A Multimodal Sensing-Based Automatic Waste Combustion Control Method and System

Abstract

This invention discloses a method and system for automatic waste combustion control based on multimodal sensing, relating to the field of combustion control technology. The method includes: real-time acquisition of multimodal operating status data of a waste incinerator via a multimodal sensing module; inputting the multimodal operating status data into an intelligent decision-making model to predict the waste combustion status and generate combustion trend information; comparing the combustion trend information with a preset combustion control target and generating optimization instructions based on the comparison results; optimizing the combustion air volume, fuel supply volume, and grate motion parameters based on the optimization instructions and the combustion trend information to generate an optimal combustion parameter set; and controlling the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator using the optimal combustion parameter set. This solves the technical problems of inaccurate incinerator control and delayed adjustment response in existing technologies, leading to low waste combustion efficiency, and achieves the technical effect of improving waste combustion efficiency.

Description

Technical Field

[0001] This invention relates to the field of combustion control technology, and more specifically to an automatic waste combustion control method and system based on multimodal sensing. Background Technology

[0002] Waste incineration, as an effective waste treatment method, can not only significantly reduce waste volume but also achieve resource utilization through energy recovery. However, the waste incineration process is affected by various factors such as the complex and diverse composition of waste, large fluctuations in feed volume, and frequent changes in the furnace combustion environment, posing significant challenges to the stability and efficiency of the combustion process. Traditional waste incinerators often rely on single sensor data or human experience for combustion regulation, resulting in limited control methods and slow response times. This makes it difficult to achieve comprehensive perception and precise control of the combustion status, leading to low combustion efficiency, high energy consumption, and difficulty in effectively controlling harmful gas emissions. Summary of the Invention

[0003] This application provides a method and system for automatic waste combustion control based on multimodal perception, which solves the technical problems of inaccurate incinerator control and sluggish adjustment response in the prior art, resulting in low waste combustion efficiency.

[0004] The first aspect of this application provides a method for automatic waste combustion control based on multimodal sensing, the method comprising:

[0005] The multimodal sensing module collects real-time multimodal operating status data of the waste incinerator, including at least temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data. This multimodal operating status data is input into an intelligent decision-making model to predict the waste combustion status and generate combustion trend information. The combustion trend information is compared with a preset combustion control target, and optimization instructions are generated based on the comparison results. Based on the optimization instructions and the combustion trend information, the combustion air volume, fuel supply volume, and grate motion parameters are optimized to generate an optimal combustion parameter set. The optimal combustion parameter set is used to control the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator.

[0006] A second aspect of this application provides an automatic waste combustion control system based on multimodal sensing, the system comprising:

[0007] Data acquisition component: Real-time acquisition of multi-modal operating status data of the waste incinerator through a multi-modal sensing module, including at least temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data; Prediction component: Inputting the multi-modal operating status data into an intelligent decision-making model to predict the waste combustion status and generate combustion trend information; Comparison component: Comparing the combustion trend information with preset combustion control targets and generating optimization instructions based on the comparison results; Optimization component: Based on the optimization instructions and the combustion trend information, optimizing the combustion air volume, fuel supply volume, and grate motion parameters to generate an optimal combustion parameter set; Control component: Controlling the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator using the optimal combustion parameter set.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] First, a multi-modal sensing module collects real-time multi-modal operating status data of the waste incinerator, including at least temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data. Next, the multi-modal operating status data is input into an intelligent decision-making model to predict the waste combustion state and generate combustion trend information. Further, the combustion trend information is compared with preset combustion control targets, and optimization instructions are generated based on the comparison results. Then, based on the optimization instructions and the combustion trend information, the combustion air volume, fuel supply volume, and grate motion parameters are optimized to generate the optimal combustion parameter set. Finally, the optimal combustion parameter set is used to control the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator. This solves the technical problems of inaccurate incinerator control and lag in adjustment response in existing technologies, leading to low waste combustion efficiency, and achieves the technical effect of improving waste combustion efficiency. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0011] Figure 1 A schematic flowchart of an automatic waste combustion control method based on multimodal sensing provided in an embodiment of this application;

[0012] Figure 2 A schematic diagram of the structure of an automatic waste combustion control system based on multimodal perception provided in an embodiment of this application.

[0013] Figure labeling: Data acquisition component 11, prediction component 12, comparison component 13, optimization component 14, control component 15. Detailed Implementation

[0014] This application provides a method and system for automatic waste combustion control based on multimodal perception, which solves the technical problems of inaccurate incinerator control and sluggish adjustment response in the prior art, resulting in low waste combustion efficiency.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0017] Example 1, as Figure 1 As shown, this application provides a method for automatic waste combustion control based on multimodal sensing, wherein the method includes:

[0018] The multimodal sensing module collects multimodal operating status data of the waste incinerator in real time, including at least temperature detection data, material layer thickness detection data, flue gas detection data, and feed rate detection data.

