Carbon asset full life cycle management method and system for industrial manufacturing industry

Abstract

The invention provides an industrial manufacturing industry-oriented carbon asset full life cycle management method and system, and the method comprises the steps: carrying out the cross-link feature fusion processing through obtaining an original carbon activity data set of an industrial manufacturing whole process, and generating carbon footprint feature data; and calling a pre-trained deep learning model to perform carbon asset state evolution analysis on the carbon footprint feature data, predicting a carbon asset evolution path, generating a carbon asset optimization strategy set based on a prediction result, including equipment energy efficiency regulation and control and a production plan recombination strategy, and pushing the carbon asset optimization strategy set to an industrial manufacturing execution system. And triggering an automatic control instruction reconstruction operation to realize automation and intelligence of carbon asset management, thereby comprehensively and accurately mastering a carbon asset state and an evolution rule, formulating a scientific and effective optimization strategy, reducing carbon emission, improving energy utilization efficiency, and improving green production capacity of an enterprise.

Description

Technical Field

[0001] The present invention relates to the field of industrial manufacturing, and in particular to a method and system for full life cycle management of carbon assets for the industrial manufacturing industry. Background Art

[0002] In the industrial manufacturing sector, carbon asset management has become a crucial means for companies to achieve green production, reduce operating costs, and enhance market competitiveness. However, traditional carbon asset management methods are often limited to data collection and analysis at a single stage, such as focusing solely on energy consumption or emissions monitoring, and lack a comprehensive consideration of carbon activity data across the entire industrial manufacturing process. This fragmented management approach makes it difficult for companies to fully and accurately grasp the status and evolution of carbon assets, making it difficult to formulate scientific and effective carbon asset optimization strategies. Furthermore, existing technologies lack the ability to predict the evolutionary paths of carbon assets, making it difficult for companies to respond promptly and accurately to market changes and policy adjustments. Summary of the Invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a carbon asset full life cycle management method for industrial manufacturing, the method comprising: Obtaining a set of raw carbon activity data for the entire industrial manufacturing process, wherein the raw carbon activity data set includes an energy consumption recording unit, a production raw material flow unit, and an emission monitoring unit; Performing cross-link feature fusion processing on the original carbon activity data set to generate carbon footprint feature data for the entire industrial manufacturing process, wherein the carbon footprint feature data includes static facility carbon emission distribution features and dynamic operation carbon emission evolution features; Calling a pre-trained deep learning model to perform carbon asset state evolution analysis on the carbon footprint feature data to generate a carbon asset evolution path prediction result; Generating a carbon asset optimization strategy set for the entire industrial manufacturing process based on the carbon asset evolution path prediction result, wherein the carbon asset optimization strategy set includes an equipment energy efficiency control strategy and a production plan reorganization strategy; The carbon asset optimization strategy set is pushed to the industrial manufacturing execution system to trigger the automatic control instruction reconstruction operation.

[0004] On the other hand, an embodiment of the present invention also provides a carbon asset full life cycle management system for the industrial manufacturing industry, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0005] Based on the above aspects, the embodiment of the present invention obtains the original carbon activity data set of the entire industrial manufacturing process and performs cross-link feature fusion processing, which can comprehensively and accurately reflect the carbon footprint characteristics of the industrial manufacturing process, including static facility carbon emission distribution characteristics and dynamic operation carbon emission evolution characteristics. It calls the pre-trained deep learning model to perform carbon asset state evolution analysis and processing on the carbon footprint feature data, and can predict the evolution path of carbon assets. The carbon asset optimization strategy set generated based on the carbon asset evolution path prediction results includes equipment energy efficiency control strategy and production plan reorganization strategy, which can optimize the company's carbon asset management in a targeted manner, reduce carbon emissions, and improve energy utilization efficiency. Finally, the carbon asset optimization strategy set is pushed to the industrial manufacturing execution system to trigger the automatic control instruction reconstruction operation, realizing the automation and intelligence of carbon asset management, and significantly improving the company's carbon asset management level and green production capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 This is a schematic diagram of the execution flow of the carbon asset full life cycle management method for the industrial manufacturing industry provided by an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of a carbon asset lifecycle management system for industrial manufacturing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0008] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a carbon asset full life cycle management method for the industrial manufacturing industry provided by an embodiment of the present invention. The carbon asset full life cycle management method for the industrial manufacturing industry is introduced in detail below.

[0009] Step S110: obtaining a set of original carbon activity data for the entire industrial manufacturing process, wherein the set of original carbon activity data includes an energy consumption recording unit, a production raw material circulation unit, and an emission monitoring unit.

[0010] In this embodiment, the energy consumption recording unit is responsible for recording the usage of various types of energy in the industrial manufacturing process. In modern industrial production, there are many types of energy, including but not limited to electricity, coal, natural gas, oil, etc. Different production equipment and production links have different energy requirements and consumption methods. For example, in a certain machinery manufacturing enterprise, metal processing equipment usually requires a large amount of electricity to drive its operation, while heat treatment processes may use coal or natural gas to provide a high-temperature environment. The energy consumption recording unit will record the consumption data of each type of energy in different time periods in detail based on time. These data exist in the form of continuous time series and can reflect the dynamic changes in energy consumption over time. In order to facilitate subsequent analysis and processing, energy consumption data are usually collected and recorded at set time intervals, such as every minute, every hour or every day, so that short-term fluctuations and long-term trends in energy consumption can be captured, so as to facilitate a deep understanding of energy utilization efficiency and carbon emissions in the industrial manufacturing process.

[0011] The production raw material flow unit focuses on the flow and transformation of raw materials throughout the entire industrial manufacturing process. In industrial production, raw materials can go through multiple links and stages, from procurement, transportation, storage, to production. The production raw material flow unit constructs a complex graph-structured data model, with raw material batches as nodes and flow relationships as edges. Each raw material batch has unique attributes and characteristics, including the type, quantity, source, and quality of the raw material. Flow relationships describe the transfer of raw materials between different links, such as transportation from suppliers to factory warehouses and material collection from warehouses to production workshops. By analyzing data from the production raw material flow unit, we can clearly understand the carbon transfer of raw materials at each link. For example, in chemical production, different raw materials undergo carbon transformation and migration during chemical reactions. Data from the production raw material flow unit can help track the fate of this carbon, thereby assessing the carbon emission contribution of raw materials throughout the entire production process.

[0012] Emission monitoring units are primarily responsible for real-time monitoring of various emissions generated during industrial manufacturing. Industrial production generates a large amount of emissions, of which greenhouse gases and waste gases are the most significant for the environment and climate. Emission monitoring units are installed near various emission sources, such as chimneys and exhaust pipes, and use monitoring equipment to measure the concentration and flow of emissions in real time. Emission concentration reflects the content of a specific pollutant per unit volume or mass of emissions, while emission flow indicates the amount of emissions emitted per unit time. By analyzing emission concentration and flow monitoring data, the carbon emissions of each emission source over different time periods can be accurately calculated. For example, in steel smelting plants, emission monitoring units monitor the concentration and flow of gases such as carbon dioxide and carbon monoxide emitted by blast furnaces, as well as exhaust gases such as sulfur dioxide and nitrogen oxides emitted by sintering machines.

[0013] Step S120: performing cross-link feature fusion processing on the original carbon activity data set to generate carbon footprint feature data for the entire industrial manufacturing process, wherein the carbon footprint feature data includes static facility carbon emission distribution features and dynamic operation carbon emission evolution features.

[0014] After obtaining a set of raw carbon activity data for the entire industrial manufacturing process, it is necessary to perform cross-process feature fusion processing to generate carbon footprint characteristic data for the entire industrial manufacturing process. Carbon footprint characteristic data includes two key aspects: static facility carbon emission distribution characteristics and dynamic operation carbon emission evolution characteristics. Static facility carbon emission distribution characteristics mainly reflect the carbon emission distribution of various fixed facilities (such as factories and equipment) in the industrial manufacturing process, while dynamic operation carbon emission evolution characteristics focus on the changing trends of carbon emissions over time during production operations.

[0015] Step S121: Input the energy consumption recording unit into the time series feature encoder, extract the energy consumption fluctuation characteristics in different time windows by processing the energy consumption data of the continuous time series, and generate the energy consumption time series feature vector. The time series feature encoder is composed of a bidirectional long short-term memory network.

[0016] In this step, the continuous time series energy consumption data from the energy consumption recording unit is input into a temporal feature encoder consisting of a bidirectional long short-term memory network. A bidirectional long short-term memory network (Bi-LSTM) is a special type of recurrent neural network that effectively processes sequential data and captures long-term dependencies within the sequence. When processing energy consumption data, the Bi-LSTM network can simultaneously consider both past and future temporal information, thereby more accurately extracting energy consumption fluctuation characteristics within different time windows.

[0017] First, energy consumption data is a continuous sequence arranged in chronological order, with each time point corresponding to an energy consumption value. To extract features within different time windows, the continuous time series data needs to be divided into multiple time windows of fixed length. The length of the time window can be adjusted based on specific analysis needs and data characteristics. For example, the time window length can be set to one hour, one day, or one week. Within each time window, a bidirectional long short-term memory network processes the energy consumption data and learns the inherent relationships and changing patterns between energy consumption values.

