A Smart Control Method for Desulfurization Systems in Coal-fired Power Plants Based on Mamdani Fuzzy Neural Network
By adopting an intelligent control method based on Mamdani fuzzy neural network, the problems of control lag and untimely disturbance response in wet desulfurization systems under wide load and variable coal quality conditions were solved, thus achieving efficient and stable operation and energy-saving goals for desulfurization systems in coal-fired power plants.
Patent Information
- Application Number
- CN202611443022.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-17
- Publication Date
- 2026-11-10
AI Technical Summary
Existing wet desulfurization systems suffer from control lag, untimely disturbance response, and difficulty in coordinating multiple actuators under wide load peak shaving and variable coal quality conditions, resulting in large fluctuations in SO2 concentration and making it difficult to simultaneously achieve emission standards and energy-saving operation of circulating pumps.
An intelligent control method based on Mamdani fuzzy neural network is adopted. By constructing and training a fuzzy neural network model, the disturbances of flue gas and slurry are identified in real time, multi-parameter collaborative optimization is performed, control quantities are generated and feedforward compensation is applied, and combined with safety threshold constraints, the real-time adjustment of limestone slurry supply, oxidation air volume and slurry circulation pump combination mode is realized.
It improves control precision and response speed, ensures stable SO2 concentration compliance, realizes multi-parameter coordinated adjustment and energy-saving operation, reduces control lag and untimely disturbance response, and ensures system safety and stability.
Smart Images

Figure CN123067166A_ABST
Abstract
Description
Technical Field
[0001] The embodiments disclosed herein relate to the field of flue gas desulfurization control technology for coal-fired power plants, and particularly to an intelligent control method, system, electronic device, and storage medium for a desulfurization system of a coal-fired power plant based on a Mamdani fuzzy neural network. Background Technology
[0002] With increasingly stringent emission standards for flue gas pollutants from coal-fired power plants, wet flue gas desulfurization (FGD) systems, as core end-of-pipe treatment devices in coal-fired power plants, have a significant impact on the unit's compliance with emission standards due to their control accuracy and response speed. At the same time, coal-fired power plants generally face the need for wide-load peak-shaving operation, requiring FGD systems to maintain stable desulfurization efficiency across a wide range of loads and under varying coal quality conditions, thus placing higher demands on the adaptive capabilities of the control system.
[0003] Existing wet desulfurization systems generally employ single-loop or cascade control schemes based on fixed PID parameters, using slurry pH or outlet SO2 concentration as the single controlled variable, and achieving closed-loop control by adjusting the limestone slurry supply. However, the inherent large time delay and strong nonlinearity of desulfurization systems make it difficult for fixed-parameter PID controllers to maintain the expected control performance under conditions of frequent load changes or sudden changes in coal quality. Furthermore, there is a significant parameter coupling relationship between limestone slurry supply, oxidation air volume, and slurry circulation pump operation mode, making it impossible for single-loop control schemes to achieve coordinated optimization of multiple parameters.
[0004] The inventors of this disclosure have noted that current wet desulfurization control schemes do not take into account the real-time identification and feedforward compensation of multi-source disturbances on the flue gas side and slurry side, nor do they incorporate the limestone slurry supply, oxidation air volume and slurry circulation pump combination into a unified multi-parameter collaborative optimization framework. This results in large fluctuations in outlet SO2 concentration under wide load switching and coal quality disturbance superposition scenarios, making it difficult to simultaneously achieve the dual goals of meeting emission standards and energy-saving operation of circulation pumps. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent control method for desulfurization systems in coal-fired power plants based on Mamdani fuzzy neural networks, which solves the technical problems of control lag, untimely disturbance response, and difficulty in coordinating multiple actuators caused by multi-parameter coupling in wet desulfurization systems of coal-fired power plants.
