Engineering material full life cycle monitoring method and device based on embedded carbon attribute

Abstract

The application discloses a kind of engineering material full life cycle monitoring method and device based on energy carbon attribute embedding, the method includes for target engineering material, constructs energy carbon embedding vector;According to the accurate distinction of effective processing state and idling or standby state of material according to fusion feature sequence, and accurately peeling net physical energy consumption of material in process;Establish the initial prior distribution of energy carbon conversion coefficient to be estimated;Using input material and the sample set of measured net energy consumption data to establish likelihood function;Based on bayesian inference method, calculate and update posterior distribution, obtain calibrated dynamic emission factor, and comprehensively calculate the full life cycle carbon emission of the material unit.This scheme has self-adaptive correction ability precision carbon monitoring system, supports enterprise to carry out material level low-carbon process optimization decision.

Description

Technical Field

[0001] This invention belongs to the field of carbon emission detection, and specifically relates to a method and device for monitoring the entire life cycle of engineering materials based on the embedding of energy and carbon properties. Background Technology

[0002] Currently, in the fields of industrial manufacturing and engineering construction, accurate monitoring and management of carbon emissions from materials and processes have become crucial to achieving objectives. However, existing technologies still face multiple challenges in material-level carbon footprint accounting: 1. Staticization of carbon emission factors and model drift issues Traditional carbon emission accounting methods primarily rely on generic static emission factors, which are typically set based on national standard libraries or industry averages. This static data processing approach has significant drawbacks: 1. Insufficient accuracy: Static factors cannot reflect in real time the differences in physical and chemical properties of different batches of materials, minor adjustments to the processing technology, and changes in the energy efficiency of the equipment itself, resulting in systematic deviations between the calculation results and the actual situation.

[0003] 2. Lack of Adaptability: Equipment aging or tool wear due to long-term operation, as well as process fine-tuning caused by environmental factors (such as temperature and humidity), can lead to concept drift between the preset emission model and actual production efficiency. Existing models lack online learning mechanisms and cannot adaptively evolve based on measured data, resulting in lower calculation accuracy the longer the system is used.

[0004] 2. Coarse granularity and difficulty in allocating energy consumption data Existing energy monitoring systems often only collect energy consumption data at the "workshop" or "production line" level, resulting in coarse-grained data. This presents a challenge for accurate energy allocation. 1. Inability to achieve material-level accounting: Energy consumption data cannot be accurately traced to specific "individual materials" or "components". The system struggles to generate detailed carbon footprint traceability reports down to the bill of materials (BOM level).

[0005] 2. The "Equal Distribution" Phenomenon: During processing, it is difficult to accurately define the timing boundary of "when processing begins" for materials. This leads to confusion between the energy consumed by equipment during "idling" or "standby" periods and the effective energy consumption generated by the material during "actual contact processing." Traditional methods struggle to accurately separate and deduct the baseline standby power to calculate the net physical energy consumption belonging solely to that material. This imprecise allocation method is known as the "equal distribution" phenomenon in energy consumption allocation.

[0006] 3. Data management of material flow and energy flow is disconnected. In engineering management practice, ERP (Equipment Management System) and EMS (Energy Management System) are like "two isolated islands".

[0007] 1. Data silos: This disconnect in management makes it difficult to establish a one-to-one correspondence between the "material flow" and the "energy flow".

[0008] 2. Lack of high-dimensional features: Material attributes (such as material, supplier, specifications) are mostly discrete and unstructured data, which cannot be directly used in computers for efficient energy-carbon similarity calculation and low-carbon process reverse inference.

[0009] In summary, existing technologies lack a precise carbon monitoring system that can deeply integrate the implicit carbon properties of materials with measured dynamic energy consumption and possess adaptive correction capabilities, making it difficult to support enterprises in making material-level low-carbon process optimization decisions. This invention is proposed to overcome the aforementioned technical obstacles. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention proposes a method for monitoring the entire lifecycle of engineering materials based on energy and carbon property embedding, comprising the following steps: Step 1: For the target engineering material, construct a structure including material entity nodes. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. ; Step 2: During the material processing, high-frequency time-series data is collected in real time. and visual flow Align the visual flow using a cross-modal attention mechanism. With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. ; Step 3: Establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; using the input materials and the measured net energy consumption data. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Step 4, based on the energy carbon embedding vector Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

[0011] This invention also discloses a life-cycle monitoring device for engineering materials based on energy and carbon property embedding, comprising: Construct a carbon attribute embedding vector module to build material entity nodes for target engineering materials. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. ; The multimodal carbon fingerprint recognition module is used to collect high-frequency time-series data in real time during material processing. and visual flow Align the visual flow using a cross-modal attention mechanism. With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. ; An adaptive correction module for the energy-carbon factor is used to establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; measured net energy consumption data obtained using the input material and the multimodal carbon fingerprint recognition module. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Dynamic carbon-nucleic acid modules are used for energy-based carbon embedding vectors. Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

