Copper product carbon footprint data collection and processing method based on full life cycle

Abstract

The application discloses a copper product carbon footprint data collection and processing method based on a whole life cycle, comprising the following steps: dividing the production and processing life cycle of a copper product into several stages, and collecting carbon emission data of the current life cycle; constructing a multivariate time series matrix X(t) according to the time sequence of the carbon emission data of the several stages, decomposing X(t) into dynamic modes, and obtaining a dominant mode subset by calculating the mode energy; obtaining a reconstructed data sequence by using the dominant mode subset, and calculating an abnormal score with X(t); when the abnormal score is greater than a preset score threshold, inputting the carbon emission data and a carbon footprint network into a pre-trained graph convolutional neural network to obtain a data abnormal stage; collecting energy consumption data and operation data of several devices in the data abnormal stage, and obtaining several abnormal devices through correlation analysis. The application relates to the field of industrial carbon emission, and solves the technical problems of difficult carbon emission data anomaly detection and complex data processing.

Description

Technical Field

[0001] The present invention belongs to the field of industrial carbon emissions and relates to data processing and analysis technology, specifically a method for collecting and processing carbon footprint data of copper products based on the entire life cycle. Background Art

[0002] With growing global attention to environmental protection, carbon emissions management has become a critical component of corporate sustainable development. Accurately collecting carbon emissions data is crucial for copper product manufacturers. This data can reveal energy efficiency, process rationality, and environmental impact within a company's production processes. By analyzing carbon emissions data, companies can identify high-carbon emission areas and implement appropriate energy-saving and emission-reduction measures to reduce production costs, comply with environmental regulations, and enhance their social responsibility and market competitiveness.

[0003] Copper production and processing involves multiple complex stages, each with diverse and dispersed carbon emission sources. Carbon emissions are generated at every stage, from raw material acquisition to final product processing. Consequently, carbon emission data comes from a wide range of sources and is complex, making data collection and integration challenging. Furthermore, existing methods lack effective organization and processing methods for handling large amounts of carbon emission time series data. This results in excessive computational effort and prolonged processing time, hindering the timely and accurate provision of analytical results to businesses, hindering the timeliness and accuracy of their decision-making.

[0004] Accurately tracking and managing carbon footprints requires ensuring the accuracy and comprehensiveness of carbon emissions data. However, when anomalies in carbon emissions data occur, existing methods lack in-depth understanding of the inherent connections between various stages of the production process, making it difficult to accurately trace the root cause of the anomaly. This makes it difficult for companies to take targeted measures to resolve these anomalies, increasing their economic losses. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method for collecting and processing carbon footprint data of copper products based on the entire life cycle, which is used to solve the technical problems in the prior art of difficult anomaly detection of carbon emission data and complex data processing.

[0006] To achieve the above objectives, the present invention provides a method for collecting and processing carbon footprint data of copper products based on the entire life cycle, comprising:

[0007] S1, divide the production and processing life cycle of copper products into several stages, and collect carbon emission data of the current life cycle through sensors to obtain carbon emission data of several stages;

[0008] S2, construct a multivariate time series matrix X(t) based on the carbon emission data of several stages in chronological order, decompose X(t) into dynamic modes, and obtain the dominant mode subset by calculating the modal energy of the dynamic modes;

[0009] S3: Use the dominant mode subset to obtain a reconstructed data sequence, calculate the anomaly score by comparing the reconstructed data sequence with X(t), and compare whether the anomaly score is greater than the preset score threshold; if yes, jump to S4; if not, mark the next life cycle as the current life cycle and jump to S1;

[0010] S4, constructing a carbon footprint network G according to several stages, inputting the carbon emission data of several stages and the carbon footprint network G into the pre-trained graph convolutional neural network to obtain the data anomaly stage;

[0011] S5, collect the energy consumption data and operation data of several devices in the data abnormality stage, and obtain several abnormal devices through correlation analysis.

[0012] Based on the above technical process, the present invention constructs a complete and efficient system for collecting, processing, early warning, and tracing anomalies in carbon emissions data. Through refined carbon emissions data collection and dynamic modal analysis, it achieves accurate tracking and anomaly detection of carbon footprints during the production and processing of copper products, effectively solving the problems of incomplete carbon emissions data collection, low processing efficiency, and difficulty in tracing anomalies in traditional methods. By constructing a carbon footprint network and applying a graph convolutional neural network, the present invention not only improves the accuracy of data processing, but also enhances the real-time nature of anomaly detection and the accuracy of tracing, providing strong technical support for the green production of copper products and carbon emissions management.

[0013] Furthermore, the multivariate time series matrix X(t)=[x1(t),x2(t),…,xn(t)] T ; Where t represents time, xi(t) represents the carbon emissions of the i-th stage at time t, i = 1, 2,…, n, and n represents the number of stages.

[0014] Furthermore, the method of obtaining the dominant mode subset by calculating the modal energy of the dynamic mode includes:

[0015] S21, decompose X(t) into several dynamic modes {Φm,λm} using a dynamic mode decomposition algorithm; where Φm represents the modal matrix and λm represents the corresponding eigenvalue;

[0016] S22, according to the formula Em=||Φm|| 2 Calculate the energy value Em of the dynamic mode and obtain several modal energies; where ||·|| represents the norm operation;

[0017] S23, sorting the modal energies in descending order, and accumulating the sorted modal energies in sequence, calculating the ratio of the accumulated value to the sum of the modal energies, and when the ratio exceeds a preset threshold, screening out the corresponding modes to obtain a dominant modal subset.