[0018] A bidirectional long short-term memory network (LSTM) consists of multiple neurons, each of which contains an input gate, a forget gate, and an output gate. These gating mechanisms control the flow and storage of information. When processing energy consumption data, the input gate determines how much of the input information at the current time step should be added to the cell state; the forget gate determines how much of the cell state from the previous time step should be retained; and the output gate determines how much of the cell state at the current time step should be output. Through these gating mechanisms, the bidirectional long short-term memory network can effectively capture the long-term dependencies and complex fluctuations in energy consumption data.

[0019] Training a bidirectional long short-term memory network requires a large amount of historical energy consumption data as training samples. By continuously adjusting the network's parameters, the network can accurately predict future energy consumption values. After training is complete, new energy consumption data is input into the trained bidirectional long short-term memory network, which then outputs the energy consumption fluctuation characteristics within each time window. These energy consumption fluctuation characteristics can be represented by a vector, the energy consumption time series feature vector. The energy consumption time series feature vector contains information across multiple dimensions, each corresponding to a specific energy consumption fluctuation characteristic, such as the energy consumption trend, fluctuation amplitude, and periodicity. Analysis of the energy consumption time series feature vector provides a deeper understanding of the dynamic changes in energy consumption in industrial manufacturing processes.

[0020] Step S122: Input the production raw material flow unit into the graph structure encoder, calculate the attention weights between each node through the graph attention network, extract the raw material carbon transfer characteristics, and generate the raw material flow graph feature vector. The graph structure encoder uses raw material batches as nodes and flow relationships as edges.

[0021] Next, the data of the production raw material flow unit is input into the graph structure encoder, which constructs a graph structure data model with raw material batches as nodes and flow relationships as edges, and uses the graph attention network (GAT) to calculate the attention weights between each node, thereby extracting the raw material carbon transfer characteristics and generating the raw material flow graph feature vector.

[0022] The Graph Attention Network (GAN) is a deep learning model based on graph-structured data that adaptively learns the importance weights between nodes in a graph. In the data for a production raw material flow unit, each raw material batch is considered a node, and the flow relationships between nodes are considered edges. The GAN analyzes node features and edge information to calculate the attention weight of each node relative to other nodes. The attention weight represents the degree of attention a node pays to other nodes; a higher weight indicates a closer relationship between the node and other nodes.

[0023] When calculating attention weights, the Graph Attention Network considers both node characteristics and edge information. Node characteristics can include attributes such as raw material type, quantity, and quality, while edge information can include flow time, distance, and transportation method. By comprehensively analyzing this information, the Graph Attention Network can more accurately capture the carbon transfer relationships between raw materials at different nodes.

[0024] Specifically, the graph attention network performs a nonlinear transformation on each node's features and then calculates a similarity score between that node and all other nodes. A higher similarity score indicates a closer relationship between the two nodes. Next, a softmax function is used to convert the similarity scores into attention weights, such that the sum of all attention weights equals 1. Finally, based on the calculated attention weights, the features of adjacent nodes are weighted and summed to obtain a new feature representation for each node.

[0025] Through multiple iterations, the graph attention network continuously updates the feature representations of nodes, thereby better capturing the characteristics of raw material carbon transfer. Ultimately, the new feature representations of each node are concatenated to produce a feature vector for the raw material flow graph. This feature vector incorporates information from multiple dimensions, reflecting the carbon transfer status and characteristics of the raw material throughout its entire flow process. Analysis of this feature vector provides a deeper understanding of the carbon emissions contribution of raw materials at different stages.

[0026] Step S123: Input the emission monitoring unit into a convolutional feature encoder to generate an emission monitoring spatial feature vector by extracting the spatial correlation pattern between emission concentration and flow rate. The convolutional feature encoder includes multiple convolution kernels of different sizes.

[0027] Subsequently, the data of the emission monitoring unit is input into a convolutional feature encoder, which contains multiple convolution kernels of different sizes to extract the spatial correlation pattern of emission concentration and flow rate and generate the emission monitoring spatial feature vector.

[0028] Data from emission monitoring units is typically collected at different spatial locations and time points, and thus exhibits certain spatial distribution characteristics. The convolutional feature encoder leverages the characteristics of convolutional neural networks (CNNs) to extract spatial correlation patterns between emission concentration and flow rate through convolution operations.

[0029] The convolution kernel is a core component of convolutional neural networks. It slides over the input data to extract features from local regions. Convolution kernels of different sizes can capture spatial features at different scales. For example, smaller kernels can extract subtle local features, while larger kernels can extract more macroscopic, global features.

[0030] In this step, the convolutional feature encoder takes the data from the emission monitoring unit as input and performs convolution operations with multiple convolution kernels of different sizes. Each convolution kernel slides over the input data, calculating a weighted sum of the local area to produce a feature map. Each element in the feature map represents the characteristic response of the input data at that location. By concatenating and fusing multiple feature maps, a richer spatial feature representation can be obtained.

[0031] After the convolution operation, a pooling operation is usually performed to reduce the dimensionality of the feature map while retaining important feature information. Pooling operations can be performed using methods such as maximum pooling or average pooling. Maximum pooling selects the maximum value in a local area as the output, while average pooling selects the average value in a local area as the output.

[0032] Finally, the feature map obtained through convolution and pooling is flattened and converted into a vector, the emission monitoring spatial feature vector. This vector contains the spatial correlation patterns and characteristic information between emission concentration and flow rate, reflecting the distribution and variation of emissions at different spatial locations. Analysis of this vector provides a deeper understanding of the emission characteristics and patterns of emissions from industrial manufacturing processes.

[0033] Step S124: Input the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector into the attention fusion layer, calculate the mutual attention weight matrix between the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector, perform weighted aggregation on the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector, and generate a fusion feature matrix containing the correlation information between the links.

[0034] After obtaining the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector, the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector are input into the attention fusion layer. By calculating the mutual attention weight matrix between them, the energy consumption time series feature vector, the raw material flow graph feature vector and the emission monitoring space feature vector are weightedly aggregated to generate a fusion feature matrix containing the correlation information between links.

[0035] The main function of the attention fusion layer is to adaptively learn the importance weights between different feature vectors, thereby better integrating the feature information of different links. In this step, the energy consumption time series feature vector reflects the characteristics of the energy consumption link, the raw material flow graph feature vector reflects the characteristics of the production raw material flow link, and the emission monitoring space feature vector reflects the characteristics of the emission monitoring link. These three links are interrelated and influence each other in the industrial manufacturing process. The attention fusion layer can capture the inherent connections and correlation information between them.

[0036] Specifically, the attention fusion layer first calculates the similarity scores between the energy consumption time series feature vector, the raw material flow graph feature vector, and the emission monitoring space feature vector. Similarity scores can be calculated using methods such as dot product and cosine similarity. Then, using the softmax function, these similarity scores are converted into a mutual attention weight matrix. The mutual attention weight matrix represents the importance of each feature vector relative to the others. Higher weights indicate a closer relationship between the feature vector and the others.

[0037] Next, based on the mutual attention weight matrix, weighted aggregation is performed on the energy consumption time series feature vector, the raw material flow diagram feature vector, and the emission monitoring spatial feature vector. Weighted aggregation can be performed using weighted addition or weighted concatenation, depending on the dimensionality of the feature vectors and the data characteristics. In this embodiment, weighted concatenation is used to concatenate the three feature vectors in a predetermined order to produce a higher-dimensional fused feature vector. Finally, the fused feature vector is reshaped to produce a fused feature matrix that contains information about the associations between the links.

[0038] The fused feature matrix contains information from multiple dimensions, reflecting the relationships and characteristics between energy consumption, production material flow, and emissions monitoring. Analyzing the fused feature matrix provides a deeper understanding of the interactions and impacts between different links in the industrial manufacturing process.

[0039] Step S125: Input the fused feature matrix into the static feature extraction branch and the dynamic feature extraction branch respectively. Use the static feature extraction branch to extract the statistical characteristics of the carbon emission distribution of equipment, raw materials, and emission links through global average pooling operation as the static facility carbon emission distribution characteristics. Use the dynamic feature extraction branch to extract the carbon emission evolution trend characteristics within the production cycle through the time convolution network as the dynamic operation carbon emission evolution characteristics, and finally splice them together to form the carbon footprint feature data.

[0040] Finally, the fused feature matrix is ​​input into the static feature extraction branch and the dynamic feature extraction branch respectively to extract the static facility carbon emission distribution characteristics and the dynamic operation carbon emission evolution characteristics, and finally spliced ​​together to form the carbon footprint feature data.

[0041] The static feature extraction branch mainly focuses on the statistical characteristics of carbon emission distribution in equipment, raw materials, and emission links. In this step, the global average pooling operation is used to extract these features. The global average pooling operation averages the fused feature matrix in the spatial dimension to obtain the average value of each channel. These average values ​​can reflect the carbon emission distribution of equipment, raw materials, and emission links in the entire industrial manufacturing process. For example, for the equipment link, the global average pooling operation can obtain the average carbon emission level of different equipment; for the raw material link, the average carbon transfer amount of different raw materials can be obtained; for the emission link, the average emission amount of different emission sources can be obtained. By analyzing these average values, the carbon emission distribution characteristics of equipment, raw materials, and emission links can be understood as static facility carbon emission distribution characteristics.