[0012] By integrating heterogeneous graph neural networks, multimodal spatiotemporal alignment, and Bayesian adaptive inference techniques, the method of this invention overcomes key technical obstacles in existing engineering material carbon emission monitoring, such as staticity, coarse granularity, and model bias, achieving the following significant technical effects: 1. It has enabled personalized and high-dimensional quantitative expression of energy and carbon properties. This invention employs embedding technology based on heterogeneous graph neural networks (HGNN) to achieve a profound digitization of the carbon properties of materials, rather than simply recording numerical values.

[0013] Addressing the issue of static data: Traditional methods rely on discrete, unstructured material properties. This invention treats the material as a high-dimensional vector, which implicitly contains the material's energy and carbon performance potential and cumulative carbon footprint under different process paths. This enables the system to achieve a technological leap from "coefficient estimation" to "vector measurement," realizing personalized and dynamic expression of energy and carbon properties.

[0014] Support for high-order computation: High-dimensional embedding vectors enable computers to directly calculate the energy-carbon similarity between materials, which lays the mathematical foundation for subsequent low-carbon material substitution recommendations and process reverse engineering.

[0015] 2. Achieve precise "fingerprint-level" identification of net energy consumption during the process. This invention solves the problems of coarse granularity and "equal distribution" in traditional energy consumption allocation by introducing a multimodal spatiotemporal alignment mechanism.

[0016] Precisely defining processing boundaries: The system employs a dual-modal fusion method combining "visual flow" (industrial camera) and "temporal flow" (high-frequency power waveform). Through the Cross-Attention Transformer mechanism, the system can align and strongly bind the visual "material contact processing action" with the "power peak change" on the electricity meter at the millisecond level.

[0017] Eliminating interference from idling losses: By accurately identifying effective processing time segments, the system can precisely integrate the actual power and accurately deduct the baseline standby power to calculate the net physical energy consumption belonging solely to that material. This completely solves the problem of confusion between idling energy consumption and actual processing energy consumption, improving the granularity and accuracy of the calculation.

[0018] 3. The model possesses adaptive evolutionary capabilities, effectively addressing concept drift. This invention introduces a parameter adaptive correction mechanism based on Bayesian inference to ensure that the carbon emission factor can be dynamically corrected as equipment status and process changes.

[0019] Dynamically updated emission factors: The system uses new measured data and a Bayesian posterior update formula to calculate the corrected mean, i.e., the dynamic emission factor, with weights dynamically tilted towards the latest measured data. This overcomes the lag defect caused by relying on static factors.

[0020] Robust drift detection: By calculating the divergence between the old and new parameter distributions, the system can distinguish between instantaneous measurement noise and persistent model biases (such as permanent increases in energy consumption due to tool wear). The global knowledge base is only formally updated when a threshold is exceeded. This ensures that the system's adaptive evolution is robust and purposeful, giving the system "life force"—its accuracy increases with longer use.

[0021] 4. Supporting high-level low-carbon decision optimization This invention combines precise accounting results with vectors, elevating the system from a simple monitoring tool to an intelligent decision support tool. It achieves precise accounting throughout the entire lifecycle: by integrating implicit carbon emissions, measured net energy consumption multiplied by a dynamic emission factor, and scientifically allocated idling losses, it realizes precise calculation of carbon emissions throughout the entire lifecycle of material units.

[0022] Low-carbon process reverse engineering: Utilizing a vector sum generative artificial intelligence (GenAI) model, the system can simulate different potential process paths and recommend optimal key technical parameters with the goal of minimizing carbon emissions. For example, the system can recommend scheduling high-energy-consuming processes to periods with the lowest grid carbon emission intensity, directly maximizing environmental benefits. Attached Figure Description

[0023] Figure 1 This is a flowchart of the life-cycle monitoring method for engineering materials based on energy and carbon property embedding proposed in this invention. Figure 2 This is a block diagram of the engineering material life cycle monitoring based on energy and carbon attribute embedding proposed in this invention; Detailed Implementation The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] This invention proposes a method for monitoring the entire life cycle of engineering materials based on energy and carbon property embedding, such as... Figure 1 As shown, it includes the following steps: Step 1: For the target engineering material, construct a structure including material entity nodes. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. The construction of the energy-carbon property embedding vector in step 1 specifically includes: defining a node set. Including material entity nodes process nodes and energy nodes Among them, material nodes initial feature vector Input its physical and chemical parameters (density, thermal conductivity, etc.); A graph neural network (GAT) based on an attention mechanism is used to calculate material nodes. and its neighboring nodes Attention coefficient between :

[0025] And weighted aggregation of neighbor features to update the embedding vector ; Represents the summation index, representing the material node. All neighbors

[0026] First, construct a heterogeneous graph to capture the complex relationships between "materials-processes-energy". .

