Intelligent control method and system for dust removal energy consumption

Abstract

The invention discloses a dedusting energy consumption intelligent control method and system, and relates to the technical field of energy consumption control, and the method comprises the steps: carrying out the filtering processing of obtained dust monitoring data, obtaining the denoised dust monitoring data, carrying out the dust scene reconstruction of the denoised dust monitoring data, and generating an initial dust field; combining the initial dust field with a preset airflow field to perform dust environment simulation to obtain a dust environment simulation result; carrying out dust removal associated feature extraction on the historical dust removal operation data; generating an energy consumption causal graph based on the dust removal associated feature set, and constructing a dust removal energy consumption matching table; and executing dedusting energy consumption mapping to obtain dedusting equipment operation data, and implementing dedusting energy consumption intelligent control. According to the method, global three-dimensional dust concentration distribution can be restored, the actual problems of sparse sensors and the like can be effectively solved, causal association between controllable operation variables and total energy consumption can be deepened, and energy waste caused by invalid operation of a fan and excessive dust removal can be remarkably reduced.

Description

Technical Field

[0001] This invention relates to the field of energy consumption control technology, and more specifically, to a method and system for intelligent control of dust removal energy consumption. Background Technology

[0002] Dust removal is a crucial step in industrial production to ensure a safe working environment, meet environmental emission standards, and maintain the normal operation of equipment. It is widely used in high-dust-generating scenarios such as welding, casting, wood processing, and cement manufacturing. Traditional dust removal systems often employ fixed airflow or simple start-stop control strategies, failing to dynamically adjust operating parameters based on actual dust generation intensity. This results in continuous high-power operation of fans under low-load conditions, leading to significant energy waste; while under sudden high-load conditions, the response lag may cause insufficient dust removal efficiency, impacting air quality and worker health. Therefore, there is an urgent need for an intelligent dust removal energy consumption control method that can significantly reduce system operating energy consumption and improve dust removal responsiveness and stability.

[0003] However, existing intelligent dust collection energy consumption control systems generally suffer from several shortcomings, including insufficient sensing capabilities, modeling methods that separate physics from data, a lack of causal basis in control strategies, and difficulty in coordinating energy efficiency and safety. Firstly, relying solely on sparse sensor data makes it impossible to accurately reconstruct the three-dimensional dust distribution across the entire area, resulting in a one-sided environmental situational awareness. Secondly, control is often based on empirical thresholds, failing to establish a causal relationship between operational variables such as fan frequency, airflow, and cleaning cycle and energy consumption. This easily leads to over-operation and energy waste, i.e., maintaining high airflow conservatively even under low dust generation conditions, failing to supply air as needed, resulting in ineffective energy consumption and equipment damage. These problems severely restrict the improvement of dust collection control in terms of accuracy, efficiency, and intelligence.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of the problems in related technologies, this invention proposes an intelligent control method and system for dust removal energy consumption to overcome the aforementioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for intelligent control of dust removal energy consumption is provided, the method comprising the following steps: S1. Filter the acquired dust monitoring data to obtain denoised dust monitoring data, and reconstruct the dust scene based on the denoised dust monitoring data to generate an initial dust field. S2. Combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results; S3. Obtain historical dust removal operation data and extract dust removal-related features from the historical dust removal operation data to obtain a dust removal-related feature set; S4. Generate an energy consumption causal graph based on the dust removal associated feature set, and construct a dust removal energy consumption matching table based on the energy consumption causal graph; S5. Based on the dust environment simulation results and the dust removal energy consumption matching table, perform dust removal energy consumption mapping to obtain dust removal equipment operation data, and implement intelligent control of dust removal energy consumption based on the dust removal equipment operation data.

[0007] Furthermore, the acquired dust monitoring data is filtered to obtain denoised dust monitoring data, and dust scene reconstruction is performed on the denoised dust monitoring data to generate an initial dust field, including: S11. Perform outlier processing on the acquired dust monitoring data to obtain cleaned dust monitoring data. Use a low-pass filter to perform noise reduction processing on the cleaned dust monitoring data to obtain denoised dust monitoring data. S12. Generative adversarial networks are used to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data to obtain enhanced dust monitoring data. S13. Combine the enhanced dust monitoring data, the three-dimensional structure grid of dust space, and the prior airflow data to perform three-dimensional dust concentration distribution reconstruction and generate the initial dust field.

[0008] Furthermore, generative adversarial networks are used to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data, resulting in enhanced dust monitoring data including: S121. Using a spatiotemporal convolutional encoder based on a generative adversarial network, the spatiotemporal feature tensor of dust is extracted from the denoised dust monitoring data, and a binary mask matrix identifying the missing dust region is generated simultaneously. S122. Input the dust spatiotemporal feature tensor into the generator, and perform decoding and spatial upsampling based on the spatiotemporal transposed convolutional network to obtain the preliminary dust concentration field. Use a discriminator with introduced physical constraint loss to iteratively optimize the preliminary dust concentration field to obtain a physically consistent preliminary dust complete field. S123. Using an attention mechanism, the initial dust full field and binary mask matrix are fused with local details and global context to obtain a semantically consistent dust transition field. Then, a semantic smoothing constraint on dust diffusion is applied to obtain enhanced dust monitoring data.

[0009] Furthermore, the enhanced dust monitoring data, the three-dimensional dust spatial structure mesh, and the prior airflow data are combined to perform a three-dimensional dust concentration distribution reconstruction, generating an initial dust field including: S131. Input the enhanced dust monitoring data and the three-dimensional structure mesh of dust space into the implicit function network, use the implicit function network to fit the observation points, and construct a continuous spatial mapping under geometric boundary constraints to obtain the preliminary three-dimensional dust implicit concentration field. S132. Using a differentiable physics solver, the initial three-dimensional implicit dust concentration field is coupled with the pre-constructed steady-state airflow velocity field to obtain the initial dust concentration evolution field. S133. Compare the initial dust concentration evolution field with the enhanced dust monitoring data point cloud, construct a variational optimization model based on the comparison results, and solve the variational optimization model to obtain the initial dust field.

