-term and short-term dependencies and global and local features, the target energy consumption value corresponding to the time series of the industrial system can be obtained more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0016] Figure 1 A schematic diagram of a flow chart of an energy consumption prediction method provided for an exemplary embodiment of the present application;

[0017] Figure 2 A flowchart of an energy consumption prediction method provided by an exemplary embodiment of the present application in an actual scenario;

[0018] Figure 3 A schematic diagram of the connection relationship between models provided for an exemplary embodiment of the present application;

[0019] Figure 4 A schematic diagram of an electronic device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0021] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation portals for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) are in compliance with relevant laws and standards.

[0022] A large amount of energy is consumed in the industrial production process, and this energy consumption is closely related to the production cost. Therefore, machine learning models are usually used to predict energy consumption, and the energy consumption prediction results are used as the basis to drive production behavior, thereby controlling energy consumption and production costs. For example, in traditional energy consumption prediction solutions based on machine learning models, machine learning models such as support vector machines (SVM), decision trees, and random forests are usually used to predict energy consumption and obtain prediction results. Usually these methods rely on manually extracted features and use simple regression models to predict future energy consumption. However, the accuracy of these existing energy consumption prediction methods based on machine learning models is low.

[0023] In response to the above technical problems, in an embodiment of the present application, machine learning and multiple neural network models can be combined to utilize the advantages of each model to perform feature extraction and corresponding feature processing. Based on long-term dependencies and short-term dependencies as well as global features and local features, the target energy consumption value of the target industrial system within the target time period can be more accurately predicted.

[0024] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0025] Figure 1 The energy consumption prediction method provided by the exemplary embodiment of the present application is as follows: Figure 1 As shown, the method may include the following steps:

[0026] Step 11: Obtain a target time series formed by at least one energy status information and at least one production line status information of a target industrial system within a target time period, where the target time series is a multi-dimensional time series.

[0027] Step 12: Input the target time series into a machine learning model based on a decision tree, extract features of the target time series, and obtain a first feature sequence; use the decision tree to screen out multiple second features whose contribution to energy consumption prediction meets preset conditions from the multiple first features of the first feature sequence to obtain a second feature sequence.

[0028] Step 13: input the second feature sequence into the long-term feature learning model and the short-term feature learning model respectively, so as to identify the long-term dependency between the global features of the second feature sequence and multiple second features through the long-term feature learning model, and perform nonlinear fitting on the second feature sequence according to the long-term dependency and the global features to obtain the first energy consumption value; and, identify the short-term dependency between the local features of the second feature sequence and multiple second features through the short-term feature learning model, and perform nonlinear fitting on the second feature sequence according to the short-term dependency and the local features to obtain the second energy consumption value.

[0029] Step 14: Using the multilayer perceptron model, perform nonlinear fitting on the second feature sequence according to the long-term dependency, short-term dependency, global features, and local features to obtain a third energy consumption value.

[0030] Step 15: Calculate a target energy consumption value of the target industrial system within a target time period according to the first energy consumption value, the second energy consumption value, and the third energy consumption value.

[0031] This embodiment can be executed by a server or any terminal device, and the terminal device can be a tablet computer, a mobile phone or a computer, etc., which is not limited in this embodiment. The target industrial system can be any industrial system, which can be composed of production equipment, automation systems, information systems, sensors and actuators, etc. Among them, the automation system can be a PLC (Programmable Logic Controller), a SCADA (Supervisory Control And Data Acquisition) system, etc., which can be used to automatically control the production process.

[0032] In this embodiment, the energy status information is used to describe the state of energy, and may include the state parameters of any energy such as water, electricity or steam. The state parameters may be flow, temperature or pressure, etc. The production line status information may be used to describe the working state of the production line in the target industrial system, and may include: the working state and output of the production line.

[0033] In this embodiment, a target time series formed by at least one type of energy status information and at least one type of production line status information of a target industrial system within a target time period may be obtained.

[0034] Among them, at least one energy status information and at least one production line status information of the target industrial system within the target time period can be obtained from the sensors, PLC and SCADA systems and other equipment in the target industrial system. Each energy status information and production line status information can have a corresponding timestamp. Therefore, according to the timestamps corresponding to the at least one energy status information and at least one production line status information within the target time period, the energy status information and production line status information from different data sources can be integrated into a common time axis, thereby forming a multi-dimensional time series, namely the target time series. Each dimension of the target time series can represent a type of information, such as water temperature, steam pressure, or the output of the production line, etc.

[0035] After the target time series is obtained, the target time series can be input into a machine learning model based on a decision tree to extract features of the target time series to obtain a first feature sequence. The first feature sequence can include multiple first features.