[0027] Node set Definition: Contains three types of nodes: a heterogeneous graph being constructed. It maps to the actual production chain structure: Material entity node This refers to specific batches of materials in the supply chain, such as "C30 concrete - batch A" or "T8 steel - batch B".

[0028] Processing technology nodes It is embodied in specific production behaviors in a factory, such as "stirring-high speed", "CNC cutting-roughing" or "heat treatment-quenching".

[0029] Energy type nodes This is manifested as the timing characteristics of the power grid, such as "electricity-peak hours" or "natural gas-boiler A". Material entity node. For example, "C30 concrete - batch A". Processing technology details. For example, "stirring-high speed". Energy node For example, "electricity - peak hours".

[0030] edge set Definition: Defines the interaction relationships between nodes, such as (Material process) or (Energy consumption in the process).

[0031] Feature initialization Initialize the feature vector for each node.

[0032] For material nodes: Input their physical and chemical parameters (density, thermal conductivity, Young's modulus).

[0033] For process nodes: Input their rated power, rated working hours, and other information.

[0034] Feature aggregation is performed using a graph neural network (GAT) based on an attention mechanism to capture meta-paths. Information flow on the internet.

[0035] For any material node , its first Layer embedding vector The update rules are as follows: Calculate attention coefficient : Measure node Its neighboring nodes Association strength:

[0036] in, It is a learnable weight matrix. It is an attention vector. This represents a vector concatenation operation; Represents material nodes The The layer embedding vector, which aggregates neighbor features weighted according to the attention coefficient, completes the first layer embedding vector. Layer embedding update: ;in ( ) represents the activation function. Indicates neighbors The previous layer embedding vector; go through After layer polymerization, the final material energy carbon embedding vector is obtained. .

[0037] To ensure The distance between vectors in high-dimensional space can reflect the energy consumption similarity of materials in the real world, and a contrastive learning mechanism with energy and carbon property constraints is introduced.

[0038] Definition of positive sample pairs: Selecting As "positive sample pairs", among them and It is a material embedding vector that is verified by multimodal carbon fingerprinting and has similar actual energy consumption data.

[0039] Optimization objective: Maximize the cosine similarity of positive sample pairs in the vector space.

[0040] loss function (InfoNCELoss): Used to train GNN weights, bringing positive samples closer and pushing negative samples further away. : This indicates the traversal of the negative sample index;

[0041] in, It is the cosine similarity function. This is the temperature coefficient.

[0042] Final output: High-dimensional carbon embedding vector of the material optimized through comparative learning. .

[0043] Step 2: During the material processing, high-frequency time-series data is collected in real time. and visual flow Align the visual flow using a cross-modal attention mechanism. With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. Step 2 specifically includes: A visual encoder is used to extract spatial feature vectors of video frames, while a temporal encoder is used to extract local trend features of high-frequency power sequences. A cross-attention mechanism is adopted, using visual features as a guide (Query) to perform weighted fusion of power features to compensate for physical delay or time deviation, thereby achieving a strong binding between visual processing actions and power peak features. Based on the fusion feature sequence Identify the start and end points of processing events. and The measured net energy consumption is calculated by integrating the difference between real-time power and baseline standby power over the effective processing time. .

[0044] Specifically, in section 2.1, regarding data input and modal feature extraction, the system simultaneously acquires and inputs two high-frequency data streams, and performs feature encoding on each: Visual flow feature extraction: Input video frame sequence captured by an industrial camera (e.g., 30fps). A visual encoder (e.g.) is used. or )extract Spatial feature vector of the image at any time :

[0045] Time-series feature extraction: High-frequency power sequences collected by smart meters as input. (e.g., 100Hz). A timing encoder (e.g.) is used. or Extract to Local trend characteristics centered :

[0046] High-frequency timing stream This signal is collected in real time by a smart meter or power sensor installed on the power supply circuit of the processing equipment. It reflects the instantaneous physical energy consumption of the equipment and requires a high sampling rate, such as 100Hz, to capture the local trend characteristics of the power waveform. Visual flow A sequence of video frames captured in real time by an industrial camera facing the processing area. This signal serves as semantic guidance, for example... The frame rate (fps) is used to identify actual physical contact, cutting, or heat treatment actions involving the material.

