Collaborative prediction and intelligent control method and system for multiple pollutants in waste incineration flue gas

Abstract

The present invention relates to a technology for controlling pollutants in flue gas from garbage incineration, and aims to provide a method and system for collaborative prediction and intelligent control of multiple pollutants in flue gas from garbage incineration. Based on a deep learning method, the present invention constructs a collaborative prediction model for four types of pollutants in flue gas from garbage incineration; then, in combination with a multi-objective optimization method, a cost index function that considers the amount of absorbent used and an environmental index function that considers the amount of pollutant emissions are constructed; a multi-objective optimization algorithm is used to solve and obtain the optimal dosage data of the absorbent corresponding to the flue gas pollutants, and finally, the opening of various absorbent delivery valves is adjusted and controlled according to the feedback of the optimal dosage data, thereby realizing intelligent control of multiple pollutants in flue gas from garbage incineration. The present invention can realize accurate collaborative prediction of multiple pollutants in flue gas from garbage incineration, effectively overcome the hysteresis problem of traditional flue gas pollutant emission monitoring equipment; minimize the pollutant control cost of the flue gas purification system, and improve the economic level of the incinerator.

Description

Technical Field

[0001] The present application relates to the field of waste incineration flue gas pollutant control, and specifically to a method and system for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas. Background Art

[0002] Waste incineration has significant advantages such as harmlessness, reduction, and resource utilization, and has become the mainstream method of domestic waste disposal at this stage. However, due to its complex composition, the flue gas after waste incineration contains hydrogen chloride (HCl), sulfur dioxide (SO 2 ), nitrogen oxides (NOx) and particulate matter (PM). The emission of these pollutants is likely to cause harm to the environment and human health, so waste incineration plants invest a lot of money in pollutant emission control.

[0003] Understanding the concentration level of pollutants in waste incineration flue gas is of great significance for achieving ultra-low pollutant emissions. At present, waste incineration plants mainly use continuous emission monitoring systems (CEMS) to monitor the emission of conventional pollutants such as NOx in flue gas. However, the monitoring results of CEMS are greatly affected by the environment and there may be a certain measurement delay. At the same time, CEMS equipment also has the disadvantage of high maintenance costs.

[0004] In recent years, data-driven machine learning model methods have been increasingly used in predicting flue gas pollutant concentrations. For example, Chinese invention patent applications for "System and method for predicting NOx emissions from circulating fluidized bed domestic waste incineration boilers" (CN106931453A) and "A data-driven prediction system for WFGD outlet SO 2 Concentration Prediction and Intelligent Optimization Method" (CN115309117A) proposed using BP neural network and artificial neural network methods to predict NOx and SO in boiler flue gas. 2 However, due to the characteristics of pollutant data, modeling complexity, differences between pollutants, and the need for training data, these methods are only targeted at NOx or SO 2 It is impossible to make coordinated predictions for pollutants such as HCl and PM (particulate matter) that also exist in the flue gas, resulting in the prediction target being too single.

[0005] In addition, the quality of pollutant absorption control by the flue gas purification system of the waste incinerator will directly affect the environmental protection cost of the incinerator and whether the final pollutant emissions can meet the standards. Under traditional control technology, the operation of the flue gas purification equipment and the amount of environmental protection materials are manually adjusted by the operating personnel based on the pollutant concentration monitoring data of the CEMS. This control strategy based on manual experience is subjective and may bring certain hysteresis, resulting in large fluctuations in operating conditions during the control period, resulting in inaccurate material control. The Chinese invention patent application "Flue gas control method, device, equipment and storage medium based on machine learning" (CN115392437A) discloses a flue gas NOx control method based on machine learning technology, which can more accurately control the output of ammonia water; however, it only controls pollutants such as NOx, and the control process only considers the cost index of the amount of ammonia water, while ignoring the environmental protection index.

[0006] If we can develop HCl, SO 2 A collaborative prediction model for four types of flue gas pollutants, NOx, and PM, and based on the collaborative prediction results of pollutant concentrations based on the model, and on the basis of comprehensive consideration of cost indicators and environmental protection indicators, intelligent control of the four types of flue gas pollutants is achieved, which will help improve the intelligent operation level of waste incinerators. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a method and system for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas.

[0008] To solve the technical problem, the solution of the present invention is:

[0009] A method for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas is provided, the method comprising the following steps:

[0010] (1) Collect all historical data of the operating parameters and flue gas pollutant emission concentrations of the waste incinerator at set time intervals within the same time period;

[0011] The incinerator operation parameter data comes from the storage database of the waste incinerator distributed control system (DCS); the flue gas pollutants include hydrogen chloride (HCl), sulfur dioxide (SO 2 ), nitrogen oxides (NOx) and particulate matter (PM), and the emission concentration data of each category are all from the storage database of the continuous emission monitoring system (CEMS);

[0012] (2) Calculating the Pearson correlation coefficient between the incinerator operating parameters and the flue gas pollutant emission concentrations in the collected samples obtained in step (1); Screening all the operating parameters in the collected samples based on the Pearson correlation coefficient with the emission concentration data of any one of the four types of flue gas pollutants, and retaining the screening results as the input features required for collaborative prediction training; The four types of flue gas pollutant emission concentration data corresponding to the samples at the same time are used as data labels corresponding to the input features of the samples at that time;

[0013] (3) for the data selected in step (2), perform mean downsampling with a larger time span; then select the incinerator operation parameter data of a portion of the samples, perform time series processing on the data, and use them as input data for training;

[0014] A multi-pollutant collaborative prediction model based on the LSTM layer structure was constructed, and the incinerator operation parameter data of the set time period was input to train the model. The collaborative prediction results of the four types of flue gas pollutant concentrations after the set time period were used as the output of the model.