[0036] Among them, the machine learning model has been pre-trained, and a decision tree is constructed inside it. The decision tree has a tree structure, and the tree structure may include root nodes, internal nodes, branches and leaf nodes; the machine learning model can screen out a variety of second features whose contribution to energy consumption prediction meets preset conditions from a variety of first features in the first feature sequence based on the decision tree to obtain a second feature sequence. In other words, the pre-trained machine learning model can use the pre-constructed decision tree to determine the contribution of each first feature to energy consumption prediction. The higher the contribution of a feature to energy consumption prediction, the higher its value / importance; the lower the contribution of a feature to energy consumption prediction, the lower its value / importance. In this way, more important features can be screened more efficiently and features with lower importance can be removed, thereby achieving the purpose of reducing the number of features and reducing the amount of calculation of other models used in subsequent steps.

[0037] Among them, the preset condition can be set to any condition according to the actual design requirements. For example, 100 first features can be sorted according to the contribution of each first feature to the energy consumption prediction. Assuming that the preset condition is in the top 50% of the sorting results, 50 second features whose contribution to the energy consumption prediction meets the preset condition can be screened out from the multiple first features in the first feature sequence, and the second feature sequence can be obtained based on these 50 second features.

[0038] The above definition of feature sequences using "first" and "second" is only used to distinguish feature sequences that have not undergone feature screening from feature sequences that have undergone feature screening, and does not define other characteristics of feature sequences. The first feature refers to the feature in the first feature sequence, and the second feature refers to the feature in the second feature sequence.

[0039] In this embodiment, the above-mentioned machine learning model can be combined with a variety of deep learning models to achieve more accurate energy consumption prediction. Taking into account that when energy consumption prediction is based on the target time series, certain long-term and short-term laws often appear in the target time series. For example, long-term laws can be generated based on seasonal factors and production cycle change factors, which can reflect the long-term trend of energy consumption; short-term laws can be generated based on operating condition change factors, which can reflect the short-term trend of energy consumption. These long-term and short-term laws are closely related to energy consumption prediction. Based on this, in this embodiment, a variety of deep learning models may include long-term feature learning models, short-term feature learning models, and multi-layer perceptron models. This will be described in detail below.

[0040] The second feature sequence is input into the long-term feature learning model and the short-term feature learning model respectively, and then the global features of the second feature sequence and the long-term dependencies between multiple second features can be identified through the long-term feature learning model. Among them, the global feature refers to the feature extracted from the entire second feature sequence that can reflect the overall nature or trend of the second feature sequence. The long-term dependency between multiple second features can be used to describe the correlation or influence between data points that are far apart in the second feature sequence. Afterwards, the second feature sequence can be nonlinearly fitted according to the long-term dependency and the global features to obtain the first energy consumption value. Among them, the long-term dependency, the global features and the second feature sequence can be input into at least one fully connected layer to perform nonlinear fitting on the second feature sequence according to the long-term dependency and the global features to obtain the first energy consumption value.

[0041] Through the short-term feature learning model, the local features of the second feature sequence and the short-term dependencies between multiple second features can be identified, and the second feature sequence is nonlinearly fitted according to the short-term dependencies and local features to obtain the second energy consumption value. Among them, the local feature refers to the feature extracted from the entire second feature sequence that can reflect the local properties or trends of the second feature sequence. The short-term dependencies between multiple second features can be used to describe the correlation or influence between data points that are closely spaced in the second feature sequence. Afterwards, the second feature sequence can be nonlinearly fitted according to the short-term dependencies and local features to obtain the second energy consumption value. Among them, the short-term dependencies, local features and the second feature sequence can be input into at least one fully connected layer to perform nonlinear fitting on the second feature sequence according to the short-term dependencies and local features to obtain the second energy consumption value.

[0042] The above respectively utilizes the long-term feature learning model and the short-term feature learning model, and performs corresponding feature processing from the long-term global dimension and the short-term local dimension, respectively, to obtain the first energy consumption value and the second energy consumption value. It should be emphasized that the embodiment of the present application does not limit the input order of "inputting the second feature sequence into the long-term feature learning model and the short-term feature learning model" and the subsequent model execution order; specifically, the second feature sequence can be first input into the long-term feature learning model and processed accordingly to obtain the first energy consumption value, and then the second feature sequence can be input into the short-term feature learning model and processed accordingly to obtain the second energy consumption value; the second feature sequence can also be first input into the short-term feature learning model and processed accordingly to obtain the second energy consumption value, and then the second feature sequence can be input into the long-term feature learning model and processed accordingly to obtain the first energy consumption value; parallel input and parallel processing can also be adopted, that is, the second feature sequence is simultaneously input into the long-term feature learning model and the short-term feature learning model, and the long-term feature learning model is used to perform corresponding processing to obtain the first energy consumption value and the short-term feature learning model is used to perform corresponding processing to obtain the second energy consumption value.

