Garbage incineration control method, device and system

Abstract

The application relates to a waste incineration control method, device and system, and relates to the technical field of waste incineration treatment. The method comprises the following steps: generating a feasible domain boundary of an operation parameter according to operation procedure text data; inputting the feasible domain boundary, a current operation parameter and current fire watching video stream data into a preset model, so as to determine a current waste incineration state according to the current fire watching video stream data by using the preset model; then determining a first control variable in the feasible domain boundary according to the current waste incineration state and the current operation parameter by using the preset model; inputting the first control variable, the current operation parameter and the current waste incineration state into an expert experience rule base for logical verification and control variable correction, so as to obtain a second control variable; controlling a waste incineration control system to generate a first target operation variable according to the second control variable; and adjusting a burning execution mechanism according to the first target operation variable.

Description

Technical Field

[0001] This invention relates to the field of waste incineration technology, and in particular to a waste incineration control method, apparatus and system. Background Technology

[0002] Currently, automatic combustion control (ACC) technology in waste-to-energy plants mainly falls into two categories. One is the conventional PID control with fuzzy logic, which relies on manually set rules and struggles to cope with the large time lag caused by drastic fluctuations in the calorific value of waste, and cannot self-learn and optimize. The other is a simple data model with manually set rules. While it can learn from historical data, the model is a black box, lacks interpretability, and is prone to generating "illusionary" commands that violate design specifications, operating procedures, and operational guidelines. For example, it may prioritize combustion efficiency over furnace negative pressure safety, posing significant safety hazards in engineering applications. Summary of the Invention

[0003] This invention provides a waste incineration control method, apparatus, and system to at least solve the problem in related technologies where control strategies struggle to simultaneously achieve compliance, safety, and combustion efficiency. The technical solution of this invention is as follows:

[0004] According to a first aspect of the present invention, a waste incineration control method is provided, applied to a waste incineration control system. The waste incineration control system includes an incineration actuator and is a distributed control system. The method includes: acquiring current operating parameters of the waste incineration control system, current furnace observation video stream data, and operating procedure text data during the waste incineration process; wherein the operating procedure text data includes parameter constraint relationships of the operating parameters in the waste incineration control system; generating a feasible domain boundary of the operating parameters based on the operating procedure text data; and inputting the feasible domain boundary, the current operating parameters, and the current furnace observation video stream data into a preset model, so as to use the preset model to first determine the feasible domain boundary based on the current furnace observation video stream data. Frequency flow data is used to determine the current waste incineration status. Then, a pre-set model is used to determine the first control variable within the feasible region boundary based on the current waste incineration status and operating parameters. The pre-set model is trained using the feasible region boundary as a penalty term in the reward or loss function, and its objective is to maximize the positively weighted result of multiple optimization goals, outputting the corresponding control variable. The first control variable, current operating parameters, and current waste incineration status are input into an expert experience rule base for logical verification and control variable calibration to obtain the second control variable. According to the second control variable, the waste incineration control system generates the first target operation variable. According to the first target operation variable, the incineration actuator is adjusted.

[0005] The current operating parameters include the current process variables, current control variables used for execution, and current operation variables output by the waste incineration control system during the waste incineration process; the current fire observation video stream data includes visual data characterizing the current actual combustion state inside the furnace.

[0006] The current waste incineration status and current operating parameters are both state sequences of multiple waste incineration statuses and parameter sequences of multiple operating parameters arranged in chronological order.

[0007] The above solution achieves the technical effect of improving combustion efficiency while ensuring operational compliance and safety by constructing an integrated architecture of a procedure constraint layer, a data-driven model, and an expert experience rule base. It utilizes the feasible domain boundary constraint model output and verifies it through expert experience.

[0008] As one implementation method, the preset model includes a visual sub-model and a reinforcement learning sub-model connected in sequence. The preset model first determines the current waste incineration status based on the current fire observation video stream data, including: using the visual sub-model to extract quantized visual features from the current fire observation video stream data; and identifying the current waste incineration status based on the extracted quantized visual features; wherein, the quantized visual features include at least one or more of the following: flame shape, flame brightness distribution, flame color distribution, and flame temperature distribution.

[0009] The above solution, by introducing a visual sub-model to process video stream data, can more intuitively and accurately identify the combustion state inside the furnace, thus overcoming the limitations of relying solely on numerical parameters to judge the operating conditions.

[0010] As one implementation method, a preset model is used to determine the first control variable within the feasible region boundary based on the current waste incineration state and current operating parameters. This includes: inputting the current waste incineration state and current operating parameters as state inputs into a reinforcement learning sub-model, so that the reinforcement learning sub-model makes decisions based on a preset reward function under the constraints of the feasible region boundary, and outputs the first control variable that maximizes the positive weighted result of multiple optimization objectives; the reward function is constructed by a weighted sum of multiple immediate reward components, including environmental indicators, economic indicators, combustion stability indicators, and safety indicators representing the excess of preset parameters; where each immediate reward component corresponds to one of the multiple optimization objectives.

[0011] The above scheme uses a reinforcement learning model to perform multi-objective optimization within the constraint boundary, which can automatically balance environmental protection, economic and stability indicators and achieve the optimal decision of control strategy.

[0012] As an implementation method, the rules stored in the expert experience rule base include: operating condition identification and state judgment rules; feedforward control and compensation rules under specific operating conditions; safety interlocking rules for waste incineration control systems; multi-objective trade-off and boundary protection rules, used to define the priority relationship between environmental protection indicators, economic indicators, combustion stability indicators and safety indicators under different operating conditions; and rules for suppressing sudden changes in control variables and verifying logical rationality.

[0013] Flame morphology includes effective flame area ratio, flame centroid offset distance, high-temperature zone circularity, and effective flame area per unit time.

[0014] The above solution, by constructing a multi-dimensional expert rule base, can cover normal operating conditions, special operating conditions, and abnormal operating conditions, providing a comprehensive basis for the logical verification of control commands.

[0015] As one implementation method, the first control variable, current operating parameters, and current waste incineration status are input into an expert experience rule base for logical verification and control variable calibration to obtain the second control variable. This includes: determining the target operating condition state corresponding to the first control variable, current operating parameters, and current waste incineration status based on operating condition identification and status judgment rules; performing a matching search in the expert experience rule base according to the target operating condition state to obtain target rules; multiple target rules are triggered based on the target operating condition state; and performing logical verification and control variable calibration on the current operating parameters and current waste incineration status according to the target rules to obtain the second control variable.

[0016] The above solution achieves precise application of expert experience by matching operating conditions with corresponding rules, thereby improving the efficiency and accuracy of verification and proofreading.

[0017] As one implementation method, the method further includes: if no target rule is found in the expert experience rule base according to the target operating condition, then the waste incineration control system is controlled to generate a second target operation variable according to the first control variable; and the incineration actuator is adjusted according to the second target operation variable.

[0018] In cases where expert rules are not covered, the above solution defaults to executing the control variables output by the model, ensuring continuous system operation and avoiding control interruptions.