[0042] The dynamic feature extraction branch focuses on the evolutionary trends of carbon emissions within the production cycle. In this step, a temporal convolutional network (TCN) is used to extract these features. A TCN is a convolutional neural network specifically designed for processing sequential data. It uses convolution operations to capture temporal dependencies within sequential data. When fed the fused feature matrix, the TCN performs convolution operations along the time dimension, extracting feature information at different time steps. By analyzing this feature information, we can understand the evolutionary trends of carbon emissions within the production cycle, such as growth trends and fluctuations. These features serve as dynamic operational carbon emission evolution features.

[0043] Finally, the static facility carbon emission distribution characteristics and the dynamic operational carbon emission evolution characteristics are combined to form carbon footprint characteristic data. Carbon footprint characteristic data contains information from multiple dimensions, reflecting both the static facility carbon emission distribution and the dynamic operational carbon emission evolution trends during the industrial manufacturing process. By analyzing carbon footprint characteristic data, we can gain a comprehensive and in-depth understanding of the carbon footprint of the entire industrial manufacturing process.

[0044] Step S130: calling a pre-trained deep learning model to perform carbon asset state evolution analysis on the carbon footprint feature data to generate a carbon asset evolution path prediction result.

[0045] After obtaining carbon footprint data for the entire industrial manufacturing process, we need to analyze and process the evolution of carbon asset status to predict the evolution path of carbon assets. In this step, we call on a pre-trained deep learning model to complete this task.

[0046] The pre-trained deep learning model is trained on a large amount of historical data. It can learn the inherent relationship between carbon footprint characteristic data and the evolution path of carbon assets. When the model is called, the carbon footprint characteristic data is used as input. The deep learning model analyzes and processes it and outputs the prediction results of the carbon asset evolution path.

[0047] Step S131: The carbon footprint feature data is downsampled by a multi-scale feature extraction module to generate a multi-scale feature set with different levels of precision, where the first scale feature reflects the global carbon emission overview information and the second scale feature reflects the detailed carbon emission information of the local link.

[0048] First, the multi-scale feature extraction module downsamples the carbon footprint feature data. Downsampling converts high-resolution feature data into low-resolution feature data, thereby reducing the data's dimensionality and complexity. In this step, the carbon footprint feature data is processed at different downsampling rates to generate a multi-scale feature set with varying degrees of sophistication.

[0049] First-scale features, obtained through a large downsampling rate, provide a global overview of carbon emissions. For example, in an automobile manufacturer, first-scale features can reflect the total carbon emissions of the entire factory over a period of time, including the comprehensive carbon emission levels of each workshop. This global information can help companies understand the overall trend and scale of carbon emissions from a macro perspective.

[0050] Second-scale features, obtained through a smaller downsampling rate, can provide detailed information about carbon emissions at specific stages. For example, in an automobile manufacturer's paint shop, second-scale features can provide a detailed picture of carbon emissions from different painting processes and equipment within the shop. This detailed local information can help companies pinpoint the specific sources and key links of carbon emissions, providing targeted guidance for subsequent energy-saving and emission-reduction measures.

[0051] Step S132: Performing spatiotemporal joint processing on the multi-scale feature set through the spatiotemporal dependency modeling module to generate an evolution feature tensor containing spatiotemporal coupling information.

[0052] Next, the spatiotemporal dependency modeling module performs spatiotemporal joint processing on the multi-scale feature set. In industrial manufacturing, carbon emissions are not only related to spatial location but also closely linked to time. Therefore, it is necessary to consider coupled information in both spatiotemporal and temporal dimensions to more accurately predict the evolution of carbon assets.

[0053] Step S1321: Input the multi-scale feature set into the spatial self-attention layer, calculate the attention weight of each spatial position and other spatial positions, and generate spatial features that reflect the interactive relationship between links.

[0054] In the spatiotemporal dependency modeling module, the multi-scale feature set is first input into the spatial self-attention layer. The spatial self-attention layer adaptively learns the importance weights between spatial locations. For each spatial location in the multi-scale feature set, the spatial self-attention layer calculates the similarity score between it and other spatial locations. Then, using the softmax function, these similarity scores are converted into attention weights. Attention weights represent the degree of attention a spatial location pays to other spatial locations. A higher weight indicates a closer relationship between the location and the others. By calculating the attention weights for all spatial locations, spatial features can be generated that reflect the interactions between links. For example, in a large factory, carbon emissions from different workshops may influence each other. The spatial self-attention layer can capture these interactions between workshops. For example, high carbon emissions from one workshop may affect the production efficiency and carbon emissions of adjacent workshops. By calculating attention weights, the degree of connection between workshops can be clearly identified.

[0055] Step S1322: Input the spatial features into the time-gated recurrent unit, process them time-step by time-step along the production time axis, capture the influence of the previous time-step features on the current time-step features, and generate time features that reflect the recursive relationship between the stages.

[0056] After obtaining the spatial features, they are input into a time-gated recurrent unit (TGRU). A TGRU is a neural network structure capable of processing sequential data, processing data time-step by time-step along the production timeline. In industrial manufacturing, carbon emissions change over time, and the carbon emission characteristics of the previous time step may influence those of the current time step. TGRUs control the flow of information through gating mechanisms, which include an input gate, a forget gate, and an output gate. The input gate determines how much of the input information in the current time step should be added to the cell state; the forget gate determines how much of the cell state from the previous time step should be retained; and the output gate determines how much of the cell state from the current time step should be output. Through these gating mechanisms, TGRUs can capture the influence of the features of the previous time step on the features of the current time step, thereby generating temporal features that reflect recursive relationships between stages. For example, in chemical production, the concentration of an intermediate product in a chemical reaction at one time step can affect the reaction rate and carbon emissions in the next time step. TGRUs can learn these temporal recursive relationships to help predict carbon emission characteristics in future time steps.

[0057] Step S1323: Fusing the spatial features and the temporal features at the element level to generate an evolution feature tensor containing spatiotemporal coupling information.

[0058] Element-level fusion of spatial and temporal features. Element-level fusion refers to performing corresponding operations on elements of spatial and temporal features at the same location. For example, each element in the spatial and temporal features is combined according to a set rule to generate an evolving feature tensor containing information about spatiotemporal coupling. This information about spatiotemporal coupling is crucial in industrial manufacturing scenarios, as carbon emissions are not only related to spatial location but also closely linked to time. By fusing spatial and temporal features, the spatiotemporal variations in carbon emissions during industrial manufacturing can be more comprehensively reflected. For example, in a factory with multiple production workshops where production processes change dynamically over time, the evolving feature tensor can simultaneously reflect the carbon emissions of different workshops at different times, as well as the mutual influence between workshops and at different time steps.

[0059] Step S133: The evolution feature tensor is subjected to multi-step decoding processing through the evolution prediction module. Combined with the constraints of the current production plan, the equipment carbon emission change trend, the raw material carbon conversion efficiency change trend and the carbon loss change trend in the emission link in the future production cycle are predicted stage by stage to generate a carbon asset evolution path prediction result including the carbon emission indicators of key nodes in each stage.

[0060] The evolution prediction module performs a multi-step decoding process on the evolution feature tensor, which contains information about spatiotemporal coupling. Multi-step decoding involves gradually analyzing and predicting the evolution feature tensor to obtain information for multiple future time steps. This process is based on the constraints of the current production plan. The current production plan specifies the operating status of equipment, raw material input plans, and production schedules, all of which affect carbon emissions. For example, if the production plan specifies that a piece of equipment must operate at a specific power level at a specific time, this will affect the carbon emissions of that equipment.

[0061] During the stage-by-stage forecasting process, trends in equipment carbon emissions, raw material carbon conversion efficiency, and carbon losses in the emissions process can be predicted over the future production cycle. Forecasting equipment carbon emissions trends can take into account factors such as equipment operating parameters, maintenance status, and the distribution of production tasks. For example, equipment operating at high load may generate more carbon emissions, while regular maintenance can improve the equipment's energy efficiency, thereby reducing carbon emissions. Forecasting raw material carbon conversion efficiency trends focuses on factors such as raw material quality, production process, and reaction conditions. Raw materials of different qualities may have different carbon conversion efficiencies under the same production process, and optimizing the production process can improve raw material carbon conversion efficiency. Forecasting carbon losses in the emissions process takes into account factors such as the operating status of emission treatment equipment, emission regulations, and leaks during the production process. For example, a malfunction in emission treatment equipment may lead to increased carbon losses in the emissions process.

[0062] By predicting these trends, we ultimately generate a carbon asset evolution path forecast that includes key carbon emission indicators at each stage. These indicators can include peak carbon emissions from equipment at specific times, carbon conversion rates for raw materials at a specific production stage, and carbon losses from emissions during a specific time period. These key carbon emission indicators can help companies understand the evolution of their carbon assets over the future production cycle.

[0063] Step S140: generating a carbon asset optimization strategy set for the entire industrial manufacturing process based on the carbon asset evolution path prediction result, wherein the carbon asset optimization strategy set includes an equipment energy efficiency control strategy and a production plan reorganization strategy.

[0064] After obtaining the carbon asset evolution path prediction results, it is necessary to generate a set of carbon asset optimization strategies for the entire industrial manufacturing process based on this. The carbon asset optimization strategy set aims to reduce carbon emissions in the industrial manufacturing process and improve the utilization efficiency of carbon assets by adjusting equipment and production plans.

[0065] Step S141: inputting the carbon asset evolution path prediction result into a strategy generation model, wherein the strategy generation model includes an equipment control subnetwork and a production planning subnetwork.