[0043] Afterwards, a multilayer perceptron (MLP) model can be used to perform nonlinear fitting on the second feature sequence according to long-term dependencies, short-term dependencies, global features, and local features to obtain a third energy consumption value. The multilayer perceptron model may include multiple fully connected layers, and the multiple fully connected layers may perform more complex nonlinear mapping on the second feature sequence according to long-term dependencies, short-term dependencies, global features, and local features, so that the obtained third energy consumption value can better fit the complex change pattern in the target time series.

[0044] Based on the above steps, the target energy consumption value of the target industrial system in the target time period can be calculated according to the first energy consumption value, the second energy consumption value and the third energy consumption value. The target energy consumption value can be calculated by weighting or averaging, and this embodiment does not limit this. Since the first energy consumption value, the second energy consumption value and the third energy consumption value use different models,

[0045] In this way, machine learning and multiple neural network models can be combined to leverage the strengths of each model for feature extraction and processing. Based on long-term and short-term dependencies as well as global and local features, the target energy consumption value corresponding to the time series of the industrial system can be obtained more accurately.

[0046] In some optional embodiments, the original energy information of the target industrial system within the target time period can be first obtained, including the original energy status information and the original production line status information; and the original information can be cleaned and regularized, including but not limited to missing value filling, outlier detection and correction, normalization, and timestamp alignment, etc., so as to obtain at least one energy status information and at least one production line status information of the target industrial system within the target time period, and then form a target time series.

[0047] In some optional embodiments, the decision tree-based machine learning model in the aforementioned embodiment may include: a random forest model and a gradient boosting model. Preferably, the gradient boosting model may be an XGBoost (Extreme Gradient Boosting, an optimized distributed gradient boosting library) model.

[0048] Based on this, step 12 in the above embodiment, "using a decision tree to select a plurality of second features whose contribution to energy consumption prediction meets preset conditions from a plurality of first features of the first feature sequence to obtain a second feature sequence" can be implemented based on the following steps:

[0049] Step 121, using the first decision tree in the random forest model, select multiple candidate features whose contribution to energy consumption prediction meets the first preset condition from the multiple first features of the first feature sequence. This step is used to preliminarily screen the multiple first features. The first preset condition can be set to any condition according to actual design requirements, and this embodiment does not limit it. Candidate features refer to features preliminarily screened from multiple first features.

[0050] Step 122: Utilize the second decision tree in the gradient boosting model to select multiple second features whose contribution to energy consumption prediction meets the second preset condition from multiple candidate features to obtain a second feature sequence. The second decision tree may be a new decision tree different from the first decision tree, or a decision tree derived from the first decision tree, which is not limited in this embodiment.

[0051] Specifically, the second decision tree in the gradient boosting model can be used to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from multiple candidate features. The tree structure of the second decision tree may have multiple nodes, and each second feature may have a corresponding node position on the second decision tree. Based on this, the feature association relationship between the multiple second features can be identified according to the node positions of the multiple second features in the second decision tree; based on the feature association relationship between the multiple second features and the multiple second features, a second feature sequence is generated. The second feature sequence can not only reflect the importance ranking of the multiple second features, but also reflect the correlation between the multiple second features.

[0052] In this way, the decision tree can be used to more efficiently perform feature screening on the first feature sequence to obtain the second feature sequence.

[0053] In some optional embodiments, the long-term feature learning model in the aforementioned embodiments may be any model with long-term feature extraction and processing, such as a Transformer model or an LSTM (Long Short-Term Memory) model, etc.; preferably, the long-term feature learning model may be a Transformer model.

[0054] The long-term feature learning model may include: a first convolutional layer and a first relationship recognition layer based on an attention mechanism. The attention mechanism, i.e., Transformer, is a method in deep learning that enables the long-term feature learning model to focus on the most relevant part when processing the second feature sequence, thereby improving the processing capability and effect of the long-term feature learning model on the long sequence data in the second feature sequence.

[0055] Based on this, in step 13 of the aforementioned embodiment, “identifying the long-term dependency relationship between the global feature of the second feature sequence and the multiple second features through the long-term feature learning model” can be implemented based on the following methods:

[0056] The first convolution layer is used to convolve the second feature sequence to obtain the global features of the second feature sequence. Among them, the convolution configuration information for global feature processing can be obtained, such as the convolution kernel size and step size, and the convolution in the first convolution layer is configured according to the convolution configuration information, so that the first convolution layer has the ability to process global features. It should be noted that a larger convolution kernel can capture a wider range of contextual information and is more suitable for extracting global features; while a smaller convolution kernel focuses on details and is more suitable for extracting local features.

[0057] Afterwards, the second feature sequence can be input into the first relationship recognition layer, and the influence weight of each second feature in the second feature sequence on other second features can be identified based on the attention mechanism; and the long-term dependency relationship can be determined based on the influence weight of each second feature on other second features. Specifically, the first relationship recognition layer can linearly transform the second feature sequence into three different representations based on the attention mechanism: query (Q), key (K) and value (V). Afterwards, for any second feature, the similarity score between the corresponding query and the key can be calculated, and the similarity score can be normalized using the Softmax function to obtain the attention weight matrix corresponding to the second feature. After calculating the attention weight matrices of multiple second features based on the attention mechanism, the influence weight of each second feature on other second features can be determined based on the attention weight matrices of multiple second features, and the long-term dependency relationship can be determined based on the influence weight of each second feature on other second features.