[0019] Multimodal sensing modules deployed in different parts of the waste incinerator collect real-time data on the incinerator's operating status, obtaining multimodal operating status data including temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data. Specifically, temperature detection data reflects the heat distribution in different combustion zones; bed thickness detection data reflects the current bed distribution characteristics; flue gas detection data reflects the current combustion sufficiency and pollutant emissions; and feed rate detection data reflects the waste supply rate.

[0020] Furthermore, the multimodal sensing module includes at least a temperature sensor array, a pressure and differential pressure sensor array, a gas composition detector, and a weight sensor. The temperature sensor array is distributed across various locations within the furnace of the waste incinerator, collecting real-time temperature distribution data for different areas of the furnace and generating temperature detection data. The pressure and differential pressure sensor array is distributed on the combustion grate surface, collecting the pressure difference above and below the grate surface and the airflow at the grate surface to determine the thickness of the material layer on the grate surface and generate material layer thickness detection data. The gas composition detector is distributed at the flue gas outlet, detecting the gas composition content in the flue gas and generating flue gas detection data. The weight sensor is distributed at the feed inlet, detecting the amount of waste fed and generating feed amount detection data.

[0021] The multimodal sensing module includes at least a temperature sensor array, a pressure and differential pressure sensor array, a gas composition detector, and a weight sensor. It is used to perceive and collect multi-dimensional operational status information within the waste incinerator in real time and comprehensively. Specifically, the temperature sensor array is evenly distributed at multiple key locations in the incinerator furnace, including the upper, middle, and grate-adjacent areas. This allows for continuous monitoring of temperature changes in different combustion zones within the furnace, generating temperature detection data characterizing the furnace's heat distribution. This data can be used to determine whether local combustion is sufficient and whether the temperature meets standards. The pressure and differential pressure sensor array is installed above and below the combustion grate. By collecting the static pressure difference above and below the grate surface and the combustion air velocity passing through the grate, it calculates the degree of obstruction of the material to the exhaust gas flow. Then, combined with a pre-defined material layer thickness calculation model, it estimates the current material layer thickness. The system monitors the distribution of waste and outputs data on the thickness of the waste layer. A gas composition detector is installed at the flue gas emission pipe or chimney outlet of the incineration system to monitor the concentration of gaseous components such as carbon monoxide (CO), carbon dioxide (CO2), oxygen (O2), nitrogen oxides (NOx), and sulfur dioxide (SO2) in the flue gas in real time, outputting flue gas detection data to assess the completeness of combustion and pollutant emission levels. A weight sensor is installed at the bottom of the feeding channel or feeding chute to detect changes in waste weight per unit time during waste disposal or feeding, thereby generating data on the amount of waste fed.

[0022] Furthermore, the pressure and differential pressure sensor array is connected to the material layer thickness calculation unit, which embeds the material layer thickness calculation formula and calculates the material layer thickness by reading the real-time sensing data of the pressure and differential pressure sensor array, thereby generating the material layer thickness detection data.

[0023] The pressure and differential pressure sensor array is electrically connected to the material layer thickness calculation unit to realize real-time calculation and evaluation of the thickness of the waste material layer on the grate surface.

[0024] The pressure and differential pressure sensor array includes multiple high-sensitivity pressure sensors deployed above and below the grate to collect static pressure of the airflow above the grate surface, air supply pressure of the air chamber below the grate, and differential pressure data. The bed thickness calculation unit is an embedded edge computing module with a pre-set bed thickness calculation formula. Specifically, the bed thickness calculation unit periodically reads the raw sensing data from the pressure and differential pressure sensor array and performs thickness estimation calculations accordingly, outputting the thickness information of the waste accumulation layer in each area of ​​the grate, thereby generating spatially continuous bed thickness detection data. The bed thickness can be calculated based on the resistance encountered by the primary airflow through combustion; the bed thickness calculation formula is as follows:

[0025] ;in, Where A is the thickness of the material layer and A is the area of ​​the grate. For primary wind density, For the air chamber pressure, Furnace pressure, PAL is the primary air flow rate in the air chamber, PAT is the primary air temperature in the air chamber, and k is a constant.

[0026] The multi-mode operation status data is input into the intelligent decision-making model to predict the waste combustion status and generate combustion trend information.