[0018] Constructing a multivariate time series matrix provides an ordered structure for the data, which not only facilitates batch processing and calculations but also effectively reduces the data dimension through dynamic modal decomposition technology, thereby reducing the amount of computation. This combination significantly improves the efficiency of data processing, enabling companies to quickly obtain analytical results and make decisions. In addition, the method of screening dominant modal subsets based on modal energy can adaptively focus on key information, avoiding the problem of information overload. The extracted dominant modal subset reveals the main variation patterns of carbon emissions, enhances the accuracy of key information analysis, helps to grasp the core characteristic trends, and provides a scientific basis for optimizing production processes.

[0019] Furthermore, the anomaly score obtained by reconstructing the data sequence and calculating X(t) includes:

[0020] S31, by formula Calculate the reconstructed data sequence in, represents the column vector of the j-th modal matrix Φj in the dominant modal subset, λj represents the corresponding eigenvalue, bj is the coefficient vector determined by the initial conditions, and k represents the number of modes in the dominant modal subset;

[0021] S32, calculating the difference between the reconstructed data sequence and X(t) at the corresponding time point to obtain the residual vector r(t);

[0022] S33, calculate the standard deviation σ of the residual vector r(t) r , by the formula AS(t)=(||r(t)|| / σ r )×w(t) is used to calculate the anomaly score, where w(t) represents the dynamic weight.

[0023] Furthermore, the method of obtaining bj includes:

[0024] S31-1, construct the initial conditions using the multivariate time series matrix at the initial moment t0 of the current life cycle, and obtain the initial conditions as follows: Where m represents the modal index, M represents the number of dynamic modes, Represents the column vector of the modal matrix Φm, bm is the coefficient vector to be solved;

[0025] S31-2, write the initial conditions into matrix form: X(t0) = Φb; where, b=[b1,b2,…,bm] T ;

[0026] S31-3, solving b = (Φ T Φ) -1 Φ T X(t0), to obtain a coefficient vector b;

[0027] S31-4, obtaining a corresponding coefficient in the coefficient vector b according to the dominant modal subset, to obtain bj.

[0028] By comparing the reconstructed data with the actual measured data, calculating the residual vector, and combining the dynamic weight to calculate the anomaly score, the accuracy of carbon emission anomaly detection can be improved, and the false positive and false negative situations can be effectively reduced. Combining dynamic modal decomposition with an anomaly scoring mechanism can quickly and accurately identify anomalies in carbon emission data, enhancing the timeliness of anomaly discovery and preventing further deterioration of problems, thereby reducing environmental risks and economic losses that may be caused by carbon emission anomalies.

[0029] Further, the method for obtaining the dynamic weight comprises:

[0030] According to the formula w(t) = a(1-a) t-t’ to calculate the dynamic weight; wherein a represents a decay coefficient for controlling the decay rate of the dynamic weight, and t' represents the current time.

[0031] In the dynamic weight formula, the time interval t-t' as the variable of the exponential part intuitively reflects the timeliness of the data. As the time interval increases, i.e. the data is more distant, the value of t-t' is smaller, so that the weight of this data point is lower. This weight allocation method based on time sequence can reasonably allocate weights according to the time sequence, making the analysis results more close to the current production actual situation. Using dynamic weight in the process of calculating the anomaly score can make the recent carbon emission data have greater influence when judging anomalies. When a sudden anomaly occurs in the production process, such as sudden equipment failure leading to a sharp increase in carbon emissions, the recent data has a higher weight, and the anomaly score will quickly rise, thereby detecting the anomaly more quickly and improving the timeliness of anomaly detection, avoiding the masking of recent anomalies due to the averaging effect of long-term historical data.

[0032] Further, the carbon footprint network G is constructed according to a plurality of stages, comprising:

[0033] S4-11, defining a plurality of stages of a copper product production and processing life cycle as nodes of the carbon footprint network, to obtain a plurality of nodes;

[0034] S4-12, defining edges and weights of the plurality of nodes according to material flow, energy flow and carbon emission transmission relationship between the plurality of stages, to obtain a plurality of edges and weights of the plurality of edges;

[0035] S4-13, based on the average carbon emissions of each stage and several equipment parameters, attribute information is added to several nodes of the carbon footprint network to obtain the carbon footprint network G(V, E, W); where V represents several nodes, E represents several edges, and W represents the weights of several edges.

[0036] The various stages of copper production and processing do not exist in isolation; rather, they are interconnected and influence each other through material flows, energy flows, and carbon emissions. Defining each stage as a node in a carbon footprint network and constructing edges and weights based on their actual relationships can transform the inherent logic of the production process into a network structure, allowing for a more comprehensive and systematic consideration of carbon emissions throughout the entire production process.

[0037] Furthermore, the step of inputting the carbon emission data of several stages and the carbon footprint network G into the pre-trained graph convolutional neural network includes:

[0038] S42-1, inputting carbon emission data of several phases and the carbon footprint network G into the pre-trained graph convolutional neural network to obtain abnormal probability values ​​of several nodes;

[0039] S42-2, marking the node whose abnormal probability value exceeds the preset probability as an abnormal node, and obtaining the corresponding data abnormal stage according to the abnormal node.

