Combustion state control method, device, equipment and medium of waste incinerator

Abstract

The application discloses a waste incinerator combustion state control method, device, equipment and medium, relates to the waste incinerator control technical field, and utilizes the waste incinerator's pretreated data to construct training sample and condition vector;The combustion state trajectory in the training sample is forwardly added noise, a noisy trajectory is obtained, the noisy trajectory and the condition vector are input into a conditional diffusion model, a predicted noise is output, the conditional diffusion model is trained based on the predicted noise and the combustion state trajectory;The current condition vector is input into the trained conditional diffusion model, and a candidate combustion state trajectory is output;Based on the candidate combustion state trajectory, an optimization objective function is constructed, the optimization objective function is solved, and a control parameter adjustment amount is obtained;The control parameter adjustment amount is converted into a control instruction, the waste incinerator is controlled in the combustion state, and stable combustion, pollution control and power generation load smooth operation of the waste incinerator under the condition that the waste composition is complex are realized.

Description

Technical Field

[0001] This invention relates to the field of waste incinerator control technology, and in particular to a method, device, equipment and medium for controlling the combustion state of a waste incinerator. Background Technology

[0002] Existing combustion control methods for waste incinerators typically include operator experience-based control, PID (Proportional-Integral-Derivative) control, model predictive control, and predictive control based on data-driven models such as neural networks, long short-term memory networks, and random forests.

[0003] Manual experience-based control or traditional PID control is prone to response lag, excessive adjustment range, or interference between multiple control variables. Some model predictive control methods require mathematical descriptions of furnace heat transfer, waste drying, volatile matter release, fixed carbon combustion, flue gas flow, and pollutant generation processes. However, the physicochemical reactions during waste incineration are complex, and the composition of the waste entering the furnace, the thickness of the waste layer on the grate, and the burnout state are difficult to observe completely. The parameters required by the model are also difficult to obtain in real time and accurately. The waste incineration process has significant uncertainties; a single predicted value cannot express multiple possible future states, nor can it reflect the corresponding risk level. When the predicted value approaches the safety boundary, the control system cannot determine the future risk of exceeding limits based on a single predicted value, which may lead to under-control or over-adjustment. Most historical data for waste incineration plants comes from relatively stable operating conditions, while high-risk conditions such as rapid drops in furnace temperature, sudden increases in carbon monoxide concentration, abrupt changes in waste calorific value, partial grate collapse, and feed blockage occur relatively infrequently. Predictive models fitted to historical data may be biased towards routine operating conditions and lack sufficient learning of dynamic changes under a few abnormal operating conditions, thus affecting prediction and early warning under abnormal conditions. In addition, there are coupling relationships between multiple control objectives of waste incinerators. It is difficult to simultaneously take into account furnace temperature stability, pollutant emissions, and stable main steam flow by relying on only a single indicator for prediction and control.

[0004] As can be seen from the above, how to solve the problems of insufficient predictive expression ability under strong disturbance and uncertain operating conditions, weak adaptability to abnormal operating conditions, difficulty in supporting risk decision-making with single-point prediction, and insufficient multi-objective collaborative control ability, fully describe the uncertainty of future combustion state, and achieve stable combustion, coordinated control of pollutants, and stable operation of power generation load in waste incinerators under complex and variable waste composition conditions are problems that need to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, equipment, and medium for controlling the combustion state of a waste incinerator. This method addresses the problems of insufficient predictive capability under strong disturbances and uncertain operating conditions, weak adaptability to abnormal operating conditions, difficulty in supporting risk decision-making with single-point predictions, and insufficient multi-objective collaborative control capability. It fully describes the uncertainty of the future combustion state, achieving stable combustion, coordinated pollutant control, and stable operation of the power generation load in the waste incinerator under complex and variable waste composition conditions. The specific solution is as follows: In a first aspect, this application discloses a method for controlling the combustion state of a waste incinerator, comprising: Historical operating data of the waste incinerator is acquired, the historical operating data is preprocessed, and training samples and condition vectors are constructed using the preprocessed data; the historical operating data includes state variables, control variables, and external disturbance variables; The combustion state trajectory in the training sample is forward-noised to obtain a noisy trajectory. The noisy trajectory and the conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and the combustion state trajectory, the conditional diffusion model is trained to obtain the trained conditional diffusion model. The real-time operation data of the waste incinerator is obtained, and the current condition vector is constructed using the real-time operation data. The current condition vector is then input into the trained conditional diffusion model to output the candidate combustion state trajectory. Based on the candidate combustion state trajectory, an optimization objective function is constructed, and the optimization objective function is solved to obtain the control parameter adjustment amount; The control parameter adjustment amount is converted into a control command, and the combustion state of the waste incinerator is controlled by the control command.

[0006] Optionally, the step of acquiring historical operating data of the waste incinerator and preprocessing the historical operating data includes: Historical operating data of waste incinerators are obtained from the operation control system, flue gas emission monitoring system, equipment control system, and manual detection data of waste incineration power plants. The historical operating data includes state variables, control variables, and external disturbance variables; The historical running data is subjected to time alignment, outlier handling, missing value handling, smoothing, and normalization to obtain preprocessed data.

[0007] Optionally, constructing training samples and conditional vectors using preprocessed data includes: Using a sliding window approach, combustion state trajectories are extracted from the preprocessed data, and training samples are constructed using these combustion state trajectories. Based on the preprocessed data, historical state sequences, historical control sequences, and disturbance features are constructed. A condition vector is constructed using the historical state sequence, the historical control sequence, and the perturbation features.

[0008] Optionally, the combustion state trajectories in the training samples are forward-noised to obtain noisy trajectories, including: The true noise is obtained by randomly sampling from a standard Gaussian distribution; Based on preset noise scheduling parameters, and using the real noise to add noise to the combustion state trajectory in the training sample, a noisy trajectory is obtained.

[0009] Optionally, based on the predicted noise and the combustion state trajectory, the conditional diffusion model is trained to obtain a trained conditional diffusion model, including: Randomly sample the preset set of diffusion steps to obtain the diffusion step numbers; The diffusion step number, noisy trajectory, and conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and combustion state trajectory, a total loss function is constructed; With the goal of minimizing the total loss function, the denoising network in the conditional diffusion model is trained to obtain the trained conditional diffusion model.

[0010] Optionally, based on the candidate combustion state trajectory, an optimization objective function is constructed, including: The candidate combustion state trajectory is inversely normalized and the engineering boundary is clipped to obtain the clipped candidate combustion state trajectory. Based on the trimmed candidate combustion state trajectories, the furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value are calculated respectively. The risk values ​​of furnace temperature, carbon monoxide, nitrogen oxides, and main steam flow fluctuation are weighted to obtain a comprehensive risk value for combustion stability. An optimization objective function is constructed using the comprehensive risk value of combustion stability.

[0011] Optionally, the control parameter adjustment amount is converted into a control command, and the combustion state of the waste incinerator is controlled using the control command, including: The control parameter adjustment amount is converted into a control set value, and a control command is generated based on the control set value; The control commands are subject to security verification. If the security check fails, an alarm message or control suggestion will be generated and sent to the client. If the safety verification passes, the control command is sent to the operation control system of the waste incinerator so that the operation control system can control the combustion status of the waste incinerator based on the control command. Determine whether the preset control termination condition is met; If the preset control termination condition is not met, the process of acquiring real-time operating data of the waste incinerator will be repeated until the preset control termination condition is triggered to complete the rolling control of the waste incinerator.