[0066] The strategy generation model consists of an equipment control subnetwork and a production planning subnetwork. The carbon asset evolution path prediction results are fed into both the equipment control subnetwork and the production planning subnetwork. The equipment control subnetwork is primarily responsible for generating equipment energy efficiency control strategies, while the production planning subnetwork is responsible for generating production plan restructuring strategies. The carbon asset evolution path prediction results include information such as the changing trends of equipment carbon emissions, raw material carbon conversion efficiency, and carbon losses in the emission process over the future production cycle.

[0067] Step S142: Utilize the equipment control subnetwork and take the equipment carbon emission change trend as input, calculate the carbon emission reduction benefits of different equipment operating parameter adjustment plans through the reinforcement learning framework, and finally output the adjustment plan with the highest carbon emission reduction benefit as the equipment energy efficiency control strategy.

[0068] The device control subnetwork uses the device carbon emission trends as input and uses a reinforcement learning framework to calculate the carbon reduction benefits of different device operating parameter adjustment schemes. Reinforcement learning is a learning method that maximizes cumulative rewards through the interaction between an intelligent agent and its environment.

[0069] Step S1421: Input the equipment carbon emission change trend into the state representation module of the reinforcement learning framework to generate a state vector reflecting the evolution law of the current equipment carbon emission level. The state vector includes the carbon emission rate change gradient and the trend persistence indicator.

[0070] First, the device's carbon emission trend is input into the state representation module of the reinforcement learning framework. The state representation module analyzes the device's carbon emission trend and generates a state vector that reflects the evolution of the device's current carbon emission level. The state vector contains two important indicators: the carbon emission rate gradient and the trend persistence indicator. The carbon emission rate gradient represents the rate of change of the device's carbon emission rate over time, reflecting the speed of the device's carbon emission growth or decline. For example, a positive and large carbon emission rate gradient indicates a rapid increase in the device's carbon emissions; a negative and large gradient indicates a rapid decrease. The trend persistence indicator indicates the duration and stability of the device's carbon emission trend. For example, if a device's carbon emissions consistently show an upward trend and the trend persistence indicator is high, this upward trend is likely to persist for an extended period. Using the carbon emission rate gradient and trend persistence indicator, the state vector comprehensively reflects the evolution of the device's current carbon emission level.

[0071] Step S1422: Based on the historical adjustment records of the device operating parameters, determine the executable adjustment direction and adjustment range combination of the device operating parameters, and construct a candidate action set in the action space.

[0072] Based on historical adjustments to equipment operating parameters, we analyze the impact of different parameter adjustment directions and adjustments on carbon emissions during past equipment operation. For example, for a motor device, historical records show that reducing the motor speed can reduce carbon emissions to a certain extent. Therefore, reducing the speed is an actionable adjustment direction. By statistically analyzing historical data, we identify a series of possible combinations of adjustment directions and adjustment ranges for the equipment operating parameters. These combinations are then used as a set of candidate actions in the action space. Each action in the candidate action set represents a possible adjustment plan for the equipment operating parameters.

[0073] Step S1423: Perform action response simulation on the equipment carbon emission change trend, predict the change in the equipment carbon emission rate after executing each candidate action, and use it as the carbon emission reduction prediction value of the candidate action; analyze the impact of the equipment operating parameter adjustment on the stability of the production process, and predict the degree of production efficiency maintenance after executing the candidate action, and use it as the production efficiency maintenance prediction value of the candidate action.

[0074] We simulate the device's carbon emission trends, simulating the device's carbon emissions after executing each candidate action. By analyzing the device's operating principles and historical data, we predict the change in the device's carbon emission rate after executing each candidate action, using this change as the predicted carbon emission reduction for that candidate action. For example, if a candidate action involves reducing the device's operating power, the simulation can predict the reduction in the device's carbon emission rate.

[0075] The impact of equipment operating parameter adjustments on production process stability is also analyzed. Adjustments to equipment operating parameters may affect production process stability, and thus production efficiency. For example, excessively reducing equipment speed may lead to product backlogs on the production line, impacting production efficiency. By analyzing and simulating the production process, the degree of production efficiency retention after executing the candidate action is predicted, and this retention level is used as the predicted production efficiency retention value for the candidate action.

[0076] Step S1424: Calculate the reward value of each state-action pair based on the predicted value of carbon emission reduction and the predicted value of production efficiency maintenance, combined with a preset weighting coefficient.

[0077] The reward value for each state-action pair is calculated based on the predicted carbon emission reduction and production efficiency maintenance values, combined with preset weighting coefficients. The preset weighting coefficients reflect the importance of carbon emission reduction and production efficiency maintenance in the optimization objective. For example, if carbon emission reduction is more important, a larger weighting coefficient can be set for the predicted carbon emission reduction value; if production efficiency stability is more important, a larger weighting coefficient can be set for the predicted production efficiency maintenance value. The reward value for each state-action pair is calculated by multiplying the predicted carbon emission reduction value by the corresponding weighting coefficient, and then adding the predicted production efficiency maintenance value multiplied by the corresponding weighting coefficient. The higher the reward value, the better the overall performance of the state-action pair in terms of carbon emission reduction and production efficiency maintenance.

[0078] Step S1425: Perform strategy evaluation on the device control subnetwork using the Monte Carlo method. Starting from the current state vector, select actions according to the current strategy and perform simulation, recording the reward value of each step until the preset simulation termination condition is reached.

[0079] The device control subnetwork's policy is evaluated using the Monte Carlo method. Monte Carlo methods are a technique for estimating the behavior of complex systems through random sampling and simulation. Starting from the current state vector, actions are selected according to the current policy and simulations are performed. During the simulation, the reward value for each step is recorded. The simulation continues until a pre-defined termination condition is reached, such as reaching a set number of simulation steps or the simulation results converge to a stable value. Through multiple simulations, the reward distribution for different state-action pairs can be obtained, allowing the effectiveness of the current policy to be evaluated.

[0080] Step S1426: Perform time-discounted accumulation on the reward values ​​recorded during the simulation process to obtain the total reward value of the round, which serves as the estimated carbon emission reduction benefit of the corresponding state-action pair.

[0081] The reward values ​​recorded during the simulation are time-discounted and accumulated. Because the importance of future rewards relative to current rewards decreases over time, rewards need to be time-discounted. The time-discount coefficient is typically less than 1, and as the number of time steps increases, the reward value is gradually discounted. The reward value for each step is multiplied by the corresponding time-discount coefficient and then accumulated to obtain the total reward value for that round. This total reward value is used as the estimated carbon reduction benefit for the corresponding state-action pair. A higher estimated carbon reduction benefit indicates that the state-action pair performs better in terms of long-term carbon reduction and maintaining production efficiency.

[0082] Step S1427: Based on the estimated carbon emission reduction benefits obtained from multiple rounds of simulation, update the parameters related to the state-action pairs in the device control subnetwork so that subsequent strategy selection is more inclined to actions with higher carbon emission reduction benefits.

[0083] Based on the estimated carbon reduction benefits from multiple simulation rounds, the parameters related to state-action pairs in the device control subnetwork are updated. By adjusting these parameters, the device control subnetwork is more likely to select actions with higher carbon reduction benefits in subsequent policy selections. For example, if a state-action pair has a higher estimated carbon reduction benefit, the probability of that state-action pair in policy selection is increased. By continuously updating parameters, the device control subnetwork can gradually learn the optimal policy, improving carbon reduction effectiveness and the stability of production efficiency.

[0084] Step S1428: Repeat the strategy evaluation and parameter update process until the parameters of the device control subnetwork converge to a stable state, and finally select the candidate action with the greatest carbon emission reduction benefit as the device operation parameter adjustment plan to generate the device energy efficiency control strategy.

[0085] The device control subnetwork is continuously optimized by repeating the policy evaluation and parameter update process. As the number of iterations increases, the parameters of the device control subnetwork gradually converge to a stable state. When the parameters converge to a stable state, the device control subnetwork has learned the optimal policy. Finally, the candidate action with the greatest carbon emission reduction benefit is selected from the candidate action set as the device operating parameter adjustment plan, which is then used as the device energy efficiency control strategy. The device energy efficiency control strategy can guide enterprises in adjusting device operating parameters to reduce carbon emissions and improve energy efficiency.

[0086] Step S143: Utilize the production planning subnetwork to take the changing trend of raw material carbon conversion efficiency and the changing trend of carbon loss in the emission link as input, evaluate the comprehensive benefits of different raw material input sequences and production step reorganization plans through a deep reinforcement learning model, and finally output the reorganization plan with the highest comprehensive benefits as the production plan reorganization strategy.

[0087] The production planning subnetwork takes the changing trends of raw material carbon conversion efficiency and carbon loss in the emission link as input, and uses a deep reinforcement learning model to evaluate the comprehensive benefits of different raw material input sequences and production step reorganization schemes.

[0088] Step S1431: Input the changing trend of raw material carbon conversion efficiency and the changing trend of carbon loss in the emission link into the state encoding module of the deep reinforcement learning model to generate a state vector containing the fluctuation characteristics of the raw material conversion rate and the changing pattern of the emission loss.