[0040] The constructed carbon footprint network G clarifies the relationships between various production stages. A graph convolutional neural network performs in-depth analysis based on the network structure and node attributes, calculating the anomaly probability for each node (i.e., production stage). By comparing this with preset probabilities, it accurately identifies anomalous nodes and locates the corresponding data anomaly stage. This more precise positioning method than traditional methods provides guidance for troubleshooting, improves the efficiency and reliability of anomaly detection, and reduces the cost of response.

[0041] Furthermore, the construction process of the pre-trained graph convolutional neural network includes:

[0042] S43-1, build and obtain graph convolution model based on deep learning algorithm;

[0043] S43-2, collect carbon emission data for several historical life cycles, and mark the carbon emission data according to abnormal conditions in several stages, to obtain a number of carbon emission data and a number of marked data;

[0044] S43-3, preprocessing the carbon emission data and the labeled data to obtain preprocessed data, and making a one-to-one correspondence between the preprocessed data and the nodes in the carbon footprint network to obtain a model data set;

[0045] S43-4, the model data set is divided into a training set, a validation set and a test set according to a preset ratio, and a loss function, a training parameter and an optimization algorithm for training of the graph convolution model are set;

[0046] S43-5, the graph convolution model is iteratively trained and verified and optimized by using the training set and the validation set, and the graph convolution model with the highest verification accuracy is saved, so as to obtain an optimal graph convolution model;

[0047] S43-6, the test set is input into the optimal graph convolution model, the test accuracy is obtained, and it is judged whether the test accuracy is greater than a preset accuracy threshold; if yes, the optimal graph convolution model is output as a pre-trained graph convolution neural network; if no, the loss function, the training parameter and the optimization algorithm of the model are adjusted, and the step S43-5 is returned.

[0048] Further, the obtaining of the plurality of abnormal devices through the correlation analysis comprises:

[0049] S5-1, collecting energy consumption data and operation data of a plurality of devices in a data anomaly stage to obtain a data set of the plurality of devices; wherein the energy consumption data comprises power consumption per unit time and fuel usage, and the operation data comprises average temperature, average voltage and average current of device operation;

[0050] S5-2, calculating a correlation coefficient of each pair of devices in the plurality of devices according to the data set by using a calculation formula of a Pearson correlation coefficient to obtain a correlation matrix S; wherein the dimension of S is p x q, and Spq represents the correlation coefficient between the pth device and the qth device;

[0051] S5-3, collecting a historical normal data set of the plurality of devices in the data anomaly stage, and calculating a plurality of standard correlation coefficients according to the historical normal data set;

[0052] S5-4, calculating a mean value μ and a standard deviation σ of the plurality of standard correlation coefficients of each pair of devices, and obtaining a correlation threshold of each pair of devices as [μ-βσ, μ+βσ] according to the mean value μ and the standard deviation σ; wherein k represents a preset adjustment coefficient;

[0053] S5-5, sequentially traversing each column data of the correlation matrix S, and counting the number of abnormal data in the pth column data which is not in the corresponding correlation threshold range to obtain the number of abnormal data;

[0054] S5-6, judging whether the number of abnormal data of the pth column is greater than a preset number threshold; if yes, marking the device p as an abnormal device; if no, continuing to traverse the next column data.

[0055] The historical normal data set contains data patterns from equipment under normal operating conditions, reflecting stable correlations between devices. Using this historical data to calculate the mean μ and standard deviation σ of the standardized correlation coefficient provides a reasonable reference for the current data anomaly phase. Because production environments and equipment operating conditions exhibit a certain degree of similarity and continuity, the statistical characteristics of historical normal data can help define a range of normal correlations. By comparing the correlation of the current device pair against this threshold, we can more accurately determine whether the current correlation is abnormal, reduce interference in anomaly judgment caused by factors such as environmental fluctuations and short-term production changes, and improve the accuracy and stability of anomaly detection.

[0056] The application of carbon footprint networks and graph convolutional neural networks, as well as equipment correlation analysis methods, can accurately locate the specific stages and equipment where carbon emission anomalies occur. Enterprises can conduct targeted, in-depth inspections and repairs, improving problem-solving efficiency and reducing the time and cost of investigating the causes of anomalies.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The production and processing of copper products involves multiple complex stages, and the carbon emission sources in each stage are diverse and dispersed. The present invention divides the production and processing life cycle into several stages and deploys sensors at each stage to achieve systematic collection of carbon emission data, providing basic data support for subsequent precise analysis, and effectively solving the problem of incomplete and inaccurate carbon emission data acquisition. In addition, the management of carbon emission data faces challenges such as large data volume and strong time series, making it difficult to effectively integrate and utilize. The present invention organizes data by constructing a multivariate time series matrix, so that the data is arranged in order in the time dimension, which is convenient for subsequent processing such as dynamic modal decomposition, and solves the problem of chaotic carbon emission data management and difficulty in effective utilization.