[0047] 2.2 Cross-modal spatiotemporal alignment and feature fusion Due to the physical delay between visual action (material contact) and power surge... Furthermore, there may be clock skew. This step employs the Cross-AttentionTransformer mechanism to perform precise alignment and weighted fusion in the feature space.

[0048] Define Query, Key, Value: based on visual features As a Query Guide the system to ensure its attention is focused on the actual processing actions; use temporal characteristics (window Power characteristics within) as Key( ) and Value( ):

[0049] in, It is a learnable projection matrix.

[0050] Calculate and fuse cross-attention weights: Calculate the visually guided power feature weighted fusion vector. :

[0051] This fusion feature By forcibly binding the "visual processing action" with the "power peak on the electricity meter", spatiotemporal alignment in the feature space was achieved, compensating for physical delay.

[0052] There is inevitably a physical delay of milliseconds between the material processing action (visual signal) and the sudden change in the actual power peak of the equipment (timing signal). To address clock skew, the system employs a cross-attention transformer mechanism for precise alignment and compensation in the feature space. (Visual features) Acting as a Query, it focuses the system's attention on the time sequence. Power mutations strongly correlated with processing actions in (Key and Value) enable strong feature binding between visual processing actions and the power peak of the meter.

[0053] 2.3 Processing Event Identification and Net Energy Consumption Integral Based on fusion feature sequences The system records the device status at each moment. Make a judgment and accurately calculate the net energy consumption of the materials. .

[0054] State classification and event detection: fusing features Input to a classifier (e.g., a fully connected layer classifier or) ), identify the current state State space At least including .

[0055] Determine the start and end points of processing: Based on the state classification results, determine the start and end times of the "effective processing" event. and .

[0056] Calculation of measured net energy consumption integral: within a defined effective processing time period Internally, for real-time power Compared with the reference standby power Integrate the difference over time to obtain the net physical energy consumption of the material. :

[0057] in, As an indicator function, integration is only performed when the process is determined to be in an "effective processing" state. This process achieves accurate deduction of idling energy consumption and baseline losses, solving the problem of coarse energy consumption allocation.

[0058] Step 3: Establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; using the input materials and the measured net energy consumption data obtained in step 2. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Specifically, section 3.1 defines the parameters and establishes the initial prior distribution, thereby establishing the energy-carbon conversion coefficient to be estimated. initial prior distribution And based on new observational data Establish the likelihood function ; Definition of parameters to be estimated: Let The energy-carbon conversion factor (Emission Factor) is a variable that the system needs to dynamically track and correct for a specific process or material category.

[0059] Prior distribution establishment During the system initialization phase, national standard libraries, industry average data, or historical data are used to build... The initial prior distribution. It is usually assumed that... Follows a Gaussian distribution:

[0060] in, As the initial estimated mean, The initial estimate variance represents the uncertainty in the initial parameter estimates.

[0061] 3.2 Likelihood function establishment and input of new observation data The system utilizes multimodal carbon fingerprint recognition in step 2 to obtain measured net energy consumption data. As a new observation sample .

[0062] New observation dataset The new observation dataset is defined as follows: .

[0063] Material feature vectors (from) (or its derived features).

[0064] Corresponding to Actual net energy consumption .

[0065] Likelihood function establishment Assume that the measurement error follows a mean of zero and a variance of . The normal distribution of the observed data is used to establish the parameters. Likelihood function:

[0066] In simplifying linear models Under the given assumptions, the likelihood function is:

[0067] 3.3 Bayesian Posterior Update and Dynamic Factor Calculation Bayes' Theorem Update: Using Bayes' Theorem, the posterior distribution of the previous iteration is updated. As the prior distribution for this time And combined with the likelihood function Update parameter distribution:

[0068] Corrected dynamic emission factors (Scalar form): The updated posterior distribution mean when the parameter is a scalar and follows a conjugate Gaussian distribution. This is the calibrated dynamic emission factor, and its calculation formula is:

[0069] This mean The system represents the best estimate of the energy-carbon conversion coefficient based on the latest measured net energy consumption data. Representing the initial estimate variance, its core function is to quantify the system's initial energy-carbon conversion coefficient. The confidence level represents the uncertainty of the system; the higher the value, the more uncertain the system is about the initial value. The higher the uncertainty, the lower the reliability of the initial prior knowledge; the smaller the value, the better the system's confidence in the initial value. The stronger the initial confidence, the more likely one is to believe The true value is very close to .

[0070] In multidimensional state updates and corrections, in complex engineering material systems, the energy-carbon conversion coefficient is typically a multidimensional vector containing multiple state variables. In this multidimensional state update and correction process, the system employs time-series inference methods such as Kalman filtering or particle filtering for parameter optimization. Kalman filtering is suitable for linear systems where both state and observation noise follow a Gaussian distribution, iteratively tracking the state through two steps: prediction and update. The mean and covariance matrices. Particle filtering: suitable for nonlinear and non-Gaussian systems, it approximates the posterior probability distribution through a set of randomly sampled particles, providing more robust dynamic correction capabilities.