[0015] (4) The flue gas purification system uses corresponding absorbents to absorb and treat the four types of flue gas pollutants; a multi-objective optimization function F(x) is constructed with the goal of controlling the amount of each absorbent; the function takes into account two types of objective functions, namely, the cost index function of the amount of absorbent used and the environmental index function considering the amount of pollutant emissions; the final optimization goal is to minimize the multi-objective optimization function F(x); the input variables of the cost index function and the environmental index function are both the flue gas multi-pollutant collaborative prediction concentration output by the multi-pollutant collaborative prediction model;

[0016] (5) According to the actual dosage range of each absorbent in engineering applications and the emission concentration limit standards of each pollutant, set the constraints of the multi-objective optimization function F(x); use the multi-objective optimization algorithm to solve the function and calculate the optimal dosage data of the absorbent corresponding to each of the four types of flue gas pollutants;

[0017] (6) The calculation result of step (5) is input into the DCS system of the waste incinerator, and the optimal dosage of each absorbent is converted into an actual analog control signal, which is then sent to the delivery valve of each absorbent in the flue gas purification system; feedback adjustment of the dosage of each absorbent is achieved by controlling the valve opening, thereby reducing the dosage of the absorbent while ensuring that the flue gas pollutants meet the environmental emission standards, thereby achieving the cost-effectiveness goal of the incinerator flue gas purification system.

[0018] As a preferred embodiment of the present invention, in step (2), operating parameters whose absolute value of the Pearson correlation coefficient with any type of pollutant emission concentration data is greater than 0.3 are selected, and these operating parameters are used as input features of the collaborative prediction model, and the remaining operating parameters are discarded.

[0019] As a preferred embodiment of the present invention, the time interval for collecting data in step (1) is 1 second, and the data screened in step (2) is downsampled to a 5-minute mean; for the prediction target of each sample after downsampling, the incinerator operating parameter data of the 12 samples before the sample are used as the input feature data of the multi-pollutant collaborative prediction model.

[0020] As a preferred embodiment of the present invention, in step (3), the multi-pollutant collaborative prediction model has 2 LSTM layers and 1 Dense layer; wherein the number of neurons in the Dense layer is 4, the input time step is 12, the output prediction step is 1, the optimizer is adam, the loss function is mse, and the maximum number of iterations is set to 100 times; the number of neurons in each LSTM layer, the type of activation function and the learning rate are used as hyperparameters of the model, and the optimal parameters are determined by a grid search method.

[0021] As a preferred embodiment of the present invention, in step (4), HCl and SO are absorbed and treated. 2 The absorbents corresponding to the four types of pollutants, NOx and PM, are slaked lime, sodium hydroxide, ammonia water and activated carbon respectively.

[0022] As a preferred solution of the present invention, in step (4), the expression of the multi-objective optimization function F(x) is as follows:

[0023] min F(x)=[f 1 (x),f 2 (x),f 3 (x),f 4 (x),g 1 (x),g 2 (x),g 3 (x),g 4 (x)]

[0024] Among them, f(x) is the cost index function, g(x) is the environmental index function, and the subscripts 1, 2, 3, and 4 refer to HCl, SO 2 , NOx and PM, four types of flue gas pollutants;

[0025] The calculation formula of the cost index function f(x) is as follows:

[0026]

[0027] Where, f(x) is the amount of absorbent; C in is the concentration of flue gas pollutants at the inlet of the flue gas purification system; C out is the flue gas pollutant synergistic predicted concentration; V is the flue gas flow rate; M Abs is the molecular molar mass of the main reactive component of the absorbent; M P is the molecular molar mass of the main component of a certain type of pollutant; η is the actual pollutant removal efficiency;

[0028] The calculation formula of the environmental protection index function g(x) is as follows:

[0029] g(x)=f(Q)

[0030] Among them, g(x) is the pollutant concentration; f is the flue gas pollutant synergistic prediction model; Q is the dosage of the controllable variable absorbent.

[0031] As a preferred solution of the present invention, in step (5), the multi-objective optimization algorithm is a particle swarm optimization algorithm, and its algorithm model is constructed by calling the pso function in the pyswarm library, the number of particles is set to 10, and the maximum number of iterations is 10.

[0032] As a preferred embodiment of the present invention, in step (6), the analog control signal is a current signal of 4 to 20 mA; when the absorbent dosage is controlled to be 0%, the valve is fully closed, corresponding to a current signal of 4 mA; when the absorbent dosage is controlled to be 100%, the valve is fully opened, corresponding to a current signal of 20 mA; in the process of sending the control signal, Modbus is used as the data communication protocol.

[0033] The present invention further provides a waste incineration flue gas multi-pollutant collaborative prediction and intelligent control system, the system comprising the following modules:

[0034] Waste incinerator distributed control system module (DCS): used to collect and store real-time operating parameter data of the waste incinerator;

[0035] Continuous Emission Monitoring System Module (CEMS): It is located at the end of the flue of the waste incinerator and is used to collect and store HCl and SO in the waste incineration flue gas. 2 Real-time emission concentration data of four types of pollutants: NOx and PM;

[0036] Multi-pollutant collaborative prediction module: used to perform the operations described in steps (1) to (5), realize data processing, model training, use the model to output the collaborative prediction concentration of multiple pollutants in flue gas and calculate the optimal dosage of absorbent;

[0037] Multi-pollutant intelligent control module: used to execute the operation described in step (6), send control signals to the opening of various absorbent delivery valves in the incinerator flue gas purification system, and reduce the flue gas purification cost of the incinerator system.