[0089] The changing trends in the raw material carbon conversion efficiency and the carbon loss trends in the emissions process are input into the state encoding module of the deep reinforcement learning model. The state encoding module analyzes and encodes these two trends, generating a state vector that contains the fluctuation characteristics of the raw material conversion rate and the variation pattern of the emission loss. The fluctuation characteristics of the raw material conversion rate reflect the changes in the carbon conversion efficiency of the raw material during the production process, such as the upward or downward trend of the conversion rate and the amplitude of the fluctuation. The variation pattern of the emission loss reflects the change pattern of the carbon loss in the emission process over time, such as whether there are periodic fluctuations and whether the trend of change is increasing or decreasing. The state vector can comprehensively describe the carbon emission characteristics of the raw materials and the emission process in the current production process.

[0090] Step S1432: Based on the standard execution process of the production plan, determine the adjustable raw material input time nodes and production step execution sequence combinations, and construct a candidate action set in the action space. The adjustment of the raw material input time nodes includes early adjustment of the time nodes and delayed adjustment of the time nodes. The adjustment of the production step execution sequence includes adjacent step sequence exchange and non-adjacent step sequence exchange.

[0091] Based on the standard execution process of the production plan, analyze which raw material input time nodes and production step execution order can be adjusted. The adjustment of raw material input time nodes includes advancing or delaying the time nodes. For example, in chemical production, early input of a certain raw material may speed up the reaction and improve carbon conversion efficiency; delayed input may avoid the occurrence of certain side reactions and reduce carbon emissions. The adjustment of the execution order of production steps includes swapping the order of adjacent steps and swapping the order of non-adjacent steps. Swapping the order of adjacent steps can change the local order of the production process, while swapping the order of non-adjacent steps can make more significant adjustments to the production process. All adjustable raw material input time nodes and production step execution orders are combined as a set of candidate actions in the action space.

[0092] Step S1433: Predict the impact of actions on the changing trends of raw material carbon conversion efficiency and carbon loss in the emission link, calculate the predicted value of carbon emission reduction after executing each candidate action, and analyze the impact of the candidate action on the production process time. Combined with the order delivery time requirements, calculate the predicted value of order delivery on time rate after executing the candidate action.

[0093] The impact of actions is predicted based on the changing trends in raw material carbon conversion efficiency and carbon losses in the emission process. By analyzing the principles of the production process and historical data, the carbon emission reduction changes after executing each candidate action are predicted, and the predicted carbon emission reduction value for each candidate action is calculated. For example, if a candidate action involves adjusting the order of raw material input, the simulation can predict that the carbon conversion efficiency of the raw materials will increase after the adjustment, thereby reducing carbon emissions.

[0094] The impact of candidate actions on production process time is also analyzed. Some candidate actions may increase or decrease production process time, which will affect order delivery time. Based on order delivery time requirements, the on-time delivery rate of the order after executing the candidate action is predicted. For example, if a candidate action increases production process time, it may cause order delivery delays, thereby reducing the on-time delivery rate.

[0095] Step S1434: Calculate the reward value of each state-action pair based on the predicted value of carbon emission reduction and the predicted value of order delivery on-time rate, combined with the first weight coefficient and the second weight coefficient.

[0096] The reward value for each state-action pair is calculated based on the predicted carbon emission reduction and on-time delivery rate, combined with the first and second weighting coefficients. The first and second weighting coefficients respectively reflect the importance of carbon emission reduction and on-time delivery in the optimization objective. The reward value for each state-action pair is calculated by multiplying the predicted carbon emission reduction by the first weighting coefficient and then adding the predicted on-time delivery rate by the second weighting coefficient. A higher reward value indicates a better overall performance of the state-action pair in terms of carbon emission reduction and on-time delivery.

[0097] Step S1435: Construct a deep Q network, whose input layer receives the state vector, the hidden layer extracts the state features through multi-layer nonlinear transformation, and the output layer outputs the comprehensive benefit prediction value corresponding to each candidate action.

[0098] A Deep Q-Network (DQN) is a neural network structure used for reinforcement learning. Its input layer receives a state vector as input data. The hidden layer processes the state vector through multiple layers of nonlinear transformations to extract state features. Nonlinear transformations increase the network's expressive power, enabling it to learn more complex features. The output layer outputs the predicted comprehensive benefit value for each candidate action. This value reflects the combined effect of executing the candidate action on carbon emissions reduction and on-time order delivery.

[0099] Step S1436: During the production plan adjustment process, the state vector, the selected candidate action, the reward value obtained, and the new state vector after executing the candidate action are recorded to form an experience data sample.

[0100] During the production plan adjustment process, relevant data information is recorded. The current state vector, the selected candidate action, the reward value obtained, and the new state vector after executing the candidate action are recorded. This data information is combined together to form an experience data sample. The experience data sample records the various states and results during the production plan adjustment process. For example, in a machinery manufacturing enterprise, when adjusting the production plan of a certain product, the state vector corresponding to the raw material carbon conversion efficiency and the carbon loss amount in the emission link during the current product production process is recorded. The candidate action selected is to advance the input time of a certain raw material, the reward value obtained after executing this action, and the new state vector corresponding to the new raw material carbon conversion efficiency and the carbon loss amount in the emission link. This information is integrated into an experience data sample.

[0101] Step S1437: Randomly sample batch data from the experience data sample and calculate the target comprehensive benefit of each sample. The target comprehensive benefit is the current reward value plus the discount factor multiplied by the maximum comprehensive benefit prediction value of the next state.

[0102] A batch of data is randomly sampled from the empirical data sample, and this batch of data contains multiple empirical data samples. For each sampled sample, its target comprehensive benefit is calculated. The target comprehensive benefit is calculated by adding the current reward value to the discount factor multiplied by the maximum comprehensive benefit prediction value of the next state. The discount factor is a coefficient less than 1, which reflects the degree of emphasis on future comprehensive benefits. The maximum comprehensive benefit prediction value of the next state refers to the maximum value of the comprehensive benefit prediction values ​​corresponding to all candidate actions predicted by the deep Q network in the new state entered after executing the current candidate action. For example, in the current sample, executing a candidate action obtains a certain current reward value. Then, based on the deep Q network's prediction of the comprehensive benefits of all candidate actions after entering the new state, the maximum value is found, multiplied by the discount factor, and added to the current reward value to obtain the target comprehensive benefit of the sample.

[0103] Step S1438: Calculate the difference between the comprehensive benefit prediction value output by the deep Q network and the target comprehensive benefit to generate a loss function value.

[0104] The comprehensive benefit prediction value corresponding to each candidate action output by the deep Q network is compared with the target comprehensive benefit calculated in step S1437, and the difference between them is calculated. The difference value can be calculated using common error calculation methods, such as mean square error. The difference values ​​of all samples are summarized and processed to generate a loss function value. The loss function value reflects the degree of deviation between the prediction result of the deep Q network and the target comprehensive benefit. If the loss function value is large, it means that there is a large difference between the prediction result of the deep Q network and the actual situation, and the network needs to be adjusted; if the loss function value is small, it means that the prediction result of the deep Q network is relatively accurate.

[0105] Step S1439: Use the adaptive optimization algorithm to backpropagate and update the parameters of the deep Q network to minimize the loss function value.

[0106] An adaptive optimization algorithm is used to update the parameters of the deep Q network. The adaptive optimization algorithm can automatically adjust the update step size and direction of the network parameters based on the changes in the loss function value, so that the loss function value gradually decreases. Backpropagation is a commonly used neural network training method. It calculates the gradient of the loss function with respect to the network parameters and propagates the gradient information from the output layer to the input layer, thereby updating the network parameters. During each backpropagation process, the adaptive optimization algorithm adjusts the update amplitude of the network parameters based on the current loss function value and gradient information, so that the network can converge to the optimal solution more quickly. For example, common adaptive optimization algorithms such as Adagrad, Adadelta, and Adam can dynamically adjust the learning rate of each parameter based on the historical update of the network parameters, thereby improving training efficiency and stability.

[0107] Step S14310: Repeat the data recording, sampling, loss calculation and parameter updating process until the comprehensive benefit prediction value of the deep Q network converges.

[0108] The process from step S1436 to step S1439 is continuously repeated, i.e., the empirical data samples during the production plan adjustment process are continuously recorded, batch data is randomly sampled from the samples, the target comprehensive benefit and loss function value are calculated, and the parameters of the deep Q network are updated using an adaptive optimization algorithm. As the number of iterations increases, the comprehensive benefit prediction value of the deep Q network gradually stabilizes. When the change in the comprehensive benefit prediction value is less than a preset threshold, the comprehensive benefit prediction value of the deep Q network is considered to have converged. This indicates that the deep Q network has learned the intrinsic relationship between production plan adjustment and comprehensive benefit and can more accurately predict the comprehensive benefit of different candidate actions.

[0109] Step S14311: For the current state vector, the comprehensive benefit prediction value of all candidate actions is obtained through the deep Q network, and the candidate action with the largest comprehensive benefit is selected as the optimal solution for the raw material input sequence and production step reorganization to generate a production plan reorganization strategy.

[0110] After the Deep Q-Network's comprehensive benefit predictions converge, the current state vector is input into the Deep Q-Network. The Deep Q-Network outputs comprehensive benefit predictions for all candidate actions. From these predictions, the candidate action with the highest comprehensive benefit prediction is selected and used as the optimal solution for reorganizing the raw material input sequence and production steps. This optimal solution takes into account factors such as the changing trends in raw material carbon conversion efficiency, the changing trends in carbon losses in the emission process, and order delivery time requirements. It can ensure smooth production and on-time order delivery while reducing carbon emissions. This optimal solution is used as a production plan reorganization strategy to guide companies in adjusting their production plans. For example, in an electronics manufacturer, based on the current state vectors of raw material carbon conversion efficiency and carbon losses in the emission process, the Deep Q-Network predicts that the candidate action of swapping the order of several production steps, which has the highest comprehensive benefit, will be the one that reorders the production steps. This production step reordering solution is then used as the production plan reorganization strategy.

