Carbon emission intelligent prediction method and system based on big data

Abstract

The invention discloses a carbon emission intelligent prediction method and system based on big data, and relates to the technical field of carbon emission monitoring, and the method comprises the steps: collecting carbon emission associated data, and generating a carbon emission feature tensor through quantum time-space coding and time-space grid alignment; constructing a dynamic causal graph network through causal entropy on the basis of the carbon emission feature tensor, and generating a causal weight matrix by combining anti-fact intervention and a Bayesian false-rejecting causal relationship; constructing a federated learning framework based on the causal weight matrix, deploying a federated aggregator at a cloud end to aggregate the encryption gradient of each edge node, and embedding a causal regular term in a federated loss function to perform joint training to generate a global carbon emission prediction model; according to the method, efficient fusion of industrial sensor data, satellite remote sensing data and other multi-source heterogeneous data is realized by utilizing quantum bit superposition state mapping and quantum entanglement state association technologies.

Description

Technical Field

[0001] The present invention relates to the field of carbon emission monitoring technology, and in particular to a carbon emission intelligent prediction method and system based on big data. Background Art

[0002] In recent years, carbon emission monitoring and forecasting technology has become a research hotspot in environmental science and industrial intelligence. Traditional carbon emission forecasting methods primarily rely on static statistical models and physical process-based simulations, such as the STIRPAT model and the LEAP model. With the development of big data technology, machine learning methods such as LSTM and random forests have been introduced to the field of carbon emission forecasting, significantly improving forecast accuracy. Furthermore, the widespread adoption of new data collection technologies, such as satellite remote sensing and IoT sensors, has provided multi-source, heterogeneous data support for carbon emission monitoring. Regarding technological convergence, the maturity of distributed computing frameworks, such as federated learning and edge computing, has provided new approaches to addressing the data silo problem.

[0003] Current carbon emission forecasting technology suffers from two key flaws: First, when it comes to multi-source data fusion, traditional methods struggle to effectively handle the spatiotemporal heterogeneity between industrial sensor data and satellite remote sensing data, resulting in inadequate feature extraction. Second, in terms of policy response mechanisms, existing models often employ static parameter update strategies, making them unable to adapt in real time to the dynamic impacts of policy adjustments such as carbon taxes. In particular, when new policies are implemented, traditional methods require retraining the entire model, resulting in delayed responses and wasted computing resources. Furthermore, most forecasting models lack causal inference mechanisms, making it difficult to explain the transmission pathways of carbon emissions, limiting the effectiveness of decision support. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a carbon emission intelligent prediction method based on big data to solve the problems of insufficient accuracy of multi-source data fusion and delayed policy response.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a carbon emission intelligent prediction method based on big data, which includes collecting carbon emission related data, generating a carbon emission feature tensor through quantum space-time coding and space-time network alignment; based on the carbon emission feature tensor, constructing a dynamic causal graph network through causal entropy, combining counterfactual intervention and Bayesian false causal relationship to generate a causal weight matrix; constructing a federated learning framework based on the causal weight matrix, deploying a federated aggregator in the cloud to aggregate the encrypted gradients of each edge node, and embedding causal regularization terms in the federated loss function for joint training to generate a global carbon emission prediction model; deploying the global carbon emission prediction model to the edge node, triggering the dynamic causal intervention mechanism when a new carbon tax policy is detected, adjusting the attention weight of the global carbon emission prediction model through incremental learning, and generating an optimized global carbon emission prediction model; based on the optimized global carbon emission prediction model, integrating operation data and satellite observation data in real time at the industrial Internet of Things terminal, and performing millisecond-level carbon emission prediction through the edge AI acceleration chip.

[0008] As a preferred solution of the big data-based intelligent carbon emission prediction method described in the present invention, the carbon emission related data includes industrial sensor data, satellite remote sensing data, logistics GPS data and policy text data.

[0009] As a preferred solution of the carbon emission intelligent prediction method based on big data described in the present invention, wherein: the carbon emission feature tensor is generated by quantum space-time coding and space-time grid alignment, the specific steps are as follows:

[0010] The carbon emission correlation data is input into quantum space-time coding, and the parallel fusion and quantization coding of multi-source data are achieved through quantum bit superposition state mapping to generate cross-dimensional quantum coded data;

[0011] The cross-dimensional quantum-encoded data is aligned in space and time, the carbon emission intensity characteristics across grid cells are correlated through quantum entangled states, and a tensor folding operation is performed to generate a carbon emission feature tensor.

[0012] As a preferred solution of the carbon emission intelligent prediction method based on big data described in the present invention, wherein: based on the carbon emission feature tensor, a dynamic causal graph network is constructed by the optimal causal entropy method, and a causal weight matrix is generated by combining counterfactual intervention and Bayesian false causal relationship. The specific steps are as follows:

[0013] Based on the carbon emission feature tensor, a dynamic causal graph network is constructed using the optimal causal entropy algorithm, and a causal topology structure containing policy intervention nodes is generated;

[0014] Based on the causal topology results, counterfactual reasoning operations are applied to the policy intervention nodes to calculate the causal effect intensity distribution of each carbon emission node under the carbon tax adjustment scenario;

[0015] Based on the causal effect intensity distribution, the pseudo causal relationship edges with confidence levels below 0.05 are eliminated through the Bayesian scoring method, and then the causal weight matrix is generated through aggregation.