[0010] Furthermore, the initial dust concentration evolution field is compared with the enhanced dust monitoring data point cloud. Based on the comparison results, a variational optimization model is constructed and solved to obtain the initial dust field, which includes: S1331. The initial dust concentration evolution field is compared with the enhanced dust monitoring data point cloud at spatial grid points one by one, and the dust concentration residual value is calculated based on the comparison results. The weighted value assignment operation is performed based on the dust concentration residual value to obtain the weighted dust residual distribution field. S1332. Based on the dust residual distribution field, perform multi-objective functional processing to obtain a variational energy expression containing data fidelity terms, physical regularization terms, and spatial smoothing terms. Then, perform regularization parameter adaptive configuration and discretization modeling on the variational energy expression to obtain a variational optimization model. S1333. Input the variational optimization model and the initial dust concentration evolution field into the sparse matrix solver, and perform optimization by minimizing the objective function to obtain the initial dust field.

[0011] Furthermore, variational optimization models include: ; In the formula, Represents the variational optimization function; This indicates the total number of observed dust data points; Indicates the index of the observed dust data points; Indicates the first Weighting coefficients for each observed dust data point; Indicates the first An observation operator for each dust data point; Indicates the first The actual observed values ​​of each dust data point; Represents the physical regularization parameter; Represents the spatial computational domain; This indicates that the physical operator acts on the field to be optimized. The result; This represents the action of the physical operator on the background field of the physical simulation. The result; This represents the smoothing regularization parameter; Indicates the field to be optimized Spatial gradient; This represents the initial dust concentration evolution field to be optimized.

[0012] Furthermore, an energy consumption causal graph is generated based on the dust removal associated feature set, and a dust removal energy consumption matching table is constructed based on the energy consumption causal graph, including: S41. Perform stationarity test and nonlinear transformation preprocessing on the dust removal correlation feature set to obtain a standardized dust removal correlation data matrix; S42. Using the causal discovery algorithm, construct causal dependencies and remove false edges from the dust removal associated data matrix to obtain the energy consumption causal graph. Then, perform causal path identification and mapping on the energy consumption causal graph to obtain the dust removal energy consumption matching table.

[0013] Furthermore, a causal discovery algorithm is used to construct causal dependencies and remove false edges from the dust removal correlation data matrix to obtain an energy consumption causal graph. Causal path identification and mapping are then performed on the energy consumption causal graph to obtain a dust removal energy consumption matching table, including: S421. Using the causal discovery algorithm, conditional independence tests and directed acyclic graph searches are performed on the dust removal associated data matrix, and the causal dependencies between variables are initially constructed to obtain the candidate causal graph structure. S422. Based on knowledge of dust removal, the candidate causal graph structure is verified for causal direction and the rationality of intervention is verified. False edges are eliminated using counterfactual consistency test to obtain the energy consumption causal graph. S423. Use graph traversal algorithm to identify energy consumption causal paths in the energy consumption causal graph, and perform rule mapping and optimization adjustment based on the energy consumption causal path identification results to obtain the dust removal energy consumption matching table.

[0014] Furthermore, a graph traversal algorithm is used to identify energy consumption causal paths in the energy consumption causal graph, and rule mapping and optimization are performed based on the energy consumption causal path identification results to obtain a dust removal energy consumption matching table, including: S4231. Use graph traversal algorithm to identify the energy consumption causal path from working condition characteristics to controllable actions in the energy consumption causal graph, and generate a structured causal rule set based on the energy consumption causal path identification results. S4232. Discretize and partition the domain of all precondition variables in the structured causal rule set to obtain a multidimensional working condition grid, and associate the rule output corresponding to each grid cell to obtain an indexed preliminary mapping matrix. S4233. Based on preset operating costs and safety constraints, a reinforcement learning strategy optimizer is used to adjust and verify the parameters of the state grid of the initial mapping matrix to obtain a dust removal energy consumption matching table.

[0015] According to another aspect of the present invention, a dust removal energy consumption intelligent control system is provided, the system comprising: The dust scene reconstruction module is used to filter the acquired dust monitoring data to obtain denoised dust monitoring data, and to reconstruct the dust scene from the denoised dust monitoring data to generate an initial dust field. The dust environment simulation module is used to combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results. The dust removal associated feature extraction module is used to acquire historical dust removal operation data and extract dust removal associated features from the historical dust removal operation data to obtain a dust removal associated feature set; The dust removal energy consumption matching module is used to generate an energy consumption cause-effect graph based on the dust removal associated feature set, and to construct a dust removal energy consumption matching table based on the energy consumption cause-effect graph. The dust removal energy consumption control module is used to perform dust removal energy consumption mapping based on the dust environment simulation results and the dust removal energy consumption matching table, obtain the dust removal equipment operation data, and implement intelligent dust removal energy consumption control based on the dust removal equipment operation data.

[0016] The beneficial effects of this invention are as follows: 1. This invention, through a collaborative mechanism of dust scene reconstruction and dust removal energy consumption matching, can not only accurately and robustly restore the three-dimensional dust concentration distribution across the entire area, effectively overcoming practical problems such as sensor sparsity, data loss, and noise interference, but also deeply reveal the causal relationship between controllable operating variables such as fan frequency, air volume adjustment, and dust removal cycle and total energy consumption. Based on this, the generated dynamic energy consumption matching table can support real-time recommendation of optimal control strategies according to operating conditions, significantly reducing energy waste caused by ineffective fan operation and excessive dust removal while ensuring dust removal efficiency and working environment safety.

[0017] 2. This invention reconstructs dust scenes, enabling the transformation of originally sparse, discrete, and noise-affected sensor data into a continuous, complete, and physically consistent three-dimensional dust concentration field. This allows for the precise depiction of the spatial distribution characteristics and dynamic evolution trends of dust within the workshop. Based on this, the system can accurately identify high-concentration areas, diffusion paths, and key influencing factors, providing a high-confidence environmental perception foundation for subsequent energy consumption modeling, causal analysis, and intelligent control. This effectively supports the on-demand, precise, and efficient operation of dust removal equipment.

[0018] 3. This invention, through dust removal energy consumption matching, enables controllable operating parameters such as fan frequency, air volume adjustment, and dust cleaning cycle to be accurately correlated with the current dust conditions, establishing a dynamic mapping relationship between environmental status, control actions, and energy consumption response. Based on the real-time reconstructed dust scenario, the optimal operating strategy can be quickly retrieved from the matching table and recommended. Under the premise of ensuring dust removal efficiency and working environment safety, it avoids ineffective high-speed operation of the fan or excessive dust cleaning, significantly reducing energy consumption and equipment wear, and achieving refined, adaptive, and energy-saving control of the dust removal process. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a dust removal energy consumption intelligent control method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a dust removal energy consumption intelligent control system according to an embodiment of the present invention.