[0038] Description of the invention principle:

[0039] The existing method for determining the concentration of pollutants in waste incineration flue gas is mainly to use the continuous emission monitoring system (CEMS) to monitor the emission of various pollutants in real time, and then the operator manually adjusts the flue gas purification system of the incinerator according to the CEMS monitoring data to achieve emission control of various pollutants. This method relies more on manual experience, and because CEMS may have a certain measurement delay, this method also has a certain hysteresis in regulating flue gas pollutants, making it difficult to operate the flue gas purification system in a timely and accurate manner to control the emission of various pollutants in the incineration flue gas.

[0040] Based on the deep learning method, the present invention uses the CEMS pollutant concentration monitoring data and the incinerator operation condition data to construct a prediction algorithm for HCl and SO in waste incineration flue gas after completing feature screening and data timing processing. 2 , NOx and PM. Then, combined with the multi-objective optimization method, a cost index function considering the amount of absorbent and an environmental index function considering the amount of pollutant emissions are constructed. Both types of objective functions accept the results of the collaborative prediction of flue gas multi-pollutants as input. After setting the constraints of the multi-objective optimization function according to the actual dosage range requirements of the absorbent and the emission concentration limit standards of the four types of pollutants, the multi-objective optimization algorithm is used to solve the optimal dosage data of the four types of absorbents corresponding to the four types of flue gas pollutants treated by the flue gas purification system. Finally, the opening of each type of absorbent delivery valve is adjusted and controlled according to the feedback of the optimal dosage data, thereby realizing intelligent control of multiple pollutants in waste incineration flue gas.

[0041] Compared with the existing single-target pollutant prediction technology, the innovation of the present invention lies in:

[0042] (1) Accurate and coordinated prediction of four types of flue gas pollutants

[0043] Using deep learning model to realize HCl and SO in waste incineration flue gas 2The collaborative prediction of four types of pollutants, NOx and PM, eliminates the need to continuously use CEMS, which has measurement delays and high maintenance costs, to monitor the concentration of flue gas pollutants in the later stage. It only needs to input the continuous operating condition data stored in the incinerator DCS system into the model to obtain the collaborative prediction results of the concentration of four types of flue gas pollutants at the same time. The collaborative prediction model uses an LSTM model that is specially designed to process and predict time series data with long-term dependencies. At the same time, since the generation of pollutants is not only affected by the state of the incinerator at a certain moment, but depends on the entire process of garbage incineration, the input data of the LSTM model is processed in a time series manner to obtain a more accurate collaborative prediction effect.

[0044] (2) Multi-objective optimization based on prediction model

[0045] In order to achieve intelligent control of multiple pollutants in waste incineration flue gas, the present invention links the control of four types of flue gas pollutants to the control of the amount of absorbent corresponding to the treatment of the four types of pollutants by the incinerator flue gas purification system, and based on the developed flue gas multi-pollutant collaborative prediction LSTM model, combined with a multi-objective optimization method, comprehensively considers the cost and environmental protection indicators in the multi-pollutant control process, and calculates the optimal dosage data of the absorbent corresponding to the treatment of the four types of flue gas pollutants.

[0046] (3) Intelligent correlation feedback control of pollutants

[0047] After the optimal dosage of absorbents corresponding to the four types of flue gas pollutants is obtained by the multi-objective optimization method, the optimal dosage result can be converted into a control signal through the DCS system of the incinerator and sent to the valve for dispensing the absorbents corresponding to the four types of flue gas pollutants in the flue gas purification system. By controlling the opening of the valve, accurate feedback control of the dosage of the four types of absorbents can be achieved. Under the level of absorbent dosage adjusted by feedback, it can not only ensure that the emissions of the four types of pollutants in the flue gas meet the standards, but also minimize the operating cost of the flue gas purification system of the incinerator, further improving the intelligent operation level of the incinerator.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention utilizes deep learning methods and can realize accurate collaborative prediction of multiple pollutants in waste incineration flue gas based on the LSTM model, effectively overcoming the hysteresis problem of traditional flue gas pollutant emission monitoring equipment.

[0050] 2. The collaborative prediction targets of the present invention are diverse, including HCl, SO 2 , NOx and PM, the coordinated prediction results of the four types of pollutants can be obtained at the same time without training prediction models for each type of pollutant separately.

[0051] 3. Based on the coupling of the flue gas multi-pollutant collaborative prediction model and the multi-objective optimization method, the present invention can quickly determine the optimization results of the dosage of various pollutant absorbents, thereby realizing intelligent correlation feedback control of multiple pollutants.

[0052] 4. The multi-pollutant intelligent control method of the present invention can ensure that the emission of various pollutants in the flue gas of the incinerator meets the emission standards while minimizing the pollutant control cost of the flue gas purification system and improving the economic efficiency of the incinerator. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly and intuitively illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only the most basic embodiments of the present invention, not all embodiments. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 This is a flow chart of a method for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas proposed by the present invention.

[0055] Figure 2 The data structure diagram of the present invention is a data structure diagram of performing time series processing on the collected data.

[0056] Figure 3 This is a schematic diagram of the intelligent control process of multiple pollutants in waste incineration flue gas according to the present invention.

[0057] Figure 4 This is a module structure diagram of the coordinated prediction and intelligent control system for multiple pollutants in waste incineration flue gas proposed by the present invention.

[0058] Figure 5 It is a heat matrix diagram of the Pearson correlation coefficient between the input characteristics of 32 incinerator operating parameters and the predicted targets of four types of flue gas pollutant concentrations in an embodiment of the present invention.

[0059] Figure 6 This is a line chart comparing the predicted values ​​and original values ​​of the concentrations of four types of flue gas pollutants by the collaborative prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The specific implementation modes of the present invention and the technical solutions in the implementation modes will be further described in detail and completely below in conjunction with the accompanying drawings and examples. Obviously, the implementation modes described are only part of the implementation modes of the present invention, not all of the implementation modes. Based on the implementation modes in the present invention, all other implementation modes obtained by ordinary technicians in the field without creative work belong to the scope of protection of the present invention.