[0019] As one implementation method, according to the target rules, logical verification and control variable calibration are performed on the current operating parameters and the current waste incineration status to obtain a second control variable, including: if multiple target rules include triggered safety interlock rules, then a warning message and an operation stop command are sent; the operation stop command is used to stop the operation of the incineration actuator; if multiple target rules do not include untriggered safety interlock rules but include triggered feedforward control and compensation rules, then based on the feedforward control and compensation rules, a preset empirical control quantity associated with the current operating parameters and the current waste incineration status is determined; the first control variable is calibrated to the preset empirical control quantity to obtain a third control variable; if multiple target rules do not include untriggered safety interlock rules or untriggered feedforward control rules... The system includes compensation rules, and includes triggered multi-objective trade-offs and boundary protection rules. Based on these rules, it determines the preset weights of one or more optimization objectives corresponding to the current operating parameters and the current waste incineration state. According to the preset weights, it determines the target weights of multiple optimization objectives. It modifies the weights of the multiple optimizations in the preset model to the target weights and calls the modified preset model to determine the fourth control variable corresponding to the current waste incineration state and the current operating parameters. If the multiple objective rules include triggered mutation suppression and logical rationality verification rules, it corrects the first, second, third, or fourth control variables according to the triggered mutation suppression and logical rationality verification rules to obtain the second control variable.

[0020] Determine whether the first control variable or the current operating parameter triggers a predefined safety interlock rule in the expert experience rule base. If so, execute the interlock action directly and terminate the subsequent optimization process.

[0021] Determine whether the current condition is special or the model confidence is below the threshold. If so, trigger the feedforward control and compensation rules in the expert experience rule base and output the experience control command as the second control variable.

[0022] For example, if the calorific value of the corresponding waste drops sharply, or pollutant emissions approach the upper limit, boundary protection rules (such as "emission value > 90% of the limit") are triggered. The multi-objective trade-off rule responds immediately: environmental protection is the primary concern, and safety risks are increasing. Therefore, the weights of objective B (environmental protection) and objective C (safety) are significantly increased, while the weight of objective A (economy) is decreased. The system instructs the vehicle to decelerate smoothly, prioritizing compliance and safety (corresponding to reducing the load on the incinerator, increasing auxiliary fuel, or adjusting airflow to ensure complete combustion and control emissions).

[0023] If the corresponding incinerator is operating stably and emissions are far below the limits, the multi-objective trade-off rule determines that "the safety risk is low and the environmental pressure is small." Therefore, the weight of objective A (economy) is dynamically increased, and the system instructs the vehicle to accelerate appropriately to save time (corresponding to the incinerator increasing its processing capacity and producing more steam).

[0024] The above solution achieves multi-level safety safeguards and efficiency improvements through a tiered arbitration mechanism that prioritizes safety, handles special working conditions, and optimizes steady-state efficiency.

[0025] As one implementation method, based on operating condition identification and state judgment rules, the target operating condition state corresponding to the first control variable, the current operating parameters, and the current waste incineration state is determined. This includes: determining the preset parameter trend associated with the current waste incineration state and the current operating parameters; the preset parameter trend includes the preset parameter change trend, the first parameter range, the second parameter range, and the third parameter range; if the first control variable is within the first parameter range, the target operating condition state is determined to be a safe abnormal operating condition; if the first control variable is not within the first parameter range, and the change trend of the first control variable does not match the preset parameter change trend, the target operating condition state is determined to be a special operating condition; if the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the first parameter range, the target operating condition state is determined to be a special operating condition; if the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the first parameter range, the target operating condition state is determined to be a special operating condition. The target operating condition is determined to be a boundary abnormal operating condition based on two parameter ranges. If the trend of the first control variable matches the trend of the preset parameter and the first control variable is within the range of the third parameter, the target operating condition is determined to be a normal operating condition. Based on the target operating condition, a matching search is performed in the expert experience rule base to obtain multiple target rules, including: if the target operating condition is determined to be a safety abnormal operating condition, then triggering a safety interlocking rule is the target rule; if the target operating condition is determined to be a special operating condition, then triggering feedforward control and compensation rules and mutation suppression and logical rationality verification rules are the target rules; if the target operating condition is determined to be a boundary abnormal operating condition, then triggering multi-objective trade-off and boundary protection rules and mutation suppression and logical rationality verification rules are the target rules.

[0026] The above solution achieves accurate identification of operating conditions by finely dividing parameter ranges and trends, thereby triggering corresponding rules and improving the system's response accuracy.

[0027] According to a second aspect of the present invention, a waste incineration control device is provided, disposed in a waste incineration control system. The waste incineration control system includes an incineration actuator, and the waste incineration control system is a distributed control system. The device includes: an acquisition unit, configured to acquire current operating parameters of the waste incineration control system, current furnace observation video stream data, and operating procedure text data during the waste incineration process; wherein the operating procedure text data includes parameter constraint relationships of the operating parameters in the waste incineration control system; a generation unit, configured to generate feasible domain boundaries of the operating parameters based on the operating procedure text data; and a control variable prediction unit, configured to input the feasible domain boundaries, current operating parameters, and current furnace observation video stream data into a preset model, so as to use the preset model to first predict the current... The system analyzes video stream data to determine the current waste incineration status. Then, using a pre-defined model, based on the current incineration status and operating parameters, it determines the first control variable within the feasible region boundary. This model is trained using the feasible region boundary as a penalty term in the reward or loss function, and its objective is to maximize the positively weighted result of multiple optimization goals, outputting the corresponding control variable. A control variable verification unit inputs the first control variable, current operating parameters, and current waste incineration status into an expert rule base for logical verification and control variable verification, resulting in a second control variable. A control execution unit, based on the second control variable, controls the waste incineration control system to generate the first target operation variable and adjusts the incineration execution mechanism according to the first target operation variable.

[0028] According to a third aspect of the present invention, a waste incineration control system is provided, wherein the waste incineration control system stores instructions that, when executed by a controller, enable the controller to perform a waste incineration control method as described in the first aspect and any possible technical solution thereof.

[0029] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of a control device, the control device is able to perform a waste incineration control method as described in the first aspect and any possible implementation thereof.

[0030] According to a fifth aspect of the present disclosure, a computer program product is provided, the computer program product including computer instructions, which, when executed on a control device, cause the control device to perform the waste incineration control method of the first aspect and any possible implementation thereof.

[0031] The technical solutions provided by the embodiments of this invention offer at least the following beneficial effects: The technical solutions provided in this application, by transforming the operational procedure text data into feasible domain boundaries as penalty terms or reward function constraints for model training, achieve "procedure internalization," ensuring that all control commands are within compliance limits and solving the problems of insufficient compliance and security in existing intelligent control algorithms. Simultaneously, by introducing visual sub-models and reinforcement learning sub-models, combined with multi-objective optimization reward functions, accurate identification of combustion states and adaptive optimization of control strategies are achieved. Furthermore, by constructing an expert experience rule base containing safety interlocking, feedforward compensation, and multi-objective trade-off rules, hierarchical arbitration and verification of model output are performed, achieving "expert fallback," enabling timely intervention when the model fails or data is abnormal, further improving the robustness and security of the system. This solution effectively integrates procedural knowledge, data-driven approaches, and expert experience, achieving safe, stable, and economical operation throughout the entire waste incineration process.

[0032] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0033] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0034] Figure 1 This is a schematic diagram illustrating a waste incineration control system according to an exemplary embodiment;

[0035] Figure 2 This is a flowchart illustrating a waste incineration control method according to an exemplary embodiment;

[0036] Figure 3 This is a block diagram illustrating a waste incineration control device according to an exemplary embodiment;

[0037] Figure 4 This is a schematic diagram of a control device according to an exemplary embodiment. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0039] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0040] Before providing a detailed description of the waste incineration control method provided in the embodiments of this application, a brief introduction to the implementation environment involved in the embodiments of this application is given.

[0041] Figure 1 This is a schematic diagram of a waste incineration control system disclosed herein. Figure 1 As shown, the waste incineration control system includes a controller 11 and at least one incineration actuator 12. The waste incineration control system can be a distributed system. The controller 11 and each incineration actuator 12 are connected via a wired network and / or a wireless network.