[0006] This disclosure provides at least one embodiment of an intelligent control method for a desulfurization system in a coal-fired power plant based on a Mamdani fuzzy neural network. The method includes: constructing and training a Mamdani fuzzy neural network model, using the inlet flue gas flow rate of the desulfurization tower, the SO2 concentration of the raw flue gas, the density of the limestone slurry, the pH value of the slurry, and the composition of the slurry as input variables, and the outlet SO2 concentration as the control target output. The Mamdani fuzzy neural network model includes a fuzzification layer, a rule inference layer, an activation intensity calculation layer, a normalization layer, and a defuzzification layer. Based on the real-time predicted output of the Mamdani fuzzy neural network model, the limestone slurry supply rate, the oxidation air flow rate, and the combination mode of the slurry circulation pump are optimized in parallel within the same control cycle to generate three... The control quantities of the three types of control parameters are: control quantities of the three types of control parameters; based on real-time sampling data of flue gas-side disturbance and slurry-side disturbance, the disturbance intensity level is identified by the Mamdani fuzzy neural network model, and feedforward compensation correction is applied to the control quantities of the three types of control parameters; the control quantities and the feedforward compensation correction are combined into a final control command, which is sent to each actuator through the command channel of the desulfurization DCS system, and a safety threshold constraint is applied to each control quantity. The flue gas-side disturbance and the slurry-side disturbance are added as additional input dimensions to the input vector of the Mamdani fuzzy neural network model, and participate in fuzzification and rule reasoning calculation together with the five types of input variables. The feedforward compensation correction is output by the independent output channel of the defuzzification layer.
[0007] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, the construction and training of Mamdani fuzzy neural network model includes: offline training of the Mamdani fuzzy neural network model using historical operating data, using the mean square error between the measured and predicted values of outlet SO2 concentration as the loss function, and iteratively updating the network weights and membership function parameters through backpropagation algorithm combined with Adam optimization algorithm; after offline training is completed, online updating of the Mamdani fuzzy neural network model using incremental learning method based on real-time operating data collected by the desulfurization DCS system, and simultaneously adaptively correcting the fuzzy rule base of the rule reasoning layer and the membership parameters of the defuzzification layer; and using the relative error between the predicted and measured values of outlet SO2 concentration as the accuracy evaluation index, triggering a retraining process when the prediction accuracy is lower than 95%.
[0008] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, the defuzzification layer uses the centroid method to calculate the accurate predicted value of SO2 concentration at the outlet. The expression is:
[0009] in, This represents the total number of fuzzy rules. For rule indexing, For the first The normalized trigger strength of the rule output by the normalization layer. For the first Each rule corresponds to the centroid value of the output fuzzy set.
[0010] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, the real-time optimization of limestone slurry supply, oxidation air volume, and slurry circulation pump combination includes: based on the deviation between the real-time sampled value of inlet SO2 concentration and the set value of outlet SO2 concentration, the Mamdani fuzzy neural network model outputs the slurry supply valve opening control quantity, which is then sent to the slurry supply valve actuator via the DCS command channel; based on the real-time monitoring data of sulfite ion concentration, with minimizing the sulfite ion concentration as the optimization objective, the Mamdani fuzzy neural network model outputs the oxidation fan inlet valve opening control quantity, which is then sent to the oxidation fan inlet valve actuator via the DCS command channel; and based on the real-time sampled values of current unit load, slurry density, and liquid-gas ratio, the Mamdani fuzzy neural network model determines the optimal number of circulating pumps in operation and the start-stop combination, and the corresponding start-stop control commands are sent to each circulating pump control loop via the DCS command channel.
[0011] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, the identification of disturbance intensity level includes: using the unit time change rate of inlet flue gas flow rate and the unit time change rate of original flue gas SO2 concentration as flue gas-side disturbance feature quantities, inputting them into the disturbance identification channel of the Mamdani fuzzy neural network model, and outputting the flue gas-side disturbance intensity level; and using the unit time change rate of slurry density and the unit time change rate of pH value as slurry-side disturbance feature quantities, inputting them into the disturbance identification channel of the Mamdani fuzzy neural network model, and outputting the slurry-side disturbance intensity level.
[0012] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, feedforward compensation correction is applied to the control amount of slurry supply regulating valve opening, the control amount of oxidation blower inlet regulating valve opening, and the combination mode of circulation pump start-stop according to the disturbance intensity level of flue gas side and the disturbance intensity level of slurry side, so that the control parameter adjustment action is completed before the disturbance effect is reflected in the outlet SO2 concentration and pH value.
[0013] For example, in the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure, the step of applying safety threshold constraints to each control quantity includes: before the final control command is issued, comparing the control quantity of slurry supply regulating valve opening, the control quantity of oxidation blower inlet regulating valve opening, and the combination mode of circulation pump start-stop with preset upper and lower limit safety thresholds respectively; and if any control quantity exceeds the corresponding safety threshold, then the control quantity is truncated to the safety threshold boundary before being issued to the actuator. The safety threshold is fixed in the form of hard coding in the desulfurization DCS system and is not affected by the online update process of the Mamdani fuzzy neural network model.