[0061] Part I Implementation of the Invention

[0062] like Figure 1 As shown in the flow chart, the method for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas proposed by the present invention specifically includes the following steps:

[0063] (1) In the same time period, the incinerator operating parameter data stored in the distributed control system (DCS) of the waste incinerator is collected at set time intervals (e.g., every 1 second), and the hydrogen chloride (HCl), sulfur dioxide (SO 2 ), nitrogen oxides (NOx) and particulate matter (PM) emission concentration historical data of four types of flue gas pollutants.

[0064] The DCS of a garbage incinerator is a comprehensive automation system used to monitor and control the operation of the incinerator. It is a general technology widely used in the industry. The system connects multiple control units, sensors, actuators, human-machine interface devices and other equipment through a distributed architecture to achieve real-time monitoring, data collection, automatic control and remote operation of the incinerator. The DCS stores many incinerator operating parameters, including furnace temperature, flue gas temperature, furnace pressure, oxygen content, steam flow, etc. The CEMS of a garbage incinerator is also a general technology widely used in the industry. Its main monitoring components are located at the end of the flue of the garbage incinerator. It consists of a gaseous pollutant monitoring module, a particulate matter monitoring module, a flue gas parameter monitoring module, a data acquisition and processing module, a data transmission module, etc. It can continuously monitor and record the composition and concentration of various pollutants in the incineration flue gas in real time, including HCl, SO 2 , NOx, PM, CO, etc. Therefore, the DCS and CEMS of the incinerator can be used to easily obtain the operating parameter data and flue gas pollutant concentration data of the incinerator.

[0065] In actual engineering application scenarios, there are bound to be some missing values ​​and outliers in the data collected by DCS and CEMS, which will affect the performance of subsequent predictions. Therefore, the statistical method of the 3σ principle and the forward filling method can be used to process the missing values ​​and outliers in the operating parameter data in the data set. The missing values ​​and outliers in the data (data values ​​outside 3 times the standard deviation) are filled with the next normal value. For the four types of pollutant concentration data, the national regulations on the pollutant concentration limit standards in the flue gas emitted by domestic waste incinerators are used as the boundaries, and the data exceeding the corresponding limit range is filled with the data within the next limit range.

[0066] (2) Calculate the Pearson correlation coefficient between the sample incinerator operating parameter data collected in step (1) and the historical data of the emission concentrations of the four types of flue gas pollutants, and select all operating parameters whose absolute value of the Pearson correlation coefficient with the emission concentration data of any one of the four types of flue gas pollutants is greater than 0.3 as input features for collaborative prediction. The emission concentrations of the four types of flue gas pollutants are the final collaborative prediction targets.

[0067] Since there are many sample incinerator operating parameters collected in step (1), not all operating parameters have a positive effect on the coordinated prediction of the emission of multiple pollutants in incineration flue gas. Therefore, the calculation of the Pearson correlation coefficient is used to assist in feature screening. The Pearson correlation coefficient (r) is used to measure the strength and direction of the linear relationship between two variables, and its value range is between [-1,1]. The calculation formula of the Pearson correlation coefficient is as follows:

[0068]

[0069] Where r is the Pearson correlation coefficient between variables x and y, n is the number of observations, and x i represents the i-th observation value of x, y i represents the i-th observation value of y.

[0070] Generally speaking, the following general interpretation can be used for the Pearson correlation coefficient: |r|<0.3 indicates no or very weak correlation, 0.3≤|r|<0.5 indicates weak correlation, 0.5≤|r|<0.7 indicates moderate correlation, 0.7≤|r|<0.9 indicates strong correlation, and |r|≥0.9 indicates extremely strong correlation. Therefore, in order to improve the performance of subsequent collaborative prediction of multiple flue gas pollutants and reduce data dimensions and model complexity, operating parameters with an absolute value of the Pearson correlation coefficient greater than 0.3 with any type of pollutant emission concentration data are selected as input features of the collaborative prediction model, and the remaining operating parameters are discarded. For the retained operating parameters, the emission concentration data of the four types of flue gas pollutants corresponding to the samples at the same time are the data labels corresponding to the input features of the samples at that time.

[0071] (3) The screened incinerator operating parameter data and four types of flue gas pollutant emission concentration data are downsampled to a larger time span (such as 5 minutes). At the same time, for the prediction target of each sample after downsampling, the incinerator operating parameter data of the 12 samples before the sample (corresponding to 60 minutes) are used as the input feature data of the constructed LSTM collaborative prediction model. The input data is time-series processed and a multi-pollutant collaborative prediction model is trained. The model can output the collaborative prediction results of the four types of flue gas pollutant concentrations after the set time period based on the incinerator operating parameter data of the set time period.

[0072] Taking the data collection time interval of 1 second in step (1) as an example, the collected data is high-frequency data at the second level. Even a short time interval will bring a huge amount of data. Therefore, the collected data is averaged according to a larger time span, for example, every 5 minutes is a time window, and finally each sample is the average of the original 5-minute data. This method can reduce the time resolution of the original data and reduce the amount of data, while retaining the overall trend and main features of the data, avoiding the loss of important information due to simply discarding data. In addition, the 5-minute mean downsampling process can also smooth out some short-term random fluctuations and noise, making it easier to focus on the long-term trend or significant changes of the data.

