Intelligent regulation and control method and system for electric dust remover

Abstract

The invention discloses an intelligent regulation and control method and system for an electric precipitator, and relates to the technical field of intelligent regulation and control, and the method comprises the steps: collecting data through a sensor, constructing an electric precipitation process mathematical model, and simplifying a deep learning model to obtain a state space model; estimating a current state parameter of the state space model based on sensor data, and recurring future output in combination with a deep learning model in a given prediction time domain and a control time domain; setting an adaptive weighted quadratic objective function of a tracking error and a control increment, and outputting a security constraint; and calculating an optimal control parameter sequence by using a genetic algorithm in combination with a target and constraints, executing first-order control, and updating control parameters in a rolling manner. According to the method, the control precision and the response stability of the electric dust remover are remarkably improved, the energy consumption of the system is effectively reduced, and the dust removal efficiency is improved. And meanwhile, the self-adaptability of the system is further enhanced, so that the system can always maintain efficient operation under different working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, and particularly relates to an intelligent control method and system for an electric dust collector. BACKGROUND

[0002] As an important air purification method, electric dust removal technology is widely used in industrial production, especially in the power, metallurgy, chemical industry and other industries, for removing dust particles in waste gas. The traditional electric dust collector uses high-voltage electric field to act on dust particles, so that the particles are charged and adsorbed by the collecting electrode, thereby achieving the purpose of purifying air. With the acceleration of industrialization and the increasingly stringent environmental protection standards, electric dust removal technology is facing more and more challenges in improving dust removal efficiency, reducing energy consumption, and ensuring equipment stability. At present, the control system of the traditional electric dust collector is mostly based on experience and timing control method. This method is difficult to cope with the dynamic changes under different working conditions, and it is difficult to realize accurate control and optimization. In order to solve this problem, in recent years, research has focused on optimizing the control system of the electric dust collector through intelligent means, such as using fuzzy control, expert system, neural network and other methods to improve control accuracy and adaptability.

[0003] However, the existing electric dust collector control technology still has some shortcomings. The traditional control system usually cannot respond in real time to the rapid changes in the electric dust removal process, such as the influence of air flow fluctuations, electric field intensity changes and other factors on dust removal efficiency, which makes the performance of the control system unstable under different load conditions. Although deep learning-based models have achieved remarkable results in some fields, in the field of intelligent control of electric dust collectors, there is a lack of systematic research on modeling of electric field and dust state, which leads to many existing deep learning models not fully considering the complex physical environment and variable working conditions of electric dust collectors. Therefore, the existing technology often cannot effectively predict and adjust in a dynamic environment, the self-adaptability of the control strategy is poor, and it is difficult to ensure the stability and efficiency of the system. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an intelligent control method for an electric dust collector to solve the problem of lack of rapid response capability to dynamic changes in the environment, and inability to accurately control factors such as electric field intensity and air flow changes in the traditional technology.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an intelligent control method for an electric dust collector, which comprises collecting data through a sensor, constructing a mathematical model of the electric dust removal process, and simplifying a state space model using a deep learning model.

[0008] Estimating the current state parameters of the state space model based on sensor data, combining a deep learning model to recursively predict future outputs within a given prediction horizon and control horizon;

[0009] Setting an adaptive weighted quadratic objective function of tracking error and control increment, and outputting safety constraints;

[0010] Using a genetic algorithm to calculate the optimal control parameter sequence combining the objective and constraints, executing the first-order control and rolling updating the control parameters.

[0011] As a preferred scheme of the intelligent control method of the electric dust collector, the simplified construction of the mathematical model of the electric dust removal process comprises: constructing a mathematical model of the electric field and dust state in the electric dust collector, and the specific modeling process comprises:

[0012] The electric field strength is a function of space, and the Poisson equation is used to describe the electric field distribution according to the electrode shape and voltage; the relationship between the current density and the electric field strength is calculated by applying voltage between the electrodes of the electric dust collector;

[0013] The dust state is composed of dust movement and dust concentration; the motion equation of the dust particle is constructed under the influence of gravity, airflow drag force and electric field force; the change of the dust particle concentration in the working space of the electric dust collector is affected by the airflow and the electric field, and the continuous medium equation is used to describe the change of the dust concentration.

