A multi-source trajectory fusion analysis coal dynamic migration tracking system and method

Abstract

The present application relates to the field of environmental monitoring and pollution prevention and control technology, and discloses a coal dynamic migration tracking system and method based on multi-source trajectory fusion analysis, the coal dynamic migration tracking method based on multi-source trajectory fusion analysis comprising: constructing a meteorological condition representation space, converting complex meteorological parameters into low-dimensional vectors; constructing a trajectory representation model based on meteorological perception, realizing deep coupling of trajectory representation and meteorological conditions; constructing a contradiction collaborative optimization framework, converting physical rationality and data fitting degree into a mutually reinforcing collaborative relationship; realizing a double-layer optimization framework, jointly optimizing trajectory fusion parameters and theoretical parameters; performing multi-source trajectory fusion and evaluation, forming a closed-loop optimization. The present application greatly improves the meteorological adaptability of the system, realizes deep coupling of physical models and data models, significantly reduces the demand for computing resources, improves the self-adaptive learning ability of the system and the stability of multi-source data fusion, and provides a new technical path for coal dynamic migration tracking.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and pollution control technology, and more specifically, to a coal combustion dynamic migration tracking system and method based on multi-source trajectory fusion analysis. Background Technology

[0002] With increasing environmental awareness, the tracking and monitoring of coal combustion emissions has become particularly important. Existing methods for tracking the dynamic migration of coal combustion generally suffer from a series of technical problems when faced with complex and ever-changing meteorological conditions: insufficient adaptability: most existing methods use fixed model structures and parameters, which cannot adapt to complex and ever-changing meteorological conditions. In particular, under extreme weather conditions such as typhoons and cold waves, the migration characteristics of coal combustion particles will change drastically, leading to a significant decrease in tracking accuracy.

[0003] The disconnect between physical and data models: Existing methods either rely too heavily on physical models, leading to insufficient flexibility, or are purely data-driven without physical constraints. The two operate independently and cannot effectively coordinate. When weather conditions change, this disconnect causes the system to fail to maintain stable tracking performance.

[0004] Multi-source trajectory fusion is unstable: different monitoring devices exhibit varying performance under different meteorological conditions. Existing fusion methods cannot dynamically adjust the weights and fusion strategies of each data source according to meteorological conditions, resulting in unstable fusion results under complex meteorological conditions.

[0005] High computational resource consumption: To cope with variable weather conditions, existing methods often employ multi-model ensembles or complex model structures, resulting in high computational complexity, a large number of parameters, and difficulty in deployment in resource-constrained environments. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a coal dynamic migration tracking system and method based on multi-source trajectory fusion analysis.

[0007] A method for tracking the dynamic migration of coal combustion from multiple sources using trajectory fusion analysis includes the following steps: A meteorological condition representation space is constructed, and meteorological parameters are transformed into low-dimensional meteorological condition vector representations through a dimensionality reduction mapping algorithm; A trajectory representation model for meteorological perception is constructed, including feature basis functions based on deep neural networks and physical constraint basis functions based on physical laws, to achieve deep coupling between trajectory representation and meteorological conditions; A framework for synergistic optimization of contradictions is constructed. Based on the mutual information calculation model, the correlation between data fit and physical rationality is quantified, and the traditional contradiction between physical rationality and data fit is transformed into a synergistic relationship that mutually reinforces each other. A two-layer optimization framework is implemented, which achieves joint parameter optimization by optimizing trajectory fusion parameters in the inner layer and theoretical parameters in the outer layer. Perform multi-source trajectory fusion, fuse multi-source trajectory data according to current meteorological conditions, evaluate the fusion results and feed the evaluation results back to the optimization framework to form a closed-loop optimization.

[0008] Preferred steps for constructing a meteorological condition representation space include: Collect multi-dimensional meteorological parameter data, including wind direction, wind speed, temperature, humidity, air pressure, and precipitation; Meteorological data are normalized, missing values ​​are handled, and time is aligned. A variational autoencoder model is used to map high-dimensional meteorological parameters to a low-dimensional manifold space; The current weather conditions are converted into a low-dimensional vector representation by using a trained variational autoencoder model.