[0058] In this way, based on the attention mechanism, the global features of the second feature sequence and the long-term dependencies between multiple second features can be more efficiently identified.

[0059] In some optional embodiments, the short-term feature learning model in the aforementioned embodiment may be any model with short-term feature extraction and processing, such as a GRU (Gated Recurrent Unit) model or a CNN (Convolutional Neural Network) model; preferably, the long-term feature learning model may be a GRU model. The short-term feature learning model may include: a second convolutional layer and a second relationship recognition layer. Based on this, in step 13 of the aforementioned embodiment, "identifying the short-term dependencies between the local features of the second feature sequence and multiple second features through the short-term feature learning model" can be implemented based on the following methods:

[0060] The second feature sequence is convolved by the second convolution layer to obtain the local features of the second feature sequence. Among them, the convolution configuration information for local feature processing can be obtained, such as the convolution kernel size and step size, etc., and the convolution in the second convolution layer is configured according to the convolution configuration information, so that the second convolution layer has the ability to process local features. It should be noted that a larger convolution kernel can capture a wider range of contextual information and is more suitable for extracting global features; while a smaller convolution kernel focuses on details and is more suitable for extracting local features. After that, the second feature sequence can be input into the second relationship recognition layer, and the second relationship recognition layer can sequentially identify short-term dependencies for any adjacent second features to obtain short-term dependencies between multiple second features. For example, there are a total of ten features in the second feature sequence, from the second feature 1 to the second feature 10, then the second relationship recognition layer can first determine the short-term dependency between the second feature 1 and the second feature 2, and then determine the short-term dependency between the second feature 2 and the second feature 3, and so on, until the short-term dependency between the second feature 9 and the second feature 10 is determined, thereby obtaining the short-term dependency between the second feature 1 to the second feature 10.

[0061] In this way, the short-term feature learning model can more efficiently identify the local features of the second feature sequence and the short-term dependencies between multiple second features.

[0062] In some optional embodiments, the multilayer perceptron model has been pre-trained to have the ability to perform nonlinear fitting on the second feature sequence according to long-term dependencies, short-term dependencies, global features, and local features. In other words, the multilayer perceptron model has pre-learned the mapping relationship between long-term dependencies, short-term dependencies, global features, and local features, and the third energy consumption value. Based on this, in the aforementioned step 14, the multilayer perceptron model not only pays attention to long-term dependencies and global features, but also pays attention to short-term dependencies and local features, so that energy consumption can be predicted more accurately to obtain the third energy consumption value.

[0063] In some optional embodiments, step 15 in the above embodiment, "calculating the target energy consumption value of the target industrial system within the target time period according to the first energy consumption value, the second energy consumption value and the third energy consumption value" can be implemented based on the following implementation methods:

[0064] Implementation method 1: According to the weight information corresponding to each energy consumption value, the first energy consumption value, the second energy consumption value and the third energy consumption value are weighted and summed to obtain the target energy consumption value. Among them, the weight information corresponding to the first energy consumption value, the second energy consumption value and the third energy consumption value can be dynamically adjusted according to the model effects (such as error metrics) of the long-term feature learning model, the short-term feature learning model and the multi-layer perceptron model. For example, the error metric of any model can be obtained by calculating the error between the target energy consumption value predicted by the model and the actual energy consumption value. If the error metric of the long-term feature learning model is lower than the error metric of the short-term feature learning model, it means that the long-term feature learning model is more accurate, and the weight of the first energy consumption value corresponding to the long-term feature learning model can be increased, while the weight of the second energy consumption value corresponding to the short-term feature learning model can be reduced.

[0065] Implementation method 2: Calculate the average of the first energy consumption value, the second energy consumption value and the third energy consumption value as the target energy consumption value.

[0066] The above embodiments 1 and 2 can be executed separately. Through the above embodiments, the target energy consumption value of the target industrial system within the target time period can be calculated more accurately based on the first energy consumption value, the second energy consumption value and the third energy consumption value.

[0067] Optionally, the target energy consumption value can be output in the form of a chart or report so that the manufacturing enterprise can manage and optimize energy consumption for the target industrial system. Optionally, energy consumption optimization suggestions can also be generated based on the target energy consumption value, such as adjusting equipment operating parameters, optimizing production scheduling, etc.

[0068] In some optional embodiments, the actual energy consumption value corresponding to the target time series may also be obtained, and an error metric index may be generated based on the actual energy consumption value and the target energy consumption value. The difference between the actual energy consumption value and the target energy consumption value may be used as the error metric index; the difference may also be corrected using a preset correction value to obtain the error metric index; the difference may also be multiplied by a preset coefficient to obtain the error metric index, and this embodiment does not limit this.