[0016] As a preferred solution of the carbon emission intelligent prediction method based on big data described in the present invention, wherein: the federated learning framework is constructed based on the causal weight matrix, a federated aggregator is deployed in the cloud to aggregate the encrypted gradients of each edge node, and a causal regularization term is embedded in the federated loss function for joint training to generate a global carbon emission prediction model. The specific steps are as follows:

[0017] In the cloud, the causal weight matrix is decomposed into a local causal graph stored in the edge nodes and a global causal topology stored in the federated aggregator. The cross-node causal paths are dynamically coordinated through a secure multi-party communication protocol to generate a causal adjacency matrix.

[0018] Based on the global causal topology, an encrypted channel is deployed in the federated aggregator to receive the encrypted causal gradients generated by each edge node based on the local causal graph.

[0019] The encrypted causal gradient is input into the federated loss function, and the regularization constraint is constructed in combination with the causal adjacency matrix. The neural network parameters and causal weight matrix of the carbon emission prediction model are synchronously updated through the joint optimization algorithm.

[0020] The optimized neural network parameters and causal weight matrix are coupled, and a global carbon emission prediction model is generated through joint representation learning of feature space projection and causal topology embedding.

[0021] As a preferred solution of the big data-based intelligent carbon emission prediction method described in the present invention, the global carbon emission prediction model is deployed to the edge node, a dynamic causal intervention mechanism is triggered when a new carbon tax policy is detected, the attention weight of the global carbon emission prediction model is adjusted through incremental learning, and an optimized global carbon emission prediction model is generated. The specific steps are as follows:

[0022] Deploy the global carbon emission prediction model to each edge node, configure the edge node to continuously collect policy text data streams, and transmit them to the federated aggregator in real time;

[0023] The federated aggregator parses the received policy data. When a new carbon tax policy is detected, it triggers the dynamic causal intervention mechanism and generates causal intervention instructions that are sent to each edge node.

[0024] Each edge node adjusts the attention weight of the global carbon emission prediction model through an incremental learning algorithm based on the causal intervention instructions, and uploads the adjustment results to the federated aggregator;

[0025] The federated aggregator aggregates the adjustment results of each edge node, jointly optimizes the updated attention weights and the causal weight matrix, and generates an optimized global carbon emission prediction model.

[0026] As a preferred solution of the carbon emission intelligent prediction method based on big data described in the present invention, wherein: based on the optimized global carbon emission prediction model, the operation data and satellite observation data are integrated in real time at the industrial Internet of Things terminal, and the millisecond-level carbon emission prediction is performed through the edge AI acceleration chip. The specific steps are as follows:

[0027] Based on the optimized global carbon emission prediction model, industrial IoT terminals are configured to synchronously collect equipment operation data and satellite observation data, and then perform spatiotemporal alignment to generate a standardized carbon emission characteristic dataset;

[0028] The global carbon emission prediction model is loaded through the edge AI acceleration chip, and the standardized carbon emission feature data set is analyzed using spatiotemporal causal reasoning to generate a carbon emission detection report.

[0029] In a second aspect, the present invention provides a carbon emission intelligent prediction system based on big data, including a carbon tensor generation module, a causal modeling module, a causal learning module, a carbon tax response module, and a carbon intelligent calculation module;

[0030] The carbon tensor generation module is used to collect carbon emission related data and generate a carbon emission feature tensor through quantum space-time coding and space-time network alignment; the causal modeling module is used to construct a dynamic causal graph network based on the carbon emission feature tensor through causal entropy, and generate a causal weight matrix by combining counterfactual intervention and Bayesian false causality; the causal learning module is used to build a federated learning framework based on the causal weight matrix, deploy a federated aggregator in the cloud to aggregate the encrypted gradients of each edge node, and embed causal regularization terms in the federated loss function for joint training to generate a global carbon emission prediction model; the carbon tax response module is used to deploy the global carbon emission prediction model to the edge node, trigger the dynamic causal intervention mechanism when a new carbon tax policy is detected, adjust the attention weight of the global carbon emission prediction model through incremental learning, and generate an optimized global carbon emission prediction model; the carbon intelligent calculation module is used to integrate operation data and satellite observation data in real time on the industrial Internet of Things terminal based on the optimized global carbon emission prediction model, and realize millisecond-level carbon emission prediction through edge AI acceleration chip.

[0031] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the carbon emission intelligent prediction method based on big data as described in the first aspect of the present invention.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent carbon emission prediction method based on big data as described in the first aspect of the present invention.

[0033] The beneficial effects of the present invention are: by utilizing quantum bit superposition state mapping and quantum entangled state correlation technology, efficient fusion of multi-source heterogeneous data such as industrial sensor data and satellite remote sensing data is achieved. Furthermore, by decomposing the causal topology into local causal graphs and global topological structures, and embedding causal regularization terms in the federated loss function, privacy-preserving collaborative training and dynamic causal intervention of cross-regional carbon emission data are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.