[0014] As a preferred scheme of the intelligent control method of the electric dust collector, the simplified construction of the mathematical model of the electric dust removal process comprises: pre-training a deep neural network, inputting the electric field model parameters and the dust state model parameters in the electric dust collector; modeling the relationship between static input and output, and learning the nonlinear mapping between input and output through a multi-layer fully connected network;

[0015] Through the nonlinear relationship extracted from the input data, the output obtained by training the deep learning model is used as the state parameter of the state space model, and the state space model is obtained by simplifying the mathematical model and performing state transition.

[0016] As a preferred scheme of the intelligent control method of the electric dust collector, the simplified construction of the mathematical model of the electric dust removal process comprises: combining the Kalman filter estimation value with the prediction value of the deep learning model to output the future state parameter;

[0017] The data obtained by the sensor is used to estimate the current state parameter in the mathematical model by using the Kalman filter method, and the estimation result is obtained The prior estimation is obtained by the deep learning model according to the historical data output;

[0018] the current state parameter and the control input parameter u t input the deep learning model, output the predicted state parameter x t+1 of the next time through the forward propagation process; preset the prediction time domain and the control time domain to determine the prediction range of the future output; set the prediction time domain of the future as the state of the future K time steps starting from the current time t; in the prediction process, the state parameter x t+k of the future K times is calculated by using the deep learning model through recursion.

[0019] the state estimation result of the Kalman filter and the prediction result x t+k of the deep learning model are weighted and fused to obtain the future state result

[0020] As a preferred scheme of the intelligent regulation and control method of the electric dust collector, wherein: the adaptive weighted quadratic objective function includes inputting the state vector x t of the current system, defining the reference state vector as x ref,t , and indicating the expected target value;

[0021] defining the tracking error e t , indicating the difference between the state space parameter x t and the expected target value x ref,t ; the control increment Δu t is the change value between the control input u t of the current time and the control input u t-1 of the last time;

[0022] constructing the objective function includes a weighted quadratic term of the tracking error and a weighted quadratic term of the control increment; wherein the weight of the tracking error term is controlled by the weighted matrix Q e , and the target is to minimize the tracking error;

[0023] the weighted quadratic term of the control increment includes that the weight of the control increment term is controlled by the weighted matrix R u , indicating the punishment degree of the control input increment; and the target is to minimize the change value of the control increment;

[0024] constraints on the control input u t are added, so that u t is between the effective range of the control input; the control increment Δu t is constrained to be always less than the maximum allowed amplitude of the control input change.

[0025] As a preferred scheme of the intelligent regulation method of the electric dust collector, wherein: the using of the genetic algorithm to combine the target and the constraint to calculate the optimal control parameter sequence includes initializing the population of the genetic algorithm, each individual in the population representing a control sequence, wherein each control input u t is a decision variable in the control time domain; generating an initial population, wherein each control input parameter u t is randomly generated within the control input range;

[0026] defining a fitness function, using the target function to evaluate the standard of the goodness of each individual; for each u t in the individual sequence, using a weighted quadratic target function of the additional constraint condition for evaluation;

[0027] using a roulette wheel selection method to select individuals with high fitness from the current population as parents for crossover and mutation operations; within the constraint condition limit, u t in the sequence, the fitness weight decreases in order;

[0028] performing crossover operation on the selected parent individuals to generate new offspring individuals; selecting two parent control sequences and generating a new control sequence through crossover operation;

[0029] performing random disturbance on the input u t in the new control sequence, and the mutated control input still needs to satisfy the control input range constraint and the increment constraint, and the optimal control input parameter sequence is obtained through iteration.

[0030] As a preferred scheme of the intelligent regulation method of the electric dust collector, wherein: the rolling update of the control parameter includes, after obtaining the optimal control sequence, performing first-order control using the first control input parameter u in the control sequence to control the operation of the electric dust collector, inputting into the control system of the electric dust collector, and adjusting the system state;

[0031] after performing the first-order control, the sensor updates the current system state according to the feedback value after execution; after the control operation is performed, the state space parameters of the electric dust collector change; according to the data provided by the sensor and the current control strategy, the state of the system is updated using the Kalman filtering algorithm and the deep learning model prediction fusion method to correct according to the actual observation value and the predicted value;

[0032] Let λ u (t) represent the weighted coefficient of the control increment in the target function, which is adjusted according to the change rate of the control input; if the change value of the control input is close to the constraint range, the weight is increased to smooth the control process; λ e(t) represents a tracking error weight in the objective function, which increases with increasing control error e (t) is adjusted;

[0033] According to the latest state estimation And the reference trajectory, recalculate the objective function, and roll the update of the optimal control sequence, recalculate the optimal solution of the objective function, update the control strategy; take the new system state As input, solve the objective function again to generate a new control input sequence.