[0042] This system is typically a Distributed Control System (DCS), responsible for the coordinated control of the entire plant. The incineration actuators are the execution ends of the waste incineration control system, and include, but are not limited to: grate drive motors (controlling waste propulsion speed), primary / secondary / tertiary air fan frequency converters (controlling airflow in each chamber and total airflow), burner valves (controlling auxiliary fuel flow), and SNCR / SCR ammonia injection regulating valves (controlling denitrification reducing agent flow), etc.

[0043] The controller executes the process steps of determining control variables and generating operation variables in the waste incineration control method according to the preset control cycle, and sends the generated operation variables to the waste incineration actuator, so that the waste incineration actuator performs the corresponding operation according to the operation variables.

[0044] The preset control cycle can be set according to process requirements, for example, a complete cycle is executed every 10 seconds or 30 seconds.

[0045] The waste incineration control method provided in this application embodiment can be applied to the aforementioned... Figure 1 The waste incineration control system in the implementation architecture shown is illustrated below. For ease of understanding, the waste incineration control method provided in this application will be described in detail below with reference to the accompanying drawings.

[0046] Figure 2This is a flowchart illustrating a waste incineration control method according to an exemplary embodiment, such as... Figure 2 As shown, the waste incineration control method includes the following steps.

[0047] S21, acquire the current operating parameters of the waste incineration control system, the current fire observation video stream data of the furnace, and the operating procedure text data during the waste incineration process.

[0048] The aforementioned current operating parameters refer to various physical quantities that reflect the operating status of the incinerator, collected in real time through the distributed control system, such as furnace temperature, furnace negative pressure, main steam flow, feeding speed, primary air volume, and secondary air volume.

[0049] In some implementations, the current fire monitoring video stream data can be real-time video images of the combustion status inside the furnace captured by an Industrial Television (ITV) system. Operating procedure text data can refer to pre-stored normative documents used to guide the waste incineration process, such as the "Waste Incineration Boiler Operating Procedures," "Safety Operation Manual," and "Emergency Response Plan." This text data contains numerous constraints on operating parameters, such as "the furnace temperature should be maintained above 850℃ and not exceeding 1100℃" and "the furnace negative pressure should be maintained between -50Pa and -100Pa." It should be understood that operating procedure text data is not limited to the documents listed above; it can also include operating manuals provided by equipment manufacturers, industry standard documents, etc., as long as they contain constraints on operating parameters.

[0050] S22, Generate the feasible domain boundary of the operating parameters based on the operating procedure text data.

[0051] This implementation step is a key step in achieving "internalization of procedures".

[0052] Since operating procedure text data is typically unstructured natural language text, computers cannot directly understand the constraint logic within it. Therefore, this embodiment employs Natural Language Processing (NLP) technology to structure the operating procedure text data. For example, Named Entity Recognition (NER) technology is used to extract key entities such as "furnace temperature," "850℃," and "1100℃" from the text, and relation extraction technology is used to identify the logical relationships between entities (such as "greater than," "less than," and "between"). Through semantic parsing, the natural language description "the furnace temperature should be maintained above 850℃ and not exceed 1100℃" is transformed into a mathematical constraint: 850℃ ≤ T_furnace ≤ 1100℃. Similarly, constraints are extracted for all relevant operating parameters, ultimately constructing a multi-dimensional feasible domain boundary. This feasible domain boundary can be represented as a set of inequality constraints or a closed hypergeometric space. In this way, the paper-based operating procedure is transformed into hard constraints that the algorithm can recognize and compute, setting a safety baseline for subsequent model training and inference.

[0053] S23, input the feasible domain boundary, current operating parameters and current fire monitoring video stream data into the preset model, so that the preset model first determines the current waste incineration state based on the current fire monitoring video stream data; then, the preset model determines the first control variable in the feasible domain boundary based on the current waste incineration state and current operating parameters.

[0054] The preset model is trained using the feasible region boundary as the penalty term of the reward function or loss function, and the preset model aims to maximize the positive weighted result of multiple optimization objectives, outputting the corresponding control variables.

[0055] The aforementioned preset model is a deep neural network model trained on a large amount of historical data, such as a reinforcement learning model or a hybrid model that combines visual processing.

[0056] To prevent the model from outputting "illusionary" instructions that violate safety regulations in pursuit of a single goal such as combustion efficiency, this implementation introduces the feasible domain boundary generated in step S22 into the training process of the preset model.

[0057] For example, in the design of the reward function for reinforcement learning, in addition to positive reward components such as combustion efficiency and environmental indicators, a penalty term is also introduced: when the control variables output by the model cause the predicted operating parameters to exceed the feasible region boundary, a large negative reward (penalty) is given. This mechanism forces the model to strictly abide by the constraints of the operating procedure while learning the patterns of historical data, thereby achieving "procedure internalization".

[0058] During the inference phase, the pre-defined model first extracts features from the current fire-watching video stream data, identifying characteristics such as flame shape, brightness, and color to determine the current combustion state (e.g., complete combustion, partial flameout, risk of coking, etc.). Subsequently, the pre-defined model, combined with the current operating parameters, seeks a control strategy that optimizes the weighted result of multiple optimization objectives (e.g., environmental indicators, economic indicators, combustion stability indicators, etc.) while satisfying the feasible domain boundary constraints, and outputs the first control variable.

[0059] The first control variable mentioned above can be a specific control command value such as the feeding speed, grate movement frequency, or damper opening.

[0060] S24. Input the first control variable, the current operating parameters, and the current waste incineration status into the expert experience rule base for logical verification and control variable calibration to obtain the second control variable.

[0061] Furthermore, although the preset model has been trained under procedural constraints, it may still encounter extreme operating conditions or data noise interference that it has not seen before in actual operation, leading to logical defects or potential risks in the output results. Therefore, this embodiment introduces an expert experience rule base as a "safety fallback" layer.

[0062] This expert experience rule base stores tacit knowledge summarized by senior operators and experts, such as logical rules like "when the oxygen level suddenly increases and the steam flow rate decreases, it is determined that the waste is cut off and the feeding should be accelerated immediately" and "when the furnace temperature drops sharply, the auxiliary burner should be started first".

[0063] After the first control variable is output, the system inputs it along with the current operating parameters into the expert rule base for matching and verification. If the first control variable conforms to the logic of the expert rules, the verification passes; if the first control variable violates certain critical safety rules or logical rationality rules (e.g., continuing to increase the feed rate when the furnace temperature is already too high), the expert rule base intervenes to correct or overwrite the first control variable, outputting a corrected second control variable. This step ensures that the system, when facing complex and changing actual operating conditions, not only conforms to procedures but also to human expert experience logic, improving the system's robustness and safety.

[0064] S25, according to the second control variable, control the waste incineration control system to generate the first target operation variable.

[0065] The second control variable is the final control command calculated by the model and verified by experts. The distributed control system (DCS) generates the specific first target operating variable based on this command, such as an analog signal (4-20mA current signal) or a switching signal.

[0066] S26, adjust the incineration actuator according to the first target operation variable.

[0067] The incineration actuators include field equipment such as feeders, grate drive motors, primary air fans, secondary air fans, induced draft fans, and dampers. The primary target control variable is transmitted to the corresponding actuator via fieldbus or hardwiring, driving the actuator to operate and thus achieving real-time regulation of the waste incineration process. For example, adjusting the feeder speed changes the waste throughput, and adjusting the damper opening changes the air distribution, ultimately achieving automatic, stable, and economical operation of the waste incineration process.