[0012] Secondly, this application discloses a combustion state control device for a waste incinerator, comprising: The preprocessing module is used to acquire historical operating data of the waste incinerator, preprocess the historical operating data, and use the preprocessed data to construct training samples and condition vectors; the historical operating data includes state variables, control variables, and external disturbance variables. The training module is used to add noise to the combustion state trajectory in the training samples to obtain a noisy trajectory, input the noisy trajectory and the conditional vector into the denoising network in the conditional diffusion model to output predicted noise, and train the conditional diffusion model based on the predicted noise and the combustion state trajectory to obtain the trained conditional diffusion model. The trajectory determination module is used to acquire real-time operating data of the waste incinerator, construct a current condition vector using the real-time operating data, and input the current condition vector into the trained conditional diffusion model to output candidate combustion state trajectories. The solution module is used to construct an optimization objective function based on the candidate combustion state trajectory, solve the optimization objective function, and obtain the control parameter adjustment amount; The combustion state control module is used to convert the control parameter adjustment amount into control commands, and use the control commands to control the combustion state of the waste incinerator.

[0013] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned method for controlling the combustion state of a waste incinerator.

[0014] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for controlling the combustion state of a waste incinerator.

[0015] Beneficial Effects: This application provides a method for controlling the combustion state of a waste incinerator, including acquiring historical operating data of the waste incinerator, preprocessing the historical operating data, and using the preprocessed data to construct training samples and condition vectors. This improves the temporal consistency and effectiveness of operating data from different sources. The combustion state trajectory in the training samples is forward-noised to obtain a noisy trajectory. The noisy trajectory and condition vector are input into a denoising network in a conditional diffusion model to output predicted noise. Based on the predicted noise and the combustion state trajectory, the conditional diffusion model is trained to obtain a trained conditional diffusion model. This improves the conditional diffusion model's ability to represent the dynamic relationship of combustion state and the uncertainty of future states. Real-time operating data of the waste incinerator is acquired, and the current condition vector is constructed using the real-time operating data. The current condition vector is then input into the trained conditional diffusion model. The model outputs candidate combustion state trajectories, enabling advance prediction of various possible combustion states and enhancing the ability to identify risks associated with changes in furnace temperature, pollutant concentration, and main steam flow. Based on these candidate combustion state trajectories, an optimization objective function is constructed and solved to obtain control parameter adjustment amounts. This reduces the risks of furnace temperature exceeding limits, pollutant exceedances, and main steam flow fluctuations, minimizing large changes and frequent adjustments to control parameters. The control parameter adjustment amounts are then converted into control commands, which are used to control the combustion state of the waste incinerator. This addresses issues such as insufficient predictive power under strong disturbances and uncertain operating conditions, weak adaptability to abnormal operating conditions, difficulty in supporting risk decision-making with single-point predictions, and insufficient multi-objective collaborative control capabilities. Ultimately, this enables stable combustion, coordinated pollutant control, and stable operation of the power generation load in waste incinerators under complex and variable waste composition conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for controlling the combustion state of a waste incinerator disclosed in this application; Figure 2 This is a schematic diagram of the combustion state control device for a waste incinerator disclosed in this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Existing combustion control methods for waste incinerators typically include operator experience-based control, PID control, model predictive control, and predictive control based on data-driven models such as neural networks, long short-term memory networks, and random forests. Manual experience-based control or traditional PID control is prone to response lag, excessive adjustment ranges, or interference between multiple control variables. Some model predictive control methods require mathematical descriptions of furnace heat transfer, waste drying, volatile matter release, fixed carbon combustion, flue gas flow, and pollutant generation processes. However, the physicochemical reactions during waste incineration are complex, and the composition of the waste entering the furnace, the thickness of the waste layer on the grate, and the burnout state are difficult to observe completely. The parameters required by the model are also difficult to obtain in real time and accurately. The waste incineration process has significant uncertainties; a single predicted value cannot represent multiple possible future states, nor can it reflect the corresponding risk level. When the predicted value approaches the safety boundary, the control system cannot determine the future risk of exceeding limits based on a single predicted value, potentially leading to under-control or over-adjustment. Most historical data for waste incineration plants comes from relatively stable operating conditions, while high-risk conditions such as rapid drops in furnace temperature, sudden increases in carbon monoxide concentration, abrupt changes in waste calorific value, localized grate collapse, and feed blockage occur relatively infrequently. Predictive models fitted to historical data may be biased towards conventional operating conditions and lack sufficient learning of the dynamic changes of a few abnormal operating conditions, thus affecting prediction and early warning under abnormal conditions. Furthermore, multiple control objectives of waste incinerators are coupled, and prediction and control based on only a single indicator cannot simultaneously ensure stable furnace temperature, stable pollutant emissions, and stable main steam flow. Therefore, addressing the issues of insufficient predictive capacity under strong disturbances and uncertain operating conditions, weak adaptability to abnormal operating conditions, inability of single-point prediction to support risk decision-making, and insufficient multi-objective coordinated control capabilities, fully describing the uncertainty of future combustion states, and achieving stable combustion, coordinated pollutant control, and stable power generation load operation of waste incinerators under complex and variable waste composition conditions are problems that need to be solved in this field.

[0020] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for controlling the combustion state of a waste incinerator, which may specifically include: Step S11: Obtain historical operating data of the waste incinerator, preprocess the historical operating data, and use the preprocessed data to construct training samples and condition vectors; the historical operating data includes state variables, control variables, and external disturbance variables.

[0021] In this embodiment, historical operating data of the waste incinerator is obtained from the operation control system, flue gas emission monitoring system, equipment control system, and manual detection data of the waste incineration power plant. The historical operating data includes state variables, control variables, and external disturbance variables. The historical operating data undergoes time alignment, outlier handling, missing value handling, smoothing, and normalization to obtain preprocessed data. Combustion state trajectories are extracted from the preprocessed data using a sliding window method, and training samples are constructed using these trajectories. Historical state sequences, historical control sequences, and disturbance features are constructed based on the preprocessed data. A condition vector is then constructed using the historical state sequences, historical control sequences, and disturbance features.

[0022] Specifically, the sources of historical operating data include the DCS (Distributed Control System), PLC (Programmable Logic Controller) control system, flue gas online monitoring system, steam and water system, grate control system, waste crane control system, CEMS (Continuous Emission Monitoring System), and necessary manual test data of the waste incineration power plant.

[0023] State variables are used to describe the current combustion state and operating results of a waste incinerator, including but not limited to furnace temperature, flue gas parameters, pollutant parameters, steam system parameters, and combustion stability parameters. Specifically: (1) The furnace temperature includes the temperature of the first flue, the temperature of the second flue, the temperature of the upper part of the furnace, the temperature of the middle part of the furnace, and the temperature of the furnace outlet; (2) Flue gas parameters include flue gas oxygen content, flue gas temperature, flue gas flow rate and furnace negative pressure; (3) Pollutant parameters include carbon monoxide concentration, nitrogen oxide concentration, hydrogen chloride concentration, sulfur dioxide concentration, dust concentration, and ammonia escape concentration; (4) Steam system parameters include main steam flow rate, main steam pressure, main steam temperature, boiler feedwater flow rate, and steam drum water level; (5) Combustion stability parameters include furnace temperature fluctuation range, carbon monoxide concentration fluctuation range, oxygen fluctuation range and main steam flow fluctuation range.

[0024] The control variables are used to describe the control actions currently performed by the waste incinerator, including but not limited to: (1) waste feed rate; (2) pusher speed or pusher stroke; (3) grate speed, including grate speed in the drying section, grate speed in the combustion section, and grate speed in the burnout section; (4) total primary air volume; (5) primary air zoning volume; (6) total secondary air volume; (7) secondary air nozzle volume or damper opening; (8) flue gas recirculation volume; (9) SNCR (Selective Non-Catalytic Reduction) ammonia injection volume or urea injection volume; (10) induced draft fan frequency, primary air fan frequency, and secondary air fan frequency.

[0025] External disturbance variables are used to describe inputs that affect the combustion state but are not entirely determined by the control system, including but not limited to: (1) estimated calorific value of waste; (2) estimated moisture content of waste; (3) weight of waste grab bucket; (4) waste pit level; (5) image recognition results of waste entering the furnace; (6) ambient temperature and ambient humidity; (7) unit load setpoint; (8) boiler load demand.