[0009] Preferred steps for constructing a trajectory representation model for meteorological sensing include: Data-driven feature basis functions are constructed using deep neural networks, implemented through multi-layer convolution and attention mechanisms, to extract effective features from raw trajectory data; Based on the physical laws governing the migration of coal combustion particulate matter, physical constraint basis functions are constructed, including fluid dynamics constraint functions, thermodynamic constraint functions, mass conservation constraint functions, and energy conservation constraint functions. A basis function weight generator is constructed to calculate the weights of characteristic basis functions and physical constraint basis functions under meteorological conditions, thereby enabling dynamic adjustment of the importance of different basis functions; The final trajectory representation is obtained by adding the product of the characteristic basis function and its weight and the product of the physical constraint basis function and its weight.

[0010] Preferred: The basis function weight generator includes: The meteorological condition coding layer processes meteorological condition vectors and extracts their key features. The attention calculation layer calculates the attention weights for each basis function based on meteorological conditions. The weight mapping layer converts attention weights into feature basis function weights and physical constraint basis function weights. The temperature control module introduces temperature parameter adjustment to improve the smoothness of the weight distribution, and adaptively adjusts according to meteorological conditions.

[0011] Preferred steps for constructing a collaborative optimization framework for contradictions include: Based on the degree of matching between multi-source trajectory data and fused trajectory, a data fit measurement function is defined; Based on the physical laws governing the migration of particulate matter from coal combustion, a physical rationality measurement function is defined. Construct a mutual information calculation model to quantify the correlation between data fit and physical plausibility; A contradictory synergistic function is constructed, which combines data fit, physical rationality, and mutual information to achieve synergistic enhancement between the two.

[0012] Preferred: The mutual information calculation model employs a neural network-based mutual information estimator, including: The joint distributed encoder takes data fit and physical plausibility as input and encodes them into a joint representation; An edge distribution encoder encodes data fit and physical plausibility separately to generate edge representations; The mutual information estimation layer estimates the mutual information between data fit and physical plausibility based on joint and marginal representations.

[0013] Preferred steps for implementing the two-layer optimization framework include: Inner layer optimization: With theoretical parameters fixed, the trajectory fusion parameters are optimized using the gradient ascent method to maximize the contradictory cooperative function; The outer layer optimization, based on the inner layer optimization results, adjusts the theoretical parameters using Bayesian optimization methods to balance exploration and utilization, and efficiently search for the optimal parameter combination. Based on the optimization results, the parameters of the trajectory representation model and the contradictory synergistic function are updated to achieve self-improvement of the system; A feedback loop module is built to feed the performance evaluation results during system operation back to the optimization framework, continuously optimizing system parameters.

[0014] Preferred steps for performing multi-source trajectory fusion include: Raw coal combustion particulate matter migration trajectory data were acquired from different monitoring devices and preprocessed. Acquire current meteorological condition data and map the meteorological condition representation space into a vector; We use a trajectory representation model based on meteorological perception and a contradictory collaborative optimization framework to fuse multi-source trajectory data; The fusion results are evaluated based on predefined evaluation metrics, and the evaluation results are fed back to the optimization framework.

[0015] Preferably, the multi-source trajectory originates from at least two of the following monitoring devices: Fixed monitoring stations are conventional monitoring stations located at key locations around coal-fired power plants, equipped with optical monitoring instruments and meteorological sensors; Mobile monitoring equipment refers to monitoring equipment installed on mobile carriers such as vehicles or drones. Remote sensing monitoring systems include satellite remote sensing and ground radar systems.

[0016] A coal combustion dynamic migration tracking system based on multi-source trajectory fusion analysis includes: The meteorological condition representation module is used to construct the meteorological condition representation space and transform complex meteorological parameters into low-dimensional vector representations. The trajectory representation module is used to construct a trajectory representation model for meteorological sensing, achieving deep coupling between trajectory representation and meteorological conditions; The contradiction-coordination optimization module is used to construct a contradiction-coordination optimization framework, which transforms physical rationality and data fit into a mutually reinforcing synergistic relationship. A two-layer optimization module is used to jointly optimize trajectory fusion parameters and theoretical parameters; The trajectory fusion module is used to perform multi-source trajectory fusion and evaluate the fusion results; The feedback loop module is used to feed the evaluation results back to the optimization framework, forming a closed-loop optimization.