[0034] In a second aspect, the present application provides an intelligent control system for an electric dust collector, comprising an electric dust collector modeling module, a deep learning prediction module, and an adaptive optimization module.

[0035] The electric dust collector modeling module is configured to construct a mathematical model of the electric field and dust state in the electric dust collector.

[0036] The deep learning prediction module is configured to simplify the complex mathematical model and estimate future output.

[0037] The adaptive optimization module is configured to calculate the optimal control sequence using a genetic algorithm and update the system state according to sensor feedback data.

[0038] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the intelligent control method for an electric dust collector according to the first aspect of the present application.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the intelligent control method for an electric dust collector according to the first aspect of the present application.

[0040] The present application has the following advantages: through the combination of deep learning and Kalman filtering, the control accuracy and response stability of the electric dust collector are significantly improved. In contrast to the frequent adjustment and large amplitude change in traditional control methods, the present application effectively reduces the energy consumption of the system and improves the dust removal efficiency by smoothing the control increment and optimizing the control sequence. At the same time, the introduction of the genetic algorithm further enhances the adaptability of the system, enabling it to maintain efficient operation under different working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0042] Figure 1 Flow chart of the intelligent control method for the electric dust collector. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0044] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0045] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0046] Reference Figure 1 For one embodiment of the present application, the embodiment provides an intelligent control method for an electric dust collector, comprising the following steps:

[0047] S1: collecting data through a sensor, constructing a mathematical model of the electric dust removal process, and simplifying a state space model using a deep learning model.

[0048] Further, the simplified construction of the mathematical model of the electric dust removal process includes constructing a mathematical model of the electric field and dust state in the electric dust collector, and the specific modeling process includes:

[0049] The electric field strength of the electric dust collector is a function of space, and the Poisson equation is used to describe the electric field distribution according to the electrode shape and voltage; the relationship between the current density and the electric field strength is calculated by applying voltage between the electrodes of the electric dust collector.

[0050] The dust state is composed of dust movement and dust concentration; the motion equation of the dust particle is constructed under the influence of gravity, airflow drag force and electric field force; the change of dust particle concentration in the working space of the electric dust collector is affected by airflow and electric field, and the continuous medium equation is used to describe the change of dust concentration.

[0051] The simplification using the deep learning model comprises pre-training a deep neural network, inputting the electric field model parameters and the dust state model parameters in the electric dust collector; modeling the relationship between the static input and output, and learning the nonlinear mapping between the input and output through a multi-layer fully connected network.

[0052] By extracting the nonlinear relationship from the input data, the output obtained by the deep learning model through training is used as the state parameter of the state space model, and the state space model is obtained by simplifying the mathematical model for state transition.

[0053] The mathematical model of the electric dust removal process can be divided into two parts: the electric field model and the dust state model. First, the two models are established to describe the electric field distribution and the motion state of the dust. The electric field is one of the most important factors in the electric dust collector. Assuming that the electric field intensity E(x, y, z) is a function in space, it depends on the shape and voltage of the electrode. The distribution of the electric field is described by using the Poisson equation:

[0054]

[0055] Where: ρ(x, y, z) is the charge density, representing the charge distribution in the electric field. ∈ is the dielectric constant, representing the conductivity of the medium. In order to relate the electric field to the current density, Ohm's law is used to obtain the relationship between the current density J(x, y, z) and the electric field intensity E(x, y, z):

[0056] J(x, y, z) = σ(x, y, z) E(x, y, z)

[0057] Where: σ(x, y, z) is the conductivity, representing the conductivity of the material to the current.

[0058] The dust state model The motion and concentration of the dust are affected by multiple factors: gravity, airflow drag force, and electric field force. Assuming that the mass of the dust particle is m, its motion equation can be expressed as:

[0059]

[0060] Where: F g = mg is the gravitational force. F d = αv is the airflow drag force, where α is the drag coefficient and v is the velocity of the particle. F e = qE is the electric field force, q is the charge of the dust particle, and E is the electric field intensity. The dust particle concentration C(x, y, z, t) changes over time in the working space of the electric dust collector, affected by the airflow and the electric field. The continuous medium equation can describe the change of the dust concentration:

[0061]

[0062] where, is the diffusion flow of dust, representing the motion and concentration change of dust particles.