[0035] Figure 1 This is a flow chart of the intelligent prediction method for carbon emissions based on big data.

[0036] Figure 2 Schematic diagram of the carbon emission intelligent prediction system based on big data.

[0037] Figure 3 Flowchart of the alignment of quantum space-time coding and space-time grid for the big data-based intelligent carbon emission prediction method.

[0038] Figure 4 Flowchart for causal modeling and weight matrix generation for the big data-based carbon emission intelligent prediction method. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0042] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a carbon emission intelligent prediction method based on big data, comprising the following steps:

[0043] S1: Collect carbon emission correlation data and generate carbon emission feature tensors through quantum space-time coding and space-time grid alignment.

[0044] S1.1: Carbon emission-related data include industrial sensor data, satellite remote sensing data, logistics GPS data, and policy text data.

[0045] It should be noted that industrial sensor data provides real-time energy consumption and operating parameters at the production line equipment level, satellite remote sensing data obtains regional carbon dioxide concentration distribution and thermal characteristics, logistics GPS data reflects the trajectory of mobile emission sources in the supply chain transportation link, and policy text data records carbon tax regulations and other regulatory information.

[0046] S1.2: Input the carbon emission related data into quantum space-time coding, realize the parallel fusion and quantization coding of multi-source data through quantum bit superposition state mapping, and generate cross-dimensional quantum coded data.

[0047] Specifically, after the carbon emission related data is input into the quantum space-time coding, the quantum bit superposition state mapping synchronously converts the different characteristic dimensions of industrial sensor data, satellite remote sensing data, logistics GPS data and policy text data into quantum state representation, and realizes the parallel fusion processing of multi-source data under the quantum computing framework. The quantization coding process establishes cross-dimensional correlation relationships through the quantum entanglement effect, and finally generates cross-dimensional quantum coded data containing equipment operating parameters, environmental monitoring indicators, logistics trajectory characteristics and policy influencing factors.

[0048] S1.3: Align the cross-dimensional quantum encoded data in space and time, associate the carbon emission intensity characteristics across grid cells through quantum entanglement states, and perform tensor folding operations to generate a carbon emission feature tensor.

[0049] The specific process involves first dividing cross-dimensional quantum-encoded data into grid cells according to a preset spatiotemporal resolution. Quantum entanglement creates quantum coherence in the carbon emission intensity characteristics within different grid cells. Quantum gate operations enable cross-grid quantum state superposition, coherently fusing the quantum state characteristics of industrial sensor data, satellite remote sensing data, logistics GPS data, and policy text data. While maintaining the quantum characteristics of each data source, quantum correlations are established between grid cells. The fused quantum state data undergoes a tensor folding operation in a specific dimension for feature reorganization, ultimately outputting a carbon emission feature tensor with a unified spatiotemporal reference frame and complete feature expression. The preset spatiotemporal resolution is based on the industrial sensor data acquisition frequency, the satellite remote sensing data spatial resolution, and the logistics GPS data update cycle, and is determined by the optimal grid division algorithm within the quantum computing framework.

[0050] S2: Based on the carbon emission feature tensor, a dynamic causal graph network is constructed through causal entropy, and a causal weight matrix is generated by combining counterfactual intervention and Bayesian false causality elimination.

[0051] S2.1: Based on the carbon emission feature tensor, a dynamic causal graph network is constructed through the optimal causal entropy algorithm, and a causal topology structure containing policy intervention nodes is generated.

[0052] The specific process involves the carbon emission feature tensor entering the optimal causal entropy algorithm processing flow. Industrial sensor data feature nodes are associated with satellite remote sensing data feature nodes in quantum state space. Logistics GPS data feature nodes are connected to industrial equipment nodes through spatiotemporal constraints. The optimal causal entropy algorithm evaluates the information transmission strength between each feature node and dynamically adjusts the causal direction between industrial equipment nodes and environmental monitoring nodes. Policy text data is processed through natural language processing to generate policy intervention nodes, which are then weightedly connected to industrial equipment nodes using a preset intervention intensity coefficient. During the iterative update process, the dynamic causal graph network maintains spatiotemporal consistency constraints between industrial equipment nodes and logistics path nodes. The resulting causal topology includes equipment operation feature nodes, environmental monitoring feature nodes, logistics trajectory feature nodes, and policy intervention feature nodes. The weights of the directed connections between these nodes reflect the causal strength of the carbon emission transmission path. The preset intervention intensity coefficient is quantitatively set based on the stringency of carbon tax regulations in the policy text data and historical policy implementation effectiveness data. The semantic information in the policy text is converted into numerical intervention parameters using a Bayesian probability model.