[0027] Multi-source operational status data, including temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data, collected and preprocessed in real time by the multimodal sensing module, are used as input features and fed into the constructed intelligent decision-making model. This intelligent decision-making model is designed based on a deep learning structure, preferably employing a graph neural network (GNN), convolutional neural network (CNN), or long short-term memory network (LSTM) with spatiotemporal modeling capabilities. Specifically, considering the acquisition locations of different sensor data and their distribution characteristics within the incinerator's spatial structure, a graph structure model based on the sensor node position relationships is preferably constructed. Multiple rounds of node state propagation and updates are performed using the graph neural network to extract high-dimensional feature representations of each region, achieving dynamic learning and trend modeling of the waste combustion state. After supervised training with historical combustion data from multiple rounds, this model possesses the ability to intelligently perceive and predict the current combustion state. The intelligent decision-making model outputs combustion trend information corresponding to the current moment, including indicators such as whether combustion is becoming unstable, whether there are drastic temperature fluctuations, whether the bed thickness is abnormal, and whether the flue gas composition exceeds limits, providing quantitative and dynamic trend references for subsequent control optimization.

[0028] Furthermore, the multi-mode operating status data is input into the intelligent decision-making model to predict the waste combustion status and generate combustion trend information, including:

[0029] A graph structure is constructed based on the data acquisition locations corresponding to the multi-mode operating status data; based on the graph structure, a graph neural network is iteratively trained to generate the intelligent decision-making model; the intelligent decision-making model is called to analyze the multi-mode operating status data and output the combustion trend information.

[0030] First, based on the physical installation locations of various sensors and the corresponding data types collected in the multi-mode operation status data, a graph structure reflecting the spatial structure of the waste incinerator's operation is constructed. Each node in the graph structure corresponds to a sensor unit or an operation parameter collection point. The node characteristics are its corresponding operation status data (such as temperature, pressure difference, gas concentration, feed rate, etc.). The edge connections are defined based on the proximity of the sensor's physical location, functional correlation, or airflow path logic, forming a graph topology that reflects the spatiotemporal interaction characteristics within the incinerator. Then, based on the constructed graph structure model, a graph neural network is used as the core architecture to perform multi-round node state propagation and... The iterative training process of feature fusion involves using historical multi-mode operating state data and their corresponding combustion trend labels as training samples. The model parameters are continuously optimized through a node feature update mechanism, ultimately generating an intelligent decision-making model with trend prediction capabilities. During actual operation, the real-time collected current multi-mode operating state data is used as the input to the graph structure nodes. The trained intelligent decision-making model is then used for forward inference, outputting the prediction results of the current combustion state, including the future short-term trends in combustion temperature, bed stability, and pollutant emissions, collectively referred to as combustion trend information, which provides a dynamic adjustment basis for combustion control.

[0031] Furthermore, based on the graph structure, iterative training of the graph neural network is performed to generate the intelligent decision-making model, including:

[0032] A first historical state prediction sample set and a second historical state prediction sample set are collected. Based on the graph structure, a graph neural network model is constructed. The first historical state prediction sample set is used to perform node feature update training to generate a node feature update mechanism. Node feature update refers to each node obtaining information from its neighboring nodes and then updating its own state by combining the node's own information. The training parameters are learning parameters based on the update of its own state by its neighboring nodes. Based on the node feature update mechanism and the graph neural network model, after performing node feature update using the second historical state prediction sample set, the combustion trend prediction training is performed using the node update data to generate the intelligent decision-making model.

[0033] First, historical multi-mode operating state data for training is collected to construct two sample sets: a first historical state prediction sample set and a second historical state prediction sample set. The first historical state prediction sample set is mainly used for learning the node feature propagation mechanism, while the second historical state prediction sample set is mainly used for training and optimizing the combustion trend output layer. Next, based on the previously constructed graph structure, a graph neural network (GNN) model is established. Each node in the model represents a state acquisition point, and its initial features are the corresponding historical state data (such as temperature, bed thickness, and flue gas concentration at a certain moment). Using the first historical state prediction sample set, the node feature propagation process is trained, and through an information aggregation mechanism, each node can obtain state information from its neighboring nodes. The system updates the features of each node by combining its original state with its own features. This process is called node feature update, and the training objective is to optimize the state representation of each node so that it better reflects the joint dynamic features of its local area and its neighborhood. The core parameters of node feature update include the neighbor node weight allocation function, the information aggregation function, and the activation function. The update strategy is determined based on the adjacent edge weights, the distance between nodes, or the functional relevance. After the node feature update mechanism is constructed, a second historical state prediction sample set is used for trend prediction training. That is, after the node state is updated, the updated node features are used as input, and the trend label is supervised and trained through the output layer. The labels may include combustion trend indicators such as furnace temperature trend changes (increase, decrease, or stability), material layer fluctuations, and flue gas emission changes. Finally, through multiple rounds of iterative optimization, the training of the intelligent decision model is completed, and the model is used to predict the current combustion trend information in real time, providing a dynamic decision basis for the optimization of combustion control strategies.