[0026] The estimated calorific value of the waste can be obtained by comprehensively estimating the weight grabbed by the waste crane, the source of the waste, the fermentation time in the waste pit, historical steam production, furnace temperature response, and the image recognition results of the waste entering the furnace. If the on-site waste image recognition is not yet available, the historical calorific value estimation model or the waste calorific value level entered by the operators can be used as alternative inputs.

[0027] All types of operational data are stored according to a uniform sampling period. The sampling period can be 5 seconds, 10 seconds, 30 seconds, 1 minute, or other time intervals suitable for the operation of the control system. For data with different sampling frequencies, resampling is used to unify all types of data to the same time scale, so that state variables, control variables, and external disturbance variables at the same moment correspond to each other.

[0028] In this embodiment, historical operational data undergoes time alignment, outlier handling, missing value handling, smoothing, and normalization, as detailed below: Time Alignment: Due to the different sampling periods of DCS, PLC, CEMS, and manual testing data, this application aligns the variables from DCS, PLC, CEMS, and manual testing data according to a unified timestamp. For high-frequency data, the average, median, or last value can be aggregated to a unified sampling period; for low-frequency data, forward padding, linear interpolation, or interpolation based on operating condition similarity can be used for padding.

[0029] Outlier Handling: Identifying sensor malfunctions, communication interruptions, abrupt changes, and data significantly exceeding physical limits. Outlier identification methods include threshold judgment, sliding window abrupt change detection, three-standard-deviation detection, or historical distribution-based anomaly detection. Identified outliers can be processed through deletion, interpolation replacement, or masking. For example, if the furnace temperature jumps from 900 degrees Celsius to 50 degrees Celsius at a certain moment, and adjacent sensors do not show similar changes, this data can be identified as an outlier, and interpolation replacement can be performed using valid data from the preceding and following time windows.

[0030] Missing value handling: For short-term missing data, linear interpolation or forward imputation can be used; for long-term missing data, variable masking can be used to mark the missing state, and a missing flag can be added to the model input to prevent the conditional diffusion model from taking the imputed data as the true observation.

[0031] Smoothing: For variables with high-frequency noise, smoothing methods such as moving average, exponential moving average, or low-pass filtering can be used. For sudden changes in carbon monoxide concentration, rapid drop in furnace temperature, or other abrupt changes that have early warning significance, excessive smoothing should be avoided to prevent weakening the characteristics of abnormal operating conditions.

[0032] Normalization: Due to the differences in the dimensions of different variables—for example, furnace temperature is typically several hundred to one thousand degrees Celsius, carbon monoxide concentration may be tens to hundreds of milligrams per cubic meter, and fan frequency can range from 0 to 50 Hz—normalization is performed on each variable. Z-score standardization can be used for normalization, and its expression is: ; in, For the standardized variable values, The original variable value, The mean of the training set, The standard deviation of the training set is given.

[0033] Normalization can also be performed using Min-Max (Minimum-Maximum) normalization, the expression of which is: ; in, The normalized variable values, To find the minimum value of this variable in the training set, This represents the maximum value of the variable in the training set.

[0034] After data preprocessing, training samples are constructed using a sliding window approach. Let the historical window length be...T The future prediction window length is H Historical operating condition sequences and future real trajectories are extracted from the preprocessed data. t - T +1 hour to t The historical operating condition sequence at any given time can be represented as: ,from t +1 hour to t + H The future true trajectory of a moment can be represented as .

[0035] Each training sample includes historical operating condition sequences and future real trajectories, as well as current control parameters. Load setpoint And estimated calorific value of waste .

[0036] In order to construct operating condition characteristics, the past T The state variables, such as furnace temperature, flue gas oxygen content, carbon monoxide concentration, nitrogen oxide concentration, main steam flow rate, and flue gas temperature, within each time step are arranged in chronological order to form a historical state sequence. Its expression is: ; in, for t The state vector at time t.

[0037] The past T The historical control sequence consists of control variables such as feed rate, grate speed, primary air volume, secondary air volume, and ammonia injection rate within each time step. Its expression is: ; in, for t Control vector at time.

[0038] The estimated calorific value of the waste, the estimated moisture content of the waste, the load setpoint, the weight of the waste grab bucket, and the image recognition features of the waste are combined to form the external disturbance feature, the expression of which is: ; in, for t External disturbance characteristics at any given time. for t Estimated calorific value of waste at any given time. for t The estimated moisture content of the waste at that time. for t The load setpoint at any given time. fort The weight of the garbage grabber at any given moment. for t Garbage image recognition features at any given time.

[0039] By combining historical state sequences, historical control sequences, and external perturbation features, a conditional vector is formed that the conditional diffusion model relies on to generate the future combustion trajectory. Its expression is: ; in, This indicates a splicing operation. It is a historical state sequence. This is a historical control sequence.

[0040] The predictive target of the conditional diffusion model is the future. H Multivariable combustion state trajectory within a time step Its expression is: ; in, y Variables may include furnace temperature, carbon monoxide concentration, nitrogen oxide concentration, flue gas oxygen content, main steam flow rate, and flue gas temperature.

[0041] Step S12: Add noise to the combustion state trajectory in the training sample to obtain a noisy trajectory. Input the noisy trajectory and the conditional vector into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and the combustion state trajectory, train the conditional diffusion model to obtain the trained conditional diffusion model.

[0042] In this embodiment, random sampling is performed on a standard Gaussian distribution to obtain real noise; based on preset noise scheduling parameters, the real noise is used to add noise to the combustion state trajectory in the training samples to obtain a noisy trajectory; the noisy trajectory and the conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise; random sampling is performed on a preset diffusion step set to obtain diffusion step numbers; the diffusion step numbers, noisy trajectories, and conditional vectors are input into the denoising network in the conditional diffusion model to output predicted noise; based on the predicted noise and the combustion state trajectory, a total loss function is constructed; with the goal of minimizing the total loss function, the denoising network in the conditional diffusion model is trained to obtain the trained conditional diffusion model.

[0043] This application employs a conditional diffusion model as the generation model for future combustion state trajectories. The conditional diffusion model includes a forward noise addition process and a backward denoising process during the training phase. The forward noise addition process gradually adds noise to the true future trajectory, making the true future trajectory gradually approach Gaussian noise; the backward denoising process, under the constraints of current and historical operating conditions, learns to recover the future combustion state trajectory from the noisy trajectory. In simpler terms, the conditional diffusion model does not directly provide a prediction value, but rather, under the constraints of current operating conditions, gradually "reconstructs" multiple possible future combustion state change paths. After training, the conditional diffusion model can generate candidate combustion state trajectories that conform to the current operating conditions from multiple sets of initial noise, thus forming a set of future combustion state trajectories.

[0044] Specifically, during the training phase, the future real trajectory is used as the noise-free trajectory. The relationship between the two can be expressed as: ; in, The noise-free trajectory before forward noise addition. From t +1 hour to t + H The true trajectory of the future at any given moment.

[0045] According to preset noise scheduling parameters , to right By gradually adding noise, a noisy trajectory can be obtained. , to . No. The forward noise addition process can be represented as: ; in, For a given first Step trajectory Time Noisy trajectory Conditional distribution, For the first Noisy trajectory, For the first Step trajectory It follows a Gaussian distribution. For the first Step noise intensity, It is the identity matrix. This represents the number of diffusion steps.

[0046] Through multi-step forward noise addition, the future true trajectory gradually becomes approximately Gaussian noise. The reverse denoising process is used to modify the conditional vector. Under constraints, from noisy trajectories Gradually recovering to the trajectory of future combustion. The reverse denoising process can be represented as: ; in, In model parameters , No. Noisy trajectory and condition vector Given the first Step trajectory The inverse conditional distribution, These are the model parameters for the conditional diffusion model. The denoised mean value predicted by the denoising network. This is the denoised variance predicted by the denoising network.