[0068] Through the above embodiments, a three-in-one control architecture of procedural constraints, data-driven operation, and expert verification was constructed. First, by transforming the operating procedures into feasible domain boundaries and using them as penalty terms for model training, the procedures were internalized, eliminating the risk of non-compliance at the source. Second, the multi-objective optimization capability of the pre-set model was utilized to achieve a balance between combustion efficiency and environmental indicators. Finally, logical verification through an expert experience rule base compensated for the shortcomings of the data model under extreme operating conditions, providing a safety net of expert experience. This layered and progressive control logic effectively solves the technical challenge in existing automatic combustion control technologies that makes it difficult to simultaneously achieve compliance, safety, and economy.

[0069] Based on the above embodiments, the specific internal architecture and working principle of the preset model will be described in detail.

[0070] The preset model consists of a visual sub-model and a reinforcement learning sub-model connected in sequence.

[0071] The visual sub-model is mainly used to process unstructured video stream data to compensate for the lag and limitations of traditional DCS data in sensing the combustion status.

[0072] Therefore, in step S23 above, a preset model is used to first determine the current waste incineration status based on the current fire observation video stream data. Specifically, this includes: extracting quantified visual features from the current fire observation video stream data using a visual sub-model; and identifying the current waste incineration status based on the extracted quantified visual features. The quantified visual features include at least one or more of the following: flame shape, flame brightness distribution, flame color distribution, and flame temperature distribution.

[0073] The visual sub-model can use a convolutional neural network (CNN) or a residual network as its backbone. For flame morphology features, the model can identify the flame outline through edge detection algorithms to determine whether the flame fills the furnace, whether there is local flameout, or whether the flame is deflected. For flame brightness and color distribution, the preset model can analyze the RGB or HSV values ​​of image pixels. For example, a bright yellowish-white flame usually indicates complete combustion, while a dark red flame may indicate oxygen deficiency or insufficient calorific value. For flame temperature distribution, the model can combine infrared thermal imaging data or use colorimetric thermometry to invert the temperature field distribution inside the furnace.

[0074] Through the extraction and fusion of the aforementioned features, the visual sub-model can output a high-dimensional feature vector, which maps to the current state of waste incineration (such as "stable combustion," "risk of off-center fire," "signs of impending material shortage," etc.), providing real-time and intuitive status information for subsequent decision-making. It should be understood that the specific network structure of the visual sub-model is not limited to the examples described above; any deep learning model capable of image feature extraction and classification falls within the scope of this invention.

[0075] Furthermore, as one implementation method, after the visual sub-model identifies the current waste incineration state, the reinforcement learning sub-model is responsible for making decisions within the feasible region boundary. The current waste incineration state and current operating parameters are used as state inputs and fed into the reinforcement learning sub-model. Under the constraints of the feasible region boundary, the reinforcement learning sub-model makes decisions based on a preset reward function and outputs a first control variable that maximizes the positive weighted result of multiple optimization objectives.

[0076] Reinforcement learning sub-models can employ algorithmic architectures such as Deep Q-Network (DQN), Policy Gradient Algorithm, or Proximal Policy Optimization (PPO).

[0077] The reward function is constructed by a weighted sum of multiple immediate reward components, including environmental indicators, economic indicators, combustion stability indicators, and safety indicators representing the excess of preset parameters; each immediate reward component corresponds to one of multiple optimization objectives.

[0078] For example, environmental indicators can be linked to flue gas emission data (such as NOx and SO2 concentrations) to encourage models to output control strategies that reduce pollutant emissions; economic indicators can be linked to main steam flow or power generation to encourage models to improve combustion efficiency; combustion stability indicators can be linked to the variance of furnace temperature or main steam pressure to penalize control strategies that cause drastic fluctuations in operating conditions; and safety indicators are used to characterize the excess of preset parameters, such as furnace temperature exceeding the preset upper limit or negative pressure being too low. When the control variables output by the model may cause parameters to go out of bounds, this indicator will give a huge negative reward, thereby forcing the model to explore within the feasible region boundary.

[0079] In this implementation, the feasible region boundary serves not only as a constraint during model training but also in the inference and decision-making phase. When generating the first control variable, the reinforcement learning sub-model treats the feasible region boundary as a hard constraint on the action space; any control variable exceeding the boundary is considered an invalid action or forcibly truncated. This mechanism ensures that the first control variable output by the model always remains within a safe and compliant range, achieving the implementation of "procedure internalization" at the decision-making level. Through this combination of multi-objective optimization and constraint mechanisms, the system can automatically find the optimal operating point for combustion efficiency while ensuring safety and environmental protection, solving the problem of traditional control methods struggling to balance multiple conflicting objectives.

[0080] Based on the above embodiments, the specific construction content of the expert experience rule base and its matching triggering mechanism are described in detail.

[0081] The expert experience rule base is the core component of this invention to realize the "expert backstop" function. The rules stored therein include: operating condition identification and status judgment rules; feedforward control and compensation rules under specific operating conditions; safety interlock rules for the waste incineration control system; multi-objective trade-off and boundary protection rules, which are used to define the priority relationship between environmental protection indicators, economic indicators, combustion stability indicators and safety indicators under different operating conditions; and rules for suppressing mutations in control variables and verifying logical rationality.

[0082] These five categories of rules constitute a layered and progressive defense system, covering all kinds of conventional and unconventional operating conditions that may occur during waste incineration.

[0083] Operating condition identification and status judgment rules serve as the entry point for the entire rule base, acting as the "eyes" and "brain" of an expert. These rules define how to determine the current combustion condition based on sensor data. For example, a rule could be expressed as: "IF furnace temperature change rate > dT / dt_max AND oxygen content increases AND main steam flow decreases, THEN state = 'fuel shortage risk'". Through these rules, the system can transform discrete parameter data into engineering-meaning operating condition status labels.

[0084] Feedforward control and compensation rules under specific operating conditions are mainly used to address operating conditions with obvious precursory characteristics and requiring rapid response. Compared with feedback control, feedforward control can issue commands before disturbances affect the system, greatly reducing control lag. For example, when the operating condition identification rule determines that "the calorific value of waste has suddenly increased," the feedforward rule will immediately output compensation commands to increase the frequency of the induced draft fan and decrease the feeding speed, without waiting for the furnace negative pressure or temperature to exceed the set range before taking action.

[0085] Safety interlock rules are the system's last line of defense and have the highest priority. These rules typically correspond to extreme operating conditions that jeopardize equipment safety or environmental compliance. For example: "IF furnace temperature > 1100℃ AND duration > t_limit, THEN trigger 'high temperature interlock,' action = stop feeding + start auxiliary fuel pump." These rules ensure that the system will not output dangerous commands even if the preset model fails or the data deviates significantly.

[0086] Multi-objective trade-offs and boundary protection rules are used to resolve multi-objective conflicts under complex operating conditions. During normal operation, the default model may prioritize economic indicators (such as improving combustion efficiency), but under boundary conditions (such as when environmental emission standards approach limits), these rules force an adjustment to the weights of each optimization objective. For example, when NOx concentration approaches the upper limit of emission standards, the rule will reduce the weight of economic indicators and increase the weight of environmental indicators, forcing the system to adopt emission reduction measures such as lowering furnace temperature and optimizing air distribution, thereby seeking a suboptimal solution while ensuring compliance.

[0087] The rules for suppressing abrupt changes in control variables and verifying logical rationality are used to prevent drastic fluctuations in control commands. Because neural network models may exhibit output jitter, these rules limit the magnitude of changes in control variables. For example: "IF current feed rate command - previous feed rate command > threshold, THEN limit the magnitude of change in this command to the threshold." Simultaneously, these rules also include logical rationality checks, such as "when the furnace temperature decreases, the air volume should not be reduced." If the model output violates this physical logic, the rule will correct it.