[0014] At least one embodiment of this disclosure also provides an intelligent control system for a desulfurization system of a coal-fired power plant based on a Mamdani fuzzy neural network. The intelligent control system includes at least one processing node, a model training and updating unit, a multi-parameter optimization unit, a disturbance identification and feedforward compensation unit, and a safety threshold control unit. The model training and updating unit is configured to construct and train a Mamdani fuzzy neural network model using the inlet flue gas flow rate of the desulfurization tower, the original flue gas SO2 concentration, the limestone slurry density, the slurry pH value, and the slurry composition as input variables, and the outlet SO2 concentration as the control target output. The Mamdani fuzzy neural network model includes a fuzzification layer, a rule inference layer, an activation intensity calculation layer, a normalization layer, and a defuzzification layer. The multi-parameter optimization unit is configured to... The system generates real-time predictive outputs and performs real-time optimization of limestone slurry supply, oxidation air volume, and slurry circulation pump combination in parallel within the same control cycle, generating control quantities for three types of control parameters. The disturbance identification and feedforward compensation unit is configured to identify the disturbance intensity level by the Mamdani fuzzy neural network model based on real-time sampled data of flue gas-side disturbance and slurry-side disturbance, and apply feedforward compensation correction to the control quantities of the three types of control parameters. The flue gas-side disturbance and the slurry-side disturbance are added as additional input dimensions to the input vector of the Mamdani fuzzy neural network model, and the feedforward compensation correction is output by the independent output channel of the defuzzification layer. The safety threshold control unit is configured to merge the control quantities and the feedforward compensation correction into a final control command, which is sent to each actuator through the command channel of the desulfurization DCS system, and a safety threshold constraint is applied to each control quantity.
[0015] At least one embodiment of this disclosure also provides an electronic device, including: at least one memory storing computer-executable instructions non-transiently; and at least one processor configured to run the computer-executable instructions, wherein the computer-executable instructions are executed by the processor to implement the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to any embodiment of this disclosure.
[0016] At least one embodiment of this disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer-executable instructions, which, when executed by at least one processor, implement the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network described in any embodiment of this disclosure.
[0017] This invention integrates the Mamdani fuzzy neural network model with the reasoning ability of fuzzy logic rules and the self-learning ability of neural networks, effectively characterizing the nonlinear characteristics and large hysteresis response characteristics of the desulfurization system. It solves the technical problem of performance degradation of fixed-parameter controllers under wide load and variable coal quality conditions, achieving the technical effect that the model prediction accuracy can sustainably meet the requirements of actual operating conditions. By optimizing the combination of limestone slurry supply, oxidation air volume, and slurry circulation pump in parallel within the same control cycle, it solves the technical problem of temporal coupling error introduced by sequential optimization of multiple parameters, achieving the technical effect of coordinated adjustment of three types of control parameters and convergence of outlet SO2 concentration to the set value. By identifying disturbances on the flue gas side and slurry side in real time and applying feedforward compensation correction, it solves the technical problem of delayed control response caused by the lag in the manifestation of disturbance effects on outlet SO2 concentration and pH value, achieving the technical effect that control parameter adjustment actions are executed before disturbance effects. By applying safety threshold truncation constraints to each control quantity before the DCS command is issued, it solves the safety risk problem of potentially outputting over-limit commands during the online update of the intelligent control algorithm, achieving the technical effect that the control quantities of each actuator are always within the preset safety range. Attached Figure Description
[0018] Figure 1 This is a flowchart of an intelligent control method for a coal-fired power plant desulfurization system based on a Mamdani fuzzy neural network, provided in an embodiment of the present invention. Figure 2 This is a block diagram of the Mamdani fuzzy neural network model structure of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in the embodiments of the present invention; Figure 3 This is a block diagram of the multi-parameter collaborative closed-loop control structure of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in the embodiments of the present invention. Detailed Implementation
[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.
[0020] like Figures 1-3 As shown, the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure includes the following steps S100 to S400.
[0021] Step S100: Construction and training of the Mamdani fuzzy neural network model.