[0021] In the picture: 1. Dust scene reconstruction module; 2. Dust environment simulation module; 3. Dust removal related feature extraction module; 4. Dust removal energy consumption matching module; 5. Dust removal energy consumption control module. Detailed Implementation

[0022] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0023] According to an embodiment of the present invention, a method and system for intelligent control of dust removal energy consumption are provided.

[0024] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for intelligent control of dust removal energy consumption is provided, the method comprising: S1. Filter the acquired dust monitoring data to obtain denoised dust monitoring data, and reconstruct the dust scene based on the denoised dust monitoring data to generate an initial dust field.

[0025] Specifically, the acquired dust monitoring data is filtered to obtain denoised dust monitoring data, and the denoised dust monitoring data is then used to reconstruct the dust scene and generate an initial dust field, including: S11. Perform outlier processing on the acquired dust monitoring data to obtain cleaned dust monitoring data. Use a low-pass filter to perform noise reduction processing on the cleaned dust monitoring data to obtain denoised dust monitoring data. S12. Generative adversarial networks are used to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data to obtain enhanced dust monitoring data.

[0026] Specifically, generative adversarial networks are used to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data, resulting in enhanced dust monitoring data including: S121. Using a spatiotemporal convolutional encoder based on a generative adversarial network, the spatiotemporal feature tensor of dust is extracted from the denoised dust monitoring data, and a binary mask matrix identifying the missing dust region is generated simultaneously. S122. Input the dust spatiotemporal feature tensor into the generator, and perform decoding and spatial upsampling based on the spatiotemporal transposed convolutional network to obtain the preliminary dust concentration field. Use a discriminator with introduced physical constraint loss to iteratively optimize the preliminary dust concentration field to obtain a physically consistent preliminary dust complete field. S123. Using an attention mechanism, the initial dust full field and binary mask matrix are fused with local details and global context to obtain a semantically consistent dust transition field. Then, a semantic smoothing constraint on dust diffusion is applied to obtain enhanced dust monitoring data.

[0027] S13. Combine the enhanced dust monitoring data, the three-dimensional structure grid of dust space, and the prior airflow data to perform three-dimensional dust concentration distribution reconstruction and generate the initial dust field.

[0028] Specifically, the enhanced dust monitoring data, the three-dimensional dust spatial structure mesh, and prior airflow data are combined to perform a three-dimensional dust concentration distribution reconstruction, generating an initial dust field including: S131. Input the enhanced dust monitoring data and the three-dimensional structure mesh of dust space into the implicit function network, use the implicit function network to fit the observation points, and construct a continuous spatial mapping under geometric boundary constraints to obtain the preliminary three-dimensional dust implicit concentration field. S132. Using a differentiable physics solver, the initial three-dimensional implicit dust concentration field is coupled with the pre-constructed steady-state airflow velocity field to obtain the initial dust concentration evolution field. S133. Compare the initial dust concentration evolution field with the enhanced dust monitoring data point cloud, construct a variational optimization model based on the comparison results, and solve the variational optimization model to obtain the initial dust field.

[0029] Specifically, the initial dust concentration evolution field is compared with the enhanced dust monitoring data point cloud. Based on the comparison results, a variational optimization model is constructed and solved to obtain the initial dust field, which includes: S1331. The initial dust concentration evolution field is compared with the enhanced dust monitoring data point cloud at spatial grid points one by one, and the dust concentration residual value is calculated based on the comparison results. The weighted value assignment operation is performed based on the dust concentration residual value to obtain the weighted dust residual distribution field. S1332. Based on the dust residual distribution field, a multi-objective functional processing is performed to obtain a variational energy expression containing data fidelity terms, physical regularization terms, and spatial smoothing terms. The variational energy expression is then subjected to adaptive configuration of regularization parameters and discretization modeling to obtain a variational optimization model.

[0030] Specifically, variational optimization models include: ; In the formula, Represents the variational optimization function; This indicates the total number of observed dust data points; Indicates the index of the observed dust data points; Indicates the first Weighting coefficients for each observed dust data point; Indicates the first An observation operator for each dust data point; Indicates the first The actual observed values ​​of each dust data point; Represents the physical regularization parameter; Represents the spatial computational domain; This indicates that the physical operator acts on the field to be optimized. The result; This represents the action of the physical operator on the background field of the physical simulation. The result; This represents the smoothing regularization parameter; Indicates the field to be optimized Spatial gradient; This represents the initial dust concentration evolution field to be optimized.

[0031] S1333. Input the variational optimization model and the initial dust concentration evolution field into the sparse matrix solver, and perform optimization by minimizing the objective function to obtain the initial dust field.