[0088] Based on the rule base described above, this embodiment further describes the rule matching and triggering mechanism. The first control variable, current operating parameters, and current waste incineration status are input into the expert experience rule base for logical verification and control variable calibration to obtain the second control variable. Specifically, this includes the following steps.

[0089] First, based on the rules for identifying operating conditions and determining the status, the target operating condition status corresponding to the first control variable, the current operating parameters, and the current waste incineration status is determined.

[0090] This process maps real-time data to a predefined operating condition space. The system calculates the changing trends of current operating parameters (such as furnace temperature and main steam pressure) and compares them with the current waste incineration status (such as "dim flame" or "flame deflection") against pre-defined operating condition features in the rule base. For example, if the current parameters show that the furnace temperature is within the normal range, but the visual characteristics of the flame show uneven brightness distribution and large fluctuations in oxygen content, the system may identify it as an "unstable combustion" operating condition.

[0091] Secondly, based on the target working condition, a matching search is performed in the expert experience rule base to obtain the target rule.

[0092] Multiple target rules are triggered based on the target operating condition.

[0093] The above matching and retrieval process employs a state-label-based indexing mechanism. The expert experience rule base maintains a "condition-rule" mapping table. Once the target condition is determined, the system directly retrieves this mapping table and the corresponding rule set. For example, if the target condition is identified as a "safety anomaly condition," the search result will directly point to the "safety interlocking rule"; if the target condition is a "boundary anomaly condition," it may simultaneously trigger the "multi-objective trade-off and boundary protection rule" and the "mutation suppression rule." This state-index-based matching method significantly improves the system's response speed compared to traversing all rules, meeting the real-time requirements of incineration control.

[0094] Finally, according to the target rules, the current operating parameters and the current waste incineration status are logically verified and the control variables are checked to obtain the second control variable.

[0095] The above implementation steps are the specific implementation at the execution level.

[0096] The control system takes the first control variable as input and substitutes it into the target rules after triggering for logical operations. If the first control variable passes the verification of all target rules (i.e., does not violate any constraints), then the second control variable becomes the first control variable; if the first control variable is intercepted or corrected by a rule, the corrected value is output. For example, if a "mutation suppression rule" is triggered, the system checks the rate of change of the first control variable; if it exceeds the limit, it is smoothed, and finally the smoothed second control variable is output. Through this mechanism, the expert experience rule base achieves accurate verification of the output of the preset model, ensuring the logical rationality and engineering safety of the control commands.

[0097] Based on the above embodiments, the specific execution logic of hierarchical arbitration and control variable calibration is described in detail. This process constructs multiple layers of defense to ensure that the system can output safe and reasonable control commands under various operating conditions.

[0098] First, the system needs to accurately identify the current operating status. Based on the operating status identification and status judgment rules, the target operating status corresponding to the first control variable, the current operating parameters, and the current waste incineration status is determined.

[0099] Specifically, the determination process includes: determining preset parameter trends associated with the current waste incineration status and current operating parameters.

[0100] The preset parameter trend includes the preset parameter change trend, the first parameter range, the second parameter range, and the third parameter range.

[0101] In practical applications, the preset parameter change trend refers to the direction and rate of change of key parameters (such as furnace temperature and main steam pressure) per unit time, for example, "the furnace temperature decreases at a rate greater than 5℃ / s". The first parameter range usually corresponds to the "danger zone", such as a furnace temperature exceeding 1100℃ or below 850℃; the second parameter range corresponds to the "boundary warning zone", such as a furnace temperature between 850℃ and 900℃; and the third parameter range corresponds to the "safe operation zone", such as a furnace temperature between 900℃ and 1000℃.

[0102] Based on the above range and trend, the system executes the following judgment logic: if the first control variable is within the range of the first parameter, the target working condition is determined to be a safe abnormal working condition.

[0103] For example, when the furnace temperature reaches 1150℃, it is determined to be an abnormal operating condition.

[0104] If the first control variable is not within the range of the first parameter, and the trend of the first control variable does not match the trend of the preset parameter, then the target working condition is determined to be a special working condition.

[0105] For example, even if the furnace temperature is within the normal range, an abnormal sudden change in the rate of temperature change (such as a sudden drop) usually means that there has been a material shortage or a sudden change in the calorific value of the waste, which is judged as a special operating condition.

[0106] If the trend of the first control variable matches the trend of the preset parameter, and the first control variable is within the range of the second parameter, the target working condition is determined to be a boundary abnormal working condition.

[0107] For example, if the furnace temperature slowly decreases and approaches the lower limit of 850°C, it is determined to be a boundary abnormal condition.

[0108] If the trend of the first control variable matches the trend of the preset parameter, and the first control variable is within the range of the third parameter, the target working condition is determined to be a normal working condition.

[0109] After determining the target operating condition, the system performs a matching search in the expert experience rule base based on the condition, triggers the corresponding target rule, and performs hierarchical arbitration processing according to the target rule.

[0110] The first level of the tiered arbitration is the "safety priority" level. If the target operating condition is determined to be an abnormal safety condition, then the safety interlock rule triggered is the target rule.

[0111] At this point, if multiple target rules include the already triggered safety interlock rules, a warning message and an operation stop command are sent; the operation stop command is used to stop the incineration actuator.

[0112] For example, when the furnace temperature exceeds the limit, the system will directly cut off the power to the feeder and start the auxiliary burner. This instruction has the highest priority and will override any optimized instructions output by the preset model, thereby ensuring equipment safety.

[0113] The second layer of tiered arbitration is the "experience-first" layer, which mainly targets special operating conditions. If the target operating condition is determined to be a special operating condition, then the feedforward control and compensation rules and the mutation suppression and logical rationality verification rules are triggered as the target rules.

[0114] At this point, if multiple target rules do not include untriggered safety interlock rules but include triggered feedforward control and compensation rules, then based on the feedforward control and compensation rules, a preset empirical control quantity associated with the current operating parameters and the current waste incineration state is determined; the first control variable is corrected to the preset empirical control quantity to obtain the third control variable.

[0115] For example, when the "material shortage" feature is detected, the preset model may still be outputting a material reduction command due to lag, while the expert rule will directly output a larger material pushing speed as a preset experience control value, thereby achieving a rapid response and avoiding large fluctuations in furnace temperature.

[0116] The core of this level lies in leveraging the speed and certainty of expert experience to compensate for the model's insufficient response under sudden operating conditions.

[0117] The third layer of the hierarchical arbitration is the "efficiency optimization" layer, which mainly targets boundary abnormal operating conditions. If the target operating condition is determined to be a boundary abnormal operating condition, then the multi-objective trade-off and boundary protection rules and the mutation suppression and logical rationality verification rules are triggered as the target rules.

[0118] At this point, if multiple target rules do not include untriggered safety interlock rules or untriggered feedforward control and compensation rules, but include triggered multi-objective trade-off and boundary protection rules, then based on the multi-objective trade-off and boundary protection rules, determine the preset weights of one or more optimization targets corresponding to the current operating parameters and the current waste incineration state; determine the target weights of multiple optimization targets according to the preset weights; modify the weights of the multiple optimizations in the preset model to the target weights, and call the preset model with modified weights to determine the fourth control variable corresponding to the current waste incineration state and the current operating parameters.

[0119] For example, when the furnace temperature approaches its lower limit, the system temporarily increases the weight of the "combustion stability index" and decreases the weight of the "economic index," then re-initiates the model. Under the guidance of the new weights, the model automatically outputs instructions such as increasing airflow and feeding speed to bring the operating conditions back to a safe range. This approach is not a simple logical overriding, but rather guides the model to autonomously find the optimal solution by dynamically adjusting the optimization objectives, ensuring both safety and preserving the model's optimization capabilities.