[0034] The combustion trend information is compared with the preset combustion control target, and optimization instructions are generated based on the comparison results.

[0035] The system's preset combustion control objectives include multiple dimensions of control indicators, such as the stable range of furnace temperature, the reasonable range of material bed thickness, the safe concentration threshold of pollutants such as CO or NOx in flue gas, and the matching relationship between feed and combustion air, among other control benchmark data. The combustion trend information output by the intelligent decision-making model is used as a predictive reference for the current state, and each indicator is compared and analyzed against these combustion control objectives to determine whether the current predicted trend meets or deviates from the preset objectives. If the combustion trend information fully meets all control objectives, the existing control parameters remain unchanged. If any trend prediction result exceeds the preset tolerance range of the control objectives, corresponding optimization instructions are generated based on the direction and degree of the deviation, combined with the priority settings of the control objectives, to prompt or drive the system to adjust the parameters.

[0036] Furthermore, the combustion trend information is compared with a preset combustion control target, and optimization instructions are generated based on the comparison result, including:

[0037] Determine whether the combustion trend information meets the preset combustion control target; if not, generate the optimization instruction.

[0038] The system determines the combustion trend information output by the intelligent decision-making model at the current moment. The combustion trend information includes at least the direction of furnace temperature change, the trend of feed layer thickness fluctuation, the trend of key pollutant concentration change in flue gas, and the dynamic evolution characteristics of the feed and combustion adaptation relationship over a period of time. The system matches and judges the above trend information item by item according to the preset combustion control objectives. The combustion control objectives include, but are not limited to, indicators such as: the stability of furnace temperature within the target temperature range, the duration of feed layer thickness within the optimal combustion range, the probability of pollutant emissions not exceeding the set threshold, and the target lower limit of combustion reaction thermal efficiency. If the combustion trend information fully meets all preset combustion control objectives, the system maintains the current operating parameters unchanged. If any trend indicator deviates from the corresponding control objective (e.g., the predicted temperature will continue to decrease or CO emissions will exceed the threshold), the system triggers the target deviation judgment mechanism to determine the deviation type and correction direction, and generates optimization instructions accordingly. The optimization instructions are used to drive the subsequent parameter optimization process, specifying the type of combustion variables to be adjusted and the direction of adjustment strategy, providing a decision basis for the subsequent generation of the optimal combustion parameter set, and realizing the active adjustment and closed-loop optimization control of the incinerator's operating status.

[0039] Based on the optimization instructions, the combustion air volume, fuel supply volume, and grate motion parameters are optimized according to the combustion trend information to generate the optimal combustion parameter set.

[0040] After receiving the optimization command generated from the combustion trend deviation analysis results, the system first extracts the predicted dynamic data related to combustion air supply, waste feeding acceleration rate, and grate operating status from the current combustion trend information. Next, it calls historical combustion control data matching the trend information time window to construct a historical combustion parameter sample set. This sample set includes combustion air volume, fuel supply rate, grate operating cycle and beat rate under different combinations, and the corresponding combustion response effects. Subsequently, based on the deviation direction and adjustment requirements specified in the optimization command, the system uses a multi-objective optimization algorithm (such as a multi-objective algorithm based on non-dominated sorting) to optimize the combustion air supply, fuel feeding rate, grate operating cycle and beat rate under different combinations. The NSGA-II algorithm is used to evaluate fitness based on the control objectives (such as combustion temperature stability, uniformity of fuel bed distribution, and compliance rate of flue gas emissions). A fitness function group containing multiple sub-objective evaluators is constructed to perform multiple rounds of evolutionary optimization iterations on the historical combustion parameter sample set. By evaluating the fitness of each parameter combination and sorting it non-dominated, inferior solutions are eliminated and superior solutions are retained. The results of multiple rounds of optimization are continuously screened and integrated, and finally the parameter solution set at the non-dominated optimal frontier is extracted. The optimal combustion parameter set is selected from it, including the combination of the optimal combustion air supply, the optimal fuel supply rate, and the optimal grate motion parameters.

[0041] Furthermore, based on the optimization instructions, the combustion air quantity, fuel supply quantity, and grate motion parameters are optimized according to the combustion trend information to generate an optimal set of combustion parameters, including:

[0042] Collect a set of historical combustion parameters corresponding to the combustion trend information; construct an fitness evaluation mechanism based on the preset combustion control target, including several sub-target evaluators; use the several sub-target evaluators to perform iterative optimization of the historical combustion parameter set based on non-dominated sorting to generate the optimal combustion parameter set.