[0071] Output: Dynamically updated emission factors, which are then used in the final carbon accounting model.

[0072] 3.4 Concept drift detection steps: Calculate the posterior distributions of the old and new parameters in the concept drift detection steps. and The Kullback-Leibler divergence between the values ​​is used to quantify the distribution difference; when the Kullback-Leibler divergence exceeds a preset threshold... When a significant drift is detected, a system alarm is triggered, and the corresponding baseline carbon embedding vector in the global knowledge base is officially updated. .

[0073] This step aims to achieve an understanding of the energy-carbon conversion coefficient by calculating statistical distances. Robust detection of distribution changes determines whether to formally integrate new calibration parameters into the global knowledge base and update the corresponding reference energy carbon embedding vector. .

[0074] Define the distribution of old and new parameters: After each time window or batch processing, the system tracks the parameters. The posterior distribution of . Definition: Posterior distribution of old parameters (benchmark):

[0075] New parameter posterior distribution (corrected):

[0076] in, and These represent the average energy-carbon conversion coefficients calculated for the new and old batches, respectively. and This represents the uncertainty variance of the corresponding distribution.

[0077] Calculate the Kullback-Leibler (KL) divergence: Use the Kullback-Leibler divergence. As an indicator quantifying the difference between the posterior distributions of old and new parameters, the KL divergence measures, in an information-theoretic sense, the divergence from the old distribution... Switch to new distribution The resulting information gain or degree of change. When and When all distributions are Gaussian, the analytical expression for the KL divergence is:

[0078] Drift detection and triggering mechanism: The calculated drift... Divergence value and preset tolerance threshold Comparison. When When this threshold is exceeded, the system determines that a significant concept drift has occurred.

[0079] Global knowledge base update and alert: When the drift determination condition is met, it indicates that a permanent, non-accidental change has occurred in the energy-carbon conversion coefficient (e.g., severe wear of equipment tools leading to a continuous decrease in energy efficiency). In this case, the system performs the following operations: Formal update of the baseline embedding vector: The system utilizes the corrected... and The parameters are then used to formally update the corresponding baseline energy carbon embedding vector in the global knowledge base. This ensures that subsequent energy-carbon similarity calculations and low-carbon process reverse engineering are based on the latest, experimentally calibrated material energy-carbon properties.

[0080] Trigger system alarms: Trigger system alarms to remind maintenance personnel to perform maintenance checks and implement predictive maintenance or process optimization suggestions.

[0081] This mechanism enables the present invention to have the ability of model adaptive evolution (Evolutionary Correction), ensuring that the calculation accuracy continues to improve as the system is used over time.

[0082] Step 4, based on the energy carbon embedding vector Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

[0083] This step is the system's final calculation output. By integrating the implicit carbon properties of the Material2Vec module, the measured energy consumption of the multimodal carbon fingerprint module, and the dynamic emission factor of the Bayesian correction module, a single material unit is calculated. carbon emissions throughout the entire life cycle .

[0084] 4.1 Integrated Carbon Accounting Model Single material unit carbon emissions throughout the entire life cycle It is broken down into implicit carbon emissions (from the upstream supply chain) and process carbon emissions (from local processing operations). It consists of two main parts.

[0085]

[0086] 4.2 Technical breakdown and traceability of each component This comprehensive accounting model ensures that every step, from material properties to final emission factors, is based on dynamic or measured data.

[0087] 4.2.1 Embedded Carbon Emissions Hidden carbon emissions Responsible for calculating the upstream carbon footprint of materials before they enter local processing procedures (such as raw material mining and transportation). Material unit The high-dimensional energy-carbon embedding vector is generated and optimized using the heterogeneous graph neural network (HGNN) in step 1. This vector implicitly contains information about the material's energy-carbon performance potential and upstream carbon footprint. The weight vector is a learnable parameter used to embed high-dimensional vectors. Linear mapping to the corresponding implicit carbon value.

[0088] 4.2.2 Process Net Energy Consumption and Dynamic Emissions Process carbon emissions calculation is performed on a material unit basis. In each process Accurately calculate real-time consumption.

[0089] Actual net energy consumption This is the measured value obtained through step 2. It represents the material. In the process Effective processing time Inside, the baseline standby power was precisely deducted. Net physical energy consumption after:

[0090] Real-time calibration of emission factors This is obtained through Bayesian adaptive correction in step 3, which is a process. In time t The real-time calibration emission factor, derived from the Bayesian inference module, is the calibrated mean. This factor is dynamic (over time). The changes reflect the actual impact of the current state of the equipment (such as aging and wear) and process fine-tuning on the energy-carbon conversion coefficient, thereby solving the deviation problem caused by relying on static emission factors.