[0053] S2.2: Based on the causal topology results, counterfactual reasoning operations are applied to the policy intervention nodes to calculate the causal effect intensity distribution of each carbon emission node under the carbon tax adjustment scenario. The expression is:

[0054]

[0055] Where j represents the target node number in the carbon emission network, Ψ j represents the causal effect intensity value of the jth node in the carbon emission network under the carbon tax adjustment scenario, E represents the mathematical expectation operator, T represents the carbon tax policy intervention variable, and t ’ represents the new tax rate after adjustment, do represents the intervention case in causal inference, Indicates that the policy variable T is forced to be set to the adjusted tax rate t for the target node j through the do operator ’ The counterfactual carbon emission forecast value is t, which represents the current carbon tax base rate. represents the counterfactual carbon emission prediction value when the policy variable T is forced to be set to the benchmark tax rate t for the target node j through the do operator, K represents the maximum propagation order, k represents the current order of network propagation, and λ k represents the attenuation coefficient of the k-th order network propagation, C k Represents the network centrality index calculated by the k-th order neighbor topology structure, N j represents the set of neighbor nodes of target node j, m represents the sequence number of the current heterogeneity adjustment factor, M represents the total number of heterogeneity adjustment factors, X m represents the mth heterogeneity regulator, represents the mth heterogeneity regulatory factor X m Flexible adjustment items.

[0056] The specific process includes: in the causal topology structure, the policy intervention node starts the counterfactual reasoning process after receiving the carbon tax adjustment instruction. The intervention node first forcibly sets the policy intervention variable from the current carbon tax base tax rate value to the adjusted new tax rate value, and generates the counterfactual carbon emissions prediction value of the target node under the adjusted tax rate scenario; at the same time, the policy intervention variable is kept at the current carbon tax base tax rate value, and the counterfactual carbon emissions prediction value of the target node under the baseline scenario is obtained.

[0057] Furthermore, the mathematical expectation operator compares the prediction differences under the two scenarios and obtains the causal effect intensity value of the target node under the carbon tax adjustment scenario; the causal effect intensity value diffuses step by step in the network according to the maximum propagation order, and the propagation intensity is adjusted according to the attenuation coefficient and network centrality index corresponding to the current propagation order during each propagation process; when the set of neighboring nodes of the target node receives the propagation effect, the heterogeneity adjustment factor corrects the propagation deviation through the elastic adjustment term according to the node characteristics; finally, a causal effect intensity distribution covering all nodes of the carbon emission network is formed, reflecting the impact of the carbon tax policy adjustment.

[0058] S2.3: Based on the causal effect intensity distribution, the pseudo causal relationship edges with confidence levels below 0.000 are eliminated through the Bayesian scoring method, and the causal weight matrix is aggregated and generated.

[0059] The specific process involves inputting the causal effect intensity distribution into the Bayesian scoring process. The Bayesian scoring method evaluates the posterior probability confidence of each directed edge in the causal topology. Causal edges with confidence levels below a preset threshold are marked as pseudo-causal edges and removed from the network. Retained causal edges are normalized based on their corresponding causal effect intensity values. The causal edge weights between industrial equipment nodes are modified based on the characteristics of equipment operating parameters. The causal edge weights between environmental monitoring nodes are adjusted to account for spatial correlation. The causal edges output by policy intervention nodes maintain the initial intervention intensity coefficient. Finally, all valid causal edges that have been screened and weighted are aggregated according to node connectivity to generate a causal weight matrix that reflects the actual carbon emission transmission path. The preset threshold is automatically determined by the inflection point of the kernel density estimate curve of the Bayesian posterior probability confidence, and is verified and adjusted based on domain knowledge.

[0060] S3: Build a federated learning framework based on the causal weight matrix, deploy a federated aggregator in the cloud to aggregate the encrypted gradients of each edge node, and embed causal regularization terms in the federated loss function for joint training to generate a global carbon emission prediction model.

[0061] S3.1: In the cloud, the causal weight matrix is decomposed into a local causal graph stored in the edge nodes and a global causal topology stored in the federated aggregator. The cross-node causal paths are dynamically coordinated through a secure multi-party communication protocol, and a causal adjacency matrix is generated.

[0062] The specific process involves dividing the causal weight matrix deployed in the cloud into two parts based on the physical location of the nodes: a local causal graph containing the direct interactions between industrial equipment is stored at the edge nodes, and a global causal topology reflecting the cross-regional carbon emission transmission paths is stored in the federated aggregator. The global causal topology is composed of the indirect relationships between nodes in the local causal graphs reported by each edge node, specifically manifested as multiple causal chains connected by environmental monitoring nodes. A secure multi-party communication protocol establishes an encrypted communication channel between the edge nodes and the federated aggregator, periodically and synchronously updating the node status change information of the local causal graph. The federated aggregator reassembles the cross-regional connection relationships in the local causal graph based on the latest data and dynamically adjusts the path weights in the global causal topology.

[0063] Furthermore, edge nodes receive global causal topology update instructions from the federated aggregator, adjusting the connection strengths of nodes in their local causal graphs with other regions. After bidirectional data coordination, the federated aggregator integrates the causal connections confirmed by all edge nodes and outputs a causal adjacency matrix containing the direct and indirect interaction strengths between the complete nodes.