[0059] The production process of copper products is complex, and there are many factors that affect carbon emissions, which makes it difficult to accurately detect and locate carbon emission anomalies. The present invention uses dynamic mode decomposition technology to obtain the dominant mode subset, and calculates the anomaly score by comparing it with the actual data. It can effectively detect anomalies in carbon emission data, solving the problem that carbon emission anomalies are difficult to detect due to the complexity of the production process. At the same time, traditional methods are difficult to determine at which stage or which equipment the carbon emission anomaly occurs when faced with complex production processes and numerous equipment. The present invention constructs a carbon footprint network and combines it with a graph convolutional neural network, as well as equipment correlation analysis, which can accurately locate the data anomaly stage and abnormal equipment, solving the problem of inaccurate and untimely positioning of carbon emission anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 A schematic diagram of the technical process of the method for collecting and processing carbon footprint data of copper products based on the entire life cycle provided by the present invention;

[0062] Figure 2 Schematic diagram of the technical process of obtaining abnormal devices through correlation analysis provided by the present invention DETAILED DESCRIPTION

[0063] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] See also Figure 1 - Figure 2 The first embodiment of the present invention provides a method for collecting and processing carbon footprint data of copper products based on the entire life cycle, including:

[0065] S1, divide the production and processing life cycle of copper products into several stages, and collect carbon emission data of the current life cycle through sensors to obtain carbon emission data of several stages;

[0066] S2, construct a multivariate time series matrix X(t) based on the carbon emission data of several stages in chronological order, decompose X(t) into dynamic modes, and obtain the dominant mode subset by calculating the modal energy of the dynamic modes;

[0067] S3: Use the dominant mode subset to obtain a reconstructed data sequence, calculate the anomaly score by comparing the reconstructed data sequence with X(t), and compare whether the anomaly score is greater than the preset score threshold; if yes, jump to S4; if not, mark the next life cycle as the current life cycle and jump to S1;

[0068] S4, constructing a carbon footprint network G according to several stages, inputting the carbon emission data of several stages and the carbon footprint network G into the pre-trained graph convolutional neural network to obtain the data anomaly stage;

[0069] S5, collect the energy consumption data and operation data of several devices in the data abnormality stage, and obtain several abnormal devices through correlation analysis.

[0070] In this embodiment, by subdividing the copper product production and processing life cycle into multiple stages and using a sensor network to realize real-time collection of carbon emission data at each stage, not only the comprehensiveness of data collection is ensured, covering every link from raw material acquisition to final product processing, but also the dynamic changes of carbon emissions in the production process can be captured in a timely manner, providing enterprises with a real-time and complete view of carbon emissions.

[0071] Specifically, during the production preparation phase, various sensors, such as equipment startup monitoring sensors, material flow sensors, power consumption sensors, and equipment vibration sensors, are installed on the equipment startup preheating process and raw material pretreatment equipment to collect equipment startup parameters and raw material processing-related data. These sensors are used to monitor the equipment startup status and energy consumption, material flow, and equipment operation stability during raw material pretreatment.

[0072] During the main production stages (such as smelting and refining), key equipment (smelting furnaces, refining furnaces, etc.) and production lines are equipped with a variety of high-precision sensors, such as furnace temperature sensors, oxygen flow sensors, exhaust gas emission monitoring sensor groups, product quality detection sensors, etc., to accurately measure furnace temperature, oxygen flow, exhaust gas emissions, and product quality parameters;

[0073] During the product processing and forming stage, equipment power sensors, mold pressure sensors, cooling medium flow sensors, and temperature sensors are installed on copper processing equipment (rolling, stretching equipment, etc.) and cooling links to monitor equipment operating power, mold pressure, cooling medium flow, and temperature changes.

[0074] During the product packaging and storage stage, packaging material usage sensors are installed in the packaging area to record the amount of packaging material used. Warehouse environment monitoring sensors (temperature and humidity sensors) and a warehouse management system are installed in the finished product warehouse to monitor the warehouse environment and record product in and out information.

[0075] Then, through the existing carbon emission calculation models and methods, the data collected by each sensor are integrated to calculate and obtain the carbon emission data for each stage. It should be noted that the division of the above several stages is only an example, and the specific division can be flexibly adjusted according to the actual production process, equipment layout, management model and other factors of the copper product manufacturer.

[0076] Next, the carbon emission data of each stage collected at the preset time interval is constructed into a multivariate time series matrix X(t) = [x1(t), x2(t), …, xn(t)] in chronological order. T; Where t represents time, xi(t) represents the carbon emissions of the i-th stage at time t, i = 1, 2, ..., n, n represents the number of stages, t = 1, 2, ..., T, T represents the time of a complete life cycle; then the dynamic mode decomposition (DMD) algorithm is used to decompose the constructed multivariate time series matrix X(t). The DMD algorithm is a data-driven method for analyzing dynamic systems, which can extract the main dynamic modes from the data. In actual calculations, through a series of complex mathematical operations (involving knowledge such as eigenvalue decomposition in linear algebra), X(t) is decomposed into several dynamic modes {Φm, λm}, where Φm represents the modal matrix and λm represents the corresponding eigenvalue;

[0077] Then according to the formula Em=||Φm|| 2 Calculate the energy value Em of the dynamic mode and obtain several modal energies; where ||·|| represents the norm operation, and the Frobenius norm can be sampled for operation;

[0078] After calculating the energy values ​​of all dynamic modes, these energy values ​​are arranged in descending order. Assume that the sorting result is E3>E1>E4>E2>… Then accumulate the sorted modal energies in sequence, and calculate the proportion of the accumulated value to the sum of the modal energies. That is, first calculate the proportion of E3 in the total energy, then calculate the proportion of E3+E1 in the total energy, then calculate the proportion of E3+E1+E4 in the total energy, and so on, until the proportion exceeds the preset threshold (in this embodiment, the preset threshold is 80%), then stop accumulating, and select the corresponding modes as the dominant modal subsets. These dominant modal subsets contain the most important dynamic change patterns in the original carbon emission data. In subsequent analysis, they can be used to simplify data processing and highlight key information, so as to better perform operations such as carbon emission anomaly detection.