[0017] The beneficial effects of this invention are as follows: Significantly improved weather adaptability: Compared with existing technologies, this implementation method can automatically adapt to different weather conditions and maintain high-precision trajectory fusion effect even under extreme weather conditions (such as typhoons, cold waves, etc.), improving tracking accuracy by more than 35%.

[0018] Synergistic enhancement of physical rationality and data fitting: Through the contradictory synergistic optimization framework, the physical model and the data model are deeply coupled, so that the fusion result can not only conform to physical laws but also effectively fit the observation data, and the proportion of physically unreasonable trajectories is reduced by 65%.

[0019] The computational resource requirements are significantly reduced: the unified theoretical framework integrates the originally separate physical and data models into one, reducing the number of parameters by 65% ​​and computational complexity by 47%, enabling the system to be deployed and run on edge computing devices.

[0020] System adaptive learning capability: Through a two-layer optimization framework and feedback loop mechanism, the system can continuously improve itself and exhibit intelligent adaptive capability similar to "weather perception" to new weather conditions, and can cope with previously unseen weather conditions without human intervention.

[0021] Improved stability of multi-source data fusion: The system can automatically adjust the weights and fusion strategies of each data source according to different meteorological conditions, so that the fusion results remain stable under complex meteorological conditions and the trajectory continuity is improved by 78%.

[0022] This invention fundamentally solves the instability problem of multi-source trajectory fusion under complex meteorological conditions by establishing a unified theoretical framework for meteorological-trajectory coupling, providing a brand-new technical path for tracking the dynamic migration of coal combustion, and promoting the development of environmental monitoring and pollution control technologies. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for tracking the dynamic migration of coal combustion through multi-source trajectory fusion analysis according to the present invention; Figure 2 This is a line graph comparing the trajectory tracking accuracy under different meteorological conditions according to the present invention. Figure 3 This is a bar chart comparing the computing resource requirements of this invention; Figure 4 This is a radar chart for multi-dimensional evaluation of the trajectory fusion performance of the present invention; Figure 5 This is a bar chart showing the proportion of physically unreasonable trajectories under different meteorological conditions according to the present invention; Figure 6 This is a line graph showing the relationship between the complexity of meteorological conditions and the adaptive weights in this invention. Detailed Implementation

[0024] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0025] Example 1 refer to Figure 1 This embodiment proposes a method for tracking the dynamic migration of coal combustion through multi-source trajectory fusion analysis, including the following steps: Step 1: Construct a meteorological condition representation space This step utilizes a dimensionality reduction mapping algorithm to process the raw meteorological data and generate a low-dimensional representation of meteorological conditions. The specific implementation is as follows: 1.1 Collect meteorological data: Collect multi-dimensional meteorological parameter data including wind direction, wind speed, temperature, humidity, air pressure, precipitation, etc.

[0026] 1.2 Data Preprocessing: Normalize meteorological data, handle missing values, and align time to ensure data quality.

[0027] 1.3 Constructing a meteorological condition mapping model: According to the technical solution of this application, a variational autoencoder model can be used to map high-dimensional meteorological parameters to a low-dimensional manifold space. This model consists of an encoder and a decoder. The encoder maps the original meteorological parameters to a low-dimensional latent space, and the decoder ensures that the low-dimensional representation can reconstruct the original meteorological information and retain key characteristics.

[0028] Specifically, the variational autoencoder model structure includes: The encoder part consists of a multi-layer fully connected network. The number of nodes in the input layer is the same as the dimension of the meteorological parameters. The intermediate layer uses the ReLU activation function. Finally, it outputs two vectors representing the mean μ and standard deviation σ of the latent space, respectively. The latent vector z is generated through reparameterization techniques. The decoder part also consists of a multi-layer fully connected network that maps the latent vector z back to the original meteorological parameter space to reconstruct the input data. Loss function: Combining reconstruction error and KL divergence, it ensures reconstruction quality while constraining the latent space distribution to be close to the standard normal distribution, thereby enhancing the model's generalization ability.

[0029] Optionally, batch normalization layers can be introduced into the encoder and decoder to improve training stability and convergence speed. Furthermore, for meteorological parameters with significant time correlation, recurrent neural network layers can be introduced into the model to better capture temporal characteristics.