[0069] Based on this, the parameters of the full-link model can be optimized according to the error metric, so that the full-link model can more accurately predict energy consumption. Among them, the full-link model may include: a machine learning model, a long-term feature learning model, a short-term feature learning model, and a multi-layer perceptron model. Among them, the machine learning model is connected to the long-term feature learning model and the short-term feature learning model respectively; the long-term feature learning model and the short-term feature learning model are connected to the multi-layer perceptron model respectively.

[0070] In this way, the energy consumption prediction accuracy of the full-link model can be improved based on the actual energy consumption values ​​and target energy consumption values ​​corresponding to the target time series.

[0071] The above energy consumption prediction method will be further explained below in combination with actual scenarios.

[0072] In actual scenarios, the GRU model can be used as a short-term feature learning model, the Transformer model can be used as a long-term feature learning model, and the MLP model can be used as a multi-layer perceptron model.

[0073] like Figure 2 As shown, in actual scenarios, it usually includes six steps: data collection, data preprocessing, feature extraction and optimization, deep learning modeling, prediction result fusion, and result output and feedback.

[0074] In the data collection process, various energy data such as water, electricity, steam, etc., as well as auxiliary data such as the working status and output of the production line can be collected from sensors, PLCs and SCADA systems in production enterprises. The frequency of data collection can be configured according to actual needs, such as collecting data every 5 minutes, every hour or every day. The collected data includes parameters such as energy flow, temperature, pressure, etc.

[0075] In the data preprocessing stage, the collected data can be cleaned and normalized, including missing value filling, outlier detection and correction, normalization, etc. For time series data, timestamp alignment is also required to ensure that the time steps of different energy consumption data are consistent.

[0076] In the feature extraction and optimization phase, the random forest model can be used to perform feature importance analysis on the target time series formed by at least one energy status information and at least one production line status information. Random forest evaluates the contribution of each feature to energy consumption prediction by training multiple decision trees and taking the average result. The importance of features can be measured by indicators such as the Gini coefficient or information gain, so as to obtain the ranking of each feature and screen out features that have a significant impact on energy consumption prediction.

[0077] For the candidate features obtained after screening, XGBoost can be further used to optimize the relationship between the features. Specifically, based on the gradient boosting tree, the complex relationship between the features can be mined while considering the importance of the features, thereby further removing redundant information, making the second feature sequence input to the subsequent model more representative and predictive.

[0078] In the deep learning modeling link, it mainly includes GRU (Gated Recurrent Unit), Transformer model and multi-layer perceptron. Among them, ① GRU network modeling short-term dependency: GRU is used to process short-term dependencies in multi-energy time series. GRU can effectively capture local changes in time series while reducing computational complexity through the gating mechanism. The GRU network accepts the feature data processed by the feature extraction module as input, and gradually models the short-term change trend of energy consumption data, such as energy consumption fluctuations caused by working condition changes. ② Transformer network modeling long-term dependency: Transformer processes long-term dependencies in energy consumption time series through the self-attention mechanism, effectively capturing the long-term change rules in the data. For example, the impact of seasonal factors, production cycle changes, etc. on energy consumption. The Transformer network can establish global dependencies in the time dimension, so that the model performs well in capturing energy consumption trends over a long span, and supports parallel processing, thereby improving training efficiency. ③ MLP nonlinear fitting: MLP is used to perform nonlinear regression on the features output by GRU and Transformer to further improve the final prediction accuracy of energy consumption. MLP performs complex nonlinear mapping on the input features through multiple fully connected layers, so that the prediction results can better fit the complex change patterns in energy consumption data.

[0079] The connection methods of the above models can be referred to Figure 3 Based on the above models, the first energy consumption value, the second energy consumption value and the third energy consumption value can be obtained respectively.

[0080] In the prediction result fusion stage, in order to improve the stability and prediction accuracy of the model, the prediction results of GRU, Transformer and MLP can be integrated to reduce the deviation of a single model. The integration process uses a weighted average method to fuse the prediction results of each model according to a certain weight. The weight can be determined by evaluating the performance of the model on the training set and the validation set to ensure that the contribution of each model can be reasonably distributed.

[0081] In the result output and feedback link, the integrated target energy consumption value can be output in the form of charts or reports to facilitate production enterprises to manage energy consumption and optimize decisions.

[0082] The energy consumption prediction method provided by the embodiment of the present application can produce the following technical effects:

[0083] ① Effectiveness and flexibility of feature extraction:

[0084] Random forest and XGBoost are used for feature extraction. Random forest can select the most critical part for energy consumption prediction from high-dimensional features by constructing multiple decision trees to evaluate the importance of features. This process reduces the noise features in the data and improves the efficiency of the model. XGBoost further performs nonlinear modeling on the features and optimizes the feature relationships to ensure that the features input to the deep learning model are refined and critical. Effective screening and optimization of features are achieved through random forest and XGBoost, making the feature set input to the subsequent deep learning model more representative and concise. This not only reduces the computational complexity, but also improves the overall prediction accuracy of the model, especially for the complex features of multi-energy data, which can better reflect the relationship between the various energy sources and their impact on energy consumption.