[0079] In step S3, the coefficient vector is first calculated based on the dominant mode subset obtained in the early stage and the corresponding initial conditions, and then the reconstructed data sequence is obtained. The residual vector is then calculated by combining the reconstructed data sequence with the actual multivariate time series matrix, and the anomaly score is further calculated in combination with the dynamic weight. Finally, the anomaly score is compared with the preset score threshold to determine the subsequent process direction.

[0080] Specifically, in the calculation of the coefficient vector, we first need to construct the initial conditions using the multivariate time series matrix at the initial moment t0 of the current life cycle: Where m represents the modal index, M represents the number of dynamic modes, Represents the column vector of the modal matrix Φm, bm is the coefficient vector to be solved; then write it in the matrix form X(t0) = Φb; where, b=[b1,b2,…,bm]T ; Then, the least squares method is used to solve the problem and obtain the complete coefficient vector. Then, the corresponding coefficients are extracted from the dominant mode subset, which will be used for the calculation of the subsequent reconstructed data sequence.

[0081] Assume that the dominant mode subset has been obtained in the previous S2 step, which contains k = 3 modes (Φ1, Φ2, Φ3), and the total number of dynamic modes obtained in the entire dynamic mode decomposition process is M = 5, that is, it is assumed that the copper product life cycle is divided into 5 stages.

[0082] Then, taking the current life cycle as an example, assume that the multivariate time series matrix collected at the starting time t0 is: X(t0) =

[23142] T ; Assume that the modal matrix Φm (m=1,2,…,5) is:

[0083]

[0084] First follow Construct the initial conditions and then write them in matrix form X(t0) = Φb, where and is the column vector of Φm, b=[b1,b2,…,b5] T ;

[0085] Then, we solve b=(Φ T Φ) -1 Φ T X(t0). Calculate Φ T Φ gets a 5×5 matrix, then finds its inverse matrix, and then compares it with Φ T Multiply X(t0) and assume the coefficient vector is: b = [0.50.30.20.10.4] T ; Since the dominant mode subset is Φ1, Φ2, Φ3, the corresponding coefficients b1 = 0.5, b2 = 0.3, b3 = 0.2 are obtained from them;

[0086] Then, through the formula Calculate the reconstructed data sequence in, represents the column vector of the j-th modal matrix Φj in the dominant modal subset, λj represents the corresponding eigenvalue, bj represents the coefficient vector, and k represents the number of modes in the dominant modal subset;

[0087] Then, calculate the residual vector And calculate the standard deviation σ of the residual vector r(t) r ;

[0088] Finally, according to the formula AS(t)=(||r(t)|| / σ r)×w(t) to calculate the anomaly score; where w(t) represents the dynamic weight, and w(t)=α(1-α) t-t’ The formula calculates the dynamic weight, α represents the attenuation coefficient, which is used to control the attenuation speed of the dynamic weight and is determined based on a lot of practical experience, and t' represents the current time;

[0089] Over time, factors such as equipment wear, process adjustments, and raw material changes can cause carbon emissions data to change. Recent carbon emissions data better reflects current production status and potential anomalies, while older data gradually becomes less representative of current production conditions. Therefore, the dynamic weighting formula introduces a time-dependent attenuation coefficient, allowing data weights to change dynamically over time. This better adapts to the dynamic nature of the production process, enabling anomaly detection and analysis to more promptly and accurately reflect current production realities.

[0090] The calculated anomaly score is compared with the preset scoring threshold. If the anomaly score is greater than the preset scoring threshold, it means that the current carbon emission data has an anomaly and it is necessary to jump to step S4 for further in-depth analysis of which production stages or equipment have problems. If the anomaly score is less than or equal to the preset scoring threshold, it indicates that the current carbon emission data is basically within the normal fluctuation range. At this time, the next life cycle is marked as the current life cycle and jump to step S1 to continue collecting new carbon emission data in order to continuously monitor and analyze carbon emissions in the production process.

[0091] In step S3, the data sequence is reconstructed using the dominant modal subset, which greatly reduces the amount of computation compared to directly processing the original high-dimensional carbon emission data. At the same time, operations such as solving the coefficient vector b based on the least squares method and subsequent anomaly score calculation are all based on mathematical models and optimization algorithms and have high computational efficiency. In addition, this method can adapt to changes in carbon emission data under different production stages and working conditions, because the dominant modal subset is dynamically selected based on actual data characteristics, and the dynamic weight can also adjust the importance of the data according to time, making the entire anomaly detection method highly adaptable and robust, and able to work effectively in complex and changing production environments.

[0092] Then, in S4, by constructing a carbon footprint network, the complex internal connections of each stage in the entire copper product production process were integrated; and using graph convolutional neural networks, the abnormal characteristics of the data were mined to provide support for the precise positioning of carbon emission problems.

[0093] Specifically, we first construct a carbon footprint network based on the production and processing of copper products, including:

[0094] S4-11, defines each stage of the copper product production and processing life cycle as a node of the carbon footprint network;

[0095] S4-11, define several stages of the copper product production and processing life cycle as nodes of the carbon footprint network, and obtain several nodes;

[0096] S4-12: Define the edges and weights of several nodes based on the material flow, energy flow, and carbon emission transfer relationships between several stages. For example, the transfer of high-temperature molten copper from the smelting stage to the refining stage is accompanied by large amounts of energy consumption and carbon emissions. Based on these actual data, the weight values ​​of the edges connecting the two stages are determined.