[0064] S3.2: Based on the global causal topology, an encrypted channel is deployed in the federated aggregator to receive the encrypted causal gradients generated by each edge node according to the local causal graph.

[0065] The specific process involves establishing an encrypted communication link between the federated aggregator and each edge node based on the global causal topology. The edge node extracts causal gradient parameters based on the causal relationships between nodes in the local causal graph and encrypts the causal gradient using homomorphic encryption. The encrypted causal gradient is then uploaded to the federated aggregator via a secure transmission channel. The federated aggregator then verifies the edge node's identity and receives the encrypted data packet.

[0066] Furthermore, the global causal topology maintained by the federated aggregator records the causal gradient update status reported by each edge node, ensuring that the specific connections in the local causal graph are not exposed during data synchronization. The received encrypted causal gradients are securely aggregated within the federated aggregator to update the connection weights of cross-regional nodes in the global causal topology while preserving the privacy of local data on edge nodes. After gradient aggregation is complete, the federated aggregator distributes the updated global causal topology adjustment information to the relevant edge nodes via an encrypted channel, ensuring consistency between the local causal graph and the global causal topology.

[0067] S3.3: Couple the optimized neural network parameters and the causal weight matrix, and generate a global carbon emission prediction model through joint representation learning of feature space projection and causal topology embedding.

[0068] The specific process involves collaboratively mapping the optimized neural network parameters and the causal weight matrix in a shared feature space. The neural network parameters capture the temporal dynamics of industrial equipment operating characteristics, while the causal weight matrix encodes the carbon emission transmission relationship between nodes. Feature space projection maps the equipment operating parameters into a low-dimensional latent space, and causal topology embedding converts the causal weight matrix into a node representation vector.

[0069] Specifically, during the joint representation learning process, the output of the feature space projection interacts with the vector embedded in the causal topology through an attention mechanism, establishing a correlation pattern between changes in device features and the causal path of carbon emissions. Regularization constraints are applied to the fully connected layer weights in the neural network parameters and the edge weights in the causal weight matrix to ensure that the prediction results conform to both the data characteristics and the causal topology. Finally, the combined output of the integrated feature representation and causal relationships is transformed through a multi-layer perceptron to generate a global carbon emissions prediction model that simultaneously reflects device operating status and cross-regional carbon flows.

[0070] S4: Deploy the global carbon emission prediction model to the edge node, trigger the dynamic causal intervention mechanism when a new carbon tax policy is detected, adjust the attention weight of the global carbon emission prediction model through incremental learning, and generate an optimized global carbon emission prediction model.

[0071] S4.1: Deploy the global carbon emission prediction model to each edge node, configure the edge node to continuously collect policy text data streams, and transmit them to the federated aggregator in real time.

[0072] The specific process involves distributing a global carbon emissions prediction model to each edge node via a secure deployment protocol. After loading the model, the edge node activates the data collection service, continuously monitoring the sensor outputs of connected industrial equipment and policy text data streams. The policy text data stream is parsed by a natural language processing component to extract features of carbon-related regulatory clauses. These features are then combined with device operational data to form a structured input. The edge node then processes the structured input using the locally deployed global carbon emissions prediction model, generating a timestamped cache of prediction results.

[0073] Furthermore, edge nodes simultaneously configure data forwarding rules and transmit policy text data streams in real time via encrypted channels to the federated aggregator. During transmission, differential privacy techniques are used to anonymize sensitive fields in policy clauses to ensure data privacy. After receiving the policy text data streams uploaded by each edge node, the federated aggregator automatically triggers a global causal topology update mechanism. This updated topology information is pushed to each edge node via a secure channel. The edge node then dynamically adjusts the causal weight parameters in its local global carbon emissions prediction model to ensure that the prediction results remain synchronized with the latest policy requirements.

[0074] S4.2: The federated aggregator parses the received policy data. When a new carbon tax policy is detected, it triggers the dynamic causal intervention mechanism and generates causal intervention instructions that are sent to each edge node.

[0075] The specific process involves receiving policy text data streams uploaded by edge nodes, initiating a natural language processing process to parse the text content, and using keyword extraction and semantic analysis techniques to identify policy clause types. Upon detecting a new policy that includes an adjustment to the carbon tax rate or a change in the scope of collection, the federated aggregator activates the dynamic causal intervention mechanism. This mechanism first matches policy clauses with intervention nodes in the global causal topology and quantifies the semantic strength of the policy text to generate an intervention intensity coefficient. The federated aggregator then weights and fuses the intervention intensity coefficient with the historical impact parameter of the corresponding intervention node to generate updated causal intervention parameters.

[0076] Furthermore, the causal intervention instruction, which includes the intervention node number, effective time, and adjusted intervention intensity coefficient, is distributed to all associated edge nodes via a secure transmission channel. Upon receiving the causal intervention instruction, the edge node modifies the parameter weights of the corresponding intervention node in the local global carbon emissions forecast model at the specified effective time, and simultaneously updates the connection strengths of the relevant nodes in the local causal graph. The federated aggregator monitors the parameter update status of each edge node to ensure consistent execution of the dynamic causal intervention mechanism across all nodes. After the intervention parameter synchronization is complete, the global causal topology records the execution log of this policy change, maintaining traceability of the causal relationship.