[0063] The simplified state-space model simplifies the above complex mathematical model by a deep learning model.

[0064] The pre-training of the deep learning model uses a deep neural network (DNN) to learn the nonlinear relationship between the electric field model and the dust state model. The input is the parameters of the electric field model and the dust state model, such as voltage, air flow speed, dust particle density, etc.

[0065] The mapping relationship between the input and the output is learned by a multi-layered fully connected network (MLP). The structure of the neural network is usually a fully connected layer composed of multiple hidden layers, and there is a nonlinear mapping between the input and the output of each layer:

[0066]

[0067] where, is the weight matrix, is the bias term, and f(·) is the activation function. The network is trained by the backpropagation algorithm to adjust the weights and bias terms to minimize the prediction error of the network.

[0068] Once the training is completed, the deep learning model can infer the state parameters x k of the state-space model according to the input data. The form of the state-space model is as follows:

[0069] x k = f θ (x k-1 , u k ) + w k

[0070] where: x k is the state parameter at the current time. u k is the control input. f θ is the nonlinear state transition function obtained by the deep learning model. w k is the process noise.

[0071] It should be noted that the training data, input data includes the parameters of the electric field model (such as voltage, current density, air flow speed) and the parameters of the dust state model (such as particle concentration, particle velocity, etc.), which are collected in real time by sensors. By inputting these data, the deep learning network can learn the nonlinear mapping relationship between input and output, thereby modeling the complex dynamic behavior of the electric dust removal process. Deep learning models can handle complex nonlinear relationships between input data, while traditional physical modeling methods often assume a certain linear relationship. This makes deep learning models more flexible and efficient in handling complex dynamic systems (such as the interaction of air flow, dust and electric field in the electric dust removal process) than traditional methods. Through this method, the deep learning model simplifies the nonlinear complexity of the electric dust removal process, making it easier to use for subsequent control strategy optimization.

[0072] S2: Estimate the current state parameters of the state space model based on sensor data, and recursively calculate the future output in combination with the deep learning model within the given prediction time domain and control time domain.

[0073] Further, the combination of the deep learning model to recursively calculate the future output includes state fusion of the Kalman filter estimation value and the prediction value of the deep learning model to output the future state parameters.

[0074] The data obtained by the sensor is used to estimate the current state parameters in the mathematical model using the Kalman filter method to obtain the estimation result Where the prior estimate is obtained from the deep learning model based on historical data output.

[0075] The current state parameters and the control input parameters u t are input into the deep learning model, and through the forward propagation process, the predicted state parameters at the next time step are output x t +1 ; The prediction range of the future output is determined by setting the prediction time domain and the control time domain; The future prediction time domain is set to start from the current time t and predict the state of the next K time steps; In the prediction process, the deep learning model is used to recursively calculate the state parameters x t+k at the next K time steps.

[0076] The state estimation result of the Kalman filter is weighted and fused with the prediction result x t+k of the deep learning model to obtain the future state result

[0077] In the control system, the current state of the system needs to be estimated and the future state needs to be predicted. By combining Kalman filtering and deep learning models, more accurate state estimation and prediction can be achieved.

[0078] State estimation of Kalman filter Kalman filter method is used to estimate the state of the current system Based on sensor observation value y k And prior state estimation

[0079]

[0080] Where: A is the state transition matrix, representing the state change from k-1 to k. B is the control input matrix, representing the influence of control input on state change. K k Is the Kalman gain, the degree of fusion of control estimation and observation value. C is the observation matrix, representing the mapping from state to observation.

[0081] Deep learning model predicts future state according to historical data:

[0082]

[0083] The estimated value of Kalman filter is fused with the predicted value of deep learning model:

[0084]

[0085] Where: Is the state estimated by Kalman filter. Is the state predicted by deep learning model. Alpha is the fusion coefficient, which determines the weight of Kalman filter and deep learning model.

[0086] It should be noted that by using the state parameter x t Output by deep learning model, the originally complex electrostatic precipitation mathematical model can be simplified into a state space model. Traditional mathematical model contains multiple physical equations (such as Poisson equation, electric field distribution, dust motion equation, etc.), which need complex numerical calculation. Through deep learning model, nonlinear relationship can be automatically extracted from input data and simplified into a state space model, and dynamic update of the model is completed through prediction ability of deep learning. This simplification is very creative, which greatly reduces the dependence on complex physical formula, and improves the adaptability of the system to external changes (such as uneven electric field, airflow change, etc.).