[0085] ②Time-dependent modeling capabilities

[0086] GRU and Transformer are used for time dependency modeling. GRU is a simplified version of LSTM, which can maintain similar time modeling capabilities while reducing the number of parameters and computational complexity, so it is faster to train and suitable for processing real-time energy consumption prediction scenarios. Transformer can better capture long-term dependencies in time series through the self-attention mechanism, and can process input data in parallel, thereby improving computational efficiency. By combining GRU with Transformer, GRU processes short-term time dependencies and Transformer processes long-term time dependencies, thereby ensuring the comprehensive modeling capability of the model in time series. The structure of GRU is relatively simple, the model complexity is low, and the parallel computing capability of Transformer makes long-term dependency modeling more efficient. Therefore, the energy consumption prediction method provided by the embodiment of the present application can more stably perform time dependency modeling when facing complex multi-energy energy consumption data, while reducing computational overhead and improving prediction efficiency.

[0087] ③Generalization and robustness

[0088] A combination of random forest, XGBoost, GRU, Transformer and MLP is used to extract features and model in different areas of their respective strengths using different types of models. At the same time, through model integration, the prediction results of different models are fused according to weights, effectively reducing the overfitting risk of a single model and enhancing the robustness of the model.

[0089] Compared with the solution of using only deep learning models, the hybrid model of the present invention has stronger generalization ability and noise resistance. Random forest and XGBoost first optimize the features of the data to reduce the interference of noise on model training, while GRU and Transformer model short-term and long-term dependencies respectively to ensure effective understanding of time series. Finally, through MLP and model integration strategy, the prediction results of different models are further balanced, making the overall model more adaptable to various data and the prediction results more robust.

[0090] ④Interpretability

[0091] Random forest and XGBoost are used for feature extraction, which can quantify and explain the importance of features. This allows the model to provide interpretable information on the impact of each feature on energy consumption while predicting energy consumption results, helping production managers understand the reasons for changes in energy consumption. Random forest and XGBoost have better interpretability, especially by evaluating the importance of features through random forest and XGBoost, which can provide a reference for energy management for production companies. This explanatory ability helps companies better understand the impact of different factors on energy consumption, so as to take targeted energy-saving measures and improve energy efficiency.

[0092] It should be noted that the execution subject of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 11 to 15 can be device A; for another example, the execution subject of steps 11 to 12 can be device A, and the execution subject of steps 13 to 15 can be device B; and so on.

[0093] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel, and the sequence numbers of the operations, such as 12, 13, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel.

[0094] It should be noted that the descriptions such as “first” and “second” in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit “first” and “second” to different types.

[0095] Figure 4 is a schematic diagram of the structure of an electronic device provided by an exemplary embodiment of the present application, and the electronic device is applicable to the energy consumption prediction method provided by the above-mentioned embodiment, such as Figure 4As shown, the electronic device may include: a memory 401 , a processor 402 , and a communication component 403 .

[0096] The memory 401 is used to store computer programs and can be configured to store various other data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc.

[0097] In some exemplary embodiments, the processor 402 is coupled to the memory 401 and is used to execute a computer program in the memory 401 to: obtain a target time series formed by at least one energy status information and at least one production line status information of a target industrial system within a target time period, wherein the target time series is a multi-dimensional time series; input the target time series into a machine learning model based on a decision tree, perform feature extraction on the target time series, and obtain a first feature sequence; use the decision tree to screen out a plurality of second features whose contribution to energy consumption prediction meets preset conditions from a plurality of first features of the first feature sequence, to obtain a second feature sequence; input the second feature sequence into a long-term feature learning model and a short-term feature learning model, respectively, to identify the second feature through the long-term feature learning model. The long-term dependency relationship between the global feature of the sequence and the multiple second features is identified, and according to the long-term dependency relationship and the global feature, the second feature sequence is subjected to nonlinear fitting to obtain a first energy consumption value; and the short-term dependency relationship between the local feature of the second feature sequence and the multiple second features is identified through the short-term feature learning model, and according to the short-term dependency relationship and the local feature, the second feature sequence is subjected to nonlinear fitting to obtain a second energy consumption value; using a multilayer perceptron model, according to the long-term dependency relationship, the short-term dependency relationship, the global feature and the local feature, the second feature sequence is subjected to nonlinear fitting to obtain a third energy consumption value; and according to the first energy consumption value, the second energy consumption value and the third energy consumption value, the target energy consumption value of the target industrial system in the target time period is calculated.