[0032] Specifically, 24 laser scattering dust sensors (model PMS7003) were deployed in a large machinery manufacturing workshop. These sensors were distributed near key dust-generating workstations such as welding, grinding, and cutting, with a sampling frequency set to once per second. The sensors continuously collected PM2.5 and PM10 particulate matter concentrations in the air, measured in μg / m³, while simultaneously recording precise timestamps, the start / stop status of relevant equipment, and environmental temperature and humidity parameters. The dust monitoring data includes particulate matter concentration information collected by the 24 sensors at key workstations in the workshop, specifically covering real-time PM2.5 and PM10 concentration values ​​in μg / m³, timestamps accurate to the second, spatial coordinates of each sensor, synchronously recorded start / stop status of production equipment, and environmental temperature and humidity parameters. This forms a multi-dimensional time-series dataset with spatiotemporal attributes and operational condition correlations, providing a complete and reliable data foundation for subsequent noise reduction processing, 3D dust field reconstruction, and intelligent control strategy generation. The original collected data contained significant impulse noise and communication packet loss. First, outlier handling is performed using a 10-second sliding time window. Outliers are identified using the interquartile range (IQR). If a data point falls outside the range of the first quartile minus 1.5 IQR to the third quartile plus 1.5 IQR, it is marked as an outlier and imputed using linear interpolation. Next, a second-order Butterworth low-pass filter with a cutoff frequency of 0.05Hz and a sampling interval of one second is applied. The filter is constructed using a Butterworth filter design function from a scientific computing library and processed using a zero-phase filter function. This effectively preserves the low-frequency components reflecting the slow diffusion trend of dust while suppressing noise interference from high-frequency sensor jitter. The final output is denoised dust monitoring data with a temporal resolution of one data point per second and a spatial dimension of 24 monitoring nodes, forming a structured time-series dataset. Finally, a spatiotemporal generative adversarial network based on a 3D convolutional neural network is constructed to perform spatiotemporal completion and semantic enhancement on the denoised data. The denoised data is resampled into 10-second time steps, and the spatial information is mapped onto an 8×8 two-dimensional grid to form the input data tensor. The encoder consists of three 3D convolutional layers with a kernel size of 3×3×3 and a stride of 1, used to extract the high-dimensional dust spatiotemporal feature tensor. Simultaneously, a binary mask matrix is ​​generated based on the actual sensor deployment locations, where the sensor locations are marked as 1 and other blank areas as 0. This mask guides the generator to focus on completing missing data regions. The generator consists of three sets of spatiotemporally transposed convolutional layers with a kernel size of 2×2×2 and an upsampling factor of 2, progressively upsampling the extracted features to restore a complete dust concentration field consistent with the input time step and spatial resolution. The discriminator, composed of a 3D convolutional neural network, is responsible for distinguishing the generated concentration field from the field filled with real data and calculating the adversarial loss.A physical constraint loss is introduced, based on the advection-diffusion equation with dust mass conservation. The airflow velocity field is derived from a pre-calculated computational fluid dynamics simulation library with a spatial resolution of 32³, mapped to an 8×8 planar grid via interpolation, and the diffusion coefficient is set to 0.1 m² / s. The gradient term is numerically approximated using the central difference method. The generator's total loss is composed of three weighted components: adversarial loss, physical constraint loss, and data fidelity loss, with weights of 1, 0.6, and 0.4, respectively. Model training uses an adaptive moment estimation optimizer with a learning rate of 1e. -4The momentum parameter is 0.5. The training dataset contains dust diffusion sequences under different operating conditions generated by computational fluid dynamics simulations. The training batch size is 8, and several training rounds are performed. After training, the generator outputs a physically consistent preliminary dust completion field. This field is spatially continuous and conforms to basic fluid transport laws. A dual-branch attention refinement network is introduced to optimize the preliminary completion field. The backbone of this network adopts a U-shaped network structure. The input is the completion field and a binary mask matrix. Through channel attention and spatial attention, the network can focus on the sensor edge transition region and the key blank area of ​​the completion, effectively fusing local detail features with global contextual information. A diffusion smoothing constraint term is added to the loss function, which forces the concentration gradient to decrease gently in areas without dust sources, avoiding sharp, non-physical abrupt changes, thereby improving the semantic rationality of the completion field. The output is enhanced dust monitoring data with a spatial resolution of 8×8, continuous temporal series, and high-quality input for 3D field reconstruction. The enhanced dust monitoring data was combined with the building information model of the workshop, which covered a spatial area of ​​60m long, 40m wide, and 8m high, constructing a 64×48×16 three-dimensional computational grid. A sinusoidal representation network was used as the architecture of the implicit function network, taking the three-dimensional spatial coordinates as input and outputting the dust concentration value at the corresponding location. The network used a sine function as the activation function, with its frequency parameter set to 30. During training, two objectives were minimized: first, to force the network's output at the sensor location to be completely consistent with the enhanced data; and second, to force the dot product of the concentration gradient and the surface normal vector at geometric boundaries such as walls and equipment to be zero, i.e., to achieve zero gradient boundary conditions. After several rounds of training, the network converged, outputting a continuous and geometrically compliant preliminary three-dimensional implicit dust concentration field. A differentiable physics solver based on the JAX framework was used for physics-driven evolution. The preliminary implicit concentration field was used as the initial condition, and a steady-state airflow velocity field pre-calculated by computational fluid dynamics software with a resolution of 32³ was input and upsampled to a 64×48×16 computational grid using interpolation methods. The solver, based on the advection-diffusion equation, performs 10 time iterations, each with a time step of one second, simulating the convective transport and diffusion of dust under airflow. This process is completed within an automatic differentiation framework, ensuring the differentiability of the entire computational graph. After iterative evolution, a physically corrected initial dust concentration evolution field is output, which already includes a reasonable dust distribution pattern guided by airflow.

[0033] The initial dust concentration evolution field and the enhanced dust monitoring data are compared one by one at corresponding nodes of the three-dimensional computational grid, and the concentration residual value at each point is calculated. Based on the accuracy index of each sensor, i.e., the measurement error of PMS7003 is ±10%, a weight is assigned to each residual value, with higher-accuracy sensors receiving larger weights, forming a weighted dust residual distribution field. Based on this residual field, an energy functional with three terms is constructed: the first term is a data fidelity term, requiring the optimized field to be highly consistent with the observed values ​​at the sensor locations, and is weighted according to the weights; the second term is a physical regularization term, requiring the optimized field to satisfy the advection-diffusion equation as much as possible to ensure adherence to physical laws; the third term is a spatial smoothing term, used to suppress non-physical oscillations and ensure the smoothness of the field. The balance weight coefficients of these three terms are determined using the L-curve method, which are 0.8, 0.15, and 0.05, respectively. This continuous functional is discretized on the three-dimensional grid, transforming it into a large sparse linear equation system. An algebraic multigrid solver is used for iterative solution, with a convergence threshold of 1e. -6 The solution yields an initial dust field that is precisely matched to the data points, physically consistent across the entire domain, and spatially smooth. This high-confidence initial dust field serves as the input for real-time prediction and intelligent control, enabling a high-precision closed-loop reconstruction of the complete and physically reliable three-dimensional environmental situation from sparse, discrete sensor data.

[0034] S2. Combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results.

[0035] Specifically, the initial dust field data was imported into computational fluid dynamics simulation software. A total number of computational grid cells were generated within the workshop's computational area, with mesh refinement at the main dust-generating equipment outlets and personnel working areas to ensure computational accuracy. The required physical model and parameters for the simulation were set. The airflow field was calculated using a Reynolds-averaged incompressible turbulence model, while the motion of dust particles was tracked using a Lagrange discrete phase model. The four air inlets at the top of the workshop were set as velocity inlet boundaries, with a wind speed set to 2.5 m / s. The four dust collector suction hood inlets were set as pressure outlet boundaries. The dust source was located at the discharge port of the No. 1 crusher, where a surface source with an area of ​​0.5 square meters was defined, with a particle ejection velocity of 1.2 m / s and an initial mass flow rate of 0.05 kg / s. The initial condition for the entire flow field was set to still air, and the initial dust field concentration values ​​were assigned to the corresponding grid cells. After completing the settings, the transient solver was started for simulation. The total simulation physics time was set to 180 seconds, and the computation time step was set to 0.1 seconds. Complete dust environment simulation results were obtained. The results showed that after 60 seconds of simulation, the dust cloud had spread to a distance of 15 meters from the source. At a typical breathing zone height of 1.6 meters, the peak concentration at a monitoring point 10 meters from the dust source reached 25 mg / m³. Throughout the simulation period, the one-minute time-weighted average concentration at this monitoring point was 12.5 mg / m³. The simulation output included a three-dimensional concentration distribution cloud map of the entire space, concentration contour maps of key sections, and time-series concentration data for all preset monitoring points. These results collectively constitute the dust environment simulation results used for subsequent intelligent control decisions.