[0047] Condition vector The diffusion model can be input in any of the following ways: (1) Condition vector Encode it as a conditional embedding vector and associate it with the noisy trajectory at each step. splicing; (2) Condition vector Input a Transformer encoder, GRU (Gated Recurrent Unit), LSTM (Long Short-Term Memory) network, or a temporal convolutional network, extract temporal context features, and then input them into a denoising network; (3) A cross-attention mechanism is adopted so that the denoising network pays attention to key time segments and key variables in the historical working condition sequence in each step of the denoising process; (4) The load setpoint, the estimated calorific value of the waste and the current control parameters are embedded into the diffusion step code as control conditions.

[0048] The denoising network can employ a one-dimensional U-Net (U-shaped Network), a temporal convolutional network, a Transformer encoder, or a multilayer perceptron structure. For multivariate time series prediction scenarios in waste incinerators, a temporal Transformer or a one-dimensional U-Net structure can be used to capture the temporal dependencies between furnace temperature, air volume, feed rate, pollutant concentration, and main steam flow.

[0049] The input to the denoising network includes the first Noisy trajectory Conditional vector And optional variable masks. The output of the denoising network is the predicted noise. Or the predicted noiseless future trajectory , or the mean and variance parameters of the inverse distribution.

[0050] This embodiment uses a training method based on predicted noise, and the denoising network outputs predicted noise. .

[0051] The model training objective is to make the predicted noise approximate the actual noise added during the forward noise addition process. The diffusion loss can be expressed as: ; in, For the spread loss, For the expected operation, This is the actual noise added during the forward noise generation process. For the denoising network according to the first Noisy trajectory Diffusion step number and condition vector Output prediction noise, For model parameters, Norm operations.

[0052] To enhance the predictive power of combustion control-related variables, a weighted loss of key variables can be introduced, the expression of which is: ; in, Weighted loss for key variables, For furnace temperature prediction error, This is the prediction error for carbon monoxide concentration. To account for the prediction error of nitrogen oxide concentration, The main steam flow prediction error To account for the oxygen content prediction error, , , , and These are the weights corresponding to the prediction error.

[0053] Based on the diffusion loss and the weighted loss of key variables, physical constraint loss and trajectory smoothing loss can be introduced to construct the total loss function. The total loss function can be expressed as: ; in, For the total loss function, For the spread loss, Weighted loss for key variables, For physical constraint loss, For trajectory smoothing loss, , and These are the weighting coefficients for the corresponding loss terms.

[0054] To avoid generating trajectories that do not conform to common engineering sense, this application introduces a physical constraint loss. For example: (1) The furnace temperature shall not be lower than the physically possible range; (2) The concentrations of carbon monoxide, nitrogen oxides, and oxygen content in flue gas shall not be less than 0; (3) The main steam flow rate shall not be less than 0; (4) Changes in air volume and oxygen content should maintain a reasonable directional relationship; (5) The change in feed rate and the change in main steam flow rate should satisfy a reasonable lag relationship.

[0055] The physical constraint loss can be expressed as: ; in, For physical constraint loss, For non-negative constraint loss, Loss due to variable range constraints. Loss is constrained by lag relationship. Loss due to variable association constraints.

[0056] To reduce unrealistic and severe jitter in the generated trajectory, a trajectory smoothing loss can be introduced, the expression of which is: ; in, For trajectory smoothing loss, For the future prediction window length, For predicting the time step index, and These are the state vectors for adjacent prediction time steps. Norm operations.

[0057] For real abnormal changes such as a sudden increase in carbon monoxide concentration or a rapid decrease in furnace temperature, the weight of trajectory smoothing loss should not be too large to avoid over-smoothing of real abnormal changes.

[0058] During offline training, data covering different operating scenarios were extracted from the historical database of the waste incineration plant. The operating scenarios included (1) stable combustion conditions; (2) low load conditions; (3) high load conditions; (4) high calorific value of waste conditions; (5) low calorific value of waste conditions; (6) high moisture content conditions; (7) sudden increase in carbon monoxide concentration conditions; (8) decrease in furnace temperature conditions; (9) high nitrogen oxide concentration conditions; and (10) fluctuation in main steam flow conditions.

[0059] The historical samples are divided into training, validation, and test sets. To avoid future data from being included in the training set, they can be divided chronologically. For example, the data from the first 60% of the time period can be used as the training set, the data from the middle 20% of the time period as the validation set, and the data from the last 20% of the time period as the test set.

[0060] Before training begins, initialize the parameters of the conditional diffusion model. Initialization parameters include (1) the number of diffusion steps. (2) Noise scheduling parameters (3) Denoising network structure; (4) Conditional encoder structure; (5) Learning rate; (6) Batch size; (7) Number of training rounds; (8) Weights of loss function.

[0061] Each round of training includes the following process: Step 1: Randomly select a batch of samples from the training set to obtain the future true trajectory. and condition vector ; Step 2: Random sampling diffusion step ; Step 3: Randomly sample Gaussian noise ; Step 4: Generate the first noise generation process based on the forward noise generation process. Noisy trajectory ; Step 5: Transfer the noisy trajectory Diffusion step number and condition vector Input denoising network; Step 6: Predict noise output from the denoising network. ; Step 7: Calculate the diffusion loss, the weighted loss of key variables, the physical constraint loss, and the trajectory smoothing loss; Step 8: Update the model parameters through backpropagation; Step 9: Calculate the prediction error, risk identification accuracy, and trajectory distribution consistency index on the validation set; Step 10: Stop training when the verification metrics meet the preset conditions or the maximum number of training rounds is reached.

[0062] After training, evaluate the conditional diffusion model on the test set. Evaluation metrics include: (1) Error in furnace temperature prediction; (2) Prediction error of carbon monoxide concentration; (3) Prediction error of nitrogen oxide concentration; (4) Prediction error of main steam flow rate; (5) Accuracy of identifying risks at the lower limit of furnace temperature; (6) Accuracy of identifying carbon monoxide exceedance risk; (7) Accuracy of identifying nitrogen oxide exceedance risk; (8) Trajectory coverage; (9) Reliability of prediction confidence intervals; (10) Early warning time.

[0063] After the conditional diffusion model passes the test, the trained conditional diffusion model is saved. The saved content includes: (1) denoising network parameters; (2) conditional encoder parameters; (3) normalization parameters; (4) variable configuration table; (5) risk threshold configuration table; (6) control constraint parameters; and (7) model version number.

[0064] Step S13: Obtain real-time operating data of the waste incinerator, construct a current condition vector using the real-time operating data, and input the current condition vector into the trained conditional diffusion model to output candidate combustion state trajectories.

[0065] In this embodiment, during the online operation phase, the latest data is obtained according to a set control cycle. T Real-time running data at each time step. The control period can be 30 seconds, 1 minute, 2 minutes, or 5 minutes. The variable types in the real-time running data correspond to the variable types used in the training phase.

[0066] Data cleaning, missing value handling, outlier handling, and smoothing are performed on real-time operational data in the same manner as during the training phase. Normalization parameters saved during the training phase are also used for normalization to ensure consistency between the processing methods for online input data and training data. A current condition vector is constructed based on the real-time historical state sequence, real-time historical control sequence, estimated waste heat value, and load setpoint. .

[0067] To generate multiple candidate combustion state trajectories, samples were taken from a standard Gaussian distribution. The initial noise of the group is expressed as: ; in, to Group 1 to Group 2 are respectively Group initial noise, For the number of diffusion steps, This represents the number of candidate trajectories. It can be 20, 50, 100 or other values.