[0098] Optionally, the decision tree-based machine learning model includes: a random forest model and a gradient boosting model; when the processor 402 uses the decision tree to screen out a plurality of second features whose contribution to energy consumption prediction meets preset conditions from a plurality of first features of the first feature sequence to obtain a second feature sequence, it is specifically used to: use the first decision tree in the random forest model to screen out a plurality of candidate features whose contribution to energy consumption prediction meets the first preset condition from a plurality of first features of the first feature sequence; use the second decision tree in the gradient boosting model to screen out a plurality of second features whose contribution to energy consumption prediction meets the second preset condition from the plurality of candidate features to obtain the second feature sequence.

[0099] Optionally, when the processor 402 uses the second decision tree in the gradient boosting model to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from the multiple candidate features to obtain the second feature sequence, it is specifically used to: use the second decision tree in the gradient boosting model to screen out multiple second features whose contribution to energy consumption prediction meets the second preset condition from the multiple candidate features; identify feature association relationships between the multiple second features based on the node positions of the multiple second features in the second decision tree; and generate the second feature sequence based on the feature association relationships between the multiple second features and the multiple second features.

[0100] Optionally, the long-term feature learning model includes: a first convolutional layer and a first relationship recognition layer based on an attention mechanism; when the processor 402 identifies the long-term dependency relationship between the global features of the second feature sequence and the multiple second features through the long-term feature learning model, it is specifically used to: use the first convolutional layer to convolve the second feature sequence to obtain the global features of the second feature sequence; input the second feature sequence into the first relationship recognition layer, and based on the attention mechanism, identify the influence weight of each second feature in the second feature sequence on other second features; determine the long-term dependency relationship according to the influence weight of each second feature on other second features.

[0101] Optionally, the short-time feature learning model includes: a second convolutional layer and a second relationship recognition layer; when the processor 402 identifies the short-term dependency relationship between the local features of the second feature sequence and the multiple second features through the short-time feature learning model, it is specifically used to: use the second convolutional layer to convolve the second feature sequence to obtain the local features of the second feature sequence; input the second feature sequence into the second relationship recognition layer, and perform short-term dependency recognition on any adjacent second features in turn to obtain the short-term dependency relationship between the multiple second features.

[0102] Optionally, when the processor 402 calculates the target energy consumption value of the target industrial system within the target time period based on the first energy consumption value, the second energy consumption value and the third energy consumption value, it is specifically used to: perform weighted summation of the first energy consumption value, the second energy consumption value and the third energy consumption value according to the weight information corresponding to each energy consumption value to obtain the target energy consumption value; or calculate the average value of the first energy consumption value, the second energy consumption value and the third energy consumption value as the target energy consumption value.

[0103] Optionally, the processor 402 is also used to obtain the actual energy consumption value corresponding to the target time series; generate an error measurement indicator based on the actual energy consumption value and the target energy consumption value; optimize the parameters of the full-link model based on the error measurement indicator, wherein the full-link model includes: the machine learning model, the long-time feature learning model, the short-time feature learning model and the multi-layer perceptron model; the machine learning model is respectively connected to the long-time feature learning model and the short-time feature learning model; the long-time feature learning model and the short-time feature learning model are respectively connected to the multi-layer perceptron model.

[0104] Further, if Figure 4 As shown, the electronic device also includes: a display 404, a power component 405, an audio component 406 and other components. Figure 4 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 4 Components shown.

[0105] An embodiment of the present application also provides a computer-readable storage medium, which, when the computer program is executed by a processor, enables the processor to implement the steps in the energy consumption prediction method.

[0106] An embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the steps in the energy consumption prediction method are executed.

[0107] In this embodiment, a machine learning model is used to extract and screen the features of the time series of the industrial system to obtain a feature sequence; the long-term feature learning model and the short-term feature learning model are used to extract the long-term dependency and global features between features, as well as the short-term dependency and local features between features; the multi-layer perceptron model is used to combine the long-term dependency, short-term dependency, global features and local features to perform nonlinear fitting on the feature sequence to obtain a third energy consumption value; and the target energy consumption value is obtained by combining various energy consumption values. In this way, machine learning and multiple neural network models are combined to utilize the advantages of each model for feature extraction and processing, and based on long-term and short-term dependencies and global and local features, the target energy consumption value corresponding to the time series of the industrial system can be obtained more accurately.

[0108] The above-mentioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0109] The above-mentioned communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0110] The above-mentioned display includes a screen, and the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0111] The power supply assembly provides power to various components of the device where the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply assembly is located.

[0112] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (Microphone, MIC), and when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal.

[0113] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-readable storage media (including but not limited to disk storage, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage, etc.) containing computer-usable program code.

[0114] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0115] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0117] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interface, network interface and memory.