[0087] S3: set adaptive weighted quadratic objective function of tracking error and control increment, output safety constraint.

[0088] Further, the adaptive weighted quadratic objective function includes inputting the state vector x t Of the current system, defining the reference state vector as x ref,t Represents the expected target value.

[0089] The tracking error e is defined as t The tracking error e is defined as t The tracking error e is defined as ref,t The control increment Δu is defined as t The control increment Δu is defined as t The control increment Δu is defined as t-1 The control increment Δu is defined as

[0090] The objective function is constructed as a weighted quadratic term of the tracking error e and a weighted quadratic term of the control increment Δu. The weight of the tracking error term is controlled by a weighting matrix Q e The objective is to minimize the tracking error e.

[0091] The weighted quadratic term of the control increment Δu includes a weight of the control increment term controlled by a weighting matrix R u The objective is to minimize the change of the control increment Δu.

[0092] Constraints are added to the control input u t to keep u t within the valid range of the control input. Constraints are also added to the control increment Δu t to keep Δu e always smaller than the maximum allowed change of the control input.

[0093] The objective function is designed to consider both the tracking error e and the control increment Δu. By adjusting the weight coefficients, a balance between the tracking accuracy and the control stability can be achieved.

[0094] The objective function J includes two parts: a weighted quadratic term of the tracking error e:

[0095] J e = w e ||x k -x ref,k || 2

[0096] where x ref,k is the desired state and w e is the weight coefficient of the tracking error e. The weighted quadratic term of the control increment Δu is expressed as:

[0097] J u = w u ||Δu k || 2

[0098] where Δu k = u k -u k-1 is the control increment and w u is the weight coefficient of the control increment. The final objective function is:

[0099] J = J e + J u

[0100] By minimizing the objective function, the control system can find the optimal control input sequence, thus achieving precise regulation of the electrostatic precipitator.

[0101] It should be noted that the fusion coefficient a is dynamically adjusted, rather than statically set. This innovation ensures that the system can adaptively adjust the weights of the deep learning model and the Kalman filter model according to different operating conditions and error situations. For example, when the noise is large or the sensor error is high, more reliance can be placed on the Kalman filter; while when the system is relatively stable and the deep learning model can provide accurate predictions, the weight of deep learning can be increased.

[0102] S4: Calculate the optimal control parameter sequence using genetic algorithm combined with objectives and constraints, perform first-order control and roll update control parameters.

[0103] Further, the calculation of the optimal control parameter sequence using genetic algorithm combined with objectives and constraints includes initializing the population of the genetic algorithm, each individual in the population representing a control sequence, wherein each control input u t is a decision variable within the control time domain; generating an initial population, wherein each control input parameter u t is randomly generated within the control input range.

[0104] Each individual in the population represents a control sequence, where each control input is a decision variable within the control time domain. The specific process is as follows: population definition: each individual in the population represents a control sequence u k = [u1, u2,..., u N ], where u k is the control input sequence within the control time domain, and N is the length of the control time domain. Initial control input random generation: each control input is randomly generated within the effective range of the control input [u min , u max ]. That is, each control input u k satisfies:

[0105] u k ∈ [u min , u max ]

[0106] After generating the initial population, the entire population is represented as a set of multiple control inputs.

[0107] Definition of fitness function and objective function Fitness function is a standard used to evaluate the goodness of an individual in genetic algorithm. Fitness function is defined as the evaluation standard of objective function, which includes the weighted quadratic objective function of tracking error and control increment. The specific steps are as follows: Objective function: Objective function includes two parts: tracking error part J e :

[0108] J e = w e ||x k -x ref,k || 2

[0109] where x k is the current system state, x ref,k is the desired target value, and w e is the weight coefficient of tracking error. Control increment part J u :

[0110] J u = w u ||Δu k || 2

[0111] where Δu k = u k -u k-1 is the control increment, and w u is the weight coefficient of control increment. The total objective function J is the weighted sum of the two parts:

[0112] J = J e + J u

[0113] Constraints: When calculating fitness, constraints are added to the objective function. The constraints of control increment and the effective range constraints of control input are included in the weighted quadratic objective function. The constraints can be expressed as:

[0114] u min ≤ u k ≤ u max

[0115] and the change constraint of control increment:

[0116] Δu k ≤ Δu max

[0117] Selection and crossover operation In each generation of genetic algorithm, roulette wheel selection method is used to select individuals with higher fitness as parent individuals for crossover and mutation operations.