[0030] 1.4 Meteorological condition vector generation: The current meteorological conditions are converted into a low-dimensional vector representation W by a trained variational autoencoder model. This vector captures the main feature patterns of the meteorological conditions for use in subsequent steps.

[0031] Figure 2 The invention demonstrates a comparison of trajectory tracking accuracy between the method of the present invention and existing methods under different meteorological conditions, intuitively reflecting the advantages of the method of the present invention under complex meteorological conditions, especially in maintaining a high tracking accuracy under extreme meteorological conditions (such as typhoons).

[0032] Step 2: Construct a trajectory representation model for meteorological sensing According to one embodiment of this application, this step constructs a unified trajectory representation model based on the meteorological-trajectory coupling representation theory, achieving deep coupling between trajectory representation and meteorological conditions. The specific implementation is as follows: 2.1 Feature Basis Function Construction: A set of data-driven feature basis functions φi(x,t) are constructed using a deep neural network. These basis functions can extract effective features from the original trajectory data. The basis functions are implemented through multi-layer convolution and attention mechanisms, which can capture the spatiotemporal dependency characteristics of the trajectory.

[0033] Specifically, feature basis function networks include: Spatiotemporal feature extraction layer: Composed of 3D convolutional layers, it processes trajectory data in the spatiotemporal dimension and extracts local spatiotemporal features; Self-attention layer: calculates the dynamic correlation between trajectory points and captures long-range dependencies; Feature integration layer: Integrates feature information at different scales through residual connections and layer normalization.

[0034] 2.2 Construction of physical constraint basis functions: Based on the physical laws of coal particulate matter migration, a set of physical constraint basis functions ψj(x,t) are constructed. These functions encode physical knowledge such as fluid dynamics model and thermodynamic model to ensure that trajectory fusion conforms to physical laws.

[0035] The physical constraint basis functions include: Fluid dynamics constraint function: Based on the Navier-Stokes equations, it describes the motion characteristics of coal-fired particulate matter in airflow; Thermodynamic constraint function: describes the effect of temperature field on the diffusion of particulate matter from coal combustion; Mass conservation constraint function: ensures that the total amount of particulate matter remains constant during migration; Energy conservation constraint function: ensures that the energy conversion of the system conforms to the laws of physics.

[0036] 2.3 Basis Function Weight Generator Construction: This application provides two mapping networks to calculate the weights of the characteristic basis functions and the weights of the physical constraint basis functions under meteorological conditions W, respectively, so as to realize the dynamic adjustment of the importance of different basis functions.

[0037] The structure of an adaptive basis function generation network includes: Meteorological condition coding layer: processes the meteorological condition vector W and extracts its key features; Attention calculation layer: Calculates the attention weights for each basis function based on meteorological conditions; Weight mapping layer: converts attention weights into feature basis function weights and physical constraint basis function weights; Temperature regulation module: Introduces temperature parameter τ to adjust the smoothness of weight distribution, and adaptively adjusts according to meteorological conditions.

[0038] In some implementations, attention computation can employ a multi-head attention mechanism to capture different aspects of the weather-basis function relationship.

[0039] 2.4 Trajectory Representation Model Integration: The above components are integrated into a unified trajectory representation model, generating the trajectory representation based on the trajectory representation calculation formula under meteorological condition W. This trajectory representation calculation process includes two parts: First, obtaining the product of the feature basis functions and their corresponding weights. Specifically, each feature basis function is multiplied by its corresponding weight coefficient, and then these product terms are summed. Second, obtaining the product of the physical constraint basis functions and their corresponding weights. Similarly, each physical constraint basis function is multiplied by its corresponding weight coefficient, and then these product terms are summed. Finally, the sums of the above two parts are added together to obtain the final trajectory representation Tr(x,t|W). This model achieves an organic combination of data-driven and physical constraint approaches through the weighted combination of feature basis functions and physical constraint basis functions.

[0040] In the above trajectory representation calculation process, the specific implementations of the feature basis function weight calculation method and the physical constraint basis function weight calculation method are as follows: The feature basis function weight calculation method is implemented through a multi-layer neural network. First, the meteorological condition vector W is nonlinearly transformed, and then the weights of each feature basis function are obtained by normalization through the softmax function. The physical constraint basis function weight calculation method adopts a similar structure, but adds prior knowledge of the correlation of physical constraints. The weights of the physical constraint basis functions are dynamically allocated according to the current meteorological conditions through an attention mechanism.