[0022] In step S100, the inlet flue gas flow rate of the desulfurization tower, the SO2 concentration of the raw flue gas, the density of the limestone slurry, the pH value of the slurry, and the composition of the slurry are used as input variables, and the outlet SO2 concentration is used as the control target output quantity. A Mamdani fuzzy neural network model is constructed. The Mamdani fuzzy neural network model integrates the rule reasoning ability of fuzzy logic and the self-learning ability of neural networks to accurately characterize the nonlinear characteristics and large hysteresis response characteristics of the desulfurization system.
[0023] For example, the Mamdani fuzzy neural network model includes a fuzzification layer, a rule inference layer, an activation intensity calculation layer, a normalization layer, and a defuzzification layer. The data transmission relationship between each layer is as follows: The fuzzification layer receives real-time sampled values of the above five types of input variables and maps each input variable to the corresponding membership function output; the rule inference layer performs logical operations on each input membership degree according to the preset fuzzy rule library to generate the activation conditions of each rule; the activation intensity calculation layer calculates the trigger intensity of each fuzzy rule; the normalization layer normalizes the trigger intensity of each rule; and the defuzzification layer uses the centroid method to convert the fuzzy inference results into an accurate predicted value of the outlet SO2 concentration, which serves as the benchmark input for subsequent control parameter optimization.
[0024] Furthermore, the deblurring layer uses the centroid method to calculate the accurate predicted value of the outlet SO2 concentration. The specific expression is: ; in, This represents the total number of fuzzy rules. For rule indexing, , For the first The normalized trigger strength of the rule output by the normalization layer. For the first Each rule corresponds to the centroid value of the output fuzzy set. The Mamdani fuzzy neural network model uses the predicted value of the outlet SO2 concentration. The output layer is the precise value output channel of the aforementioned deblurring layer.
[0025] In at least one example of the embodiments of this disclosure, the training process of the Mamdani fuzzy neural network model includes: Offline training of the Mamdani fuzzy neural network model was conducted using historical operating data from the desulfurization system. The historical data included multi-dimensional time-series samples of inlet flue gas flow rate, raw flue gas SO2 concentration, limestone slurry density, pH value, and slurry composition under different load ranges and coal quality conditions. The measured outlet SO2 concentration was used as the basis for the training. As a supervision label, with predicted values Compared with measured values Mean square error between As the loss function, where The number of training samples. For sample index, The network weights and membership function parameters are iteratively updated by combining the backpropagation algorithm with the Adam optimization algorithm. After offline training is completed, the Mamdani fuzzy neural network model is updated online using real-time operating data collected by the DCS of the desulfurization system. The network weights are continuously corrected using an incremental learning method, so that the Mamdani fuzzy neural network model can continuously adapt to the desulfurization operation characteristics under the current load and coal quality conditions. The relative error between the predicted and measured values of SO2 concentration at the outlet by the Mamdani fuzzy neural network model is used as the accuracy evaluation index. The prediction accuracy of the Mamdani fuzzy neural network model is required to be no less than 95%. If this threshold is not reached, a retraining process is triggered.
[0026] It should be noted that during the online update process, the Mamdani fuzzy neural network model adaptively corrects the fuzzy rule base of the rule reasoning layer and the membership parameters of the defuzzification layer simultaneously, rather than only updating the neural network connection weights. This ensures that the matching between the fuzzy rules and the actual working conditions remains effective when switching between different working conditions.
[0027] Step S200: Multi-parameter collaborative closed-loop control based on the Mamdani fuzzy neural network model.
[0028] In step S200, based on the real-time prediction output of the Mamdani fuzzy neural network model, the three core control parameters of limestone slurry supply, oxidation air volume and slurry circulation pump combination mode are optimized in real time. The three parameters are adjusted in a coordinated manner to form a closed-loop control of all parameters.
[0029] For example, the optimization process for the three types of core control parameters is as follows: Limestone slurry supply optimization: Based on the deviation between the real-time sampled value of SO2 concentration at the inlet and the set value of SO2 concentration at the outlet, the Mamdani fuzzy neural network model outputs the control quantity of the slurry supply regulating valve opening. This control quantity is sent to the slurry supply regulating valve actuator through the DCS command channel to adjust the supply flow rate of limestone slurry so that the outlet SO2 concentration converges to the set value. Oxidation air volume optimization: Based on real-time monitoring data of sulfite ion concentration, with the goal of minimizing sulfite ion concentration, the Mamdani fuzzy neural network model outputs the control quantity of the opening of the oxidation fan inlet damper. This control quantity is sent to the oxidation fan inlet damper actuator via the DCS command channel to adjust the oxidation air supply and ensure that sulfite ions in the slurry are fully oxidized. Optimization of slurry circulation pump combination: Based on the real-time sampling values of the current unit load, slurry density, and liquid-to-gas ratio, the optimal number of operating pumps and the start-stop combination of each pump are determined by the Mamdani fuzzy neural network model. The corresponding start-stop control commands are sent to the control loop of each circulation pump through the DCS command channel, so as to achieve energy-saving operation of the circulation pumps while ensuring desulfurization efficiency.