[0097] S4-13, based on the average carbon emissions of each stage and several equipment parameters, attribute information is added to each node of the carbon footprint network to obtain the carbon footprint network G(V, E, W); where V represents several nodes, E represents several edges, and W represents the weights of several edges.

[0098] The various stages of the copper production and processing process do not exist in isolation; rather, they are interconnected and mutually influenced through material flows, energy flows, and carbon emissions. Defining each stage as a node in a carbon footprint network and constructing edges and weights based on their actual relationships transforms the inherent logic of the production process into a network structure, enabling a more comprehensive and systematic consideration of carbon emissions throughout the entire production process. This network-based representation facilitates the discovery of deeper connections between stages, avoiding the limitations of analyzing carbon emissions data for each stage individually and providing a richer information base for accurately locating anomalous stages.

[0099] Then the carbon emission data of each stage and the carbon footprint network G are input into the pre-trained graph convolutional neural network to obtain the data anomaly stage, including:

[0100] S42-1: Input the carbon emission data of several stages and the carbon footprint network G into the pre-trained graph convolutional neural network. The neural network model undergoes internal graph convolution operations, feature extraction, and classification processes, outputting the abnormality probability value of each node to obtain the abnormality probability values ​​of several nodes.

[0101] S42-2, mark the node whose abnormal probability value exceeds the preset probability as an abnormal node. For example, in this embodiment, the node whose abnormal probability value exceeds 0.5 is marked as an abnormal node, and then the corresponding stage is obtained according to the abnormal node, that is, the data abnormal stage.

[0102] In S4, the construction process of the pre-trained graph convolutional neural network specifically includes:

[0103] Based on the PyTorch deep learning framework, we select appropriate graph convolutional layers, activation functions and other components to build a graph convolutional model structure and obtain a graph convolutional neural network.

[0104] Collect carbon emission data from multiple historical life cycles. For each stage of each cycle, mark the carbon emission data based on whether there are any abnormalities in the actual production process. For example, if an equipment failure occurs during the smelting stage of a certain cycle, causing a sudden and substantial increase in carbon emissions that exceeds the normal fluctuation range, the carbon emission data corresponding to the smelting stage will be marked as abnormal (1 can be used to indicate abnormality and 0 to indicate normality), and a number of carbon emission data and corresponding marked data will be obtained;

[0105] Preprocess the collected carbon emission data and annotated data. Common preprocessing operations include data normalization and missing value processing. Then, the preprocessed data is mapped one-to-one with several nodes in the carbon footprint network to ensure that the data of each node at different time points can be accurately matched, thereby constructing a model dataset.

[0106] The model dataset is divided into training, validation, and test sets in a preset ratio of 7:2:1. At the same time, the loss function, training parameters (including learning rate, number of iterations, batch size, etc.), and optimization algorithm (such as the AdamW optimization algorithm, which is used to update the model's weight parameters to minimize the loss function) for graph convolutional model training are set.

[0107] The graph convolution model is iteratively trained and validated using the training set and validation set. In each iteration, the data in the training set is input into the model, the loss value is calculated according to the loss function, and then the weight parameters of the model are updated through the optimization algorithm to make the model continuously fit the data. After each round of training, the model is validated with the validation set to evaluate the accuracy of the model on the validation set (such as accuracy, recall rate, precision, etc.), and the F1 value is used as the main accuracy indicator. The validation accuracy is recorded, that is, the model state when the F1 value is the highest, and the graph convolution model with the highest validation accuracy is obtained;

[0108] The test set is input into the saved graph convolution model with the highest verification accuracy to obtain the test accuracy. Assuming the preset F1 threshold is 90%, that is, if the test accuracy is greater than 90%, the model performance is considered to meet the requirements, and this optimal graph convolution model is output as the pre-trained graph convolutional neural network; if the test accuracy is less than or equal to 90%, the model's loss function (for example, replacing it with another type of loss function), training parameters (such as adjusting the learning rate, increasing the number of iterations, etc.), and optimization algorithm are adjusted, and then iterative training and verification optimization are repeated until a pre-trained model that meets the accuracy requirements is obtained.

[0109] Convolutional neural networks (GCNNs) have powerful capabilities for processing graph-structured data. They can automatically learn the complex feature representations of nodes and edges in carbon footprint networks. They perform analysis by propagating information across the network and integrating information about nodes and their neighbors. Feeding carbon emissions data and the carbon footprint network together into the GCNN leverages its unique characteristics, enabling it to better capture complex patterns in the data and thus improve the accuracy of identifying data anomalies.

[0110] After identifying the abnormal stage in the copper product production process, we further analyze the correlation between equipment to quickly locate the specific abnormal equipment, guide subsequent accurate and efficient maintenance and repair work, and effectively ensure the smooth operation of the production process and the compliance control of carbon emissions.

[0111] Specifically, in S5, energy consumption data and operating data of several devices during the data anomaly phase are first collected to obtain data sets for several devices. The energy consumption data includes the degree of electricity consumption and fuel usage per unit time, and the operating data includes the average temperature, average voltage, and average current of the device during operation.

[0112] Then, using the Pearson correlation coefficient calculation formula, the correlation coefficient of each pair of devices in the data set is calculated to obtain the correlation matrix S. The dimension of S is p × q, and Spq represents the correlation coefficient between the p-th device and the q-th device.