[0091] 4.2.3 Scientifically allocate idling losses To achieve end-to-end carbon accounting and avoid including all idling losses in a single material, the model introduces a scientific allocation mechanism. The allocation items include: , : Indicates process Total idling energy consumption. : Indicates material In the process Effective processing time. The total operating time of the equipment throughout the entire accounting period. This allocation item ensures the energy consumption during idling. The allocation is reasonably weighted according to the time occupied by the materials, avoiding the "iron rice bowl" allocation defect in traditional accounting.

[0092] Final output: As materials The material-level carbon footprint traceability report, also known as the core data of the Digital Carbon Passport (DPP), is used to support upper-level applications such as supply chain optimization and reverse engineering of low-carbon processes.

[0093] This method can also be further applied to low-carbon process reverse engineering and technical parameter optimization, specifically including: obtaining the energy-carbon embedding vector of the material to be optimized. Low-carbon process reverse deduction and parameter optimization based on energy carbon embedding vector.

[0094] This embodiment details how to utilize the energy-carbon embedding vector generated in step 1. and Bayesian corrected dynamic emission factor This enables intelligent evaluation and optimization suggestions for existing processing routes, aiming to minimize carbon emissions from material units. .

[0095] Material information acquisition and feature extraction to be optimized Obtain the carbon embedding vector : Obtain materials to be optimized High-dimensional carbon embedding vector This vector was generated by a heterogeneous graph neural network (HGNN) and optimized through contrastive learning (InfoNCELoss). This vector implicitly contains material... Energy and carbon performance potential and energy consumption sensitivity under different process routes.

[0096] Get current technical parameters: Get material In the target process (e.g., heat treatment process) The current set of key technical parameters in ) Processing rate Temperature curve Timing scheduling (such as peak hours and trough hours).

[0097] High-dimensional space simulation based on generative artificial intelligence (GenAI) The system utilizes GenAI models (e.g., models based on graph variational autoencoders (GVAE) or reinforcement learning RL) in Simulation and generation of process paths in vector space.

[0098] Simulation space definition: Defines the set of potential process paths. Each path Corresponding to a new set of technical parameters This represents an adjustable process scheme while maintaining product quality.

[0099] GenAI Model Input and Output: The input to the GenAI model is... and the current graph structure The model's task is to generate technical parameters that can significantly reduce carbon emissions. :

[0100] GenAI, based on high-dimensional vector space, Energy carbon tendency, simulating different process nodes and energy nodes Combined, to discover low-energy areas.

[0101] Predicting energy consumption and carbon emissions: for each simulated path (corresponding parameters) The system predicts its net energy consumption. and total carbon emissions Predictive models utilize With calibrated dynamic emission factors Perform the calculation:

[0102] Recommended Optimization Objectives and Key Technical Parameters Optimize the objective function: The system's goal is to find an optimal path. This makes the material In the process Total carbon emissions Minimize:

[0103] This optimization process is simultaneously constrained by product quality and process constraints.

[0104] Recommended key technical parameters: The system follows the optimal path Extract the corresponding technical parameters As a recommendation, these parameters can significantly reduce energy consumption and carbon emissions. The recommended key technical parameters include, but are not limited to: Improved processing speed According to the material Based on the hardness embedding characteristics, an optimal processing rate is recommended to minimize the energy consumption per unit time.

[0105] The system follows the optimal path Extract the corresponding technical parameters As a recommendation, these parameters can significantly reduce energy consumption and carbon emissions. The recommended key technical parameters include, but are not limited to: Improved processing speed According to the material Hardness embedding features We recommend an optimal processing rate to minimize energy consumption per unit time.

[0106] Heat treatment timing optimization (Carbon strength oriented): According to energy nodes Real-time carbon intensity (e.g., electricity) This intensity is determined by real-time calibration of emission factors. Implicit and corrective measures. The system recommends scheduling high-energy-consuming processes (such as heat treatment and high-temperature drying) to the time period when the grid's carbon emission intensity is lowest.

[0107] The optimization objective can be formulated as follows:

[0108] Temperature or pressure profile adjustment: It is recommended to fine-tune the specific temperature or pressure profile to match the current material. Reduce energy consumption sensitivity and decrease sudden energy consumption and idling energy consumption during processing.

[0109] Temperature or pressure profile adjustment: It is recommended to fine-tune the specific temperature or pressure profile to match the current material. Energy consumption sensitivity.