[0043] Based on the current combustion trend information, the system retrieves multimodal sensing data and control parameter records from the associated historical operating cycles, collecting a corresponding historical combustion parameter set. This historical combustion parameter set covers combustion air flow, fuel supply rate, grate operating cycle time, primary / secondary air distribution ratio, etc., under different combustion environments during historical periods, along with corresponding combustion response results (such as furnace temperature stability, material bed thickness uniformity, and pollutant concentration in flue gas). An fitness assessment mechanism is constructed based on preset combustion control targets, which typically include, but are not limited to, furnace temperature stability targets, thermal efficiency targets, emission compliance targets, and waste burnout rate targets. The fitness assessment mechanism includes several sub-target evaluators, each evaluating and scoring a specific sub-target. The system forms a family of multi-objective fitness functions. It uses several sub-objective evaluators to calculate the fitness of each set of parameters in the historical combustion parameter set. In each iteration, non-dominated sorting (such as the NSGA-II algorithm) is used to sort and filter the multi-objective fitness, determining the parameter combinations in the current solution space that are at the Pareto front—that is, solutions that perform well on multiple evaluation dimensions simultaneously. Dominated solutions that are degraded by multiple objectives are eliminated. Genetic crossover and mutation operations are performed on the remaining non-dominated solutions to expand the candidate parameter space, and sub-objective evaluation and non-dominated screening are performed again. This process iterates until a preset convergence condition or iteration round is reached. Finally, the parameter set with the best overall fitness is selected from the obtained global non-dominated front solution set as the optimal combustion parameter set under the current combustion state.

[0044] Furthermore, the fitness evaluation mechanism is used to perform iterative optimization of the historical combustion parameter set based on non-dominated ranking to generate the optimal combustion parameter set, including:

[0045] Step 1: Use the several sub-objective evaluators to evaluate the fitness of each solution in the historical combustion parameter set, generating various fitness sets; Step 2: Calculate the fitness difference between any two solutions using the same sub-objective evaluator based on the fitness sets, generating a fitness difference distribution; Step 3: Call the non-dominated stratification mechanism to perform dominance relationship stratification on each solution according to the fitness difference distribution, constructing a dominance relationship structure graph; perform a preset non-dominated level threshold screening on the dominance relationship structure graph to generate a screening population; after parameter expansion of the screening population, repeat steps 1 to 3, continuing iteration until a preset number of iterations is reached; globally fuse the dominance relationship structure graphs obtained from the iterations; extract the optimal non-dominated solution from the fused dominance relationship structure graph to generate the optimal combustion parameter set.

[0046] Based on multiple pre-set sub-objective evaluators (such as furnace temperature fluctuation range evaluator, burnout efficiency evaluator, NO... xEmissions assessors, thermal efficiency assessors, etc., evaluate the fitness of each candidate parameter group (i.e., a solution) in the historical combustion parameter set, obtaining multiple fitness scores, which constitute the fitness set of that solution. Each fitness score represents the performance of that parameter group under a specific evaluation objective. The fitness sets of all candidate solutions are compared pairwise. Under the same sub-objective evaluation dimension, the fitness difference between any two solutions is calculated, resulting in a fitness difference distribution matrix, which measures the degree of dominance and relative performance differences between solutions. Based on the above difference distribution, a non-dominated stratification mechanism is invoked. According to the Pareto optimality principle, the dominance relationship between candidate solutions is determined, and the candidate solutions are stratified according to the number of times they are dominated, constructing a dominance relationship structure diagram. The first layer is the set of optimal solutions not dominated by any other solution, and so on, constructing the second, third, and so on sets of inferior solutions. The dominance relationship structure graph is filtered based on a preset non-dominated level threshold (e.g., retaining the first two or three layers) to generate a screening population for the current round of optimization. Subsequently, parameter expansion operations are performed on the candidate solutions in the screening population, and crossover and mutation mechanisms in genetic algorithms (such as simulated binary crossover SBX, non-uniform mutation, etc.) are used to generate a new set of candidate solutions, which is then used as the input for the next round of iteration. Steps one to three are repeated. After each round of iteration, the system merges the current dominance relationship structure graph with the historical structure graphs from previous rounds, retains the global dominance relationship, and continuously optimizes the search space until the preset number of iterations or convergence criteria are met. Finally, the optimal solution of the non-dominated frontier is extracted from the merged global dominance relationship structure graph as the optimal combustion parameter set.