[0030] In at least one example of the embodiments of this disclosure, the optimization calculation of the three types of control parameters is performed in parallel within the same control cycle, and the control output of each parameter is based on the latest prediction value of the Mamdani fuzzy neural network model within the current control cycle, so as to avoid the timing coupling error introduced by the parameter sequential optimization.
[0031] Step S300: Adaptive adjustment for multiple perturbations.
[0032] In step S300, based on the real-time sampling data of the flue gas side disturbance and the slurry side disturbance, the Mamdani fuzzy neural network model automatically identifies the disturbance intensity and adjusts the control quantities of the three types of control parameters in step S200 in real time to offset the impact of the disturbance on the outlet SO2 concentration and slurry pH value in advance.
[0033] For example, the perturbation adaptive adjustment process includes: Flue gas side disturbance identification: The rate of change of inlet flue gas flow rate and the rate of change of original flue gas SO2 concentration are used as flue gas side disturbance characteristics and input into the disturbance identification channel of the Mamdani fuzzy neural network model. The Mamdani fuzzy neural network model outputs the flue gas side disturbance intensity level according to the amplitude and changing trend of the two types of disturbance characteristics. Slurry-side disturbance identification: The rate of change of slurry density and the rate of change of pH value per unit time are used as slurry-side disturbance features and input into the disturbance identification channel of the Mamdani fuzzy neural network model. The Mamdani fuzzy neural network model outputs the slurry-side disturbance intensity level based on the amplitude and changing trend of the two types of disturbance features. Control parameter feedforward correction: Based on the disturbance intensity level output from steps a and b, feedforward compensation correction is applied to the limestone slurry supply valve opening control, oxidation blower inlet valve opening control, and circulating pump start / stop combination in step S200, so that the control parameter adjustment action is completed before the disturbance effect is reflected in the outlet SO2 concentration and pH value.
[0034] In at least one example of the embodiments of this disclosure, under low load conditions, the feedforward compensation correction amount output by the Mamdani fuzzy neural network model causes the control amount of the slurry supply regulating valve opening and the control amount of the oxidation blower inlet regulating valve opening to decrease synchronously, and the number of circulating pumps in operation is reduced accordingly; under high load conditions, the above three types of control amounts are increased synchronously, realizing adaptive adaptation across the entire operating range.
[0035] It should be noted that the flue gas side and slurry side disturbance identification channels share the same rule inference layer of the Mamdani fuzzy neural network model. The disturbance features of the flue gas side and the disturbance features of the slurry side are added as additional input dimensions into the input vector of the Mamdani fuzzy neural network model. Together with the five types of basic input variables defined in step S100, they participate in the fuzzification and rule inference calculation. The feedforward compensation correction is output by the independent output channel of the defuzzification layer, which is independent of the control quantity optimization output channel in step S200 and does not generate computational interference.
[0036] Step S400: DCS system linkage execution and safety threshold control.
[0037] In step S400, the control quantities and feedforward compensation correction quantities output from steps S200 and S300 are combined into a final control command, which is directly sent to each actuator through the command channel of the desulfurization DCS system. This achieves a closed-loop operation of Mamdani fuzzy neural network model calculation, parameter optimization, and command execution. At the same time, safety threshold constraints are applied to each control quantity to ensure that the slurry supply, oxidation air volume, and circulating pump operating status are within the preset safety range.
[0038] For example, the safety threshold control process includes: before the final control command is issued, comparing the control quantity of the slurry supply regulating valve opening, the control quantity of the oxidation blower inlet regulating valve opening, and the combination mode of the circulating pump start / stop with the preset upper and lower safety thresholds respectively; if any control quantity exceeds the corresponding safety threshold, the control quantity is truncated to the safety threshold boundary before being issued to the actuator; the safety threshold is determined by the rated parameters of the desulfurization system equipment and process constraints, and is fixed in the DCS system in the form of hard coding, and is not affected by the online update process of the Mamdani fuzzy neural network model.