[0118] Roulette wheel selection method: The fitness F of each individual is calculatedi = -J i ( negative value because we want to minimize the objective function), the selection probability of each individual is calculated:

[0119]

[0120] Then the selection is made using the probability, and individuals with high fitness are selected as parents for crossover and mutation. Crossover operation: select two parent individuals u k1 and u k2 , generate two new control sequences u' k1 and u' k2 through the crossover operation:

[0121] u' k1 = au k1 + (1 - a)u k2

[0122] u' k2 = (1 - a)u k1 + au k2

[0123] where a e [0, 1] is the crossover coefficient, which controls the way of generating new control sequences.

[0124] In order to increase diversity and avoid local optimal solution, mutation operation is also needed in genetic algorithm. Mutation operation randomly perturbs the input in the newly generated control sequence. The specific steps are as follows: for each control input u' k in the newly generated control sequence, perform random perturbation operation:

[0125] u' k = u' k + d k

[0126] where d k is a random perturbation term, and d k satisfies the effective range constraint of control input [u min , u max ] and the constraint of control increment. The mutated control input u' k still needs to satisfy the control input range constraint and the increment constraint:

[0127] u min ≤ u' k ≤ u max

[0128] Au k ≤ Au max

[0129] The rolling update control parameters, after obtaining the optimal control sequence, use the first control input parameter u1 to perform the first-stage control. Through the execution of the control input, the state space parameters of the electric precipitator change.

[0130] First-stage control execution: according to the optimal control sequence obtained by optimization, the first control input u1 is used to control the electric precipitator. The control input u1 is input into the control system, and the operation of the electric precipitator is adjusted to optimize its state.

[0131] After the first-stage control is executed, the sensor updates the current system state according to the feedback value after execution. Assuming that the new system state is x new,k According to the results of the control operation, the state space parameters of the system change. Kalman filter and deep learning model fusion: use the fusion method of Kalman filter algorithm and deep learning model for state correction. Kalman filter is used to estimate the current state, and the estimated value is fused with the predicted value of the deep learning model. The predicted state Recursion through the deep learning model:

[0132]

[0133] Then the estimated result of the Kalman filter is weighted and fused with the predicted result of the deep learning model to obtain the updated state. Dynamic adjustment of the objective function: according to the new state estimation and the reference trajectory, the weight in the objective function is updated. In particular, when the change of the control increment is close to the constraint range, the weight needs to be increased to ensure the smoothness of the control process. The tracking error weight is dynamically adjusted with the increase of the control error to better track the target trajectory.

[0134] Define the fitness function, use the objective function to evaluate the standard of the goodness of each individual; for each u t in the individual sequence, use the weighted quadratic objective function with additional constraints to evaluate.

[0135] Use the roulette wheel selection method to select individuals with high fitness from the current population as parents for crossover and mutation operations; within the constraint condition limit, u t The fitness weight decreases in order.

[0136] Perform crossover operation on the selected parent individuals to generate new offspring individuals; select two parent control sequences and, and generate a new control sequence through crossover operation.

[0137] Perform random disturbance on the input u t in the new control sequence, and the mutated control input still needs to satisfy the control input range constraint and the increment constraint, and the optimal control input parameter sequence is obtained through iteration.

[0138] The rolling update control parameter includes, after obtaining the optimal control sequence, performing the first-stage control using the first control input parameter in the control sequence to control the operation of the electric precipitator, input into the control system of the electric precipitator, and adjust the system state.

[0139] After performing the first-stage control, the sensor updates the current system state according to the feedback value after the performance; the state space parameters of the electric precipitator change after the control operation is performed; and the state of the system is updated according to the data provided by the sensor and the current control strategy The prediction fusion method uses the Kalman filtering algorithm and the deep learning model to correct according to the actual observation value and the predicted value.

[0140] The subsequent parameters in the sequence are selected for regulation, and if they do not conform to the correction result, the target function weight is modified to regenerate the future prediction sequence.

[0141] Let lambda u (t) represent the weighting coefficient of the control increment in the target function, which is adjusted according to the change rate of the control input; if the change value of the control input is close to the constraint range, the weight is increased to smooth the control process; lambda e (t) represents the tracking error weight in the target function, which is increased with the increase of the control error, and the value of lambda e (t) is adjusted;

[0142] According to the latest state estimation and the reference trajectory, the target function is recalculated, and the rolling update of the optimal control sequence is performed, the optimal solution of the target function is recalculated, and the control strategy is updated; the new system state is input again to solve the target function to generate a new control input sequence.