[0047] Furthermore, the non-dominated hierarchical mechanism includes complete domination of the plurality of sub-objective estimators and high-level domination of at least one sub-objective estimator, wherein high-level domination means that the fitness difference between any two solutions on at least one sub-objective estimator is higher than a preset threshold.

[0048] Complete dominance means that one solution is superior to another solution in all sub-objective evaluator dimensions; while advanced dominance means that in at least one sub-objective evaluator dimension, its fitness value has a significant advantage over the fitness value of another solution that exceeds a preset numerical threshold.

[0049] Specifically, if there are n sub-objective estimators, for any two candidate solutions A and B, let the fitness function of the i-th estimator be f. i Then: when , (Assuming all objectives are minimization problems), then solution A is said to completely dominate solution B; if there exists at least one evaluator , making ,(in If the threshold value is the preset threshold value for the j-th sub-objective evaluator, then solution A is said to have high dominance over solution B.

[0050] In each round of optimization, the system prioritizes retaining solution groups that simultaneously satisfy both complete domination and high-level domination. When constructing the domination relationship structure graph, it prioritizes assigning higher non-dominated levels. By integrating complete domination and high-level domination rules, the convergence speed of the solution space is improved, and the accuracy and reliability of finding performance-balanced solutions under a multi-objective evaluation system are also enhanced.

[0051] The combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator are controlled by the optimal combustion parameter set.

[0052] The optimal combustion parameter set includes combustion air supply parameters, fuel addition parameters, and grate operating parameters, which respectively control the combustion fan, fuel supply device, and grate drive mechanism. The combustion fan adjusts the air volume and pressure according to the combustion air supply parameters to ensure sufficient and uniform oxygen supply at different combustion stages; the fuel supply device controls the waste feeding rate and interval according to the fuel addition parameters to achieve fuel supply that matches the actual calorific value requirement; the grate drive mechanism adjusts the grate movement frequency, propulsion speed, and propulsion mode according to the grate operating parameters to achieve uniform spreading and complete combustion of waste in the incinerator.

[0053] The system achieves stable combustion conditions, reasonable furnace temperature control, and controlled emission concentration through precise linkage control of the three key actuators mentioned above. This enables intelligent and dynamic regulation of the waste combustion process, effectively improving incineration efficiency and reducing environmental impact.

[0054] In summary, the embodiments of this application have at least the following technical effects:

[0055] First, a multi-modal sensing module collects real-time multi-modal operating status data of the waste incinerator, including at least temperature detection data, bed thickness detection data, flue gas detection data, and feed rate detection data. Next, the multi-modal operating status data is input into an intelligent decision-making model to predict the waste combustion state and generate combustion trend information. Further, the combustion trend information is compared with preset combustion control targets, and optimization instructions are generated based on the comparison results. Then, based on the optimization instructions and the combustion trend information, the combustion air volume, fuel supply volume, and grate motion parameters are optimized to generate the optimal combustion parameter set. Finally, the optimal combustion parameter set is used to control the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator. This solves the technical problems of inaccurate incinerator control and lag in adjustment response in existing technologies, leading to low waste combustion efficiency, and achieves the technical effect of improving waste combustion efficiency.

[0056] Example 2 is based on the same inventive concept as the multimodal sensing-based automatic waste combustion control method in the previous examples, such as... Figure 2As shown, this application provides an automatic waste combustion control system based on multimodal sensing, wherein the system includes:

[0057] Data acquisition component 11: Real-time acquisition of multi-modal operating status data of the waste incinerator through a multi-modal sensing module, including at least temperature detection data, material layer thickness detection data, flue gas detection data, and feed rate detection data; Prediction component 12: Inputting the multi-modal operating status data into an intelligent decision-making model to predict the waste combustion status and generate combustion trend information; Comparison component 13: Comparing the combustion trend information with a preset combustion control target and generating optimization instructions based on the comparison results; Optimization component 14: Based on the optimization instructions and the combustion trend information, optimizing the combustion air volume, fuel supply volume, and grate motion parameters to generate an optimal combustion parameter set; Control component 15: Controlling the combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator using the optimal combustion parameter set.

[0058] Furthermore, the data acquisition component 11 is used to perform the following methods:

[0059] The multimodal sensing module includes at least a temperature sensor array, a pressure and differential pressure sensor array, a gas composition detector, and a weight sensor. The temperature sensor array is distributed across various locations within the furnace of the waste incinerator to collect real-time temperature distribution data in different areas of the furnace and generate temperature detection data. The pressure and differential pressure sensor array is distributed on the combustion grate surface to collect the pressure difference above and below the grate surface and the airflow on the grate surface, determining the thickness of the material layer on the grate surface and generating material layer thickness detection data. The gas composition detector is distributed at the flue gas outlet to detect the gas composition content in the flue gas and generate flue gas detection data. The weight sensor is distributed at the feed inlet to detect the amount of waste fed and generate feed amount detection data.