[0068] For each set of initial noise, the trained conditional diffusion model in the current conditional vector Execution under constraints After a reverse denoising step, a future candidate combustion state trajectory is obtained, the expression of which is: ; in, For the first The noise-free trajectory obtained by reverse denoising of the initial noise group For the first Article from Time to The trajectory of the candidate combustion state at time t, For candidate trajectory index, For the current moment, Predict the length of the future prediction window.

[0069] Will The candidate combustion state trajectories form the set of future combustion state trajectories, and their expression is: ; in, for The set of future combustion state trajectories at each moment. to Articles 1 to 1 A candidate combustion state trajectory.

[0070] Each candidate combustion state trajectory includes the future... Predicted values ​​for furnace temperature, carbon monoxide concentration, nitrogen oxide concentration, flue gas oxygen content, main steam flow rate, and flue gas temperature within each time step. Multiple candidate combustion state trajectories are used to represent various possible development paths of the future combustion state under the current operating conditions.

[0071] After generating the candidate combustion state trajectory, each candidate combustion state trajectory is denormalized to restore the predicted value to the corresponding actual engineering dimension; at the same time, engineering boundary trimming is performed, for example, the carbon monoxide concentration or air volume less than 0 is trimmed to a range not less than 0, and the furnace temperature is kept within a reasonable physical range to obtain the trimmed candidate combustion state trajectory.

[0072] Step S14: Based on the candidate combustion state trajectory, construct an optimization objective function, solve the optimization objective function, and obtain the control parameter adjustment amount.

[0073] In this embodiment, the candidate combustion state trajectory is inversely normalized and the engineering boundary is trimmed to obtain the trimmed candidate combustion state trajectory. Based on the trimmed candidate combustion state trajectory, the furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value are calculated respectively. The furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value are weighted to obtain the comprehensive combustion stability risk value. The comprehensive combustion stability risk value is used to construct an optimization objective function.

[0074] Let the safe threshold for furnace temperature be... For example, 850 degrees Celsius. For the first... A candidate combustion state trajectory, if in the future H The furnace temperature at any moment within a time step is lower than the furnace temperature safety threshold. If the candidate combustion trajectory is deemed to have a risk of reaching the furnace temperature limit, then the furnace temperature risk value can be expressed as: ; in, This is the furnace temperature risk value. The number of candidate trajectories. This is an indicator function; it takes the value 1 if the condition in the indicator function is true, and 0 otherwise. For the first i The furnace temperature sequence in the candidate combustion state trajectory.

[0075] Set the carbon monoxide concentration limit as follows: For the first i Candidate combustion state trajectories to determine future H Does the carbon monoxide concentration within each time step exceed the carbon monoxide concentration limit? The carbon monoxide risk value can be expressed as: ; in, This is the carbon monoxide risk value. For the first i The sequence of carbon monoxide concentrations in the candidate combustion state trajectories.

[0076] Let the nitrogen oxide concentration limit be... For the first i Candidate combustion state trajectories to determine future H Does the nitrogen oxide concentration within each time step exceed the nitrogen oxide concentration limit? The risk value for nitrogen oxides can be expressed as: ; in, The risk value for nitrogen oxides. For the first i The sequence of nitrogen oxide concentrations in the candidate combustion state trajectories.

[0077] Assume the allowable fluctuation range of the main steam flow is For the first i Candidate combustion state trajectories are used to calculate future combustion state trajectories. H The difference between the maximum and minimum main steam flow rates within a given time step; if this difference exceeds the allowable fluctuation range of the main steam flow rate. If so, it is determined that there is a risk of main steam flow fluctuation. The risk value of main steam flow fluctuation can be expressed as: ; in, The main steam flow rate fluctuation risk value. For the first i The sequence of main steam flow rates in the candidate combustion state trajectory.

[0078] The comprehensive risk value for combustion stability is obtained by weighting the risk values ​​of furnace temperature, carbon monoxide, nitrogen oxides, and main steam flow fluctuation, and its expression is as follows: ; in, The overall risk value for combustion stability. This is the furnace temperature risk value. This is the carbon monoxide risk value. The risk value for nitrogen oxides. The main steam flow rate fluctuation risk value. , , and These are the risk weights corresponding to the risk values. Each risk weight can be set according to operational objectives, environmental protection requirements, and unit load demands.

[0079] As an optional implementation, a dioxin formation risk indicator can also be introduced. This indicator may not directly predict the dioxin concentration, but can be calculated based on the duration of the furnace temperature being below a preset threshold, the carbon monoxide concentration, the oxygen content of the flue gas, the temperature of the flue gas cooling zone, the amount of activated carbon injected, and the inlet temperature of the bag filter.

[0080] The control objective of risk constraint control is to reduce combustion risk within the future prediction time window, while reducing excessive or frequent control actions. The control objectives include (1) reducing the risk of furnace temperature below the limit; (2) reducing the risk of carbon monoxide exceeding the limit; (3) reducing the risk of nitrogen oxide exceeding the limit; (4) reducing the risk of main steam flow fluctuation; (5) maintaining a stable adjustment range of control parameters; and (6) meeting equipment safety boundaries and process operation constraints.

[0081] Risk-constrained control optimization can output adjustments for feed rate, grate velocity, primary air volume, secondary air volume, primary and secondary air ratio, and SNCR ammonia injection rate. These control parameter adjustments can be expressed as: ; in, for The amount of control parameter adjustment at any given time. Adjustment amount for feed rate This is the adjustment amount for the grate speed. This is the primary air volume adjustment amount. This is the secondary air volume adjustment amount. This refers to the adjustment amount for the ratio of primary air to secondary air. Adjustment amount for SNCR ammonia injection rate.

[0082] Control variables need to satisfy equipment and process constraints, for example: ; in, This is the current feed rate. This is the lower limit of the feed rate. This is the upper limit of the feed rate; ; in, The current grate speed, This is the lower limit of grate speed. This is the upper limit of grate speed; ; in, This is the current primary air volume. This is the lower limit of the primary air volume. This is the upper limit of the primary air volume; ; in, For the current secondary air volume, This is the lower limit of secondary air volume. This is the upper limit of the secondary air volume; ; in, This represents the current SNCR ammonia injection rate. This is the lower limit of ammonia injection rate for SNCR. This is the upper limit of ammonia injection rate for SNCR.

[0083] It can also limit the adjustment range within a unit control cycle, and its expression is: ; To prevent excessive control actions from causing disturbances in the furnace.

[0084] Risk constraint control optimization also needs to meet the allowable upper limits for furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value, the expressions of which are: ; in, , , and These are the risk values ​​for furnace temperature, carbon monoxide, nitrogen oxides, and main steam flow fluctuations, respectively. , , and These represent the maximum allowable risk values.

[0085] When constructing the optimization objective function using the comprehensive risk value of combustion stability, we can simultaneously consider each risk value, the adjustment range of control parameters, and the changes in control parameters between adjacent control cycles. The optimization objective function can be expressed as: ; in, To optimize the objective function, This represents the adjustment amount of the control parameter to be solved. This refers to the adjustment amount of the control parameters in the previous control cycle. This is an item used to limit the range of control movements. This is a term used to limit changes in control actions between adjacent control cycles. Norm operations.

[0086] When solving the objective function, the method of enumerating candidate control actions can be adopted. Within the allowable range of feed rate, grate speed, air volume and ammonia injection rate, multiple candidate control actions are generated. Each candidate control action is used as a condition input, and the corresponding future risk is predicted by the conditional diffusion model. Then, the control action with lower risk and smaller adjustment range is selected.

[0087] Alternatively, a gradient optimization approach can be used, where the control variables are treated as variables to be optimized, and the adjustment amount of the control parameters is solved by iteratively optimizing the gradient of the objective function; or a model predictive control approach can be used, where the control sequence for several future steps is solved within a rolling time window, and only the first control action is executed; or a rule fusion approach can be used, which combines the risk value output by the conditional diffusion model with expert rules to determine the adjustment amount of the control parameters.