[0110] Output: The system outputs structured suggestions for adjusting low-carbon processes, achieving this from the perspective of material properties ( From actual production decisions () This completes the closed loop, achieving the goals of predictive optimization and supply chain optimization.

[0111] The method of this invention employs a variety of technologies, including graph neural networks, multimodal recognition, and Bayesian inference. The data-driven, adaptively evolving, and precise carbon monitoring closed loop established between these modules, as well as the resulting synergistic effect, are not simply a combination of each other.

[0112] The first synergistic effect of this invention is reflected in the real-time calibration of emission factors. Step 2 involves the analysis of high-frequency time-series streams. and visual flow Perform cross-modal attention alignment to accurately strip and calculate the net physical energy consumption of material cells. .These Data (i.e., new observation samples) The precise input in step 3 then drives the Bayesian inference method to update the energy-carbon conversion coefficient. The posterior distribution. This collaborative mechanism ensures that the accounting no longer relies on a static average factor, but instead obtains a real-time calibrated emission factor with weights dynamically tilted towards the latest measured data. This ensures that the emission factors are highly accurate for specific single materials and single processes.

[0113] During model evolution and feedback updates of the energy carbon embedding vector, the second synergistic effect of this invention is reflected in the model's adaptive evolutionary capability. The Bayesian inference module further introduces a mechanism based on... The concept of divergence drift detection mechanism constitutes the key feedback loop of the system. The system quantifies the posterior distribution of the old and new parameters. and The difference between them robustly determines whether the carbon conversion coefficient has suffered a persistent model bias (e.g., equipment aging or permanent changes in material properties). Only when Divergence exceeds preset threshold Only then does the system formally trigger the processing of the corresponding energy-carbon embedding vector in the global knowledge base. The update. This enables the material generated via heterogeneous graph neural networks (HGNNs). Vectors are no longer static descriptions of attributes, but have evolved into dynamic, adaptive material energy and carbon states calibrated through measured feedback. This closed-loop feedback mechanism is the core innovation of this invention, ensuring that the longer the system is used, the higher the calculation accuracy becomes.

[0114] Intelligent decision optimization empowered by dynamic features, and the corrected dynamic features Combined with precise accounting (Include The contributions of these factors, along with their contributions to generative artificial intelligence (GenAI) models, serve as input for the reverse engineering of low-carbon processes. This deep synergy enables... Able to utilize The vector-embedded energy and carbon potential is efficiently simulated in high-dimensional space to discover low-energy consumption areas, and optimal key technical parameters are recommended with the goal of minimizing carbon emissions. This process achieves a complete and efficient closed loop from high-precision data acquisition and model adaptive correction to final intelligent decision support, which is a key technological advantage supporting material-level low-carbon process optimization decisions.

[0115] This invention also discloses a life-cycle monitoring device for engineering materials based on energy and carbon property embedding, comprising: Construct a carbon attribute embedding vector module to build material entity nodes for target engineering materials. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. ; The multimodal carbon fingerprint recognition module is used to collect high-frequency time-series data in real time during material processing. and visual flow Align the visual flow using a cross-modal attention mechanism. With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. ; An adaptive correction module for the energy-carbon factor is used to establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; measured net energy consumption data obtained using the input material and the multimodal carbon fingerprint recognition module. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Dynamic carbon-nucleic acid modules are used for energy-based carbon embedding vectors. Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

[0116] It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0117] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0118] It should be understood that the above-described memory is exemplary but not restrictive. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0119] This application also provides a computer-readable storage medium for storing computer programs.

[0120] Optionally, the computer-readable storage medium can be applied to the terminal device in the embodiments of this application, and the computer program causes the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0121] This application also provides a computer program product, including computer program instructions.

[0122] Optionally, the computer program product can be applied to the terminal device in the embodiments of this application, and the computer program instructions cause the computer to execute the corresponding processes implemented by the mobile terminal / terminal device in the various methods of the embodiments of this application. For the sake of brevity, they will not be described in detail here.

[0123] This application also provides a computer program.

[0124] Optionally, the computer program can be applied to the vehicle autonomous driving device in the embodiments of this application. When the computer program is run on a computer, it causes the computer to execute the corresponding processes implemented by the terminal device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0126] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0128] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for monitoring the entire life cycle of engineering materials based on energy and carbon attribute embedding, characterized in that, Includes the following steps: Step 1: For the target engineering material, construct a structure including material entity nodes. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. ; Step 2: During the material processing, high-frequency time-series data is collected in real time. and visual flow ; Align the visual flow using a cross-modal attention mechanism With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. ; Step 3: Establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; using the input materials and the measured net energy consumption data. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Step 4, based on the energy carbon embedding vector Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

2. The method according to claim 1, wherein, The construction of the energy carbon property embedding vector in step 1 specifically includes: For material nodes process nodes and energy nodes Initialize the feature vector, where the initial feature vector of the material node includes the input physicochemical parameters; A graph neural network (GAT) based on an attention mechanism is used to calculate the attention coefficients between nodes based on meta-paths, and the node embedding vector is iteratively updated by weighted aggregation of neighbor features. Employing a contrastive learning mechanism constrained by energy and carbon properties, this approach optimizes the loss function to maximize the embedding vectors of materials with similar energy and carbon properties, as verified by measured net energy consumption. Similarity between them, material embedding vectors Positive sample pairs are formed in the loss function.