[0120] In the above-mentioned processing at each level, if multiple target rules include triggered mutation suppression and logical rationality verification rules, the first control variable, the second control variable, the third control variable, or the fourth control variable is corrected according to the triggered mutation suppression and logical rationality verification rules to obtain the second control variable.

[0121] For example, regardless of whether the output of the previous level is the third or fourth control variable, the mutation suppression rule will check its magnitude of change. If a feed rate command jumps directly from 20% to 80%, the rule will smooth it out by increasing it by 10% each time, executing it over multiple cycles, thereby avoiding mechanical shock to the actuator.

[0122] Furthermore, this embodiment also includes a fallback logic. If, based on the target operating condition, no matching target rule is found in the expert experience rule base, then the waste incineration control system generates a second target operation variable according to the first control variable; and the incineration actuator is adjusted according to the second target operation variable.

[0123] This typically occurs under normal operating conditions and when no boundary protection or mutation suppression rules are triggered. Specifically, if no target rule is found in the expert experience rule base based on the target operating condition, the waste incineration control system generates a second target operating variable according to the first control variable; and the incineration actuator is adjusted according to the second target operating variable.

[0124] At this point, the system fully trusts the calculation results of the preset model and directly executes the first control variable output by the model. This design embodies the control concept of "data-driven as the primary approach and expert rules as a supplement," maximizing the optimization potential of the artificial intelligence model while ensuring safety.

[0125] To more intuitively illustrate the beneficial effects of the above embodiments of the present invention, this embodiment provides a detailed description in conjunction with a practical application scenario of a waste-to-energy incineration plant. This application scenario involves a mechanical grate furnace with a daily processing capacity of 400 tons, equipped with a complete distributed control system (DCS) and an industrial television (ITV) monitoring system. This embodiment applies the above-mentioned waste incineration control method to the automatic combustion control (ACC) system of this incinerator.

[0126] In the preparation phase before system commissioning, step S21 is executed first to acquire the operating procedure text data. Specifically, technicians import existing paper or electronic documents such as the plant's "Waste Incineration Boiler Operating Procedures" and "Safety Operation Manual" into the system. The system uses Natural Language Processing (NLP) technology to automatically extract parameter constraints from these unstructured texts. For example, the system identifies the key clause "the main control temperature of the furnace should be maintained above 850℃" from the procedures and converts it into a mathematical feasible domain boundary: T_furnace ≥ 850℃. In this embodiment, this boundary is set as a first-level alarm boundary, meaning that when the predicted temperature is below this boundary, the model's reward function will be significantly negatively penalized, thus internalizing this safety procedure requirement during the training phase.

[0127] Subsequently, the system executes the model training step in S23. Historical DCS operation data from the incinerator over the past 12 months is selected as training samples, with a sampling period of 1 second. After preprocessing such as cleaning and normalization, the data is used to train a pre-defined model based on a Long Short-Term Memory (LSTM) network architecture. During model training, the generated feasible region boundary is introduced as a penalty term in the loss function. Specifically, if the control variables output by the model cause the predicted furnace temperature to be lower than 850℃, the loss function will add a large penalty coefficient, forcing the model parameters to update in the direction that satisfies the constraints. This training method ensures that while pursuing optimization goals such as combustion efficiency, the model strictly adheres to the safety baseline defined by the operating procedures.

[0128] Meanwhile, during the construction of the expert experience rule base, specific expert rules were incorporated for the "waste feed shortage" condition that had previously occurred in the incinerator. This rule is defined as: "IF oxygen level surge >3% AND main steam flow rate decrease >5t / h, THEN determine waste feed shortage, immediately trigger pusher acceleration." This rule corresponds to the feedforward control and compensation rule for the specific condition described in Example 3. When the system detects the above parameter characteristics during operation, it will directly trigger this rule and output a preset experience control value, instead of solely relying on the model's calculation results, thus achieving rapid response to sudden operating conditions.

[0129] After the system is officially put into operation, it executes real-time control cycles. Real-time operating parameters such as furnace temperature, main steam flow, and oxygen content are collected via the DCS interface, and the current fire observation video stream data of the furnace is obtained through the ITV system. The visual sub-model analyzes the video stream in real time, extracting flame morphology and brightness distribution features to identify the current combustion state. The reinforcement learning sub-model, combined with the current operating parameters, outputs the first control variable under the feasible domain boundary constraints. Subsequently, the system inputs the first control variable into the expert experience rule base for logical verification. For example, in a certain control cycle, the model outputs the instruction to "reduce the feed rate," but the expert rule base, considering the current state of "sudden increase in oxygen and decrease in steam flow," determines it as a "feed interruption" condition, and thus corrects the instruction to "accelerate the pusher," obtaining the second control variable and issuing it for execution.

[0130] After three months of continuous operation and statistical analysis, the system has achieved significant technical results. Compared with manual control or traditional PID control before the upgrade, the fluctuation rate of main steam flow has been reduced by 10%, indicating a more stable combustion process and smoother boiler operation. Simultaneously, due to the system's internalized procedures and expert support capabilities, the number of manual interventions by operators has been reduced by 90%, greatly reducing the workload of operators and achieving the initial goal of a "lights-out factory."

[0131] More importantly, no violations of operating procedures occurred during system operation, verifying the comprehensive advantages of this invention in terms of safety, stability, and economy. It should be understood that the aforementioned processing capacity of 400 t / d, temperature boundary of 850℃, and specific data on reduced volatility are merely a preferred embodiment of this invention and not a limitation on the scope of protection of this invention. In other application scenarios, the specific parameter boundaries and model structure can be adaptively adjusted according to different furnace types and process requirements.

[0132] To achieve the above functions, the waste incineration control device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art will readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0133] This disclosure also provides an embodiment such as Figure 3The waste incineration control device shown is applied to a waste incineration control system. The waste incineration control system includes an incineration actuator and is a distributed control system. The device includes the following units: an acquisition unit 31, a generation unit 32, a control variable prediction unit 33, a control variable calibration unit 34, and a control execution unit 35.

[0134] The acquisition unit 31 is used to acquire the current operating parameters of the waste incineration control system, the current fire observation video stream data of the furnace, and the operating procedure text data during the waste incineration process; wherein, the operating procedure text data includes the parameter constraint relationship of the operating parameters in the waste incineration control system.

[0135] The generation unit 32 is used to generate the feasible domain boundary of the running parameters based on the running procedure text data.

[0136] The control variable prediction unit 33 is used to input the feasible region boundary, current operating parameters, and current fire monitoring video stream data into the preset model. The preset model first determines the current waste incineration state based on the current fire monitoring video stream data. Then, the preset model determines the first control variable within the feasible region boundary based on the current waste incineration state and current operating parameters. The preset model is trained using the feasible region boundary as the penalty term of the reward function or loss function, and the preset model outputs the corresponding control variable with the objective of maximizing the positive weighted result of multiple optimization objectives.

[0137] The control variable calibration unit 34 is used to input the first control variable, the current operating parameters and the current waste incineration status into the expert experience rule base for logical verification and control variable calibration to obtain the second control variable.

[0138] The control execution unit 35 is used to control the waste incineration control system to generate a first target operation variable according to the second control variable; and to adjust the incineration execution mechanism according to the first target operation variable.

[0139] As one implementation method, the preset model includes a visual sub-model and a reinforcement learning sub-model connected in sequence; the control variable prediction unit 33 is specifically used to: extract quantized visual features from the current fire-watching video stream data using the visual sub-model; and identify the current waste incineration status based on the extracted quantized visual features; wherein, the quantized visual features include at least one or more of the following: flame shape, flame brightness distribution, flame color distribution, and flame temperature distribution.