[0088] Step S15: Convert the control parameter adjustment amount into a control command, and use the control command to control the combustion state of the waste incinerator.

[0089] In this embodiment, the control parameter adjustment amount is converted into a control setpoint, and a control command is generated based on the control setpoint. The control command is then subjected to a security check. If the security check fails, an alarm message or control suggestion is generated and sent to the client. If the security check passes, the control command is sent to the operation control system of the waste incinerator, so that the operation control system can control the combustion state of the waste incinerator based on the control command. It is then determined whether the current condition for the preset control termination is met. If the current condition for the preset control termination is not met, the process of acquiring the real-time operation data of the waste incinerator is repeated until the preset control termination condition is triggered, thereby completing the rolling control of the waste incinerator.

[0090] After the risk constraint control optimization module outputs the control parameter adjustment values, it converts these adjustments into control setpoints that can be recognized by the DCS or PLC. For example: ; ; ; ; .

[0091] Before issuing control commands, a safety check is performed. The safety check may include: determining whether the control setpoint exceeds the equipment's allowable range; determining whether the control change rate exceeds the allowable upper limit; determining whether the control command conflicts with the existing interlock protection logic; determining whether the control command triggers the manual confirmation condition; and determining whether the waste incinerator is under maintenance, start-up, shutdown, or in an abnormal interlock state.

[0092] If a control command fails the safety verification, it will not be issued directly. Instead, an alarm message will be output, or a control suggestion will be sent to the client and submitted for confirmation by the operator. If the control command passes the safety verification, it will be issued to the operation control system of the waste incinerator, which will then adjust the corresponding control parameters according to the command.

[0093] This embodiment supports suggestion mode, semi-automatic mode, and automatic mode. In suggestion mode, only control suggestions are output, which are then executed after confirmation by the operator. In semi-automatic mode, low-risk, small-amplitude control actions are executed automatically, while high-risk or large-amplitude control actions require manual confirmation. In automatic mode, control commands are automatically sent to the DCS or PLC within safety constraints.

[0094] After the control command is executed, real-time operating data continues to be collected in the next control cycle, and the control effect is evaluated. When there is a significant deviation between the actual operating state and the candidate combustion state trajectory, the current condition vector is updated, the candidate combustion state trajectory is regenerated, and the risk value is recalculated. If the preset control termination condition is not met, the process of real-time data acquisition, current condition vector construction, candidate combustion state trajectory generation, risk calculation, control parameter solving, safety verification, and control command issuance is repeated, thus forming a rolling closed-loop control.

[0095] In one specific implementation, when both the furnace temperature risk value and the carbon monoxide risk value exceed the corresponding threshold, there may be a risk of insufficient furnace temperature and incomplete combustion in the future. In this case, the grate speed can be appropriately reduced to extend the residence time of waste in the combustion zone; the ratio of primary and secondary air can be optimized to improve burnout; the feeding rhythm can be appropriately adjusted to reduce the concentration of wet waste entering the furnace; and, if necessary, the secondary air volume can be adjusted to improve furnace disturbance and mixing.

[0096] In another specific implementation, when the risk value of nitrogen oxides is higher than the corresponding threshold and the risk value of carbon monoxide is lower, the amount of ammonia injected into the SNCR can be adjusted, the secondary air volume and injection position can be optimized, the furnace temperature control target can be appropriately adjusted, and excessive air supply can be avoided to prevent nitrogen oxides from rising.

[0097] In another specific implementation, when the risk value of main steam flow fluctuation is higher than the corresponding threshold, the feed rate change can be smoothed, the sudden change of grate speed can be limited, the slope of air volume change can be adjusted, and the load setting change can be gradually processed.

[0098] The above embodiments generate multiple candidate combustion state trajectories within a future time window through a conditional diffusion model, transforming the future combustion state from a single prediction result into multiple possible paths; based on the multiple candidate combustion state trajectories, the risk values ​​of furnace temperature limit, pollutant exceedance, and main steam flow fluctuation are statistically analyzed, enabling the prediction results to be used for risk constraint control; based on the risk values, equipment and process constraints, and control action smoothing requirements, the control parameter adjustment amount is solved, and closed-loop control is formed through safety verification and rolling feedback.

[0099] This application addresses the core challenges of waste incineration processes, namely "uncertainty of future states, scarcity of abnormal samples, strong coupling of multiple variables, and conflicting control objectives," by constructing a novel intelligent control mechanism capable of multi-trajectory probability prediction and risk constraint control.

[0100] To address the problem that existing data-driven models such as neural networks, support vector machines, random forests, and long short-term memory networks often rely on single-point predictions and struggle to represent future uncertainties, this invention utilizes the multi-trajectory generation capability of the conditional diffusion model to transform traditional single-prediction results into probabilistic predictions of multiple possible future states. Specifically, this invention uses the conditional diffusion model to learn the dynamic relationships between multiple variables in historical operating data, including furnace temperature, feed rate, grate speed, primary air volume, secondary air volume, flue gas oxygen content, carbon monoxide concentration, nitrogen oxide concentration, and main steam flow rate. Using current and historical operating conditions, load setpoints, estimated waste calorific value, and current control parameters as conditional information, it generates multiple candidate combustion state trajectories within the future prediction time window. Unlike traditional single-point prediction methods, this solution can provide multiple possible future operating paths and further calculate the probability of furnace temperature falling below a preset threshold, the probability of pollutant exceedances, the risk of main steam flow fluctuations, and the risk of combustion stability. In other words, this solution not only focuses on "what the future predicted value will be," but also on "how likely instability or exceedances will occur in the future."

[0101] At the control level, this solution optimizes control parameters such as feed rate, grate speed, primary air volume, secondary air volume, air volume ratio, and SNCR ammonia injection rate based on risk assessment results formed from multiple predicted trajectories. This allows the control system to intervene in advance before risks actually occur. For example, when the model predicts an increased probability of the furnace temperature falling below the safety threshold within the next few minutes, and simultaneously increases the risk of a sudden rise in carbon monoxide concentration, the system can adjust the grate speed and secondary air ratio in advance, rather than waiting until the furnace temperature has significantly decreased or the carbon monoxide level has exceeded the limit before taking remedial action. In this way, this solution transforms traditional "post-event corrective control" into "pre-emptive control," converts single deterministic prediction into multi-trajectory probabilistic prediction, and further transforms the prediction results into executable risk constraint control strategies.

[0102] Therefore, the technical solution proposed in this application is mainly to address the problems of insufficient predictive expression ability, weak adaptability to abnormal operating conditions, inability of single-point prediction to support risk decision-making, and insufficient multi-objective collaborative control ability of existing waste incinerator combustion prediction and control methods. By introducing the multi-trajectory generation capability of the conditional diffusion model, this application can more fully describe the uncertainty of future combustion state, and uses the risks of furnace temperature lower limit, pollutant exceedance, and steam load fluctuation as control constraints, providing a more robust technical path for waste incinerators to achieve stable combustion, collaborative pollutant control, and stable power generation load operation under complex waste composition and calorific value fluctuation conditions.

[0103] The core technical point of this application is: 1. Multi-trajectory probability prediction mechanism: The system generates multiple candidate combustion state trajectories within a future time window using a conditional diffusion model, instead of outputting only a single prediction value, enabling the system to express the uncertainty of the future operating state of the waste incinerator.

[0104] 2. Conditional diffusion model for furnace combustion prediction: The conditional diffusion model is introduced into the combustion state prediction scenario of waste incinerator. Historical operating conditions, current control parameters, estimated calorific value of waste and load setpoint are used as conditional information to generate a future combustion state trajectory that meets the constraints of the current operating conditions.