[0077] S4.3: Each edge node adjusts the attention weight of the global carbon emission prediction model through an incremental learning algorithm based on the causal intervention instructions, and uploads the adjustment results to the federated aggregator.

[0078] Specifically, after receiving a causal intervention instruction from the federated aggregator, the edge node parses the instruction to identify the intervention node and the intervention intensity coefficient. Based on this received intervention intensity coefficient, an incremental learning algorithm performs a gradient update on the attention weights associated with the intervention node in the global carbon emissions prediction model. This update preserves the underlying distribution characteristics of the model's original parameters and adjusts only the attention weight parameters relevant to the policy intervention. The edge node verifies the adjusted global carbon emissions prediction model output using locally cached industrial equipment operating data to ensure that the forecast results align with the expected impact of the new policy.

[0079] Furthermore, after parameter adjustment, the edge node extracts the updated attention weight parameters and, after homomorphic encryption, generates a model adjustment report. This encrypted adjustment report is uploaded to the federated aggregator via a secure channel. The report contains the difference in attention weights before and after the adjustment, as well as the verification loss value. The federated aggregator receives model adjustment reports uploaded by all edge nodes, compares the differences in execution of the same intervention instruction across different nodes, and verifies the consistency of global policy execution. After completing the attention weight adjustment, the global carbon emissions prediction model maintained by the edge node continues to process real-time input device operation data and outputs carbon emissions prediction results that incorporate the latest policy impacts.

[0080] S4.4: The federated aggregator aggregates the adjustment results of each edge node, jointly optimizes the updated attention weights and the causal weight matrix, and generates an optimized global carbon emission prediction model.

[0081] The specific process involves the federated aggregator collecting the encrypted adjustment results uploaded by each edge node. It first decrypts the updated attention weight parameters of each node and aggregates the weight updates from different edge nodes using a federated averaging algorithm to generate a unified attention weight adjustment scheme. The aggregated attention weights are then co-optimized with the causal weight matrix maintained by the global causal topology. The optimization process uses an alternating direction multiplication method to alternately update the attention weights and causal weight matrix parameters, ensuring consistent representation of the two in the feature space.

[0082] Furthermore, during the joint optimization process, attention weight adjustments are propagated to the relevant nodes of the causal weight matrix via graph convolution operations, and the updated information of the causal weight matrix is fed back to the attention weight parameters. The optimized attention weights retain their sensitivity to policy intervention characteristics, and the causal weight matrix maintains an accurate depiction of the carbon emission transmission path. The federated aggregator re-encodes the jointly optimized attention weights and causal weight matrix parameters and generates a distributable version of the optimized global carbon emission prediction model through a secure deployment protocol. The optimized global carbon emission prediction model integrates the local adjustment experience of each edge node while maintaining the constraints of the global causal topology. The final output prediction results simultaneously reflect changes in device operating status and the impact of policy interventions.

[0083] S5: Based on the optimized global carbon emission prediction model, the industrial IoT terminal integrates operation data and satellite observation data in real time, and uses the edge AI acceleration chip to perform millisecond-level carbon emission prediction.

[0084] S5.1: Based on the optimized global carbon emission prediction model, configure the industrial Internet of Things terminal to synchronously collect equipment operation data and satellite observation data, and perform spatiotemporal alignment to generate a standardized carbon emission characteristic dataset.

[0085] The specific process involves deploying the optimized global carbon emissions prediction model to an IIoT terminal. The terminal then initiates a multi-source data acquisition protocol to simultaneously acquire real-time operating parameters of connected industrial equipment and satellite remote sensing observation data from a designated area. Industrial equipment operating data includes current and voltage readings from energy consumption monitors and production status signals collected by process sensors. Satellite observation data provides vertical profiles of atmospheric composition in the target area. During data collection, the IIoT terminal activates the spatiotemporal alignment service, using a global positioning clock to synchronize the timestamps of the equipment operating data and matching the spatial coverage of the satellite observation data based on geographic coordinates.

[0086] After completing the spatial and temporal benchmark unification, the IIoT terminals executed a feature engineering process, converting device current and voltage readings into power features, parsing satellite atmospheric composition data into concentration gradient features, and encoding process status signals into operating condition classification features. These features were then standardized according to a unified dimension, with power features cross-validated against concentration gradient features, and operating condition classification features annotated as metadata. The resulting standardized carbon emissions feature dataset, comprising time-aligned device operation features, spatially aligned satellite observation features, and verified operating condition labels, provides multi-dimensional input for the global carbon emissions prediction model.

[0087] S5.1: Load the global carbon emission prediction model through the edge AI acceleration chip, use spatiotemporal causal reasoning to analyze the standardized carbon emission feature data set, and generate a carbon emission detection report.