[0073] In addition, the generation of pollutants is not only affected by the state of the incinerator at a certain moment, but depends on the entire process during the incineration of garbage. Therefore, for the prediction model, compared with the conventional data input method of using the input features at a certain moment to predict the pollutant concentration at the corresponding moment, it is more reasonable to use time series data for prediction, which makes it easier for the model to capture the patterns and trends of data changes over time, thereby improving the accuracy and stability of the model prediction. Figure 2 As shown in the data structure diagram, since the collected data has been downsampled to a 5-minute average, each data point corresponds to the original 5-minute average data. At this time, the data at each time point is combined with the data of the previous 11 consecutive time points to form a time series data input model with a length of 12 to predict the concentration of the four types of flue gas pollutants at the next time point, thereby realizing the time series processing of the model input data. In addition, before formally training the collaborative prediction model, the input data needs to be normalized to scale the numerical range of the data to a unified scale to avoid the impact of numerical differences between different features. At the same time, the data set is divided into training set, validation set and test set in a certain proportion according to the chronological order.

[0074] The multi-pollutant collaborative prediction model based on the LSTM layer structure has a structure of 2 LSTM layers + 1 Dense layer, the number of neurons in the Dense layer is 4, the input time step is 12, the output prediction step is 1, the optimizer is adam, the loss function is mse, and the maximum number of iterations is set to 100. The number of neurons in each LSTM layer, the type of activation function, and the learning rate are used as hyperparameters of the model, and the optimal parameters are determined using the grid search method.

[0075] The multi-pollutant collaborative prediction model in the present invention is a recursive neural network (RNN) model that can effectively process time series data, and is particularly suitable for sequence prediction problems with time dependency. The model can capture the patterns and trends of incinerator operating parameters changing over time, and provide an accurate basis for the future collaborative prediction of multiple pollutants in incineration flue gas. The LSTM layer refers to a long short-term memory layer, which mainly determines which information should be retained or forgotten through forgetting gates and memory units. Each LSTM layer contains multiple neurons, which can capture the time dependency in time series data, and the specific number of neurons will be determined by hyperparameter optimization. The Dense layer refers to a fully connected layer, which is usually used to integrate the feature outputs from the previous layer to provide the final prediction. In this model, the number of neurons in the Dense layer is set to 4, corresponding to the prediction targets of the output of four types of pollutants. At the same time, the model accepts the input data of the first 12 time steps (i.e., the first 60 minutes, one data point every 5 minutes) each time to form a time series, and predicts the data of one time step (i.e., one 5 minutes in the future) each time. The hyperparameters of the model, such as the number of neurons in the LSTM layer, the type of activation function, and the learning rate, have an important influence on the prediction performance of the model. To find the best hyperparameter combination, this model uses a grid search method. Grid search can systematically try various hyperparameter combinations to select a set of parameters that make the model perform best.

[0076] The model performance is evaluated using two indicators: mean square error (MSE) and mean absolute error (MAE).

[0077] (4) To control the waste incinerator flue gas purification system to treat HCl, SO 2 Taking the amount of absorbent corresponding to the four types of pollutants, NOx and PM as the target, a multi-objective optimization function F(x) is constructed, including two types of objective functions: a cost index function that considers the amount of absorbent used and an environmental index function that considers the amount of pollutant emissions. Both the cost index function and the environmental index function use the synergistic prediction concentration of multiple flue gas pollutants as input variables. The synergistic prediction concentration of multiple flue gas pollutants refers to the synergistic prediction results of the four types of flue gas pollutant concentrations output by step (3). The final optimization goal is to minimize the optimization function F(x), and the formula is as follows:

[0078] min F(x)=[f 1 (x),f 2 (x),f 3 (x),f 4 (x),g 1 (x),g 2 (x),g 3 (x),g 4 (x)]

[0079] Where f(x) is the cost index function, g(x) is the environmental index function, and subscripts 1, 2, 3, and 4 refer to HCl, SO 2 , NOx and PM: four types of flue gas pollutants.

[0080] The treatment of HCl, SO 2 The absorbents corresponding to the four types of pollutants, NOx and PM, are slaked lime, sodium hydroxide, ammonia water and activated carbon respectively.

[0081] The present invention takes the adjustment of the dosage of four types of absorbents as the control target, thereby linking the incinerator's control over the four types of flue gas pollutants to the control of the dosage of the four types of absorbents. The purpose is to directly influence and reduce the emission of corresponding flue gas pollutants by accurately adjusting the dosage of each type of absorbent, so as to ensure that the emitted flue gas pollutants are always within the compliance standards.

[0082] The calculation formula of the cost index function f(x) is as follows:

[0083]

[0084] Where f(x) is the amount of absorbent used; C in is the concentration of flue gas pollutants at the inlet of the flue gas purification system; C out is the flue gas pollutant synergistic predicted concentration; V is the flue gas flow rate; M Abs is the molecular molar mass of the main reactive component of the absorbent; M P is the molecular molar mass of the main component of a certain type of pollutant; η is the actual pollutant removal efficiency.

[0085] The calculation formula of the environmental protection index function g(x) is as follows:

[0086] g(x)=f(Q)

[0087] Where g(x) is the pollutant concentration; f is the flue gas pollutant synergistic prediction model; Q is the amount of controllable variable absorbent.

[0088] By combining the collaborative prediction model with the multi-objective optimization process, it is possible to meet both cost control and environmental protection requirements. Specifically, the collaborative prediction model provides collaborative prediction results of the concentrations of four types of pollutants in the flue gas of the incinerator, providing a basis for subsequent decision-making and control. These prediction results are directly used as input variables in the multi-objective optimization and provided to the two indicator functions of cost and environmental protection, thereby linking the collaborative prediction model with the multi-objective optimization process. Based on the pollutant concentration output by the collaborative prediction model, the system can dynamically adjust the operation strategy of the incinerator, that is, the amount of pollutant absorbent added. In this way, the optimization model can balance cost and environmental protection requirements and find the optimal solution.

[0089] (5) According to the actual dosage range requirements of the absorbents corresponding to the treatment of the four types of flue gas pollutants in engineering applications and the national emission concentration limit standards for the four types of pollutants, the constraints of the multi-objective optimization function F(x) are set, and the multi-objective optimization algorithm is used to solve the multi-objective optimization function F(x) to calculate the optimal dosage data of the absorbents corresponding to the treatment of the four types of flue gas pollutants.