[0105] 3. Transforming “predicted values” into “risk probabilities”: Based on the statistical analysis of multiple candidate trajectories, the probability of furnace temperature falling below the threshold, carbon monoxide exceeding the limit, nitrogen oxide exceeding the limit, and main steam flow fluctuation is calculated, transforming traditional prediction results into risk indicators that can be used for control decisions.

[0106] 4. Furnace temperature low limit risk constraint control: The probability of the furnace temperature falling below the preset threshold is used as a control constraint. The feed rate, grate speed, primary air volume and secondary air volume are adjusted in advance to avoid lag compensation after the furnace temperature drops significantly.

[0107] 5. Pollutant Exceedance Risk Constraint Control: The probability of future exceedances of pollutants such as carbon monoxide and nitrogen oxides is used as a control constraint. The air volume ratio, grate speed and SNCR ammonia injection amount are optimized in advance to achieve predictive control of pollutant emissions.

[0108] 6. Steam load fluctuation risk constraint control: The future fluctuation risk of main steam flow is included in the control target, so that combustion control not only focuses on furnace temperature and pollutants, but also takes into account the stability of power generation load.

[0109] 7. Robust control for fluctuating calorific value of waste: By constructing conditional inputs based on estimated calorific value of waste, estimated moisture content of waste, weight of waste grab bucket, or image recognition features of waste, the model's adaptability to complex waste composition, sudden changes in calorific value, and uneven feeding conditions is improved.

[0110] 8. Abnormal operating condition candidate trajectory generation: Utilizing the generation capability of the diffusion model, candidate future trajectories are generated under low-sample, high-risk operating conditions such as rapid furnace temperature drop, sudden increase in carbon monoxide, sudden change in waste calorific value, and feed blockage, thereby enhancing the system's ability to identify and warn of low-frequency abnormal operating conditions.

[0111] 9. Prediction-Evaluation-Control Closed Loop: Establish a complete closed-loop control process of "condition diffusion prediction - risk indicator calculation - risk constraint optimization - control command output - feedback update" to realize the transformation from ex-post correction to advance prediction control.

[0112] 10. Multi-objective collaborative optimization control: Under the same control framework, furnace temperature stability, pollutant emission compliance, main steam flow stability and control action range are considered simultaneously to avoid the deterioration of other operating indicators due to the optimization of a single indicator.

[0113] 11. Smooth Control Action Constraints: In the risk constraint control optimization, constraints on the adjustment range of control parameters and the change range of adjacent cycles are added to reduce the secondary disturbances caused by frequent fluctuations in feed, air volume, grate speed and ammonia injection.

[0114] 12. Compatible with existing DCS / PLC systems: This application can be used as an upper-level intelligent predictive control module, operating in parallel with existing DCS, PLC, PID or model predictive control systems, outputting control suggestions or control setpoints, facilitating engineering transformation and implementation.

[0115] To address the weakness of existing methods in adapting to abnormal, boundary, and limited-sample disturbance conditions, this invention utilizes the conditional diffusion model's ability to generate and characterize complex disturbance conditions. This allows the model to learn the distribution characteristics of different combustion states based on historical data and generate corresponding candidate future trajectories for scenarios such as sudden changes in waste calorific value, uneven feeding, rapid decreases in furnace temperature, and sudden increases in CO. This approach enhances the system's ability to identify low-frequency, high-risk conditions, enabling control strategies to move beyond reliance on empirical rules under normal stable conditions and better address low-frequency but highly damaging abnormal disturbances encountered in actual operation.

[0116] To address the challenge of existing combustion optimization control methods simultaneously achieving stable combustion, pollutant control, and stable power generation load, this invention introduces a risk-constrained control mechanism based on conditional diffusion prediction results. The system comprehensively determines adjustments to control parameters such as feed rate, grate speed, primary air volume, secondary air volume, air-to-weight ratio, and SNCR ammonia injection rate based on the risks of lower furnace temperature limits, pollutant exceedances, and steam load fluctuations calculated from multiple candidate trajectories. This control mechanism aims to reduce future risks and constrains the magnitude of control actions, ensuring that the control results can both proactively suppress furnace temperature drops and pollutant exceedances while avoiding frequent or excessive adjustments that could cause new disturbances to the stable operation of the incinerator.

[0117] In summary, this invention addresses the problem of existing combustion prediction methods' inability to express future uncertainties through a novel prediction generation method called "conditional diffusion model"; it solves the problem of traditional single-point prediction's difficulty in assessing operational risks through a novel prediction result format called "multiple candidate combustion state trajectories"; and it addresses the problems of slow response and insufficient multi-objective coordination capabilities in existing control methods through a novel control decision-making mechanism called "risk constraint control". Therefore, this invention achieves a shift from post-event corrective control to proactive predictive control, from deterministic single-point prediction to probabilistic risk prediction, and from single-index adjustment to multi-objective collaborative optimization control.

[0118] This technology can be applied to municipal solid waste incineration power plants, industrial solid waste incineration systems, and other incineration control scenarios with fluctuations in fuel composition and constraints on pollutant emissions. It is particularly suitable for intelligent waste incineration power generation systems that have high requirements for furnace temperature stability, pollutant emission compliance rate, main steam load stability, and automation control level.

[0119] See Figure 2 As shown in the figure, an embodiment of the present invention discloses a combustion state control device for a waste incinerator, which may specifically include: Preprocessing module 11 is used to acquire historical operating data of the waste incinerator, preprocess the historical operating data, and use the preprocessed data to construct training samples and condition vectors; the historical operating data includes state variables, control variables and external disturbance variables; Training module 12 is used to add noise to the combustion state trajectory in the training sample to obtain a noisy trajectory, input the noisy trajectory and the conditional vector into the denoising network in the conditional diffusion model to output predicted noise, and train the conditional diffusion model based on the predicted noise and the combustion state trajectory to obtain the trained conditional diffusion model. The trajectory determination module 13 is used to acquire real-time operating data of the waste incinerator, construct a current condition vector using the real-time operating data, and input the current condition vector into the trained conditional diffusion model to output candidate combustion state trajectories. The solution module 14 is used to construct an optimization objective function based on the candidate combustion state trajectory, solve the optimization objective function, and obtain the control parameter adjustment amount. Combustion state control module 15 is used to convert the control parameter adjustment amount into control commands, and use the control commands to control the combustion state of the waste incinerator.

[0120] In some specific embodiments, the preprocessing module 11 may specifically include: The historical operation data acquisition module is used to acquire historical operation data of the waste incinerator from the operation control system, flue gas emission monitoring system, equipment control system, and manual detection data of the waste incineration power plant; the historical operation data includes state variables, control variables, and external disturbance variables; The data processing module is used to perform time alignment, outlier handling, missing value handling, smoothing, and normalization on the historical running data to obtain preprocessed data.

[0121] In some specific embodiments, the preprocessing module 11 may specifically include: The training sample construction module is used to extract combustion state trajectories from the preprocessed data in a sliding window manner, and to construct training samples using the combustion state trajectories. The sequence and feature construction module is used to construct historical state sequences, historical control sequences, and perturbation features based on the preprocessed data. The condition vector construction module is used to construct a condition vector using the historical state sequence, the historical control sequence, and the perturbation features.

[0122] In some specific embodiments, the training module 12 may specifically include: The real noise acquisition module is used to randomly sample from a standard Gaussian distribution to obtain real noise; The forward noise module is used to add forward noise to the combustion state trajectory in the training sample based on preset noise scheduling parameters and the real noise, so as to obtain a noisy trajectory.

[0123] In some specific embodiments, the training module 12 may specifically include: The diffusion step number acquisition module is used to randomly sample a preset set of diffusion steps to obtain the diffusion step number; The prediction noise output module is used to input the diffusion step number, noisy trajectory, and conditional vector into the denoising network in the conditional diffusion model to output the prediction noise. The total loss function construction module is used to construct the total loss function based on the predicted noise and combustion state trajectory; The conditional diffusion model training module is used to train the denoising network in the conditional diffusion model with the goal of minimizing the total loss function, so as to obtain the trained conditional diffusion model.