[0111] Step S144: Integrate the equipment energy efficiency control strategy and the production plan reorganization strategy into a carbon asset optimization strategy set.

[0112] The equipment energy efficiency control strategies and production plan reorganization strategies obtained through the above steps are integrated to form a set of carbon asset optimization strategies. The equipment energy efficiency control strategy mainly adjusts the operating parameters of the equipment to improve the equipment's energy utilization efficiency and reduce carbon emissions; the production plan reorganization strategy focuses on optimizing the raw material input time nodes and the execution sequence of production steps, thereby reducing carbon emissions in the overall production process. Integrating these two strategies can comprehensively optimize the carbon assets of the entire industrial manufacturing process from both the equipment and production planning levels. For example, in an automobile manufacturing company, the equipment energy efficiency control strategy may include adjusting the operating power of the stamping equipment and optimizing the heating temperature of the painting equipment; the production plan reorganization strategy may include adjusting the input time of raw materials such as steel and plastic, and changing the execution sequence of production steps such as welding and assembly. These strategies are integrated into a set of carbon asset optimization strategies.

[0113] Step S150: Pushing the carbon asset optimization strategy set to the industrial manufacturing execution system to trigger an automatic control instruction reconstruction operation.

[0114] The integrated carbon asset optimization strategy set is pushed to the industrial manufacturing execution system to trigger the reconstruction operation of the automatic control instructions, thereby realizing the automated adjustment of the industrial manufacturing process and achieving the goal of reducing carbon emissions and optimizing carbon assets.

[0115] Step S151: Convert the operating parameter adjustment direction, adjustment range and time node information in the equipment energy efficiency control strategy into the equipment control protocol format of the industrial manufacturing execution system, and generate equipment control data including the equipment identifier and parameter adjustment instructions.

[0116] First, the operating parameter adjustment direction, adjustment range, and time point information in the equipment energy efficiency control strategy are processed. The industrial manufacturing execution system has its own set equipment control protocol format, and the information in the equipment energy efficiency control strategy needs to be converted according to this equipment control protocol format. The equipment control protocol specifies the data transmission format, encoding method, and other details. The generated equipment control data contains a device identifier and parameter adjustment instructions. The device identifier uniquely identifies each device in the industrial manufacturing process. For example, each machine tool and each motor has a corresponding device identifier. The parameter adjustment instructions specify which operating parameters of the device to adjust, whether to increase or decrease the adjustment, the adjustment range, and the time point at which the adjustment should be made. For example, for an injection molding machine, the equipment control data will include the device identifier of the injection molding machine and specific instructions for adjusting its operating parameters such as injection pressure and injection speed. It also specifies that these adjustments should be made at the beginning of a specific production batch.

[0117] Step S152: Convert the raw material input time node adjustment information and the production step execution sequence adjustment information in the production plan reorganization strategy into a production scheduling protocol format, and generate production scheduling data including a production order number and process adjustment instructions.

[0118] For the production plan reorganization strategy, the information on adjusting raw material input timing nodes and production step execution sequence is converted into a production scheduling protocol format. The production scheduling protocol is a protocol used in industrial manufacturing execution systems to schedule production tasks and resources. It specifies the management of production orders, the scheduling of processes, and other aspects. The generated production scheduling data contains production order numbers and process adjustment instructions. The production order number identifies each specific production order, which corresponds to a certain number of product production tasks. The process adjustment instructions specify adjustments to raw material input timing nodes, such as advancing or delaying the input of a certain raw material; and adjustments to the execution sequence of production steps, such as swapping the order of two adjacent processes or adjusting the order of non-adjacent processes. For example, in a furniture manufacturing company, production scheduling data may include the number of a furniture production order, as well as instructions for adjusting the execution sequence of processes such as wood cutting, assembly, and painting, and the input times of raw materials such as board and paint.

[0119] Step S153: The equipment control data and production scheduling data are transmitted to the equipment control module and the production scheduling module of the industrial manufacturing execution system, so that after the equipment control module parses the equipment control data, it modifies the control parameter configuration of the corresponding equipment, triggering the equipment operation status adjustment operation, and after the production scheduling module parses the production scheduling data, it updates the raw material delivery time and process execution sequence in the production schedule, triggering the reorganization operation of the raw material flow path and the production step sequence.

[0120] The generated equipment control data and production scheduling data are transmitted to the equipment control module and production scheduling module of the industrial manufacturing execution system, respectively. After receiving the equipment control data, the equipment control module can parse it. This parsing process extracts the device identifier and parameter adjustment instructions based on the rules of the device control protocol. It then locates the corresponding device based on the device identifier and modifies the control parameter configuration for that device. For example, the operating power of a particular device can be adjusted from the original configured value to the new target value. This control parameter modification triggers an adjustment to the device's operating status, causing the device to operate according to the new parameters, thereby optimizing its energy efficiency. The production scheduling module also parses the production scheduling data, extracting the production order number and process adjustment instructions. Based on the production order number, it locates the corresponding production schedule and updates the raw material delivery time and process execution order in the production schedule. For example, the delivery time of a particular raw material can be advanced from the original planned time, or the execution order of two processes can be swapped. After updating the production schedule, the reorganization of the raw material flow path and the production step sequence is triggered. The raw materials will flow according to the new time nodes and paths, and the production steps will be executed in the new order, thereby optimizing the production plan and reducing carbon emissions.

[0121] Step S154: The actual values ​​of the equipment operating parameters and the actual execution order of the production steps are obtained in real time through the status acquisition module of the industrial manufacturing execution system, and compared and analyzed with the target parameters of the carbon asset optimization strategy set to generate feedback data including adjustment deviations.

[0122] The status acquisition module of the industrial manufacturing execution system collects the actual values ​​of equipment operating parameters and the actual sequence of production steps in real time. These values ​​can be collected by sensors installed on the equipment, such as motor current, voltage, and speed. The actual sequence of production steps can be recorded by monitoring equipment on the production line or through the production management system. These values ​​are compared and analyzed with the target parameters in the carbon asset optimization strategy. The target parameters are the ideal parameter values ​​and sequence defined in the equipment energy efficiency control strategy and the production plan reorganization strategy. This comparative analysis calculates the deviations between the actual and target values, such as the difference between the actual values ​​of equipment operating parameters and the adjusted target values, and the difference between the actual sequence of production steps and the adjusted target sequence. This deviation information is collated and summarized to generate feedback data containing the adjusted deviations. This feedback data can be used to evaluate the effectiveness of the carbon asset optimization strategy and identify implementation issues. For example, if the feedback data shows that there is a large deviation between the actual operating power of a certain device and the target power, it means that there may be a problem with the device control, and further inspection of the device or adjustment of the control parameters is required; if the actual execution order of the production steps is inconsistent with the target order, there may be an error in the production scheduling process, and the production scheduling module needs to be adjusted.

[0123] Furthermore, for example, the method may further include: Step S210: Collect a sample carbon activity data set of the entire historical industrial manufacturing process and the corresponding actual carbon asset evolution path data, and construct a training data set containing input carbon footprint feature data and output evolution path labels.

[0124] To pre-train the deep learning model, it is necessary to collect a set of sample carbon activity data from the entire historical industrial manufacturing process and the corresponding actual carbon asset evolution path data. This sample carbon activity data set, similar to the original carbon activity data set obtained in step S110, contains data from energy consumption recording units, production raw material flow units, and emission monitoring units. This data is collected from historical industrial manufacturing processes and records energy consumption, production raw material flow, and emission monitoring over different time periods. The actual carbon asset evolution path data is the actual evolution of carbon assets throughout the entire industrial manufacturing process, as recorded in historical data. For example, it includes changes in carbon emissions from equipment at different time points, changes in carbon conversion efficiency of raw materials, and changes in carbon losses in the emission process. The sample carbon activity data set is processed and input carbon footprint feature data is generated according to the method in step S120. The actual carbon asset evolution path data is used as the output evolution path label. The input carbon footprint feature data and the output evolution path label are paired to construct a training dataset. For example, a chemical company collects energy consumption, raw material flow, and emission monitoring data for each production batch over the past few years to generate corresponding carbon footprint characteristics. Simultaneously, the actual evolution path of equipment carbon emissions, raw material carbon conversion, and emission losses for each production batch is recorded as evolution path labels. These data are paired to form a training dataset for deep learning model training.

[0125] Step S220: Initialize the convolution kernel parameters of the multi-scale feature extraction module, the attention head parameters and gated recurrent unit parameters of the spatiotemporal dependency modeling module, and the decoding layer parameters of the evolution prediction module.