[0090] The constraints of the multi-objective optimization function are mainly used to limit the solutions in the optimization process, so that the final solution not only achieves the optimal value of the objective function, but also meets various constraints in practical applications. These constraints ensure that the optimization results are feasible in reality and meet the specific requirements of system operation. Only solutions that meet these constraints are considered feasible solutions. For example, in engineering problems, the amount of absorbent used must be within the actual allowable range and cannot exceed the capacity of the equipment or the economically unacceptable level. At the same time, in multi-objective optimization involving environmental protection, constraints ensure that the emission concentration of flue gas pollutants does not exceed the legal limit to avoid violating environmental regulations.

[0091] The multi-objective optimization algorithm is a particle swarm optimization algorithm. The algorithm model is constructed by calling the pso function in the pyswarm library. The number of particles is set to 10 and the maximum number of iterations is 10.

[0092] Particle Swarm Optimization (PSO) is an optimization algorithm based on swarm intelligence. It simulates the behavior of a flock of birds foraging or a school of fish swimming to find the optimal path. PSO explores the search space through a group of individuals called "particles". Each particle represents a potential solution. The particle continuously adjusts its position based on its own experience and the experience of the group to find the global optimal solution. The algorithm first randomly generates a group of particles, and the position of each particle represents a possible solution. Each particle also has an initial velocity, which determines its direction of movement and step size in the search space. The position of each particle is then evaluated by an objective function, which determines the quality of the solution, that is, the "fitness" of the particle. After that, the algorithm will perform velocity updates and position updates to update the individual optimum and the global optimum. Finally, the process of velocity and position updates is repeated until the maximum number of iterations is reached or convergence to a satisfactory solution.

[0093] The velocity of each particle is updated using the following formula:

[0094]

[0095] In the formula, v i (t+1) is the velocity of particle i at time t+1, v i (t) is the velocity of particle i at time t, x i (t) is the position of particle i at time t, is the personal optimal position of particle i, g best is the global optimal position, w is the inertia weight, c 1 and c 2 is the learning factor, which controls the ability of the particle to learn from the personal optimal position and the global optimal position, r 1 and r 2 It is a random number used to introduce randomness, usually between [0,1].

[0096] The particle's position information is updated based on the velocity:

[0097] x i (t+1)=x i (t)+v i (t+1)

[0098] In the formula, x i (t+1) is the position of particle i at time t+1, x i (t) is the position of particle i at time t, v i (t+1) is the velocity of particle i at time t+1, which is the updated velocity mentioned above.

[0099] Pyswarm is a classic Python library that is specifically used to implement the particle swarm optimization algorithm. It provides an easy-to-use pso function to perform optimization tasks. The function defines the objective function through the func parameter, the lower and upper boundaries of the search space through the lb and ub parameters, and the number of particles and the maximum number of iterations through the swarmsize and maxiter parameters. The number of particles determines the breadth of the solution space explored by the algorithm in each iteration. More particles can increase the probability of finding the global optimal solution, but it will increase the computational complexity. The maximum number of iterations determines the length of time the algorithm runs. A larger number of iterations allows the particle swarm to have more opportunities to converge to the global optimal solution.

[0100] After multiple iterations, the particle swarm optimization algorithm will eventually find one (or more) optimal solutions, which represent the pollutant absorbent dosage configuration that can minimize costs and meet environmental protection requirements under given constraints. These optimal solutions are the optimal dosage data of absorbents corresponding to the treatment of four types of flue gas pollutants.

[0101] (6) The optimal dosage optimization results of various absorbents obtained by the multi-objective optimization algorithm are converted into actual analog control signals through the DCS system of the waste incinerator. The control signals are then sent to various absorbent delivery valves in the flue gas purification system to control the valve opening to achieve feedback adjustment of the dosage of absorbents for the four types of flue gas pollutants. Ultimately, while ensuring that the flue gas pollutants meet the environmental emission standards, the dosage of absorbents is reduced, thereby achieving the cost-effectiveness of the incinerator flue gas purification system.

[0102] The analog control signal is a current signal of 4 to 20 mA; when controlling the absorbent dosage to 0%, the valve is fully closed, corresponding to a current signal of 4 mA; when controlling the absorbent dosage to 100%, the valve is fully opened, corresponding to a current signal of 20 mA; in the process of sending the control signal, Modbus is used as the data communication protocol.

[0103] like Figure 3 As shown in the flow chart, the control target of the present invention is to optimize the optimal dosage of pollutant absorbent based on the multi-objective optimization algorithm, so as to effectively control the concentration of four types of pollutants in the waste incineration flue gas, ensure that it meets the environmental emission standards, and optimize the dosage of absorbent to reduce costs. The feedback control strategy is mainly used to convert the optimization results into actual control signals, so as to adjust the operating parameters of the system and adjust the dosage of absorbent in real time.

[0104] By integrating the multi-objective optimization results (optimal dosage of four types of absorbents) with the controller of the actual flue gas purification system, the controller is the distributed control system DCS of the waste incinerator. In addition, it is necessary to ensure that the computer or server running the LSTM model and the optimization algorithm can communicate with the controller, which involves the use of the industrial communication protocol Modbus. The flue gas purification system usually includes some electronic valves, which are mainly used to adjust the dosage of various pollutant absorbents. The electronic valve needs to receive a control signal, usually an analog signal, such as a 4-20mA current signal or a 0-10V voltage signal. By writing the corresponding control logic program in the DCS system in advance, the function of converting the received absorbent dosage data calculated by the optimization algorithm into the corresponding analog output signal is realized. Then the analog control signal is output and sent to the electronic valve to realize the dynamic adjustment of the opening of the electronic valve, thereby realizing the feedback adjustment of the pollutant absorbent dosage.