[0041] Figure 3 By comparing the differences between the method of this invention and existing methods in three key computing resource indicators—computation time, number of parameters, and memory usage—it is verified that the method of this invention significantly reduces the computing resource requirements, enabling the system to be deployed and run on edge computing devices.

[0042] Step 3: Construct a framework for collaborative optimization of contradictions It should be understood that this step constructs a contradictory synergistic optimization framework, transforming the traditional contradiction between physical rationality and data fit into a mutually reinforcing synergistic relationship. The specific implementation is as follows: 3.1 Data Fit Measurement: Based on the degree of matching between multi-source trajectory data and fused trajectory, a data fit measurement function is defined, which comprehensively considers factors such as trajectory point position deviation and velocity direction consistency.

[0043] 3.2 Quantification of physical rationality: Based on the physical laws of coal particulate matter migration, a physical rationality measurement function is defined. This function checks whether the fusion trajectory conforms to physical constraints such as fluid dynamics laws and energy conservation.

[0044] 3.3 Construction of Mutual Information Computation Model: A mutual information computation model is constructed to quantify the correlation between data fit and physical rationality, and to evaluate the degree of synergy between the two.

[0045] The mutual information calculation model employs a neural network-based mutual information estimator, the specific structure of which includes: Joint Distributive Encoder: Takes data fit and physical plausibility as inputs and encodes them into a joint representation; Edge distribution encoder: Encodes data fit and physical plausibility separately to generate edge representations; Mutual information estimation layer: Based on joint representation and marginal representation, estimate the mutual information between data fit and physical plausibility.

[0046] 3.4 Construction of the Contradictory Coordination Function: The contradictory coordination function is constructed, and its calculation process includes three parts: First, the product of the data fit and its weights is used as the first part, specifically by multiplying the adaptive weights of meteorological conditions by the data fit metric; second, the product of physical rationality and its weights is used as the second part, specifically by multiplying the value of 1 minus the adaptive weights of meteorological conditions by the physical rationality metric; finally, the mutual information of the data fit and physical rationality is multiplied by the mutual information adjustment coefficient as the third part. The three parts are then added together to obtain the final contradictory coordination function value. This function achieves better trajectory fusion results by balancing data fit and physical rationality and promoting their synergistic enhancement.

[0047] In the aforementioned calculation of the contradictory collaborative function, the specific implementation of the meteorological condition adaptive weight calculation method is as follows: First, the meteorological condition vector W is input into a two-layer neural network. The intermediate layer uses the ReLU activation function, and the output layer uses the sigmoid activation function, restricting the output value to between 0 and 1. This adaptively adjusts the weight balance between the data fit term and the physical rationality term according to different meteorological conditions. The mutual information calculation method is implemented through a neural network mutual information estimator. This estimator estimates the mutual information between the data fit and physical rationality through contrastive learning methods of joint encoding and marginal encoding.

[0048] Figure 4 The method of this invention is comprehensively evaluated from five dimensions: trajectory continuity, physical rationality, data fit, meteorological adaptability, and real-time performance, demonstrating the comprehensive advantages of the method of this invention in each dimension.

[0049] Step 4: Implement the two-layer optimization framework This application further provides a two-layer optimization framework to achieve joint optimization of trajectory fusion parameters and theoretical parameters. The specific implementation is as follows: 4.1 Inner Layer Optimization: With theoretical parameters fixed, optimize trajectory fusion parameters to maximize the contradictory cooperative function H. This optimization process iteratively updates the feature basis function weights and physical constraint basis function weights using the gradient ascent method until convergence.

[0050] 4.2 Outer Layer Optimization: Based on the inner layer optimization results, theoretical parameters (such as the mutual information adjustment coefficient γ, the number of basis functions n and m, etc.) are adjusted to further improve the overall system performance. This optimization adopts the Bayesian optimization method, balancing exploration and utilization, and efficiently searching for the optimal parameter combination.