[0118] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0119] Computer readable media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0120] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0121] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for predicting energy consumption, characterized in that: include: Acquire a target time series formed by at least one energy state information and at least one production line state information of a target industrial system within a target time period, wherein the target time series is a multi-dimensional time series; The target time series is input into a machine learning model based on a decision tree, and features are extracted from the target time series to obtain a first feature sequence; and a plurality of second features whose contribution to energy consumption prediction meets preset conditions are screened out from a plurality of first features of the first feature sequence using the decision tree to obtain a second feature sequence; The second feature sequence is input into a long-term feature learning model and a short-term feature learning model respectively, so as to identify the long-term dependency relationship between the global features of the second feature sequence and the multiple second features through the long-term feature learning model, and perform nonlinear fitting on the second feature sequence according to the long-term dependency relationship and the global features to obtain a first energy consumption value; and the short-term dependency relationship between the local features of the second feature sequence and the multiple second features is identified through the short-term feature learning model, and the second feature sequence is nonlinearly fitted according to the short-term dependency relationship and the local features to obtain a second energy consumption value; Using a multilayer perceptron model, according to the long-term dependency, the short-term dependency, the global feature, and the local feature, nonlinearly fit the second feature sequence to obtain a third energy consumption value; A target energy consumption value of the target industrial system within the target time period is calculated according to the first energy consumption value, the second energy consumption value, and the third energy consumption value.

2. The method according to claim 1, characterized in that Machine learning models based on decision trees include: random forest model and gradient boosting model; Using the decision tree, multiple second features whose contribution to energy consumption prediction meets preset conditions are screened out from multiple first features of the first feature sequence to obtain a second feature sequence, including: Using the first decision tree in the random forest model, multiple candidate features whose contribution to energy consumption prediction meets the first preset condition are screened out from the multiple first features of the first feature sequence; A second decision tree in the gradient boosting model is used to screen out a plurality of second features whose contribution to energy consumption prediction meets a second preset condition from the plurality of candidate features to obtain the second feature sequence.

3. The method according to claim 2, characterized in that Using the second decision tree in the gradient boosting model, multiple second features whose contribution to energy consumption prediction meets the second preset condition are screened out from the multiple candidate features to obtain the second feature sequence, including: Using the second decision tree in the gradient boosting model, multiple second features whose contribution to energy consumption prediction meets the second preset condition are screened out from the multiple candidate features; Identifying feature association relationships between the multiple second features according to node positions of the multiple second features in the second decision tree; The second feature sequence is generated according to the feature association relationship between the multiple second features and the multiple second features.

4. The method according to claim 1, characterized in that The long-term feature learning model includes: a first convolutional layer and a first relationship recognition layer based on an attention mechanism; Identifying the long-term dependency relationship between the global feature of the second feature sequence and the plurality of second features by the long-term feature learning model includes: Using the first convolutional layer, convolving the second feature sequence to obtain global features of the second feature sequence; The second feature sequence is input into the first relationship recognition layer, and based on the attention mechanism, the influence weight of each second feature in the second feature sequence on other second features is identified; and the long-term dependency relationship is determined according to the influence weight of each second feature on other second features.

5. The method according to claim 1, characterized in that The short-term feature learning model includes: a second convolutional layer and a second relationship recognition layer; Identifying the short-term dependency relationship between the local features of the second feature sequence and the multiple second features by the short-term feature learning model includes: Using the second convolutional layer, convolving the second feature sequence to obtain local features of the second feature sequence; The second feature sequence is input into the second relationship recognition layer, and short-term dependency recognition is performed on any adjacent second features in turn to obtain the short-term dependency relationship between the multiple second features.

6. The method according to any one of claims 1 to 5, characterized in that: Calculating a target energy consumption value of the target industrial system within the target time period according to the first energy consumption value, the second energy consumption value, and the third energy consumption value includes: According to the weight information corresponding to each energy consumption value, the first energy consumption value, the second energy consumption value and the third energy consumption value are weighted and summed to obtain the target energy consumption value; or, An average value of the first energy consumption value, the second energy consumption value, and the third energy consumption value is calculated as the target energy consumption value.

7. The method according to any one of claims 1 to 5, characterized in that: Also includes Obtaining actual energy consumption values ​​corresponding to the target time series; generating an error metric based on the actual energy consumption value and the target energy consumption value; According to the error metric, the parameters of the full-link model are optimized, wherein the full-link model includes: the machine learning model, the long-time feature learning model, the short-time feature learning model and the multi-layer perceptron model; the machine learning model is respectively connected to the long-time feature learning model and the short-time feature learning model; the long-time feature learning model and the short-time feature learning model are respectively connected to the multi-layer perceptron model.

8. An electronic device, characterized in that: include: A memory and a processor; wherein the memory is used to: store one or more computer instructions; and the processor is used to execute the one or more computer instructions to: execute the steps in the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: When the computer program is executed by a processor, the processor is enabled to implement the steps of the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, enables the processor to implement the steps of the method according to any one of claims 1 to 7.

Images