[0036] S3. Obtain historical dust removal operation data and extract dust removal-related features from the historical dust removal operation data to obtain a dust removal-related feature set.

[0037] Specifically, S3 includes: S31, acquiring historical dust removal operation data, standardizing and segmenting the historical dust removal operation data to obtain a structured time-series dataset with marked operating conditions, and using time-series feature engineering to extract dust removal-related features from the time-series dataset with marked operating conditions to obtain a dust removal-related feature set.

[0038] S4. Generate an energy consumption causal graph based on the dust removal associated feature set, and construct a dust removal energy consumption matching table based on the energy consumption causal graph.

[0039] Specifically, an energy consumption causal graph is generated based on the dust removal associated feature set, and a dust removal energy consumption matching table is constructed based on the energy consumption causal graph, including: S41. Perform stationarity test and nonlinear transformation preprocessing on the dust removal correlation feature set to obtain a standardized dust removal correlation data matrix; S42. Using the causal discovery algorithm, construct causal dependencies and remove false edges from the dust removal associated data matrix to obtain the energy consumption causal graph. Then, perform causal path identification and mapping on the energy consumption causal graph to obtain the dust removal energy consumption matching table.

[0040] Specifically, a causal discovery algorithm is used to construct causal dependencies and remove false edges from the dust removal correlation data matrix to obtain an energy consumption causal graph. Then, causal path identification and mapping are performed on the energy consumption causal graph to obtain a dust removal energy consumption matching table, including: S421. Using the causal discovery algorithm, conditional independence tests and directed acyclic graph searches are performed on the dust removal associated data matrix, and the causal dependencies between variables are initially constructed to obtain the candidate causal graph structure. S422. Based on knowledge of dust removal, the candidate causal graph structure is verified for causal direction and the rationality of intervention is verified. False edges are eliminated using counterfactual consistency test to obtain the energy consumption causal graph. S423. Use graph traversal algorithm to identify energy consumption causal paths in the energy consumption causal graph, and perform rule mapping and optimization adjustment based on the energy consumption causal path identification results to obtain the dust removal energy consumption matching table.

[0041] Specifically, a graph traversal algorithm is used to identify energy consumption causal paths in the energy consumption causal graph, and rule mapping and optimization are performed based on the energy consumption causal path identification results to obtain a dust removal energy consumption matching table, including: S4231. Use graph traversal algorithm to identify the energy consumption causal path from working condition characteristics to controllable actions in the energy consumption causal graph, and generate a structured causal rule set based on the energy consumption causal path identification results. S4232. Discretize and partition the domain of all precondition variables in the structured causal rule set to obtain a multidimensional working condition grid, and associate the rule output corresponding to each grid cell to obtain an indexed preliminary mapping matrix. S4233. Based on preset operating costs and safety constraints, a reinforcement learning strategy optimizer is used to adjust and verify the parameters of the state grid of the initial mapping matrix to obtain a dust removal energy consumption matching table.