[0126] Before training a deep learning model, the parameters of each module must be initialized. The multi-scale feature extraction module primarily uses convolution operations to downsample the carbon footprint feature data, so the convolution kernel parameters must be initialized. The convolution kernel parameters determine the method and effect of the convolution operation; different convolution kernel parameters extract different features. The spatiotemporal dependency modeling module includes a spatial self-attention layer and a temporal gated recurrent unit. The attention head parameters control how the attention weights in the spatial self-attention layer are calculated, while the gated recurrent unit parameters control how the temporal gated recurrent unit processes sequence data. The evolution prediction module uses a decoding layer to perform multi-step decoding of the evolution feature tensor, so the decoding layer parameters must be initialized. These parameters determine the decoding method and output format. Initialization parameters can be random or based on prior knowledge. For example, for the convolution kernel parameters, Gaussian distribution can be used for random initialization to make the initial value of the convolution kernel have a certain degree of randomness and diversity; for the attention head parameters and gated recurrent unit parameters, some initial values ​​can be set based on experience to ensure that the initial state of the model can work normally.

[0127] Step S230: Input the carbon footprint feature data in the training data set into the deep learning model to generate a predicted evolution path result.

[0128] The carbon footprint feature data from the constructed training dataset is input into the deep learning model. The deep learning model processes the input carbon footprint feature data according to its internal structure and algorithm. First, the multi-scale feature extraction module downsamples the carbon footprint feature data to generate a multi-scale feature set with varying degrees of refinement. Then, the spatiotemporal dependency modeling module performs spatiotemporal joint processing on the multi-scale feature set to generate an evolutionary feature tensor containing spatiotemporal coupling information. Finally, the evolutionary prediction module performs multi-step decoding on the evolutionary feature tensor and, combined with the constraints of the current production plan, generates a predicted evolutionary path. This predicted evolutionary path includes predictions of the changing trends in equipment carbon emissions, raw material carbon conversion efficiency, and carbon losses in the emission process over the future production cycle. For example, it is predicted that over a period of time, the carbon emissions of a particular piece of equipment will gradually decrease, the carbon conversion efficiency of a particular raw material will improve, and carbon losses in the emission process will remain at a low level.

[0129] Step S240: Calculate the loss value between the predicted evolution path result and the evolution path label. The loss value measures the similarity difference between the predicted evolution path result and the evolution path label using a dynamic time warping algorithm.

[0130] In order to evaluate the prediction accuracy of the deep learning model, it is necessary to calculate the loss value between the predicted evolution path result and the evolution path label. The dynamic time warping algorithm is used here to measure the similarity difference between the two. The dynamic time warping algorithm is an algorithm for comparing the similarity of two time series. It can flexibly match the two sequences on the time axis to find the optimal matching path. In this embodiment, the predicted evolution path result and the evolution path label are both time series data, representing the carbon asset evolution path predicted by the deep learning model and the actual carbon asset evolution path, respectively. The dynamic time warping algorithm calculates the difference between the predicted evolution path result and the evolution path label at each time point, and then finds an optimal matching path through dynamic programming so that the sum of the differences along this path is minimized. The minimum sum of differences is used as the loss value. The smaller the loss value, the more similar the predicted evolution path result is to the evolution path label, and the higher the prediction accuracy of the deep learning model; the larger the loss value, the larger the prediction result of the deep learning model is compared with the actual situation, and the deep learning model needs to be adjusted.

[0131] Step S250: Adopting the adaptive optimization algorithm to back-propagate and update the network parameters of the multi-scale feature extraction module, the spatiotemporal dependency modeling module and the evolution prediction module until the loss value converges to a stable range, thereby completing the pre-training of the deep learning model.

[0132] An adaptive optimization algorithm is used to update the network parameters of the deep learning model. This algorithm automatically adjusts the update step size and direction based on changes in the loss value. Backpropagation is a commonly used neural network training method. It calculates the gradient of the loss function with respect to the network parameters and propagates the gradient information back from the output layer to the input layer, thereby updating the network parameters. During each backpropagation step, the adaptive optimization algorithm adjusts the network parameters of the multi-scale feature extraction module, the spatiotemporal dependency modeling module, and the evolution prediction module based on the current loss value and gradient information. As the number of iterations increases, the loss value gradually decreases. When the loss value converges to a stable range, meaning that the change in the loss value is less than a preset threshold, the deep learning model is considered to have reached a good state and pre-training is complete. At this point, the network parameters of the deep learning model are able to effectively capture the intrinsic relationship between carbon footprint feature data and the carbon asset evolution path, and can accurately predict the carbon asset evolution path within the future production cycle. For example, after multiple training iterations, the loss value gradually decreases from an initial high value and stabilizes within a small range, indicating that the model's predictive accuracy is continuously improving and the pre-training goal has been achieved.

[0133] Figure 2A schematic diagram illustrates exemplary hardware and software components of a carbon asset lifecycle management system 100 for industrial manufacturing, provided in some embodiments of the present application, that can implement the concepts of the present application. For example, a processor 120 can be used in the carbon asset lifecycle management system 100 for industrial manufacturing to perform the functions described in the present application.

[0134] The carbon asset lifecycle management system 100 for industrial manufacturing can be a general-purpose server or a special-purpose server, both of which can be used to implement the carbon asset lifecycle management method for industrial manufacturing described herein. Although only one server is shown in this application, for convenience, the functions described herein can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0135] For example, the carbon asset life cycle management system 100 for the industrial manufacturing industry may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the carbon asset life cycle management system 100 for the industrial manufacturing industry may also include program instructions stored in ROM, RAM, or other types of non-temporary storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The carbon asset life cycle management system 100 for the industrial manufacturing industry also includes an I / O interface 150 between the computer and other input and output devices.

[0136] For ease of explanation, only one processor is described in the carbon asset life cycle management system 100 for industrial manufacturing. However, it should be noted that the carbon asset life cycle management system 100 for industrial manufacturing in this application may also include multiple processors, so the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the carbon asset life cycle management system 100 for industrial manufacturing executes step A and step B, it should be understood that step A and step B may also be executed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0137] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned carbon asset full life cycle management method for the industrial manufacturing industry is implemented.

[0138] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A carbon asset life cycle management method for industrial manufacturing, characterized by: The method comprises: Obtaining a set of raw carbon activity data for the entire industrial manufacturing process, wherein the raw carbon activity data set includes an energy consumption recording unit, a production raw material flow unit, and an emission monitoring unit; Performing cross-link feature fusion processing on the original carbon activity data set to generate carbon footprint feature data for the entire industrial manufacturing process, wherein the carbon footprint feature data includes static facility carbon emission distribution features and dynamic operation carbon emission evolution features; Calling a pre-trained deep learning model to perform carbon asset state evolution analysis on the carbon footprint feature data to generate a carbon asset evolution path prediction result; Generating a carbon asset optimization strategy set for the entire industrial manufacturing process based on the carbon asset evolution path prediction result, wherein the carbon asset optimization strategy set includes an equipment energy efficiency control strategy and a production plan reorganization strategy; The carbon asset optimization strategy set is pushed to the industrial manufacturing execution system to trigger the automatic control instruction reconstruction operation.

2. The carbon asset full life cycle management method for industrial manufacturing according to claim 1 is characterized in that: The performing of cross-link feature fusion processing on the original carbon activity data set to generate carbon footprint feature data for the entire industrial manufacturing process includes: The energy consumption recording unit is input into a time series feature encoder, which extracts energy consumption fluctuation characteristics in different time windows by processing the continuous time series energy consumption data to generate an energy consumption time series feature vector. The time series feature encoder is composed of a bidirectional long short-term memory network. The production raw material flow unit is input into the graph structure encoder, and the attention weight between each node is calculated through the graph attention network, the raw material carbon transfer feature is extracted, and the raw material flow graph feature vector is generated. The graph structure encoder uses raw material batches as nodes and flow relationships as edges; The emission monitoring unit is input into a convolutional feature encoder, and an emission monitoring spatial feature vector is generated by extracting a spatial correlation pattern between emission concentration and flow rate. The convolutional feature encoder includes multiple convolution kernels of different sizes. The energy consumption time series feature vector, the raw material flow graph feature vector, and the emission monitoring space feature vector are input into the attention fusion layer. By calculating the mutual attention weight matrix between the energy consumption time series feature vector, the raw material flow graph feature vector, and the emission monitoring space feature vector, the energy consumption time series feature vector, the raw material flow graph feature vector, and the emission monitoring space feature vector are weightedly aggregated to generate a fusion feature matrix containing the correlation information between the links; The fused feature matrix is ​​input into the static feature extraction branch and the dynamic feature extraction branch respectively. The static feature extraction branch is used to extract the statistical characteristics of the carbon emission distribution of equipment, raw materials, and emission links through global average pooling operation as the static facility carbon emission distribution characteristics. The dynamic feature extraction branch is used to extract the carbon emission evolution trend characteristics within the production cycle through the time convolution network as the dynamic operation carbon emission evolution characteristics, and finally spliced ​​to form the carbon footprint characteristic data.

3. The carbon asset full life cycle management method for industrial manufacturing according to claim 1 is characterized in that: The deep learning model includes a multi-scale feature extraction module, a spatiotemporal dependency modeling module, and an evolution prediction module. The pre-trained deep learning model is called to perform carbon asset state evolution analysis and processing on the carbon footprint feature data to generate a carbon asset evolution path prediction result, including: The carbon footprint feature data is downsampled using a multi-scale feature extraction module to generate a multi-scale feature set with different levels of refinement. The first-scale feature reflects the global carbon emissions overview, while the second-scale feature reflects the detailed carbon emissions of the local link. The spatiotemporal dependency modeling module performs spatiotemporal joint processing on the multi-scale feature set to generate an evolutionary feature tensor containing spatiotemporal coupling information; The evolutionary feature tensor is subjected to multi-step decoding processing through the evolutionary prediction module. Combined with the constraints of the current production plan, the changing trends of equipment carbon emissions, raw material carbon conversion efficiency and carbon loss in the emission link in the future production cycle are predicted stage by stage, and the carbon asset evolution path prediction results are generated, which include the carbon emission indicators of key nodes in each stage.