[0088] The specific process involves loading the global carbon emissions prediction model onto the edge AI accelerator chip and then initiating the model inference engine to process the standardized carbon emissions feature dataset. Spatiotemporal causal reasoning analysis first performs a temporal convolution operation on the device operating characteristics to extract energy consumption patterns at different time scales. Simultaneously, a spatial graph convolution operation is performed on the satellite observation features to establish a spatial correlation network between monitoring points. The causal attention mechanism is implemented on the parallel computing unit of the edge AI accelerator chip, using a multi-head attention layer to calculate the dynamic weight distribution between device and satellite features.

[0089] Furthermore, the causal weight matrix in the global carbon emissions prediction model constrains feature interactions, ensuring that the causal relationship between changes in device operating status and fluctuations in atmospheric composition conforms to physical laws. The edge AI accelerator chip's dedicated computing architecture distributes computational tasks such as feature extraction, spatiotemporal correlation analysis, and causal reasoning to different processing units for parallel execution. The resulting carbon emissions monitoring report includes real-time device-level emission intensity, regional carbon diffusion heat maps, and warnings for abnormal emission events. The report format complies with international carbon emissions monitoring standards.

[0090] This embodiment also provides a carbon emission intelligent prediction system based on big data, including: a carbon tensor generation module, a causal modeling module, a causal learning module, a carbon tax response module and a carbon intelligent calculation module; a carbon tensor generation module is used to collect carbon emission related data, and generate a carbon emission feature tensor through quantum space-time coding and space-time network alignment; a causal modeling module is used to construct a dynamic causal graph network based on the carbon emission feature tensor through causal entropy, and generate a causal weight matrix by combining counterfactual intervention and Bayesian false causal relationship; a causal learning module is used to construct a federated learning framework based on the causal weight matrix, and deploy a federated aggregator in the cloud to The encrypted gradients of each edge node are aggregated, and causal regularization terms are embedded in the federated loss function for joint training to generate a global carbon emission prediction model; the carbon tax response module is used to deploy the global carbon emission prediction model to the edge node, trigger the dynamic causal intervention mechanism when a new carbon tax policy is detected, adjust the attention weight of the global carbon emission prediction model through incremental learning, and generate an optimized global carbon emission prediction model; the carbon intelligent calculation module is used to integrate operation data and satellite observation data in real time on the industrial Internet of Things terminal based on the optimized global carbon emission prediction model, and perform millisecond-level carbon emission prediction through the edge AI acceleration chip.

[0091] This embodiment also provides a computer device, which is suitable for the case of a carbon emission intelligent prediction method based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the carbon emission intelligent prediction method based on big data proposed in the above embodiment.

[0092] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0093] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent carbon emission prediction method based on big data proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0094] In summary, the present invention achieves efficient fusion of multi-source heterogeneous data such as industrial sensor data and satellite remote sensing data by utilizing quantum bit superposition state mapping and quantum entangled state correlation technology. Furthermore, by decomposing the causal topology into local causal graphs and global topological structures, and embedding causal regularization terms in the federated loss function, privacy-preserving collaborative training and dynamic causal intervention of cross-regional carbon emission data are achieved.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions 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 solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A carbon emission intelligent prediction method based on big data, characterized by: include, Collect carbon emission correlation data and generate carbon emission feature tensors through quantum space-time coding and space-time grid alignment; Based on the carbon emission feature tensor, a dynamic causal graph network is constructed through causal entropy, and a causal weight matrix is generated by combining counterfactual intervention and Bayesian false causal relationship elimination. A federated learning framework is built based on the causal weight matrix. A federated aggregator is deployed in the cloud to aggregate the encrypted gradients of each edge node. A causal regularization term is embedded in the federated loss function for joint training to generate a global carbon emissions prediction model. Deploy the global carbon emission prediction model to edge nodes. When a new carbon tax policy is detected, a dynamic causal intervention mechanism is triggered. The attention weight of the global carbon emission prediction model is adjusted through incremental learning to generate an optimized global carbon emission prediction model. Based on the optimized global carbon emission prediction model, operation data and satellite observation data are integrated in real time on the industrial Internet of Things terminal, and millisecond-level carbon emission prediction is performed through the edge AI acceleration chip.

2. The method for intelligent carbon emission prediction based on big data according to claim 1, characterized in that: The carbon emission related data includes industrial sensor data, satellite remote sensing data, logistics GPS data and policy text data.

3. The method for intelligent carbon emission prediction based on big data according to claim 2, characterized in that: The carbon emission characteristic tensor is generated by quantum space-time coding and space-time grid alignment. The specific steps are as follows: The carbon emission correlation data is input into quantum space-time coding, and the parallel fusion and quantization coding of multi-source data are achieved through quantum bit superposition state mapping to generate cross-dimensional quantum coded data; The cross-dimensional quantum-encoded data is aligned in space and time, the carbon emission intensity characteristics across grid cells are correlated through quantum entangled states, and a tensor folding operation is performed to generate a carbon emission feature tensor.