[0105] like Figure 4 As shown in the module structure diagram, the present invention proposes a multi-pollutant coordinated prediction and intelligent control system for waste incineration flue gas, including the following modules:

[0106] Waste incinerator distributed control system module (DCS): used to collect and store real-time operating parameter data of the waste incinerator;

[0107] Continuous Emission Monitoring System Module (CEMS): It is located at the end of the flue of the waste incinerator and is used to collect and store HCl and SO in the waste incineration flue gas. 2 Real-time emission concentration data of four types of pollutants: NOx and PM;

[0108] Multi-pollutant collaborative prediction module: used to realize data processing, model training, use the model to output the collaborative prediction concentration of multiple pollutants in flue gas and calculate the optimal dosage of absorbent;

[0109] Multi-pollutant intelligent control module: used to send control signals to the opening of various absorbent delivery valves in the incinerator flue gas purification system, reducing the flue gas purification cost of the incinerator system.

[0110] Among them, the garbage incinerator distributed control system module and the flue gas emission continuous monitoring system module belong to the basic data layer of the system, which are used to monitor and collect the operating parameter data and pollutant concentration emission data of the incinerator. The multi-pollutant collaborative prediction module belongs to the calculation and prediction application layer implemented in the form of software programs, which is used to realize data processing, model training, and use the model to output the collaborative prediction concentration of multiple pollutants in flue gas and calculate the optimal dosage of absorbent. The multi-pollutant intelligent control module belongs to the equipment control layer of the system, which is used to receive the collaborative prediction results of the model prediction layer, couple the multi-objective optimization algorithm, and realize the feedback control of the incinerator flue gas purification system by adjusting the amount of absorbents for four types of pollutants. The four modules together constitute the multi-pollutant collaborative prediction and intelligent control system for garbage incineration flue gas. The modules work closely together to realize real-time monitoring, prediction and control of flue gas during garbage incineration, so as to achieve the goal of optimizing pollutant emission control and reducing operating costs. Among them, the development of software programs belongs to the skills that technicians in this field are proficient in, and because it does not belong to the innovative content of the present invention, it will not be elaborated in detail.

[0111] Part II: A specific application example

[0112] In this example, the DCS stored operating parameter data and CEMS monitored HCl and SO for a total of 36 days in a waste incineration plant in Zhejiang Province were collected. 2 The sampling period is 1 second, and there are 3,116,417 samples in total. Among them, the operating parameters include 221 parameters such as main steam pressure, main steam temperature, incineration furnace temperature, and induced draft fan outlet pressure.

[0113] Then, the missing values ​​and outliers in the collected data are processed by the 3σ criterion and forward filling method. Among them, the determination of the outliers of pollutant emission concentration is based on the pollutant concentration limit standard in the flue gas emitted by domestic waste incinerators. In this embodiment, HCl, SO 2 The emission limits for the four types of pollutants, NOx and PM, are 0 to 10 mg / Nm 3 , 0~50mg / Nm 3 , 0~90mg / Nm 3 and 0~10mg / Nm 3 .

[0114] After completing the processing of missing values ​​and outliers, the Pearson correlation coefficient is used to screen out the operating parameters with strong correlation with the four types of flue gas pollutant concentration data from a total of 221 incinerator operating parameters. In this embodiment, after calculation and statistics, 32 incinerator operating parameters are finally determined as input features for collaborative prediction. The Pearson correlation coefficient heat matrix between the 32 input features and the four types of flue gas pollutant concentration prediction targets is shown in the figure below: Figure 5 shown.

[0115] In order to reduce the model complexity and data dimension, and shorten the subsequent model training time, the input data is processed by 5-minute mean downsampling, that is, the original data is collected every 5 minutes instead of every 1 second. In addition, for the pollutant prediction target of each sample, the input feature data of the sample 60 minutes before the corresponding time point (corresponding to 12 sample data points after downsampling) is used for prediction, so as to perform time series processing on the input data of the model. After downsampling and time series processing, the original 1-hour sample data dimension changed from 3600×32 to 12×32, greatly reducing the data dimension. In this embodiment, after this step is completed, 13958 samples remain. Subsequently, the sample data is row normalized. This embodiment adopts the z-score normalization method to make the original sample data become a distribution with a mean of 0 and a standard deviation of 1.

[0116] In this embodiment, the LSTM collaborative prediction model is built based on Python (V3.9.7) and TensorFlow (V2.8.0), and the model structure is 2 LSTM layers + 1 Dense layer, the Dense layer has 4 neurons, the input time step is 12, the output prediction step is 1, the optimizer adam is selected, and mse is selected as the loss function. The maximum number of iterations is set to 100 times. The hyperparameters to be determined in the model include the number of neurons in each LSTM layer, the type of activation function, and the learning rate. In this embodiment, the values ​​of the above hyperparameters determined by grid search are shown in Table 1 below:

[0117] Table 1 LSTM collaborative prediction model hyperparameter settings

[0118] Hyperparameters value Number of LSTM 1 layer units 64 Number of LSTM 2 layer units 64 Activation Function ReLU Learning Rate 0.005

[0119] In addition, in this embodiment, before model training, the data set is divided into a training set, a validation set, and a test set in a 7:2:1 ratio in chronological order.