[0140] As one implementation method, the control variable prediction unit 33 is specifically used to: input the current waste incineration state and current operating parameters as state inputs to the reinforcement learning sub-model, so that the reinforcement learning sub-model can make decisions based on a preset reward function under the constraints of the feasible region boundary, and output the first control variable that maximizes the positive weighted result of multiple optimization objectives; the reward function is constructed by the weighted sum of multiple instant reward components, including environmental indicators, economic indicators, combustion stability indicators and safety indicators representing the excess of preset parameters; wherein each instant reward component corresponds to one of the multiple optimization objectives.

[0141] As one implementation method, the rules stored in the expert experience rule base include: operating condition identification and status judgment rules; feedforward control and compensation rules under specific operating conditions; safety interlock rules for waste incineration control systems; multi-objective trade-off and boundary protection rules, used to define the priority relationship between environmental protection indicators, economic indicators, combustion stability indicators and safety indicators under different operating conditions; and rules for suppressing mutations and verifying the logical rationality of control variables.

[0142] As one implementation method, the control variable calibration unit 34 is specifically used for: determining the target operating condition state corresponding to the first control variable, the current operating parameters, and the current waste incineration state based on the operating condition identification and state judgment rules; performing a matching search in the expert experience rule base according to the target operating condition state to obtain the target rule; multiple target rules are triggered based on the target operating condition state; and performing logical verification and control variable calibration on the current operating parameters and the current waste incineration state according to the target rule to obtain the second control variable.

[0143] As one implementation, the control execution unit 35 is also used to control the waste incineration control system to generate a second target operation variable according to the first control variable if no target rule is found in the expert experience rule base based on the target operating condition; and to adjust the incineration execution mechanism according to the second target operation variable.

[0144] As one implementation method, the control execution unit 35 is further specifically used for: if multiple target rules include triggered safety interlock rules, sending a warning message and an operation stop command; the operation stop command is used to stop the operation of the incineration actuator; if multiple target rules do not include untriggered safety interlock rules but include triggered feedforward control and compensation rules, determining a preset empirical control quantity related to the current operating parameters and the current waste incineration state based on the feedforward control and compensation rules; calibrating the first control variable to the preset empirical control quantity to obtain a third control variable; if multiple target rules do not include untriggered safety interlock rules or untriggered feedforward control and compensation rules, but include triggered multi-target rules... Based on the balance and boundary protection rules, the preset weights of one or more optimization objectives corresponding to the current operating parameters and the current waste incineration state are determined. According to the preset weights, the target weights of multiple optimization objectives are determined. The weights of the multiple optimizations in the preset model are modified to the target weights, and the preset model with modified weights is called to determine the fourth control variable corresponding to the current waste incineration state and the current operating parameters. If the multiple objective rules include triggered mutation suppression and logical rationality verification rules, the first control variable, the second control variable, the third control variable, or the fourth control variable is corrected according to the triggered mutation suppression and logical rationality verification rules to obtain the second control variable.

[0145] As one implementation method, the control execution unit 35 is further specifically used to: determine a preset parameter trend associated with the current waste incineration state and the current operating parameters; the preset parameter trend includes a preset parameter change trend, a first parameter range, a second parameter range, and a third parameter range; if the first control variable is within the first parameter range, the target operating condition is determined to be a safe abnormal operating condition; if the first control variable is not within the first parameter range, and the change trend of the first control variable does not match the preset parameter change trend, the target operating condition is determined to be a special operating condition; if the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the second parameter range, the target operating condition is determined to be a boundary abnormal operating condition; if the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the third parameter range, the target operating condition is determined to be a normal operating condition.

[0146] Among them, based on the target working condition, a matching search is performed in the expert experience rule base to obtain multiple target rules, including:

[0147] If the target operating condition is determined to be a safety abnormal operating condition, then the safety interlocking rule is triggered as the target rule; if the target operating condition is determined to be a special operating condition, then the feedforward control and compensation rule and the mutation suppression and logical rationality verification rule are triggered as the target rule; if the target operating condition is determined to be a boundary abnormal operating condition, then the multi-objective trade-off and boundary protection rule and the mutation suppression and logical rationality verification rule are triggered as the target rule.

[0148] It should be understood that the "unit" described in this embodiment can be a physical module implemented by hardware circuits (such as ASIC, FPGA), a functional module implemented by a processor executing software programs, or a combination of the two.

[0149] Regarding the apparatus in the above embodiments, the specific manner in which each unit module performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0150] Figure 4 This is a schematic diagram of a control device provided in this application. Figure 4 The control device 50 may include at least one first processor 501 and a memory 503 for storing processor-executable instructions. The first processor 501 is configured to execute the instructions in the memory 503 to implement the waste incineration control method in the following embodiments.

[0151] In addition, the control device 50 may also include a communication bus 502, at least one communication interface 504, an input device 506, and an output device 505.

[0152] The first processor 501 may be a processor (central processing unit, CPU), a microprocessor unit, an ASIC, or one or more integrated circuits for controlling the execution of programs according to the present application.

[0153] The communication bus 502 may include a path for transmitting information between the aforementioned components.

[0154] Communication interface 504 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0155] Input device 506 is used to receive input signals and output device 505 is used to output signals.

[0156] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, digital universal discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory may exist independently and be connected to the processing unit via a bus. Memory may also be integrated with the processing unit.

[0157] The memory 503 stores instructions for executing the scheme of this application, and the execution is controlled by the first processor 501. The first processor 501 executes the instructions stored in the memory 503 to realize the functions of the method of this application.

[0158] In a specific implementation, as one example, the first processor 501 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 in the CPU.

[0159] In a specific implementation, as one example, the control device 50 may include multiple processors, such as... Figure 4 The first processor 501 and the second processor 507 are described. Each of these processors can be a single-core processor or a multi-core processor. A processor here can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0160] The control device, such as Figure 4 The diagram shows a first processor 501 and a memory 503 for storing executable instructions of the first processor 501. The first processor 501 is configured to execute the executable instructions to implement the waste incineration control method as described in any of the possible embodiments above. Since the same technical effects can be achieved, further details are omitted here to avoid repetition.

[0161] This application also provides a computer-readable storage medium, which, when executed by a processor of a waste incineration control device or equipment, enables the waste incineration control device or equipment to perform a waste incineration control method as described in any of the above possible embodiments. The same technical effects can be achieved, and to avoid repetition, further details are omitted here.

[0162] This application also provides a computer program product, including a computer program or instructions, which are executed by a processor as a waste incineration control method according to any of the possible implementations described above. It achieves the same technical effects, and to avoid repetition, will not be described again here.

[0163] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0164] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for controlling waste incineration, characterized in that, An application to a waste incineration control system, wherein the waste incineration control system includes an incineration actuator and is a distributed control system, the method comprising: The current operating parameters of the waste incineration control system, the current fire monitoring video stream data of the furnace, and the operating procedure text data are obtained during the waste incineration process; wherein, the operating procedure text data includes the parameter constraint relationships of the operating parameters in the waste incineration control system; Based on the operation procedure text data, generate the feasible domain boundary of the operation parameters; The feasible region boundary, the current operating parameters, and the current fire monitoring video stream data are input into a preset model, which includes a visual sub-model and a reinforcement learning sub-model connected in sequence. The visual sub-model extracts quantified visual features from the current fire monitoring video stream data. Based on the extracted quantified visual features, the current waste incineration state is identified. The current waste incineration state and the current operating parameters are used as state inputs and input into the reinforcement learning sub-model. Under the constraints of the feasible region boundary, the reinforcement learning sub-model makes a decision based on a preset reward function and outputs a first control variable that maximizes the positive weighted result of multiple optimization objectives. The preset model is trained using the feasible region boundary as a penalty term in the reward function or loss function, and it outputs the corresponding control variable with the objective of maximizing the positive weighted result of multiple optimization objectives. The first control variable, the current operating parameters, and the current waste incineration status are input into the expert experience rule base for logical verification and control variable calibration to obtain the second control variable. According to the second control variable, the waste incineration control system generates the first target operation variable; Adjust the incineration actuator according to the first target operation variable.