[0143] It should be noted that according to the new state estimation and the reference trajectory, the weight in the target function is updated. In particular, when the change of the control increment is close to the constraint range, the weight needs to be increased to ensure the smoothness of the control process. The tracking error weight is dynamically adjusted with the increase of the control error to better track the target trajectory. According to the latest state estimation, the target function is recalculated, and the control sequence is updated by the genetic algorithm. The new control sequence can be optimized by updating the optimal solution of the target function. Each time the rolling update is performed, the current system state is used as input to solve the target function again to generate a new control input sequence.

[0144] The embodiment also provides an intelligent regulation and control system of an electric precipitator, which comprises an electric precipitator modeling module, a deep learning prediction module, and a self-adaptive optimization module.

[0145] The electric precipitator modeling module is configured to construct a mathematical model of the electric field and the dust state in the electric precipitator.

[0146] The deep learning prediction module is used to simplify complex mathematical models and estimate future outputs.

[0147] The adaptive optimization module uses a genetic algorithm to calculate an optimal control sequence and updates system states according to sensor feedback data.

[0148] The embodiment also provides a computer device suitable for the intelligent regulation and control method of the electric dust collector, which includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the intelligent regulation and control method of the electric dust collector proposed in the above embodiment.

[0149] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0150] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to realize the intelligent regulation and control method of the electric dust collector proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage device, a flash memory, a magnetic disk or an optical disk.

[0151] In summary, the present application significantly improves the control accuracy and response stability of the electric dust collector by combining deep learning with Kalman filtering. In contrast to the frequent adjustments and large changes that occur in traditional control methods, the present application effectively reduces the energy consumption of the system and improves the dust removal efficiency by smoothing the control increment and optimizing the control sequence. At the same time, the introduction of genetic algorithm further enhances the adaptability of the system, enabling it to maintain efficient operation under different working conditions.

[0152] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for intelligent control of an electrostatic precipitator, characterized in that: This includes collecting data through sensors, constructing a mathematical model of the electrostatic precipitator process, and simplifying it using a deep learning model to obtain a state-space model; The current state parameters of the state-space model are estimated based on sensor data, and the future output is recursively inferred by combining a deep learning model within the given prediction and control time domains. Define an adaptive weighted quadratic objective function for tracking error and control increment, and output safety constraints; The optimal control parameter sequence is calculated by combining the objective and constraints using a genetic algorithm, and the first-order control is executed while the control parameters are updated on a rolling basis.

2. The intelligent control method for an electrostatic precipitator as described in claim 1, characterized in that: The simplified mathematical model for constructing the electrostatic precipitator process includes constructing a mathematical model of the electric field and dust state inside the electrostatic precipitator. The specific modeling process includes: Assuming the electric field strength of the electrostatic precipitator is a function of space, the electric field distribution is described using the Poisson equation based on the electrode shape and voltage. By applying a voltage between the electrodes of the electrostatic precipitator, the relationship between current density and electric field strength is calculated. The dust state is composed of dust movement and dust concentration; dust particles are affected by gravity, airflow drag force and electric field force, which constitute the motion equation of dust particles; within the working space of the electrostatic precipitator, the change of dust particle concentration is affected by the combined effect of airflow and electric field, and the change of dust concentration is described by the continuous medium equation.

3. The intelligent control method for an electrostatic precipitator as described in claim 2, characterized in that: The simplification using deep learning models includes: pre-training a deep neural network, inputting electric field model parameters and dust state model parameters into the electrostatic precipitator; modeling the relationship between static input and output, and learning the nonlinear mapping between input and output through a multi-layer fully connected network; By extracting nonlinear relationships from the input data, the output obtained by training the deep learning model is used as the state parameters of the state-space model, and the mathematical model is simplified to perform state transitions to obtain the state-space model.