[0051] 4.3 Model parameter self-calibration: Based on the optimization results, the parameters of the trajectory representation model and the contradictory cooperative function are updated to achieve the self-improvement of the system.

[0052] 4.4 Feedback Loop Construction: Construct a feedback loop module to feed back the performance evaluation results during system operation to the optimization framework, continuously optimize system parameters, and improve the system's adaptability to different meteorological conditions.

[0053] Therefore, through the above-mentioned feedback loop mechanism, the implementation method of this application can achieve continuous self-improvement of the system and adapt to new meteorological conditions.

[0054] Figure 5 The invention demonstrates the proportion of physically unreasonable trajectories generated by the method of the present invention and existing methods under different meteorological conditions, and verifies that the method of the present invention, through a contradictory collaborative optimization framework, effectively reduces the proportion of physically unreasonable trajectories and improves the physical rationality of the trajectory fusion results.

[0055] Step 5: Multi-source trajectory fusion and evaluation In addition, this step utilizes the constructed meteorological adaptive framework to achieve the fusion and evaluation of multi-source trajectory data. The specific implementation is as follows: 5.1 Multi-source data acquisition: Obtain raw coal particulate matter migration trajectory data from different monitoring devices and perform preprocessing.

[0056] 5.2 Meteorological Condition Acquisition: Acquire current meteorological condition data and map it to vector W using the meteorological condition representation space from step 1.

[0057] 5.3 Trajectory Fusion Execution: Utilizing the trajectory representation model based on meteorological perception and the contradictory collaborative optimization framework, multi-source trajectory data are fused according to the current meteorological conditions W to generate a unified migration trajectory of coal combustion particulate matter.

[0058] 5.4 Evaluation of Fusion Results: Based on predefined evaluation indicators (such as trajectory continuity, physical rationality, consistency with reference data, etc.), the fusion results are evaluated and the evaluation results are fed back to the optimization framework to form a closed-loop optimization.

[0059] It is evident that, through this closed-loop feedback mechanism, the implementation method of this application can continuously improve system performance.

[0060] Figure 6 The invention demonstrates the changing trend of the adaptive weights of meteorological conditions in its method as the complexity of meteorological conditions increases, verifying that the system can automatically adjust the weight balance between data fit and physical rationality according to the complexity of meteorological conditions, thereby achieving meteorological adaptive capability.

[0061] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for tracking the dynamic migration of coal combustion using multi-source trajectory fusion analysis, characterized in that, Includes the following steps: A meteorological condition representation space is constructed by collecting multi-dimensional meteorological parameter data, including wind direction, wind speed, temperature, humidity, air pressure, and precipitation. The meteorological data is normalized, missing values ​​are handled, and time alignment is performed. A variational autoencoder model is used to map the high-dimensional meteorological parameters to a low-dimensional manifold space. The variational autoencoder model includes an encoder part and a decoder part. The encoder part consists of a multi-layer fully connected network. The number of nodes in the input layer is the same as the dimension of the meteorological parameters. The intermediate layers use the ReLU activation function. Finally, two vectors are output to represent the mean μ and standard deviation σ of the latent space, respectively. A latent vector z is generated through reparameterization techniques. The decoder part also consists of a multi-layer fully connected network. It maps the latent vector z back to the original meteorological parameter space and reconstructs the input data. The loss function combines the reconstruction error and KL divergence. Through the trained variational autoencoder model, the current meteorological conditions are converted into a low-dimensional vector representation W. A trajectory representation model based on meteorological perception is constructed. A data-driven feature basis function is built using a deep neural network, implemented through multi-layer convolution and attention mechanisms, to extract effective features from the original trajectory data. The feature basis function network includes a spatiotemporal feature extraction layer, a self-attention layer, and a feature integration layer. Based on the physical laws of coal combustion particulate matter migration, physical constraint basis functions are constructed, including fluid dynamics constraint functions, thermodynamic constraint functions, mass conservation constraint functions, and energy conservation constraint functions. A basis function weight generator is constructed to calculate the weights of the feature basis functions and the physical constraint basis functions under meteorological conditions, enabling dynamic adjustment of the importance of different basis functions. The trajectory representation calculation process includes two parts: First, each feature basis function is multiplied by its corresponding weight coefficient, and then these products are summed. Second, each physical constraint basis function is multiplied by its corresponding weight coefficient, and then these products are summed. Finally, the sums of the two parts are added together to obtain the final trajectory representation Tr(x,t|W). A framework for contradiction-coordination optimization is constructed. Based on the matching degree between multi-source trajectory data and fused trajectory, a data fit measurement function is defined. Based on the physical laws of coal particulate matter migration, a physical rationality measurement function is defined. A mutual information calculation model is constructed to quantify the correlation between data fit and physical rationality, and a contradiction-coordination function is constructed. A two-layer optimization framework is implemented. The inner layer optimizes the fixed theoretical parameters and optimizes the trajectory fusion parameters through the gradient ascent method to maximize the contradictory cooperative function. The outer layer optimizes the theoretical parameters based on the inner layer optimization results and adjusts them through the Bayesian optimization method. According to the optimization results, the parameters of the trajectory representation model and the contradictory cooperative function are updated. A feedback loop module is constructed to feed back the performance evaluation results during the system operation to the optimization framework and continuously optimize the system parameters. Perform multi-source trajectory fusion, fuse multi-source trajectory data according to current meteorological conditions, evaluate the fusion results and feed the evaluation results back to the optimization framework to form a closed-loop optimization.