[0039] In at least one example of the embodiments of this disclosure, the control algorithm of the Mamdani fuzzy neural network model is embedded in the controller of the desulfurization DCS system in the form of a functional module. The control cycle is synchronized with the scanning cycle of the DCS system. The calculation results of the Mamdani fuzzy neural network model are updated and the command is triggered before the end of each scanning cycle, without the need for manual intervention.
[0040] The intelligent control system for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network provided in at least one embodiment of this disclosure includes at least one DCS controller, a model training and update unit, a multi-parameter optimization unit, a disturbance identification and feedforward compensation unit, and a safety threshold control unit.
[0041] For a detailed explanation of the model training and update unit, please refer to the description of step S100 in the above embodiment of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network; it will not be repeated here. For a detailed explanation of the multi-parameter optimization unit, please refer to the description of step S200 in the above embodiment of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network; it will not be repeated here. For a detailed explanation of the disturbance identification and feedforward compensation unit, please refer to the description of step S300 in the above embodiment of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network; it will not be repeated here. For a detailed explanation of the safety threshold control unit, please refer to the description of step S400 in the above embodiment of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network; it will not be repeated here.
[0042] The technical effect of the intelligent control system for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network is the same as that of the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network mentioned above, and will not be repeated here.
[0043] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.
Claims
1. A smart control method for a desulfurization system of a coal-fired power plant based on a Mamdani fuzzy neural network, characterized in that, include: Using the flue gas flow rate at the inlet of the desulfurization tower, the SO2 concentration of the raw flue gas, the density of the limestone slurry, the pH value of the slurry, and the composition of the slurry as input variables, and the SO2 concentration at the outlet as the control target output, a Mamdani fuzzy neural network model is constructed and trained. The Mamdani fuzzy neural network model includes a fuzzification layer, a rule inference layer, an activation intensity calculation layer, a normalization layer, and a defuzzification layer. Based on the real-time prediction output of the Mamdani fuzzy neural network model, the limestone slurry supply, oxidation air volume and slurry circulation pump combination are optimized in real time within the same control cycle to generate control quantities for three types of control parameters. Based on the real-time sampling data of flue gas-side disturbance and slurry-side disturbance, the Mamdani fuzzy neural network model identifies the disturbance intensity level and applies feedforward compensation correction to the control quantities of the three types of control parameters. as well as The control quantity and the feedforward compensation correction quantity are combined into a final control command, which is then sent to each actuator via the command channel of the desulfurization DCS system. A safety threshold constraint is also applied to each control quantity. The flue gas-side disturbance and the slurry-side disturbance are added as additional input dimensions to the input vector of the Mamdani fuzzy neural network model, and together with the five types of input variables, they participate in fuzzification and rule reasoning calculations. The feedforward compensation correction is output by the independent output channel of the defuzzification layer.
2. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 1, characterized in that, The construction and training of the Mamdani fuzzy neural network model includes: The Mamdani fuzzy neural network model was trained offline using historical operating data. The mean square error between the measured and predicted values of SO2 concentration at the outlet was used as the loss function. The network weights and membership function parameters were iteratively updated by combining the backpropagation algorithm with the Adam optimization algorithm. After offline training, the Mamdani fuzzy neural network model is updated online using incremental learning based on real-time operating data collected by the desulfurization DCS system. Simultaneously, the membership parameters of the fuzzy rule base in the rule inference layer and the defuzzification layer are adaptively corrected. The relative error between the predicted and measured SO2 concentration at the outlet is used as the accuracy evaluation index. When the prediction accuracy is lower than 95%, the retraining process is triggered.
3. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 1, characterized in that, The deblurring layer uses the centroid method to calculate the accurate predicted value of the outlet SO2 concentration. The expression is: ; in, This represents the total number of fuzzy rules. For rule indexing, For the first The normalized trigger strength of the rule output by the normalization layer. For the first Each rule corresponds to the centroid value of the output fuzzy set.
4. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 1, characterized in that, The real-time optimization of the combination of limestone slurry supply, oxidation air volume, and slurry circulation pump includes: Based on the deviation between the real-time sampled value of SO2 concentration at the inlet and the set value of SO2 concentration at the outlet, the Mamdani fuzzy neural network model outputs the control quantity of the slurry supply regulating valve opening, which is then sent to the slurry supply regulating valve actuator via the DCS command channel. Based on real-time monitoring data of sulfite ion concentration, and with minimizing sulfite ion concentration as the optimization objective, the Mamdani fuzzy neural network model outputs the control quantity for the opening of the oxidation fan inlet damper, which is then sent to the oxidation fan inlet damper actuator via the DCS command channel; and Based on the real-time sampled values of the current unit load, slurry density, and liquid-to-gas ratio, the optimal number of circulating pumps and the start-stop combination mode are determined by the Mamdani fuzzy neural network model. The corresponding start-stop control commands are then sent to the control loops of each circulating pump via the DCS command channel.
5. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 1, characterized in that, The identified disturbance intensity levels include: The rate of change of inlet flue gas flow rate and the rate of change of original flue gas SO2 concentration are used as flue gas-side disturbance characteristics, input into the disturbance recognition channel of the Mamdani fuzzy neural network model, and the output is the flue gas-side disturbance intensity level; and The per-unit-time change rate of slurry density and the per-unit-time change rate of pH value are used as slurry-side disturbance characteristics, which are input into the disturbance recognition channel of the Mamdani fuzzy neural network model to output the slurry-side disturbance intensity level.
6. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 5, characterized in that, Based on the flue gas side disturbance intensity level and the slurry side disturbance intensity level, feedforward compensation correction is applied to the slurry supply regulating valve opening control amount, the oxidation blower inlet regulating valve opening control amount, and the circulating pump start-stop combination mode, so that the control parameter adjustment action is completed before the disturbance effect is reflected in the outlet SO2 concentration and pH value.
7. The intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to claim 1, characterized in that, The application of safety threshold constraints to each control variable includes: Before the final control command is issued, the control values for the slurry supply regulating valve opening, the oxidation blower inlet regulating valve opening, and the circulating pump start / stop combination are compared with the preset upper and lower safety thresholds; and If any control quantity exceeds the corresponding safety threshold, the control quantity is truncated to the safety threshold boundary before being sent to the actuator. The safety threshold is fixed in the desulfurization DCS system in the form of hard coding and is not affected by the online update process of the Mamdani fuzzy neural network model.
8. An intelligent control system for a desulfurization system in a coal-fired power plant based on a Mamdani fuzzy neural network, characterized in that, The method for implementing the intelligent control method for a coal-fired power plant desulfurization system based on a Mamdani fuzzy neural network as described in any one of claims 1-7 includes: The model training and update unit is configured to use the flue gas flow rate at the desulfurization tower inlet, the SO2 concentration of the raw flue gas, the density of the limestone slurry, the pH value of the slurry, and the composition of the slurry as input variables, and the SO2 concentration at the outlet as the control target output quantity to construct and train the Mamdani fuzzy neural network model. The Mamdani fuzzy neural network model includes a fuzzification layer, a rule inference layer, an activation intensity calculation layer, a normalization layer, and a defuzzification layer. The multi-parameter optimization unit is configured to perform real-time optimization of limestone slurry supply, oxidation air volume and slurry circulation pump combination mode in parallel within the same control cycle based on the real-time prediction output of the Mamdani fuzzy neural network model, and generate control quantities for three types of control parameters. The disturbance identification and feedforward compensation unit is configured to identify the disturbance intensity level using the Mamdani fuzzy neural network model based on real-time sampled data of flue gas-side and slurry-side disturbances, and apply feedforward compensation corrections to the control quantities of the three types of control parameters. The flue gas-side and slurry-side disturbances are added as additional input dimensions to the input vector of the Mamdani fuzzy neural network model, and the feedforward compensation correction is output through an independent output channel of the defuzzification layer. The safety threshold control unit is configured to merge the control quantity and the feedforward compensation correction quantity into a final control command, which is then sent to each actuator via the command channel of the desulfurization DCS system, and a safety threshold constraint is applied to each control quantity.
9. An electronic device, characterized in that, include: At least one memory stores computer-executable instructions non-transitory; At least one processor, configured to run the computer-executable instructions, The computer-executable instructions are executed by the processor to implement the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer-executable instructions, which, when executed by at least one processor, implement the intelligent control method for desulfurization system of coal-fired power plant based on Mamdani fuzzy neural network according to any one of claims 1 to 7.