3. The method according to claim 1, wherein, Step 2 specifically includes: A visual encoder is used to extract spatial feature vectors of video frames, while a temporal encoder is used to extract local trend features of high-frequency power sequences. A cross-attention mechanism is adopted, using visual features as a guide (Query) to perform weighted fusion of power features to compensate for physical delay or time deviation, thereby achieving a strong binding between visual processing actions and power peak features. Based on the fusion feature sequence Identify the start and end points of processing events. and The measured net energy consumption is calculated by integrating the difference between real-time power and baseline standby power over the effective processing time. .

4. The method according to claim 1, wherein, The adaptive correction of the energy carbon factor in step 3 specifically includes: Establish the energy-carbon conversion coefficient to be estimated. initial prior distribution And based on new observational data Establish the likelihood function ; Update the energy-carbon conversion coefficient to be estimated using Bayes' theorem. posterior distribution Thus, the corrected mean can be calculated. ; In multidimensional state updates and corrections, Kalman filtering or particle filtering time-series inference methods are used for parameter optimization.

5. The method according to claim 1, wherein, The method in step 3 further includes a concept drift detection step: Calculate the posterior distributions of the old and new parameters in the concept drift detection step. and The Kullback-Leibler divergence between the values ​​is used to quantify the distribution difference; when the Kullback-Leibler divergence exceeds a preset threshold... When a significant drift is detected, a system alarm is triggered, and the corresponding baseline carbon embedding vector in the global knowledge base is officially updated. .

6. According to the method of claim 5, if a significant drift is determined, then the equipment has aged or the material properties have been permanently changed.

7. The method according to claim 1, wherein, In step 4, the dynamic carbon accounting obtains the life-cycle carbon emissions by weighted summing of the following three items. Based on embedding vectors Calculated initial implicit carbon emissions from materials; measured net energy consumption. With adaptively corrected dynamic emission factors The product of the product and the proportion based on time or output, which is used to scientifically allocate the shared idling loss.

8. The method according to claim 1, characterized in that: in, The method is further applied to low-carbon process reverse engineering and technical parameter optimization, specifically including: obtaining the energy-carbon embedding vector of the material to be optimized. The vector implicitly represents the energy and carbon performance potential of the material under different process pathways; using a generative artificial intelligence (GenAI) model, based on the vector... Vectors are used to perform high-dimensional space simulations of different potential process paths; based on the simulation results, key technical parameters required for the material to minimize energy consumption and carbon emissions in a specific target process are recommended.

9. The method according to claim 1, characterized in that: in, The key technical parameters include, but are not limited to, processing rate, heat treatment sequence, and temperature or pressure heat treatment sequence adjustment.

10. A life-cycle monitoring device for engineering materials based on energy and carbon property embedding, characterized in that, include: Construct a carbon attribute embedding vector module to build material entity nodes for target engineering materials. Processing technology nodes and energy type nodes heterogeneous graph Based on preset metapath By aggregating node neighbor features using a graph neural network mechanism, and through learned energy-carbon weights and nonlinear transformations, material properties are mapped into high-dimensional continuous energy-carbon embedding vectors. ; The multimodal carbon fingerprint recognition module is used to collect high-frequency time-series data in real time during material processing. and visual flow ; Align the visual flow using a cross-modal attention mechanism With the high-frequency timing stream Extracted features, generate fused feature sequences According to the fusion feature sequence Accurately distinguish between the effective processing state and the idling or standby state of materials, and accurately separate and measure the net physical energy consumption of materials in the process. ; An adaptive correction module for the energy-carbon factor is used to establish the energy-carbon conversion coefficient to be estimated. The initial prior distribution; measured net energy consumption data obtained using the input material and the multimodal carbon fingerprint recognition module. A likelihood function is established using the sample set; the posterior distribution is calculated and updated based on Bayesian inference to obtain the calibrated dynamic emission factor. ; Dynamic carbon-nucleic acid modules are used for energy-based carbon embedding vectors. Implied carbon emissions, and corrected dynamic emission factors Compared with the measured net energy consumption The carbon emissions of this material unit throughout its entire life cycle are calculated comprehensively. .

Images