2. The method for tracking the dynamic migration of coal combustion based on multi-source trajectory fusion analysis according to claim 1, characterized in that, The basis function weight generator includes: The meteorological condition coding layer processes meteorological condition vectors and extracts their key features. The attention calculation layer calculates the attention weights for each basis function based on meteorological conditions. The weight mapping layer converts attention weights into feature basis function weights and physical constraint basis function weights. The temperature control module introduces temperature parameter adjustment to improve the smoothness of the weight distribution, and adaptively adjusts according to meteorological conditions.

3. The method for tracking the dynamic migration of coal combustion based on multi-source trajectory fusion analysis according to claim 1, characterized in that, The mutual information calculation model employs a neural network-based mutual information estimator, including: The joint distributed encoder takes data fit and physical plausibility as input and encodes them into a joint representation; An edge distribution encoder encodes data fit and physical plausibility separately to generate edge representations; The mutual information estimation layer estimates the mutual information between data fit and physical plausibility based on joint and marginal representations.

4. The method for tracking the dynamic migration of coal combustion based on multi-source trajectory fusion analysis according to claim 1, characterized in that, The steps for performing multi-source trajectory fusion include: Raw coal combustion particulate matter migration trajectory data were acquired from different monitoring devices and preprocessed. Acquire current meteorological condition data and map the meteorological condition representation space into a vector; We use a trajectory representation model based on meteorological perception and a contradictory collaborative optimization framework to fuse multi-source trajectory data; The fusion results are evaluated based on predefined evaluation metrics, and the evaluation results are fed back to the optimization framework.

5. The method for tracking the dynamic migration of coal combustion based on multi-source trajectory fusion analysis according to claim 1, characterized in that, The multi-source trajectory is derived from at least two of the following monitoring devices: Fixed monitoring stations are conventional monitoring stations located at key locations around coal-fired power plants, equipped with optical monitoring instruments and meteorological sensors; Mobile monitoring equipment, monitoring equipment installed on mobile carriers such as vehicles or drones; Remote sensing monitoring systems include satellite remote sensing and ground radar systems.

6. A coal combustion dynamic migration tracking system based on multi-source trajectory fusion analysis, used to execute the coal combustion dynamic migration tracking method based on multi-source trajectory fusion analysis as described in any one of claims 1-5, characterized in that, include: The meteorological condition representation module is used to construct the meteorological condition representation space and transform complex meteorological parameters into low-dimensional vector representations. The trajectory representation module is used to construct a trajectory representation model for meteorological sensing, achieving deep coupling between trajectory representation and meteorological conditions; The contradiction-coordination optimization module is used to construct a contradiction-coordination optimization framework, which transforms physical rationality and data fit into a mutually reinforcing synergistic relationship. A two-layer optimization module is used to jointly optimize trajectory fusion parameters and theoretical parameters; The trajectory fusion module is used to perform multi-source trajectory fusion and evaluate the fusion results; The feedback loop module is used to feed the evaluation results back to the optimization framework, forming a closed-loop optimization.

Images