[0042] Specifically, the historical dust removal operation data acquired typically spans six months, with a sampling frequency of one minute. Data sources include distributed control systems, sensor networks, and electricity meters. Raw fields include: environmental variables reflecting operating conditions, such as inlet dust concentration PM_in, which typically fluctuates between 0 and 500 mg / m³; duct negative pressure P characterizing system resistance, normally between -800 and -2000 Pascals; flue gas temperature T, typically between 70 and 120 degrees Celsius; and the binary equipment operating status signal On_Off. Key controllable action variables include: the fan inverter frequency F, which determines power input, with an adjustment range of 30 to 50 Hz; the pulse jet interval Interval, controlling cleaning intensity, set between 20 and 60 seconds; and the bypass valve opening Valve, adjusting airflow distribution from 0% (fully closed) to 100% (fully open). The core target variable is real-time power consumption Power, read from a smart meter, with a value of 78.5 kW. Z-Score standardization is employed, which involves subtracting the mean over the entire time span from each continuous variable sequence and dividing by the standard deviation. For example, the mean of PM_in might be 180 mg / m³ with a standard deviation of 95; a standardized value of 150 mg / m³ would be approximately -0.32. This aims to eliminate numerical scale differences caused by different physical units. To capture dynamic patterns rather than isolated instantaneous states, a sliding time window of 10 minutes with a step size of 1 minute is introduced. Within each window, the mean, standard deviation, and linear regression slope of key sequences such as PM_in, P, and Power are calculated. Subsequently, K-Score standardization is applied. The Means clustering algorithm performs unsupervised learning on these multi-dimensional window features, determining the optimal number of clusters based on metrics such as the silhouette coefficient. In practice, it typically clusters four typical operating conditions: Condition A represents low-load stable operation, characterized by a PM_in window mean below 80 mg / m³ with a standard deviation of less than 15, and a power mean stable at around 45 kW; Condition B represents high-load stable operation, with a PM_in window mean above 200 mg / m³ and a power mean above 85 kW; Condition C represents a load increase transition process, with a significantly positive slope within the PM_in window; and Condition D represents a load decrease transition process, with a significantly negative slope. After clustering, each minute of the original data stream is labeled with its corresponding window's operating condition label, resulting in a structured time-series dataset rich in contextual information.Deep correlation features are extracted by constructing lag features and interaction features. Specifically, for key variables such as PM_in and F, first-order lag PM_in_lag1, second-order lag PM_in_lag2, and fan frequency lag F_lag1 are created to encode the causal impact of historical states on the current system. Simultaneously, physically meaningful interaction features are constructed, such as the load factor Load_Factor, calculated as the product of PM_in and F, to approximate the dust load being processed at that moment; and the instantaneous energy efficiency index Efficiency, calculated as the ratio of PM_in to Power, to measure the processing capacity per unit of energy consumption. Furthermore, operating condition labels are transformed into binary features such as Regime_A and Regime_B through one-hot encoding, and further multiplied with continuous variables such as PM_in to generate PM_in. The interaction term of Regime_B models the hypothesis that "the impact of dust concentration on operation may differ under high load conditions." By calculating the mutual information between all candidate features, including the original variables and newly constructed indicators, and the target variable Power, the top 25 most important features are selected to form a dust removal-related feature set. This set is a data matrix with higher dimension and greater information density. The feature set is then tested for stationarity, non-stationary sequences are subjected to first-order differencing, and features with nonlinear relationships are logarithmically transformed. Global standardization is then performed again, outputting a clean and well-ordered standardized dust removal-related data matrix. The PC algorithm, with conditional independence testing at its core, begins with a complete graph where all variables have undirected edges. For each pair of variables, such as fan frequency F and energy consumption Power, statistical independence is tested under progressively increasing condition sets, such as an empty set, containing only PM_in, and containing both PM_in and P. Fisher's Z test, suitable for continuous Gaussian variables, is used, with a significance threshold p-value of 0.05. If, given PM_in, F and Power are found to be independent, the direct edge between them is deleted, implying that their correlation may be caused by the common cause PM_in. After a series of tests and rule orientation based on the V-structure, a candidate causal graph structure is obtained, which may contain some directed edges and many undirected edges. Domain-knowledge-based causal graph refinement is initiated by introducing inviolable domain rules: the inlet dust concentration PM_in, as an environmental input, should point to the fan frequency F and the pipeline negative pressure P, not the other way around; the fan frequency F, as the main control action, should directly point to the real-time energy consumption Power and the pipeline negative pressure P; time-lag variables, such as PM_in_lag1, can only point to variables at the current time. Based on these rules, the direction of all edges in the candidate graph is forcibly verified and corrected. To further eliminate spurious associations, a counterfactual consistency test is implemented. For example, for an edge in the candidate graph pointing from flue gas temperature T to Power, two high-precision gradient boosting tree models M1 and M2 are trained to predict Power. M1 uses all features, while M2 removes feature T. Counterfactual intervention is performed on a large dataset sample. Specifically, in M2, the value of T is fixed to the average of the entire dataset. The distribution of predicted Power is then observed to see if it systematically shifts from the distribution predicted in M1 using the true value of T. If the shift is not significant, it indicates that the effect of T on Power is likely due to confounding effects from its association with related variables such as PM_in. Therefore, this edge is removed from the graph.After dual filtering using data-driven and knowledge-guided approaches, a robust energy consumption causal graph is obtained. This graph clearly reveals the effects of PM_in on Power through two main paths: one a direct influence, and the other an indirect influence through F. F has a direct causal effect on P, but P has a back-curving effect on F based on control logic. This is typically represented by a bidirectional edge in the graph or by introducing a control error variable. The abstract energy consumption causal graph is then transformed into an operational dynamic dust removal energy consumption matching table. Specifically, a graph traversal algorithm is used to identify energy consumption causal paths. Starting from characteristic nodes such as PM_in and Regime_B, a depth-first search is performed to find all paths leading to the target node Power and the controllable action node F. The identified critical paths may include: the main path PM_in, F, Power; auxiliary paths PM_in, P, F, Power representing pressure difference-based back-curving regulation; and the operating condition path Regime_B, F, Power. Each path is transformed into a structured causal rule, whose parameters are quantified by fitting a linear or nonlinear structural equation model to the graph structure. For example, for the main path, the fitting results show that when PM_in is less than 50 mg / m³, maintaining F at 35 Hz can predict a power of approximately 40 kW; when PM_in is in the range of 150 to 300 mg / m³, F should be increased to 45 Hz, corresponding to a predicted power of approximately 75 kW. This generates a quantified set of structured causal rules. State space discretization and preliminary table construction involve selecting the two most intuitive and easily monitored dimensions: the discretized PM_in level and the operating condition label Regime, which serve as the row and column indices of the matching table. PM_in is divided into four levels: low [0,50), medium [50,150), high [150,300), and ultra-high [300,∞), with units of mg / m³. The operating condition labels are derived from clustering: A, B, C, and D. This forms a 4x4 two-dimensional decision grid with 16 cells. The content of each cell is mapped and filled from the rule set. For example, the cell located at (PM_in=high, Regime=B) contains: recommended fan frequency F=48 Hz, injection interval Interval=15 seconds, expected energy consumption Power=85 kW, along with a confidence score calculated based on historical data coverage and prediction error. This results in an indexed preliminary mapping matrix, i.e., a static matching table. Reinforcement learning-based policy fine-tuning is then implemented. A comprehensive cost function is defined: total cost = 0.6. Normalized energy consumption +0.3 Equipment wear and tear costs +10.0 Exceeding the limit penalty. Equipment wear cost is approximated by the square of the fan frequency variation; the exceeding penalty is a large constant penalty term triggered when the outlet concentration exceeds the environmental standard of 20 mg / m³ in the simulation. A simulation environment is constructed based on historical data or simplified physical equations, such as damper equations and filter models, to simulate the system response under any given control action. State s is the index of the matching table, i.e., PM_in level, Regime. Action a is a fine-tuning of the recommended parameters in the table, for example, allowing the agent to choose adjustments of -3, 0, or +3 Hz based on the recommended F=48 Hz. A neural network agent is trained using a proximal policy optimization algorithm, allowing it to explore millions of time steps in the simulation environment. The agent's learning objective is not to minimize instantaneous energy consumption, but rather to minimize the long-term discounted total cost. After training, the agent's optimal policy is used to revise the static table. For example, the agent might discover that, under PM_in=high and Regime=B conditions, slightly reducing F from 48 Hz to 47 Hz, while causing a slight 1% increase in instantaneous energy consumption, significantly reduces fan stress changes and the potential risk of filter bag damage, resulting in lower long-term overall costs. Based on this, the recommended parameters for that cell are updated. The output dynamic dust removal energy consumption matching table not only encodes causal relationships in historical data but also integrates optimization strategies for long-term economic efficiency and safety. It becomes the core knowledge hub connecting dust environment perception and energy-saving intelligent control, enabling operators to directly query and obtain a set of globally optimized equipment operating parameter settings based on real-time monitored concentration ranges and system operating modes.