2. The method according to claim 1, wherein the quantified visual features include at least one or more of flame morphology, flame brightness distribution, flame color distribution, and flame temperature distribution.

3. The method according to claim 2, characterized in that, The first control variable, determined by the preset model based on the current waste incineration state and the current operating parameters within the feasible region boundary, includes: The current waste incineration state and the current operating parameters are used as state inputs and fed into the reinforcement learning sub-model. Under the constraints of the feasible region boundary, the reinforcement learning sub-model makes decisions based on a preset reward function, outputting a first control variable that maximizes the positive weighted result of multiple optimization objectives. The reward function is constructed by a weighted sum of multiple immediate reward components, including environmental indicators, economic indicators, combustion stability indicators, and safety indicators representing the excess of preset parameters. Each immediate reward component corresponds to one of the multiple optimization objectives.

4. The method according to claim 3, characterized in that, The rules stored in the expert experience rule base include: operating condition identification and status judgment rules, used to determine the current combustion operating condition based on sensor data; feedforward control and compensation rules under specific operating conditions, used to deal with operating conditions with obvious precursor characteristics and requiring rapid response; safety interlock rules of the waste incineration control system, used to deal with extreme operating conditions that endanger equipment safety or environmental compliance; multi-objective trade-off and boundary protection rules, used to define the priority relationship between environmental indicators, economic indicators, combustion stability indicators and safety indicators under different operating conditions; and rules for suppressing sudden changes in control variables and verifying logical rationality, used to prevent drastic fluctuations in control commands.

5. The method according to claim 4, characterized in that, The first control variable, the current operating parameters, and the current waste incineration status are input into an expert experience rule base for logical verification and control variable calibration to obtain a second control variable, including: Based on the operating condition identification and status judgment rules, the target operating condition status corresponding to the first control variable, the current operating parameters, and the current waste incineration status is determined. Based on the target operating condition, a matching search is performed in the expert experience rule base to obtain the target rule corresponding to the target operating condition. According to the target rules, the current operating parameters and the current waste incineration status are logically verified and the control variables are checked to obtain the second control variable.

6. The method according to claim 5, characterized in that, The method further includes: If, based on the target operating condition, no matching target rule is found in the expert experience rule base, then the waste incineration control system generates a second target operating variable according to the first control variable. Adjust the incineration actuator according to the second target operating variable.

7. The method according to claim 5 or 6, characterized in that, The second control variable is obtained by performing logical verification and control variable calibration on the current operating parameters and the current waste incineration state according to the target rule, including: If the multiple target rules include a triggered safety interlock rule, then a warning message and an operation stop command are sent; the operation stop command is used to stop the operation of the incineration actuator. If the multiple target rules do not include untriggered safety interlock rules but include triggered feedforward control and compensation rules, then based on the feedforward control and compensation rules, a preset empirical control quantity associated with the current operating parameters and the current waste incineration state is determined; the first control variable is corrected to the preset empirical control quantity to obtain the third control variable; If the multiple target rules do not include untriggered safety interlock rules or untriggered feedforward control and compensation rules, but include triggered multi-objective trade-off and boundary protection rules, then based on the multi-objective trade-off and boundary protection rules, determine the preset weights of one or more optimization targets corresponding to the current operating parameters and the current waste incineration state; determine the target weights of multiple optimization targets according to the preset weights; modify the weights of the multiple optimizations in the preset model to the target weights, and call the preset model with modified weights to determine the fourth control variable corresponding to the current waste incineration state and the current operating parameters; If the multiple target rules include triggered mutation suppression and logical rationality verification rules, the first control variable, the second control variable, the third control variable, or the fourth control variable are corrected according to the triggered mutation suppression and logical rationality verification rules to obtain the second control variable.

8. The method according to claim 7, characterized in that, The step of determining the target operating condition state corresponding to the first control variable, the current operating parameters, and the current waste incineration state based on the operating condition identification and state judgment rules includes: A preset parameter trend is determined that is associated with the current waste incineration state and the current operating parameters. The preset parameter trend includes a preset parameter change trend, a first parameter range, a second parameter range, and a third parameter range. If the first control variable is within the first parameter range, the target operating condition is determined to be a safe abnormal operating condition. If the first control variable is not within the first parameter range, and the change trend of the first control variable does not match the preset parameter change trend, the target operating condition is determined to be a special operating condition. If the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the second parameter range, the target operating condition is determined to be a boundary abnormal operating condition. If the change trend of the first control variable matches the preset parameter change trend, and the first control variable is within the third parameter range, the target operating condition is determined to be a normal operating condition. The step of performing a matching search in the expert experience rule base based on the target working condition state to obtain multiple target rules includes: If the target operating condition is determined to be a safety anomaly, then the safety interlocking rule is triggered as the target rule; if the target operating condition is determined to be a special operating condition, then the feedforward control and compensation rule and the mutation suppression and logical rationality verification rule are triggered as target rules; if the target operating condition is determined to be a boundary anomaly, then the multi-objective trade-off and boundary protection rule and the mutation suppression and logical rationality verification rule are triggered as target rules.

9. A waste incineration control device, characterized in that, An application in a waste incineration control system, the waste incineration control system including an incineration actuator, and the waste incineration control system being a distributed control system, the device comprising: The acquisition unit is used to acquire the current operating parameters of the waste incineration control system, the current fire monitoring video stream data of the furnace, and the operating procedure text data during the waste incineration process; wherein, the operating procedure text data includes the parameter constraint relationships of the operating parameters in the waste incineration control system; The generation unit is used to generate the feasible domain boundary of the running parameters based on the running procedure text data. A control variable prediction unit is used to input the feasible region boundary, the current operating parameters, and the current fire monitoring video stream data into a preset model. The preset model includes a visual sub-model and a reinforcement learning sub-model connected in sequence. The visual sub-model extracts quantified visual features from the current fire monitoring video stream data. Based on the extracted quantified visual features, the current waste incineration state is identified. The current waste incineration state and the current operating parameters are used as state inputs and input into the reinforcement learning sub-model. Under the constraints of the feasible region boundary, the reinforcement learning sub-model makes a decision based on a preset reward function and outputs a first control variable that maximizes the positive weighted result of multiple optimization objectives. The preset model is trained using the feasible region boundary as a penalty term in the reward function or loss function, and the preset model aims to maximize the positive weighted result of multiple optimization objectives, outputting the corresponding control variable. The control variable verification unit is used to input the first control variable, the current operating parameters, and the current waste incineration status into the expert experience rule base for logical verification and control variable verification to obtain the second control variable; The control execution unit is used to control the waste incineration control system to generate a first target operation variable according to the second control variable; and to adjust the incineration execution mechanism according to the first target operation variable.

10. A waste incineration control system, characterized in that, The waste incineration control system includes an incineration actuator and a controller, and the waste incineration control system is a distributed control system. The waste incineration control system stores instructions, and when the instructions in the waste incineration control system are executed by the controller, the controller is able to perform the waste incineration control method as described in any one of claims 1 to 8.

Images