4. The intelligent control method for an electrostatic precipitator as described in claim 3, characterized in that: The method of combining the deep learning model to recursively predict the future output includes fusing the Kalman filter estimate with the prediction of the deep learning model to output the future state parameters. Using data acquired from sensors, the Kalman filter method is used to estimate the current state parameters in the mathematical model, yielding the estimation results. Among them, the prior estimate is obtained by the deep learning model based on historical data; Set the current state parameters and control input parameter u t Input a deep learning model, and through the forward propagation process, output the predicted state parameters x for the next time step. t+1 The prediction time domain and control time domain are preset to determine the prediction range of future outputs; the future prediction time domain is set to start from the current time t and predict the state for the next K time steps; during the prediction process, the state parameters x for the next K time steps are recursively calculated using a deep learning model. t+k ; The state estimation results of the Kalman filter The prediction result x of the deep learning model t+k Perform weighted fusion to obtain the future state result.

5. The intelligent control method for an electrostatic precipitator as described in claim 4, characterized in that: The adaptive weighted quadratic objective function includes the input current system state vector x. t Define the reference state vector as x ref,t Indicates the expected target value; Define tracking error e t , representing the state-space parameter x t With the expected target value x ref,t The difference between them; control increment Δu t It is the control input u at the current moment. t and the control input u from the previous moment t-1 The amount of change between; The objective function is constructed by including a weighted quadratic term for the tracking error and a weighted quadratic term for the control increment; the weights of the tracking error term are determined by the weighting matrix Q. e The goal of control is to minimize tracking error; The weighted quadratic term of the control increment includes the weights of the control increment term determined by the weighting matrix R. u Control refers to the degree of punishment imposed on the increment of control input; The goal is to minimize the change in the control increment; Add control input u t The constraints make u t Between the effective range of the control input; for the control increment Δu t The constraint is always applied and must be less than the maximum permissible range of change in the control input.

6. The intelligent control method for an electrostatic precipitator as described in claim 5, characterized in that: The step of using a genetic algorithm to calculate the optimal control parameter sequence by combining the objective and constraints includes initializing the genetic algorithm population, where each individual in the population represents a control sequence, and each control input u t It is a decision variable within the control time domain; Generate an initial population, where each control input parameter u t Randomly generated within the control input range; Define a fitness function and use an objective function to evaluate the criteria for each individual's performance; for each individual's sequence, u t The evaluation is performed using a weighted quadratic objective function with additional constraints. Using the roulette wheel selection method, individuals with high fitness are selected from the current population as parents for crossover and mutation operations; within the constraints, the u in the sequence... t Fitness weights decrease sequentially. Perform crossover on the selected parent individuals to generate new offspring individuals; select two parent control sequences and perform crossover to generate a new control sequence. For the input u in the new control sequence t Perform random perturbations, and ensure that the mutated control input still meets the control input range constraints and incremental constraints, and iterate to obtain the optimal control input parameter sequence.

7. The intelligent control method for an electrostatic precipitator as described in claim 6, characterized in that: The rolling update of control parameters includes, after obtaining the optimal control sequence, performing first-order control and using the first control input parameter in the control sequence. To control the operation of the electrostatic precipitator, input the data into the electrostatic precipitator's control system, and adjust the system status; After the first-order control is executed, the sensors update the current system state based on the feedback values. After the control operation is executed, the state-space parameters of the electrostatic precipitator change; the system state is then updated based on the data provided by the sensors and the current control strategy. The method of using Kalman filtering algorithm and deep learning model prediction fusion is corrected based on actual observations and predictions; Let λ u (t) represents the weighting coefficient of the control increment in the objective function, which is adjusted according to the rate of change of the control input; when the change in the control input approaches the constraint range, the weight is increased to smooth the control process; λ e (t) represents the tracking error weight in the objective function. If the control error increases with time, then λ is increased. e Adjust the (t) value; Based on the latest state estimate Based on the reference trajectory, recalculate the objective function and perform rolling updates of the optimal control sequence. Recalculate the optimal solution of the objective function and update the control strategy; then apply the new system state. As input, the objective function is solved again to generate a new control input sequence.

8. An intelligent control system for an electrostatic precipitator, based on the intelligent control method for an electrostatic precipitator according to any one of claims 1 to 7, characterized in that: It includes an electrostatic precipitator modeling module, a deep learning prediction module, and an adaptive optimization module; The electrostatic precipitator modeling module is used to construct a mathematical model of the electric field and dust state inside the electrostatic precipitator. The deep learning prediction module is used to simplify complex mathematical models and estimate future outputs. The adaptive optimization module uses a genetic algorithm to calculate the optimal control sequence and updates the system state based on sensor feedback data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent control method for the electrostatic precipitator according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent control method for the electrostatic precipitator according to any one of claims 1 to 7.

Images