4. The carbon asset full life cycle management method for industrial manufacturing according to claim 3 is characterized in that: The spatiotemporal joint processing of the multi-scale feature set by the spatiotemporal dependency modeling module to generate an evolution feature tensor containing spatiotemporal coupling information includes: The multi-scale feature set is input into the spatial self-attention layer, the attention weight of each spatial position and other spatial positions is calculated, and the spatial features reflecting the interaction relationship between links are generated; The spatial features are input into the time-gated recurrent unit and processed time-step by time-step along the production time axis to capture the influence of the previous time-step features on the current time-step features and generate time features that reflect the recursive relationship between the stages. The spatial features and temporal features are fused at the element level to generate an evolutionary feature tensor containing spatiotemporal coupling information.

5. The carbon asset full life cycle management method for industrial manufacturing according to claim 3 is characterized in that: The pre-training process of the deep learning model includes: Collect sample carbon activity data sets from the entire historical industrial manufacturing process and the corresponding actual carbon asset evolution path data, and construct a training dataset containing input carbon footprint feature data and output evolution path labels; Initialize the convolution kernel parameters of the multi-scale feature extraction module, the attention head parameters and gated recurrent unit parameters of the spatiotemporal dependency modeling module, and the decoding layer parameters of the evolution prediction module; Input the carbon footprint feature data in the training dataset into the deep learning model to generate the predicted evolution path results; Calculate the loss value between the predicted evolution path result and the evolution path label, and measure the similarity difference between the predicted evolution path result and the evolution path label using a dynamic time warping algorithm; The network parameters of the multi-scale feature extraction module, the spatiotemporal dependency modeling module, and the evolution prediction module are updated by back propagation using an adaptive optimization algorithm until the loss value converges to a stable range, thereby completing the pre-training of the deep learning model.

6. The carbon asset full life cycle management method for industrial manufacturing according to claim 1 is characterized in that: The carbon asset optimization strategy set for the entire industrial manufacturing process is generated based on the carbon asset evolution path prediction result, including: Inputting the carbon asset evolution path prediction result into a strategy generation model, wherein the strategy generation model includes an equipment control subnetwork and a production planning subnetwork; The equipment control subnetwork uses the equipment carbon emission change trend as input, calculates the carbon emission reduction benefits of different equipment operating parameter adjustment plans through a reinforcement learning framework, and ultimately outputs the adjustment plan with the highest carbon emission reduction benefit as the equipment energy efficiency control strategy; The production planning subnetwork uses the changing trends of raw material carbon conversion efficiency and carbon loss in the emission link as inputs, and uses a deep reinforcement learning model to evaluate the comprehensive benefits of different raw material input sequences and production step reorganization plans, ultimately outputting the reorganization plan with the highest comprehensive benefits as the production plan reorganization strategy; The equipment energy efficiency control strategy and production plan reorganization strategy are integrated into a carbon asset optimization strategy set.

7. The carbon asset life cycle management method for industrial manufacturing according to claim 6 is characterized in that: The device control subnetwork uses the device carbon emission change trend as input, calculates the carbon emission reduction benefits of different device operating parameter adjustment plans through a reinforcement learning framework, and ultimately outputs the adjustment plan with the highest carbon emission reduction benefit as the device energy efficiency control strategy, including: Inputting the device carbon emission change trend into the state representation module of the reinforcement learning framework to generate a state vector reflecting the evolution law of the current device carbon emission level, wherein the state vector includes the carbon emission rate change gradient and the trend persistence indicator; Based on the historical adjustment records of equipment operating parameters, determine the executable adjustment direction and adjustment range combination of equipment operating parameters, and construct a set of candidate actions in the action space; Conduct action response simulations on the equipment's carbon emission change trends, predict the change in the equipment's carbon emission rate after executing each candidate action, and use this as the predicted carbon emission reduction value for that candidate action. Analyze the impact of equipment operating parameter adjustments on production process stability, and predict the degree of production efficiency retention after executing that candidate action, and use this as the predicted production efficiency retention value for that candidate action. Calculating a reward value for each state-action pair based on the predicted carbon emission reduction value and the predicted production efficiency maintenance value in combination with a preset weighting coefficient; The device control subnetwork is evaluated using the Monte Carlo method. Starting from the current state vector, actions are selected according to the current strategy and simulation is performed. The reward value of each step is recorded until the preset simulation termination condition is reached. The reward values ​​recorded during the simulation are accumulated with time discount to obtain the total reward value of the round, which is used as the estimated carbon emission reduction benefit of the corresponding state-action pair; Based on the estimated carbon reduction benefits obtained from multiple rounds of simulation, the parameters related to the state-action pairs in the device control subnetwork are updated so that subsequent policy selection is more inclined towards actions with higher carbon reduction benefits. The strategy evaluation and parameter update process is repeated until the parameters of the device control subnetwork converge to a stable state. Finally, the candidate action with the greatest carbon emission reduction benefit is selected as the device operation parameter adjustment plan to generate the device energy efficiency control strategy.

8. The carbon asset full life cycle management method for industrial manufacturing according to claim 6 is characterized in that: The production planning subnetwork uses the changing trend of raw material carbon conversion efficiency and the changing trend of carbon loss in the emission link as input, evaluates the comprehensive benefits of different raw material input sequences and production step reorganization plans through a deep reinforcement learning model, and ultimately outputs the reorganization plan with the highest comprehensive benefit as the production plan reorganization strategy, including: The changing trends of raw material carbon conversion efficiency and carbon loss in the emission process are input into the state encoding module of the deep reinforcement learning model to generate a state vector containing the fluctuation characteristics of the raw material conversion rate and the changing pattern of the emission loss; Based on the standard execution process of the production plan, determine the adjustable raw material input time nodes and production step execution sequence combinations, and construct a candidate action set in the action space. The adjustment of the raw material input time nodes includes time node advance adjustment and time node delay adjustment. The adjustment of the production step execution sequence includes the sequence swapping of adjacent steps and the sequence swapping of non-adjacent steps. Predict the impact of actions on the changing trends of raw material carbon conversion efficiency and carbon loss in the emission link, calculate the predicted carbon emission reduction value after executing each candidate action, and analyze the impact of candidate actions on production process time. Combined with order delivery time requirements, calculate the predicted order delivery on-time rate after executing the candidate action; Calculate the reward value of each state-action pair based on the predicted value of carbon emission reduction and the predicted value of order delivery on-time rate, combined with the first weight coefficient and the second weight coefficient; Construct a deep Q network, whose input layer receives the state vector, the hidden layer extracts state features through multi-layer nonlinear transformation, and the output layer outputs the comprehensive benefit prediction value corresponding to each candidate action; During the production plan adjustment process, the state vector, the selected candidate action, the reward value obtained, and the new state vector after executing the candidate action are recorded to form an experience data sample; Randomly sample batches of data from the empirical data sample and calculate the target comprehensive benefit of each sample, where the target comprehensive benefit is the current reward value plus the discount factor multiplied by the maximum comprehensive benefit prediction value of the next state; Calculate the difference between the comprehensive benefit prediction value output by the deep Q network and the target comprehensive benefit to generate the loss function value; Use the adaptive optimization algorithm to back-propagate and update the parameters of the deep Q network to minimize the loss function value; Repeating the data recording, sampling, loss calculation, and parameter updating process until the comprehensive benefit prediction value of the deep Q network converges; For the current state vector, the comprehensive benefit prediction values ​​of all candidate actions are obtained through the deep Q network, and the candidate action with the largest comprehensive benefit is selected as the optimal solution for the raw material input sequence and production step reorganization to generate a production plan reorganization strategy.

9. The carbon asset life cycle management method for industrial manufacturing according to claim 1 is characterized in that: The step of pushing the carbon asset optimization strategy set to the industrial manufacturing execution system to trigger the automatic control instruction reconstruction operation includes: Convert the operating parameter adjustment direction, adjustment range, and time node information in the equipment energy efficiency control strategy into the equipment control protocol format of the industrial manufacturing execution system, and generate equipment control data containing equipment identifiers and parameter adjustment instructions; Convert the raw material input time node adjustment information and production step execution sequence adjustment information in the production plan reorganization strategy into the production scheduling protocol format, and generate production scheduling data containing the production order number and process adjustment instructions; Transmitting equipment control data and production scheduling data to the equipment control module and production scheduling module of the industrial manufacturing execution system, so that the equipment control module, after parsing the equipment control data, modifies the control parameter configuration of the corresponding equipment, triggering the equipment operation status adjustment operation, and the production scheduling module, after parsing the production scheduling data, updates the raw material delivery time and process execution sequence in the production schedule, triggering the reorganization operation of the raw material flow path and production step sequence; The status acquisition module of the industrial manufacturing execution system obtains the actual values ​​of equipment operating parameters and the actual execution order of production steps in real time, compares and analyzes them with the target parameters of the carbon asset optimization strategy set, and generates feedback data including adjustment deviations.

10. A carbon asset life cycle management system for industrial manufacturing, characterized by: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the carbon asset full life cycle management method for industrial manufacturing as described in any one of claims 1 to 9.

Images