[0060] Furthermore, the data acquisition component 11 is used to perform the following methods:

[0061] The pressure and differential pressure sensor array is connected to the material layer thickness calculation unit. The material layer thickness calculation unit embeds the material layer thickness calculation formula and calculates the material layer thickness by reading the real-time sensing data of the pressure and differential pressure sensor array, thereby generating the material layer thickness detection data.

[0062] Furthermore, the prediction component 12 is used to perform the following method:

[0063] A graph structure is constructed based on the data acquisition locations corresponding to the multi-mode operating status data; based on the graph structure, a graph neural network is iteratively trained to generate the intelligent decision-making model; the intelligent decision-making model is called to analyze the multi-mode operating status data and output the combustion trend information.

[0064] Furthermore, the prediction component 12 is used to perform the following method:

[0065] A first historical state prediction sample set and a second historical state prediction sample set are collected. Based on the graph structure, a graph neural network model is constructed. The first historical state prediction sample set is used to perform node feature update training to generate a node feature update mechanism. Node feature update refers to each node obtaining information from its neighboring nodes and then updating its own state by combining the node's own information. The training parameters are learning parameters based on the update of its own state by its neighboring nodes. Based on the node feature update mechanism and the graph neural network model, after performing node feature update using the second historical state prediction sample set, the combustion trend prediction training is performed using the node update data to generate the intelligent decision-making model.

[0066] Furthermore, the optimization component 14 is used to perform the following method:

[0067] Collect a set of historical combustion parameters corresponding to the combustion trend information; construct an fitness evaluation mechanism based on the preset combustion control target, including several sub-target evaluators; use the several sub-target evaluators to perform iterative optimization of the historical combustion parameter set based on non-dominated sorting to generate the optimal combustion parameter set.

[0068] Furthermore, the optimization component 14 is used to perform the following method:

[0069] Step 1: Use the several sub-objective evaluators to evaluate the fitness of each solution in the historical combustion parameter set, generating various fitness sets; Step 2: Calculate the fitness difference between any two solutions using the same sub-objective evaluator based on the fitness sets, generating a fitness difference distribution; Step 3: Call the non-dominated stratification mechanism to perform dominance relationship stratification on each solution according to the fitness difference distribution, constructing a dominance relationship structure graph; perform a preset non-dominated level threshold screening on the dominance relationship structure graph to generate a screening population; after parameter expansion of the screening population, repeat steps 1 to 3, continuing iteration until a preset number of iterations is reached; globally fuse the dominance relationship structure graphs obtained from the iterations; extract the optimal non-dominated solution from the fused dominance relationship structure graph to generate the optimal combustion parameter set.

[0070] Furthermore, the optimization component 14 is used to perform the following method:

[0071] The non-dominated hierarchical mechanism includes complete domination of the plurality of sub-objective estimators and high-level domination of at least one sub-objective estimator, wherein high-level domination means that the fitness difference between any two solutions on at least one sub-objective estimator is higher than a preset threshold.

[0072] Furthermore, the comparison component 13 is used to perform the following method:

[0073] Determine whether the combustion trend information meets the preset combustion control target; if not, generate the optimization instruction.

[0074] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0075] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0076] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A waste automatic combustion control method based on multimodal sensing, characterized in that, The method includes: The multimodal sensing module collects multimodal operating status data of the waste incinerator in real time, including at least temperature detection data, material bed thickness detection data, flue gas detection data, and feed rate detection data. The multi-mode operating status data is input into the intelligent decision-making model to predict the waste combustion status and generate combustion trend information, including: A graph structure is constructed based on the data acquisition locations corresponding to the multi-mode operation status data; Based on the graph structure, iterative training of the graph neural network is performed to generate the intelligent decision-making model; The intelligent decision-making model is invoked to analyze the multi-mode operating status data and output the combustion trend information; The combustion trend information is compared with a preset combustion control target, and optimization instructions are generated based on the comparison results. Based on the optimization instructions, the combustion air volume, fuel supply volume, and grate motion parameters are optimized according to the combustion trend information to generate an optimal set of combustion parameters, including: Collect a set of historical combustion parameters corresponding to the combustion trend information; An fitness evaluation mechanism is constructed based on the preset combustion control target, including several sub-target evaluators; The historical combustion parameter set is iteratively optimized using the aforementioned sub-objective evaluators based on non-dominated sorting to generate the optimal combustion parameter set; The combustion fan, fuel supply device, and grate drive mechanism in the waste incinerator are controlled by the optimal combustion parameter set.