[0124] In some specific embodiments, the solving module 14 may specifically include: The trajectory trimming module is used to perform inverse normalization processing and engineering boundary trimming on the candidate combustion state trajectory to obtain the trimmed candidate combustion state trajectory. The risk value calculation module is used to calculate the furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value based on the trimmed candidate combustion state trajectory. The comprehensive risk value determination module is used to weight the furnace temperature risk value, the carbon monoxide risk value, the nitrogen oxide risk value, and the main steam flow fluctuation risk value to obtain a comprehensive risk value for combustion stability. An optimization objective function construction module is used to construct an optimization objective function using the comprehensive risk value of combustion stability.

[0125] In some specific embodiments, the combustion state control module 15 may specifically include: The control command generation module is used to convert control parameter adjustment amounts into control setpoints and generate control commands based on the control setpoints. A security verification module is used to perform security verification on the control commands; The failure handling module is used to generate and send alarm information or control suggestions to the client if the security check fails. The control command issuing module is used to issue the control command to the operation control system of the waste incinerator if the safety verification passes, so that the operation control system can control the combustion status of the waste incinerator based on the control command. The termination condition judgment module is used to determine whether the preset control termination condition is met. The rolling control module is used to repeatedly execute the process of acquiring real-time operating data of the waste incinerator if the preset control termination condition is not met, until the preset control termination condition is triggered, so as to complete the rolling control of the waste incinerator.

[0126] Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the combustion state control method for a waste incinerator performed by an electronic device as disclosed in any of the foregoing embodiments.

[0127] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0128] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0129] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the data 223 in the memory 22. It can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the combustion state control method for the waste incinerator executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the combustion state control device of the waste incinerator from external devices, and may also include data collected by its own input / output interface 25.

[0130] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0131] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the combustion state control method steps of the waste incinerator disclosed in any of the foregoing embodiments.

[0132] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0133] The present invention provides a detailed description of a method, apparatus, equipment, and medium for controlling the combustion state of a waste incinerator. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for controlling the combustion state of a waste incinerator, characterized in that, include: Historical operating data of the waste incinerator is obtained, the historical operating data is preprocessed, and training samples and conditional vectors are constructed using the preprocessed data. The historical operating data includes state variables, control variables, and external disturbance variables; The combustion state trajectory in the training sample is forward-noised to obtain a noisy trajectory. The noisy trajectory and the conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and the combustion state trajectory, the conditional diffusion model is trained to obtain the trained conditional diffusion model. The real-time operation data of the waste incinerator is obtained, and the current condition vector is constructed using the real-time operation data. The current condition vector is then input into the trained conditional diffusion model to output the candidate combustion state trajectory. Based on the candidate combustion state trajectory, an optimization objective function is constructed, and the optimization objective function is solved to obtain the control parameter adjustment amount; The control parameter adjustment amount is converted into a control command, and the combustion state of the waste incinerator is controlled by the control command. Based on the predicted noise and combustion state trajectory, the conditional diffusion model is trained to obtain the trained conditional diffusion model, including: randomly sampling a preset diffusion step set to obtain the diffusion step number; The diffusion step number, noisy trajectory, and conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and combustion state trajectory, a total loss function is constructed. The denoising network in the conditional diffusion model is trained with the goal of minimizing the total loss function to obtain the trained conditional diffusion model.

2. The method for controlling the combustion state of a waste incinerator according to claim 1, characterized in that, The acquisition of historical operating data of the waste incinerator and the preprocessing of the historical operating data include: Historical operating data of the waste incinerator is obtained from the operation control system, flue gas emission monitoring system, equipment control system, and manual detection data of the waste incineration power plant; the historical operating data includes state variables, control variables, and external disturbance variables. The historical running data is subjected to time alignment, outlier handling, missing value handling, smoothing, and normalization to obtain preprocessed data.

3. The method for controlling the combustion state of a waste incinerator according to claim 1, characterized in that, The process of constructing training samples and conditional vectors using preprocessed data includes: Using a sliding window approach, combustion state trajectories are extracted from the preprocessed data, and training samples are constructed using these combustion state trajectories. Based on the preprocessed data, historical state sequences, historical control sequences, and disturbance features are constructed. A condition vector is constructed using the historical state sequence, the historical control sequence, and the perturbation features.

4. The method for controlling the combustion state of a waste incinerator according to claim 1, characterized in that, Adding noise to the combustion state trajectories in the training samples to obtain noisy trajectories includes: The true noise is obtained by randomly sampling from a standard Gaussian distribution; Based on preset noise scheduling parameters, and using the real noise to add noise to the combustion state trajectory in the training sample, a noisy trajectory is obtained.

5. The method for controlling the combustion state of a waste incinerator according to claim 1, characterized in that, Based on the candidate combustion state trajectories, an optimization objective function is constructed, including: The candidate combustion state trajectory is inversely normalized and the engineering boundary is clipped to obtain the clipped candidate combustion state trajectory. Based on the trimmed candidate combustion state trajectories, the furnace temperature risk value, carbon monoxide risk value, nitrogen oxide risk value, and main steam flow fluctuation risk value are calculated respectively. The risk values ​​of furnace temperature, carbon monoxide, nitrogen oxides, and main steam flow fluctuation are weighted to obtain a comprehensive risk value for combustion stability. An optimization objective function is constructed using the comprehensive risk value of combustion stability.

6. The method for controlling the combustion state of a waste incinerator according to any one of claims 1 to 5, characterized in that, The control parameter adjustment is converted into a control command, and the combustion state of the waste incinerator is controlled using the control command, including: The control parameter adjustment amount is converted into a control set value, and a control command is generated based on the control set value; The control commands are subject to security verification. If the security check fails, an alarm message or control suggestion will be generated and sent to the client. If the safety verification passes, the control command is sent to the operation control system of the waste incinerator so that the operation control system can control the combustion status of the waste incinerator based on the control command. Determine whether the preset control termination condition is met; If the preset control termination condition is not met, the process of acquiring real-time operating data of the waste incinerator will be repeated until the preset control termination condition is triggered to complete the rolling control of the waste incinerator.

7. A combustion state control device for a waste incinerator, characterized in that, include: The preprocessing module is used to acquire historical operating data of the waste incinerator, preprocess the historical operating data, and use the preprocessed data to construct training samples and condition vectors. The historical operating data includes state variables, control variables, and external disturbance variables; The training module is used to add noise to the combustion state trajectory in the training samples to obtain a noisy trajectory, input the noisy trajectory and the conditional vector into the denoising network in the conditional diffusion model to output predicted noise, and train the conditional diffusion model based on the predicted noise and the combustion state trajectory to obtain the trained conditional diffusion model. The trajectory determination module is used to acquire real-time operating data of the waste incinerator, construct a current condition vector using the real-time operating data, and input the current condition vector into the trained conditional diffusion model to output candidate combustion state trajectories. The solution module is used to construct an optimization objective function based on the candidate combustion state trajectory, solve the optimization objective function, and obtain the control parameter adjustment amount; The combustion state control module is used to convert the control parameter adjustment amount into control commands, and use the control commands to control the combustion state of the waste incinerator. Based on the predicted noise and combustion state trajectory, the conditional diffusion model is trained to obtain the trained conditional diffusion model, including: randomly sampling a preset diffusion step set to obtain the diffusion step number; The diffusion step number, noisy trajectory, and conditional vector are input into the denoising network in the conditional diffusion model to output predicted noise. Based on the predicted noise and combustion state trajectory, a total loss function is constructed. The denoising network in the conditional diffusion model is trained with the goal of minimizing the total loss function to obtain the trained conditional diffusion model.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the combustion state control method for a waste incinerator as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the combustion state control method for a waste incinerator as described in any one of claims 1 to 6.

Images