4. The method for intelligent carbon emission prediction based on big data according to claim 3, characterized in that: Based on the carbon emission feature tensor, the optimal causal entropy method is used to construct a dynamic causal graph network, and the causal weight matrix is generated by combining counterfactual intervention and Bayesian false causal relationship. The specific steps are as follows: Based on the carbon emission feature tensor, a dynamic causal graph network is constructed using the optimal causal entropy algorithm, and a causal topology structure containing policy intervention nodes is generated; Based on the causal topology results, counterfactual reasoning operations are applied to the policy intervention nodes to calculate the causal effect intensity distribution of each carbon emission node under the carbon tax adjustment scenario; Based on the causal effect intensity distribution, the pseudo causal relationship edges with confidence levels below 0.05 are eliminated through the Bayesian scoring method, and then the causal weight matrix is generated through aggregation.

5. The method for intelligent prediction of carbon emissions based on big data according to claim 4, characterized in that: The federated learning framework is constructed based on the causal weight matrix. A federated aggregator is deployed in the cloud to aggregate the encrypted gradients of each edge node. The causal regularization term is embedded in the federated loss function for joint training to generate a global carbon emission prediction model. The specific steps are as follows: In the cloud, the causal weight matrix is decomposed into a local causal graph stored in the edge nodes and a global causal topology stored in the federated aggregator. The cross-node causal paths are dynamically coordinated through a secure multi-party communication protocol to generate a causal adjacency matrix. Based on the global causal topology, an encrypted channel is deployed in the federated aggregator to receive the encrypted causal gradients generated by each edge node based on the local causal graph. The encrypted causal gradient is input into the federated loss function, and the regularization constraint is constructed in combination with the causal adjacency matrix. The neural network parameters and causal weight matrix of the carbon emission prediction model are synchronously updated through the joint optimization algorithm. The optimized neural network parameters and causal weight matrix are coupled, and a global carbon emission prediction model is generated through joint representation learning of feature space projection and causal topology embedding.

6. The method for intelligent carbon emission prediction based on big data according to claim 5, characterized in that: The global carbon emission prediction model is deployed to the edge node, and the dynamic causal intervention mechanism is triggered when a new carbon tax policy is detected. The attention weight of the global carbon emission prediction model is adjusted through incremental learning, and an optimized global carbon emission prediction model is generated. The specific steps are as follows: Deploy the global carbon emission prediction model to each edge node, configure the edge node to continuously collect policy text data streams, and transmit them to the federated aggregator in real time; The federated aggregator parses the received policy data. When a new carbon tax policy is detected, it triggers the dynamic causal intervention mechanism and generates causal intervention instructions that are sent to each edge node. Each edge node adjusts the attention weight of the global carbon emission prediction model through an incremental learning algorithm based on the causal intervention instructions, and uploads the adjustment results to the federated aggregator; The federated aggregator aggregates the adjustment results of each edge node, jointly optimizes the updated attention weights and the causal weight matrix, and generates an optimized global carbon emission prediction model.

7. The method for intelligent prediction of carbon emissions based on big data according to claim 6, characterized in that: Based on the optimized global carbon emission prediction model, the operation data and satellite observation data are integrated in real time at the industrial Internet of Things terminal, and millisecond-level carbon emission prediction is performed through the edge AI acceleration chip. The specific steps are as follows: Based on the optimized global carbon emission prediction model, industrial IoT terminals are configured to synchronously collect equipment operation data and satellite observation data, and then perform spatiotemporal alignment to generate a standardized carbon emission characteristic dataset; The global carbon emission prediction model is loaded through the edge AI acceleration chip, and the standardized carbon emission feature data set is analyzed using spatiotemporal causal reasoning to generate a carbon emission detection report.

8. A carbon emission intelligent prediction system based on big data, based on the carbon emission intelligent prediction method based on big data according to any one of claims 1 to 7, characterized in that: Including carbon tensor generation module, causal modeling module, causal learning module, carbon tax response module and carbon intelligent calculation module; The carbon tensor generation module is used to collect carbon emission correlation data and generate carbon emission feature tensors through quantum space-time coding and space-time grid alignment; The causal modeling module is used to construct a dynamic causal graph network based on the carbon emission feature tensor through causal entropy, and to generate a causal weight matrix by combining counterfactual intervention and Bayesian false causality elimination; The causal learning module is used to build a federated learning framework based on the causal weight matrix. A federated aggregator is deployed in the cloud to aggregate the encrypted gradients of each edge node. The causal regularization term is embedded in the federated loss function for joint training to generate a global carbon emission prediction model. The carbon tax response module is used to deploy the global carbon emission prediction model to edge nodes. When a new carbon tax policy is detected, the dynamic causal intervention mechanism is triggered. The attention weight of the global carbon emission prediction model is adjusted through incremental learning to generate an optimized global carbon emission prediction model. The carbon intelligent calculation module is used to integrate operation data and satellite observation data in real time on the industrial Internet of Things terminal based on the optimized global carbon emission prediction model, and perform millisecond-level carbon emission prediction through the edge AI acceleration chip.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the carbon emission intelligent prediction method based on big data according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the carbon emission intelligent prediction method based on big data according to any one of claims 1 to 7 are implemented.

Images