2. The automatic waste combustion control method based on multimodal sensing as described in claim 1, characterized in that, The multimodal sensing module includes at least a temperature sensor array, a pressure and differential pressure sensor array, a gas composition detector, and a weight sensor; The temperature sensor array is distributed at various locations in the furnace of the waste incinerator to collect the temperature distribution in different areas of the furnace in real time and generate temperature detection data; the pressure and differential pressure sensor array is distributed on the combustion grate surface to collect the pressure difference above and below the grate surface and the air volume on the grate surface, determine the thickness of the material layer on the grate surface, and generate material layer thickness detection data; the gas composition detector is distributed at the flue gas emission port to detect the content of gas components in the flue gas and generate flue gas detection data; and the weight sensor is distributed at the feed inlet to detect the amount of waste fed and generate feed amount detection data.

3. The automatic waste combustion control method based on multimodal sensing as described in claim 2, characterized in that, The pressure and differential pressure sensor array is connected to the material layer thickness calculation unit. The material layer thickness calculation unit embeds the material layer thickness calculation formula and calculates the material layer thickness by reading the real-time sensing data of the pressure and differential pressure sensor array, thereby generating the material layer thickness detection data.

4. The automatic waste combustion control method based on multimodal sensing as described in claim 1, characterized in that, Based on the graph structure, iterative training of the graph neural network is performed to generate the intelligent decision-making model, including: Collect the first historical state prediction sample set and the second historical state prediction sample set; Based on the graph structure, a graph neural network model is constructed. The first historical state prediction sample set is used to perform node feature update training to generate a node feature update mechanism. Node feature update refers to each node obtaining information from its neighboring nodes and then updating its own state by combining the node's own information. The training parameters are learning parameters based on updating the state of its neighboring nodes. Based on the node feature update mechanism and the graph neural network model, the intelligent decision-making model is generated by using the second historical state prediction sample set to perform combustion trend prediction training after node feature update.

5. The automatic waste combustion control method based on multimodal sensing as described in claim 1, characterized in that, The fitness evaluation mechanism is used to perform iterative optimization of the historical combustion parameter set based on non-dominated ranking to generate the optimal combustion parameter set, including: Step 1: Use the aforementioned sub-target evaluators to evaluate the fitness of each solution in the historical combustion parameter set, and generate various fitness sets; Step 2: Based on the fitness sets, calculate the fitness difference for any two solutions using the same sub-objective evaluator to generate a fitness difference distribution; Step 3: Invoke the non-dominated hierarchical mechanism to hierarchically classify the dominance relationships of each solution according to the fitness difference distribution, and construct a dominance relationship structure graph; The dominance relationship structure graph is filtered by a preset non-dominance level threshold to generate a screening population. After the parameters of the screening population are expanded, steps one to three are repeated to continue iterating until the preset number of iterations is reached. The dominance relationship structure graphs obtained by iteration are globally fused, and the optimal non-dominance solution is extracted from the fused dominance relationship structure graph to generate the optimal combustion parameter set.

6. The automatic waste combustion control method based on multimodal sensing as described in claim 5, characterized in that, The non-dominated hierarchical mechanism includes complete domination of the plurality of sub-objective estimators and high-level domination of at least one sub-objective estimator, wherein high-level domination means that the fitness difference between any two solutions on at least one sub-objective estimator is higher than a preset threshold.

7. The automatic waste combustion control method based on multimodal sensing as described in claim 1, characterized in that, The combustion trend information is compared with a preset combustion control target, and optimization instructions are generated based on the comparison result, including: Determine whether the combustion trend information meets the preset combustion control target; If not, generate the optimization instructions.

8. A waste automatic combustion control system based on multimodal sensing, characterized in that, The system is used to implement the automatic waste combustion control method based on multimodal sensing as described in any one of claims 1-7, the system comprising: Data acquisition component: Real-time acquisition of multi-modal operating status data of waste incinerator through multi-modal sensing module, including at least temperature detection data, material bed thickness detection data, flue gas detection data and feed rate detection data; Prediction component: Inputs the multi-mode operation status data into the intelligent decision-making model to predict the waste combustion status and generate combustion trend information; Comparison component: compares the combustion trend information with the preset combustion control target, and generates optimization instructions based on the comparison results; Optimization component: Based on the optimization instructions, optimize the combustion air volume, fuel supply volume and grate motion parameters according to the combustion trend information to generate the optimal combustion parameter set; Control components: control the combustion fan, fuel supply device and grate drive mechanism in the waste incinerator with the optimal combustion parameter set.

Images