[0113] We further collected data from the normal operating phase of these devices over several historical copper production cycles to obtain historical normal data sets for several devices. These data sets were also categorized and organized according to their corresponding energy consumption and operating data. Then, based on these historical normal data sets, we calculated the correlation coefficient between each pair of devices under historical normal conditions using the Pearson correlation coefficient formula. This yielded several standard correlation coefficients, which reflect the degree of correlation between devices during normal production and provide a reference for determining the correlation threshold for the current abnormal phase.

[0114] For each pair of devices, the mean μ and standard deviation σ are calculated based on several calculated standard correlation coefficients. The correlation threshold for each pair of devices is then calculated as [μ-βσ, μ+βσ]. Here, k represents the preset adjustment coefficient, typically set to 2. Because the production environment and equipment operating conditions have a certain degree of similarity and continuity, the statistical characteristics of historical normal data (mean and standard deviation) can help determine a reasonable correlation threshold range. This can reduce interference with abnormality judgment caused by factors such as environmental fluctuations and short-term production changes, making the determination of abnormal devices more accurate.

[0115] Next, traverse each column of the correlation matrix S in turn. For the p-th column of data, compare each element (i.e., the correlation coefficient Spq between the p-th device and each other device) with the corresponding correlation threshold [μ-βσ, μ+βσ]. If the value of Spq is not within this threshold range, it is regarded as an abnormal data point. By comparing and judging each element one by one, the number of data in the p-th column that are not within the corresponding correlation threshold range is counted. This number is also the number of data with abnormal correlation between device p and other devices. It reflects, to a certain extent, the degree of abnormality of the device in the entire device association network;

[0116] Then, the number of abnormal data in the pth column is compared with the preset number threshold. Assume that the preset number threshold is 5, and there are 10 devices in total, that is, each device has 10 correlation coefficients (including with itself); if the number of abnormal data is greater than 5, then the device p is marked as an abnormal device; otherwise, continue to traverse the next column of data until the comparative analysis of all devices is completed, and all abnormal devices are identified, so as to accurately screen out the problematic devices from a large number of devices, and provide a clear target object for the subsequent targeted solution of abnormal problems in the production process.

[0117] The above correlation analysis process accurately identifies devices that exhibit deviations from normal operating conditions during the data anomaly phase. By sequentially traversing each column of the correlation matrix, counting the number of data points that fall outside the correlation threshold and comparing them with a preset threshold, we can filter out truly abnormal devices from the vast pool, avoiding mistakenly labeling devices with normal fluctuations as abnormal and reducing the possibility of missing abnormal devices.

[0118] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0119] Working principle of the present invention:

[0120] The present invention divides the production and processing life cycle of copper products into multiple stages, and uses a sensor network to realize real-time collection of carbon emission data in each stage, thereby realizing comprehensive collection of carbon footprint data of copper products; then constructs a multivariate time series matrix, and uses dynamic mode decomposition technology to improve the efficiency and pertinence of data processing, and screens out the dominant mode subset by calculating the modal energy, highlighting the key information in the data, and avoiding the information overload problem that occurs when processing large-scale and complex data; then calculates the anomaly score by reconstructing the data to achieve overall early warning of carbon emission data, and then uses the carbon footprint network and graph convolutional neural network for precise positioning, quickly narrowing the investigation scope to a specific production stage; finally, a correlation analysis is performed on the energy consumption and operating data of the equipment in the data anomaly stage, and anomalies are discovered from the perspective of equipment collaboration, and abnormal equipment is accurately identified; the present invention constructs a complete and efficient carbon emission data collection, processing, early warning and anomaly traceability system, which provides solid data support and technical guarantee for copper product manufacturers to accurately control carbon emission status and promote sustainable development.

[0121] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for collecting and processing carbon footprint data of copper products based on the entire life cycle, characterized in that: include: S1, divide the production and processing life cycle of copper products into several stages, and collect carbon emission data of the current life cycle through sensors to obtain carbon emission data of several stages; S2, construct a multivariate time series matrix X(t) based on the carbon emission data of several stages in chronological order, decompose X(t) into dynamic modes, and obtain the dominant mode subset by calculating the modal energy of the dynamic modes; S3, using the dominant mode subset to obtain a reconstructed data sequence, and calculating the anomaly score by comparing the reconstructed data sequence with X(t), and comparing whether the anomaly score is greater than a preset score threshold; If yes, jump to S4; if no, mark the next life cycle as the current life cycle and jump to S1; The abnormality score obtained by reconstructing the data sequence and calculating X(t) includes: S31, by formula Calculate the reconstructed data sequence ;in, represents the column vector of the jth modal matrix Φj in the dominant modal subset, represents the corresponding eigenvalue, b j is the coefficient vector determined by the initial conditions, k represents the number of modes in the dominant mode subset; S32, calculating the difference between the reconstructed data sequence and X(t) at the corresponding time point to obtain the residual vector r(t); S33, calculate the standard deviation σ of the residual vector r(t) r , by the formula AS(t)=(||r(t)|| / σ r )×w(t) is used to calculate the anomaly score, where w(t) represents the dynamic weight. S4, constructing a carbon footprint network G according to several stages, inputting the carbon emission data of several stages and the carbon footprint network G into the pre-trained graph convolutional neural network to obtain the data anomaly stage; S5, collect the energy consumption data and operation data of several devices in the data abnormality stage, and obtain several abnormal devices through correlation analysis.

2. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 is characterized in that: The multivariate time series matrix X(t)=[x1(t),x2(t),…,xn(t)] T ; Where t represents time, xi(t) represents the carbon emissions of the i-th stage at time t, i=1,2,…,n, and n represents the number of stages.

3. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 is characterized in that: The method of obtaining the dominant mode subset by calculating the modal energy of the dynamic mode includes: S21, decompose X(t) into several dynamic modes {Φm,λm} using a dynamic mode decomposition algorithm; where Φm represents the modal matrix and λm represents the corresponding eigenvalue; S22, according to the formula Em=||Φm|| 2 Calculate the energy value Em of the dynamic mode and obtain several modal energies; where || || represents norm operation; S23, sorting the modal energies in descending order, and accumulating the sorted modal energies in sequence, calculating the ratio of the accumulated value to the sum of the modal energies, and when the ratio exceeds a preset threshold, screening out the corresponding modes to obtain a dominant modal subset.

4. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 is characterized in that: The b j Ways to obtain include: S31-1, construct the initial conditions using the multivariate time series matrix at the initial moment t0 of the current life cycle, and obtain the initial conditions as follows: ; Where m represents the mode index, M represents the number of dynamic modes, represents the column vector of the modal matrix Φm, b m is the coefficient vector to be solved; S31-2, write the initial conditions in matrix form: X(t0)=Φb; where Φ=[ , ,…, ],b=[b1,b2,…,b m ] T ; S31-3, solve by least square method: b=(Φ T Φ) -1 Φ T X(t0), get the coefficient vector b; S31-4, obtain the corresponding coefficient in the coefficient vector b according to the dominant mode subset, and obtain b j .

5. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 is characterized in that: The dynamic weight is obtained by: According to w(t)=α(1-α) t-t’ The dynamic weight is calculated by the formula; where α represents the attenuation coefficient, which is used to control the attenuation speed of the dynamic weight, and t' represents the current time.

6. The method for collecting and processing copper product carbon footprint data based on the entire life cycle according to claim 1 is characterized in that: The carbon footprint network G is constructed according to several stages, including: S4-11, define several stages of the copper product production and processing life cycle as nodes of the carbon footprint network, and obtain several nodes; S4-12, defining the edges and weights of several nodes according to the material flow, energy flow and carbon emission transmission relationship between several stages, and obtaining several edges and their weights; S4-13, based on the average carbon emissions of each stage and several equipment parameters, attribute information is added to several nodes of the carbon footprint network to obtain the carbon footprint network G(V, E, W); where V represents several nodes, E represents several edges, and W represents the weights of several edges.

7. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 is characterized in that: The carbon emission data of several stages and the carbon footprint network G are input into the pre-trained graph convolutional neural network, including: S42-1, inputting carbon emission data of several phases and the carbon footprint network G into the pre-trained graph convolutional neural network to obtain abnormal probability values ​​of several nodes; S42-2, marking the node whose abnormal probability value exceeds the preset probability as an abnormal node, and obtaining the corresponding data abnormal stage according to the abnormal node.

8. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1 or 7, characterized in that: The construction process of the pre-trained graph convolutional neural network includes: S43-1, build and obtain graph convolution model based on deep learning algorithm; S43-2, collect carbon emission data for several historical life cycles, and mark the carbon emission data according to abnormal conditions in several stages, to obtain a number of carbon emission data and a number of marked data; S43-3, preprocessing the carbon emission data and the labeled data to obtain preprocessed data, and making a one-to-one correspondence between the preprocessed data and the nodes in the carbon footprint network to obtain a model data set; S43-4, dividing the model dataset into a training set, a validation set, and a test set according to a preset ratio, and setting the loss function, training parameters, and optimization algorithm for graph convolutional model training; S43-5, use the training set and validation set to iteratively train and validate the graph convolution model, and save the graph convolution model with the highest validation accuracy to obtain the optimal graph convolution model; S43-6, input the test set into the optimal graph convolution model, obtain the test accuracy, and determine whether the test accuracy is greater than the preset accuracy threshold; if yes, output the optimal graph convolution model as the pre-trained graph convolution neural network; if not, adjust the model's loss function, training parameters and optimization algorithm, and jump to S43-5.

9. The method for collecting and processing carbon footprint data of copper products based on the entire life cycle according to claim 1, characterized in that: The correlation analysis yields several abnormal devices, including: S5-1, collecting energy consumption data and operating data of several devices during the data anomaly phase to obtain data sets for the several devices; wherein the energy consumption data includes the degree of electricity consumption and fuel usage per unit time, and the operating data includes the average temperature, average voltage, and average current of the devices during operation; S5-2, using the Pearson correlation coefficient calculation formula, calculate the correlation coefficient of each pair of devices in the data set to obtain the correlation matrix S; where the dimension of S is p × q, and Spq represents the correlation coefficient between the p-th device and the q-th device; S5-3, collecting historical normal data sets of several devices during the data abnormality phase, and calculating several standard correlation coefficients based on the historical normal data sets; S5-4, calculating the mean μ and standard deviation σ of several standard correlation coefficients for each pair of devices, and obtaining the correlation threshold of each pair of devices as [μ-βσ, μ+βσ] based on the mean μ and standard deviation σ; where k represents a preset adjustment coefficient; S5-5, traverse each column of the correlation matrix S in turn, count the number of data in the p-th column that are not within the corresponding correlation threshold range, and obtain the number of abnormal data; S5-6, determine whether the number of abnormal data in the p-th column is greater than a preset number threshold; if yes, mark the device p as an abnormal device; if not, continue to traverse the next column of data.

Images