[0043] S4. Based on the dust environment simulation results and the dust removal energy consumption matching table, perform dust removal energy consumption mapping to obtain dust removal equipment operation data, and implement intelligent control of dust removal energy consumption based on the dust removal equipment operation data.

[0044] Specifically, the energy consumption mapping and intelligent control based on the dust environment simulation results and the dust removal energy consumption matching table involves analyzing the three-dimensional concentration field data, key monitoring point time-series curves, and statistical reports generated by the dust environment simulation. It automatically identifies pollution hotspots with concentrations exceeding the 10 mg / m³ alarm threshold and calculates their geometric center coordinates and peak concentrations. For example, it identifies the peak concentration of Hotspot_A located at coordinates (15, 10, 1.5) as 22.5 mg / m³. At the same time, it analyzes the preset virtual monitoring point V1, such as the concentration data of 5 meters downwind of the crusher over the past 5 minutes, and obtains a medium-to-high concentration, slowly increasing load-dominant operating condition with an average concentration of 15.2 mg / m³ and a trend slope of +0.3 mg / m³ / min. Combined with the global average load value of 8.7 mg / m³ calculated from the height layer of the entire workshop's breathing zone, the results are structured and output as a decision feature vector containing the main hotspot concentration, location, dominant operating condition, global load, and trend. Using this feature vector as input, multi-level retrieval and interpolation mapping are performed on the dynamic dust removal energy consumption matching table: the recommended parameters are located based on the load increase condition and medium load level as the first-level index, such as the fan frequency benchmark of 42Hz and the pulse interval benchmark of 25 seconds. Then, the concentration compensation rule is triggered based on the hot spot concentration of 22.5mg / m³, and the concentration compensation rule is increased by 2Hz. The feedforward compensation is triggered based on the upward trend, and the frequency is increased by 1Hz. The weight of the hot spot position is considered for comprehensive calculation. After safety boundary verification, such as limiting the frequency change rate to ≤3Hz / cycle, a set of specific dust removal equipment operation parameter instruction set is finally generated, such as {fan frequency: 45.0Hz, fan frequency: 40.0Hz, pulse valve group interval: 20s, A zone air outlet valve opening: 85%} and expected energy consumption of 78kW. The industrial control system sends commands to field equipment for execution and simultaneously initiates multimodal closed-loop feedback monitoring: smart meters collect actual power in real time (e.g., 80.5kW) and compare it with the predicted value; real dust sensors near Hotspot_A track the concentration decrease curve to verify the control effect (e.g., the target is to reduce the concentration to below 12mg / m³ within 5 minutes); and equipment status such as fan current, vibration, and filter bag pressure differential are monitored. Short-term adaptive adjustments are made based on real-time feedback. If the concentration decrease does not meet expectations, the frequency is fine-tuned via a built-in PID controller (e.g., gradually increasing from 45.0Hz to 46.5Hz). In the long term, all control process data is stored in a historical database as samples of status, actions, and rewards, and a reinforcement learning strategy is periodically triggered to optimize the process. New data is used to update the recommended parameter values ​​in the dynamic dust removal energy consumption matching table, thus forming a complete intelligent control closed loop from environmental perception, intelligent decision-making, precise execution to continuous evolution. This ensures stable dust concentration compliance while achieving continuous optimization of system energy efficiency.

[0045] like Figure 2 As shown, according to another embodiment of the present invention, a dust removal energy consumption intelligent control system is provided, the system comprising: The dust scene reconstruction module 1 is used to filter the acquired dust monitoring data to obtain denoised dust monitoring data, and to reconstruct the dust scene on the denoised dust monitoring data to generate an initial dust field. Dust environment simulation module 2 is used to combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results. The dust removal associated feature extraction module 3 is used to acquire historical dust removal operation data and extract dust removal associated features from the historical dust removal operation data to obtain a dust removal associated feature set; Dust removal energy consumption matching module 4 is used to generate an energy consumption causal graph based on the dust removal associated feature set, and to construct a dust removal energy consumption matching table based on the energy consumption causal graph; The dust removal energy consumption control module 5 is used to perform dust removal energy consumption mapping based on the dust environment simulation results and the dust removal energy consumption matching table, obtain the dust removal equipment operation data, and implement intelligent dust removal energy consumption control based on the dust removal equipment operation data.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent control of dust removal energy consumption, characterized in that, The method includes: S1. Filter the acquired dust monitoring data to obtain denoised dust monitoring data, and reconstruct the dust scene based on the denoised dust monitoring data to generate an initial dust field. S2. Combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results; S3. Obtain historical dust removal operation data and extract dust removal-related features from the historical dust removal operation data to obtain a dust removal-related feature set; S4. Generate an energy consumption causal graph based on the dust removal associated feature set, and construct a dust removal energy consumption matching table based on the energy consumption causal graph; S5. Based on the dust environment simulation results and the dust removal energy consumption matching table, perform dust removal energy consumption mapping to obtain dust removal equipment operation data, and implement intelligent control of dust removal energy consumption based on the dust removal equipment operation data.

2. The intelligent dust removal energy consumption control method according to claim 1, characterized in that, The process of filtering the acquired dust monitoring data to obtain denoised dust monitoring data, and then reconstructing the dust scene from the denoised dust monitoring data to generate an initial dust field includes: S11. Perform outlier processing on the acquired dust monitoring data to obtain cleaned dust monitoring data. Use a low-pass filter to perform noise reduction processing on the cleaned dust monitoring data to obtain denoised dust monitoring data. S12. Generative adversarial networks are used to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data to obtain enhanced dust monitoring data. S13. Combine the enhanced dust monitoring data, the three-dimensional structure grid of dust space, and the prior airflow data to perform three-dimensional dust concentration distribution reconstruction and generate the initial dust field.

3. The intelligent control method for dust removal energy consumption according to claim 2, characterized in that, The process of using a generative adversarial network to perform spatiotemporal completion and semantic enhancement on the denoised dust monitoring data results in enhanced dust monitoring data, including: S121. Using a spatiotemporal convolutional encoder based on a generative adversarial network, the spatiotemporal feature tensor of dust is extracted from the denoised dust monitoring data, and a binary mask matrix identifying the missing dust region is generated simultaneously. S122. Input the dust spatiotemporal feature tensor into the generator, and perform decoding and spatial upsampling based on the spatiotemporal transposed convolutional network to obtain the preliminary dust concentration field. Use a discriminator with introduced physical constraint loss to iteratively optimize the preliminary dust concentration field to obtain a physically consistent preliminary dust complete field. S123. Using an attention mechanism, the initial dust full field and binary mask matrix are fused with local details and global context to obtain a semantically consistent dust transition field. Then, a semantic smoothing constraint on dust diffusion is applied to obtain enhanced dust monitoring data.

4. The intelligent control method for dust removal energy consumption according to claim 2, characterized in that, The step of combining enhanced dust monitoring data, three-dimensional dust spatial structure mesh, and prior airflow data to perform three-dimensional dust concentration distribution reconstruction and generate an initial dust field includes: S131. Input the enhanced dust monitoring data and the three-dimensional structure mesh of dust space into the implicit function network, use the implicit function network to fit the observation points, and construct a continuous spatial mapping under geometric boundary constraints to obtain the preliminary three-dimensional dust implicit concentration field. S132. Using a differentiable physics solver, the initial three-dimensional implicit dust concentration field is coupled with the pre-constructed steady-state airflow velocity field to obtain the initial dust concentration evolution field. S133. Compare the initial dust concentration evolution field with the enhanced dust monitoring data point cloud, construct a variational optimization model based on the comparison results, and solve the variational optimization model to obtain the initial dust field.

5. The intelligent dust removal energy consumption control method according to claim 4, characterized in that, The process involves comparing the initial dust concentration evolution field with the enhanced dust monitoring data point cloud, constructing a variational optimization model based on the comparison results, and solving the variational optimization model to obtain the initial dust field, which includes: S1331. The initial dust concentration evolution field is compared with the enhanced dust monitoring data point cloud at spatial grid points one by one, and the dust concentration residual value is calculated based on the comparison results. The weighted value assignment operation is performed based on the dust concentration residual value to obtain the weighted dust residual distribution field. S1332. Based on the dust residual distribution field, perform multi-objective functional processing to obtain a variational energy expression containing data fidelity terms, physical regularization terms, and spatial smoothing terms. Then, perform regularization parameter adaptive configuration and discretization modeling on the variational energy expression to obtain a variational optimization model. S1333. Input the variational optimization model and the initial dust concentration evolution field into the sparse matrix solver, and perform optimization by minimizing the objective function to obtain the initial dust field.

6. The intelligent dust removal energy consumption control method according to claim 5, characterized in that, The variational optimization model includes: ； In the formula, Represents the variational optimization function; This indicates the total number of observed dust data points; Indicates the index of the observed dust data points; Indicates the first Weighting coefficients for each observed dust data point; Indicates the first An observation operator for each dust data point; Indicates the first The actual observed values ​​of each dust data point; Represents the physical regularization parameter; Represents the spatial computational domain; This indicates that the physical operator acts on the field to be optimized. The result; This represents the action of the physical operator on the background field of the physical simulation. The result; This represents the smoothing regularization parameter; Indicates the field to be optimized Spatial gradient; This represents the initial dust concentration evolution field to be optimized.

7. The intelligent dust removal energy consumption control method according to claim 1, characterized in that, The step of generating an energy consumption causal graph based on a dust removal correlation feature set, and constructing a dust removal energy consumption matching table based on the energy consumption causal graph, includes: S41. Perform stationarity test and nonlinear transformation preprocessing on the dust removal correlation feature set to obtain a standardized dust removal correlation data matrix; S42. Using the causal discovery algorithm, construct causal dependencies and remove false edges from the dust removal associated data matrix to obtain the energy consumption causal graph. Then, perform causal path identification and mapping on the energy consumption causal graph to obtain the dust removal energy consumption matching table.

8. The intelligent control method for dust removal energy consumption according to claim 7, characterized in that, The process involves using a causal discovery algorithm to construct causal dependencies and remove false edges from the dust removal associated data matrix, resulting in an energy consumption causal graph. This graph is then used for causal path identification and mapping to obtain a dust removal energy consumption matching table, which includes: S421. Using the causal discovery algorithm, conditional independence tests and directed acyclic graph searches are performed on the dust removal associated data matrix, and the causal dependencies between variables are initially constructed to obtain the candidate causal graph structure. S422. Based on knowledge of dust removal, the candidate causal graph structure is verified for causal direction and the rationality of intervention is verified. False edges are eliminated using counterfactual consistency test to obtain the energy consumption causal graph. S423. Use graph traversal algorithm to identify energy consumption causal paths in the energy consumption causal graph, and perform rule mapping and optimization adjustment based on the energy consumption causal path identification results to obtain the dust removal energy consumption matching table.

9. The intelligent dust removal energy consumption control method according to claim 8, characterized in that, The process of using a graph traversal algorithm to identify energy consumption causal paths in the energy consumption causal graph, and then performing rule mapping and optimization adjustments based on the identified energy consumption causal paths, results in a dust removal energy consumption matching table, including: S4231. Use graph traversal algorithm to identify the energy consumption causal path from working condition characteristics to controllable actions in the energy consumption causal graph, and generate a structured causal rule set based on the energy consumption causal path identification results. S4232. Discretize and partition the domain of all precondition variables in the structured causal rule set to obtain a multidimensional working condition grid, and associate the rule output corresponding to each grid cell to obtain an indexed preliminary mapping matrix. S4233. Based on preset operating costs and safety constraints, a reinforcement learning strategy optimizer is used to adjust and verify the parameters of the state grid of the initial mapping matrix to obtain a dust removal energy consumption matching table.

10. A dust removal energy consumption intelligent control system, used to implement the dust removal energy consumption intelligent control method according to any one of claims 1-9, characterized in that, The system includes: The dust scene reconstruction module is used to filter the acquired dust monitoring data to obtain denoised dust monitoring data, and to reconstruct the dust scene from the denoised dust monitoring data to generate an initial dust field. The dust environment simulation module is used to combine the initial dust field with the preset airflow field to simulate the dust environment and obtain the dust environment simulation results. The dust removal associated feature extraction module is used to acquire historical dust removal operation data and extract dust removal associated features from the historical dust removal operation data to obtain a dust removal associated feature set; The dust removal energy consumption matching module is used to generate an energy consumption cause-effect graph based on the dust removal associated feature set, and to construct a dust removal energy consumption matching table based on the energy consumption cause-effect graph. The dust removal energy consumption control module is used to perform dust removal energy consumption mapping based on the dust environment simulation results and the dust removal energy consumption matching table, obtain the dust removal equipment operation data, and implement intelligent dust removal energy consumption control based on the dust removal equipment operation data.

Images