[0120] After the model training is completed, the mean square error MSE and mean absolute error MAE are used to evaluate the prediction performance of the model. The total MAE and MSE results of the multi-pollutant collaborative prediction model for waste incineration flue gas developed in this embodiment for the prediction of four pollutants on the validation set and the test set, that is, the average MAE and MSE results of the collaborative prediction of the concentrations of four types of pollutants are shown in Table 2 below:

[0121] Table 2 MAE and MSE results of LSTM collaborative prediction model

[0122]

[0123] In this example, in order to more clearly understand the synergistic prediction effect of the model on the four types of pollutants, the following Figure 6 The model shows a line chart comparing the predicted values ​​and original values ​​of the concentrations of four types of flue gas pollutants. In order to make the changing trend of pollutants more intuitive, the data is processed by sliding mean, and the sliding window size is set to 6 and the sliding step is set to 1. As can be seen from the figure, the model has a good collaborative prediction effect on the four types of flue gas pollutants, and can basically predict the fluctuations in the concentrations of the four types of pollutants in a timely manner. Therefore, the accuracy of the model's collaborative prediction results can provide a good guarantee for the subsequent intelligent control of pollutants.

[0124] Since the control of flue gas pollutants involves constraints in many aspects such as cost budget, production scheduling and environmental protection requirements, it is very suitable to be solved by using multi-objective optimization methods. The control of flue gas pollutants in the present invention mainly considers two objectives: one is the cost target of the amount of absorbent corresponding to the four types of pollutants, and the other is the environmental target of ensuring the compliance of pollutant emissions. In this embodiment, the cost index function f(x) and environmental index function g(x) are constructed for each of the four types of flue gas pollutants. Compared with HCl, SO 2 Compared with PM, the control of NOx in waste incineration flue gas is more complicated and cumbersome, and the control processes of the four types of flue gas pollutants are relatively similar. Therefore, this embodiment will mainly take NOx pollutant control as an example to explain the subsequent steps.

[0125] In the flue gas purification system of the garbage incinerator, the absorbent corresponding to the treatment of NOx pollutants is ammonia water. Therefore, the control of NOx by the incinerator is linked to the control of the amount of ammonia water. At the same time, to simplify the calculation, only NO is considered in NOx. The cost indicator function f constructed 3 (x) are as follows:

[0126]

[0127] In the formula, f 3 (x) is the cost of ammonia injection for treating NOx in waste incineration flue gas, is the NOx concentration at the flue gas purification system inlet, is the predicted NOx concentration, V is the flue gas flow rate, and M NO NH 3 and the molecular molar mass of NO, η is the actual NOx removal efficiency, which is taken as 0.8 in this embodiment.

[0128] The constructed environmental index function g 3 (x) are as follows:

[0129] g 3 (x) = f(Q)

[0130] In the formula, g 3 (x) is the concentration of NOx in the flue gas from waste incineration, f refers to the collaborative prediction model, and Q is the controllable variable ammonia dosage.

[0131] The cost index function of NOx control As can be seen from f in the environmental index function, both functions use the NOx concentration prediction results of the flue gas multi-pollutant collaborative prediction model as input, thereby realizing the association between the prediction model and multi-objective optimization. Based on the above two objective functions, the final objective function is constructed as follows:

[0132] minF(x)=k 1 *f 3 (x)+k 2 *g 3 (x)

[0133] The constraints set are as follows:

[0134]

[0135] In the formula, k 1 and k 2 They are the weight coefficient of the cost function and the weight coefficient of the environmental protection function, which are mainly used to adjust the weight distribution between the two objective functions. In this embodiment, both coefficients are 1. At the same time, in order to prevent the control variable from being adjusted too much, the adjustment range of the controllable variable Q is controlled within 5L / h each time.

[0136] Subsequently, the particle swarm optimization method based on Python was used to complete the construction and solution of the optimization model. The number of particles was set to 10 and the maximum number of iterations was 10. Through iterative calculation, the optimal dosage data of ammonia absorbent that met the constraints was obtained. The optimal ammonia dosage data obtained by multi-objective optimization was then converted into an analog control signal and sent to the ammonia dosing control electronic valve in the incinerator flue gas purification system to control the opening of the electronic valve to control the actual flow of ammonia dosing, thereby achieving effective and low-cost control of NOx pollutants in flue gas.

[0137] Correspondingly, this embodiment also builds a waste incineration flue gas multi-pollutant collaborative prediction and intelligent system according to the above-mentioned waste incineration flue gas multi-pollutant collaborative prediction and intelligent control method, the system includes a waste incinerator distributed control system module, a flue gas emission continuous monitoring system module, a multi-pollutant collaborative prediction module and a multi-pollutant intelligent control module, and the four modules cooperate with each other to jointly complete the target task of waste incineration flue gas multi-pollutant collaborative prediction and intelligent control. In this embodiment, Python is mainly used as the system program development language, and the system program is compiled in the MSVC201932bit environment. At the same time, Qt is used as the program development framework, Modbus is used as the data communication protocol, and Visual Studio Code (VS Code) is used as the integrated development environment of the system program.

[0138] In order to verify the effectiveness of the multi-pollutant collaborative prediction and intelligent control system for waste incineration flue gas, in this embodiment, the constructed system is installed on the waste incinerator object applied in this embodiment, and the operation data of the incineration plant is collected for 120 consecutive hours, and the collected data is used to perform an optimization control test on the amount of ammonia water. The results show that the average amount of ammonia water spraying before optimization control is 11.79L / h, and the average amount of ammonia water spraying after optimization is 10.63L / h, and the amount of ammonia water spraying is reduced by 9.84%. While ensuring that the pollutant emissions meet the standards, it can effectively reduce the environmental protection material input cost of the flue gas purification system, achieving the effect of killing two birds with one stone.

[0139] In summary, the method for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas proposed in the present invention can not only realize the accurate collaborative prediction of four types of pollutants in waste incineration flue gas, but also can realize effective intelligent control of four types of flue gas pollutants based on the developed collaborative prediction model and coupled multi-objective optimization method; it not only ensures that the emission of various types of flue gas pollutants meets the current national standards, but also can reduce the cost of absorbent placement of the flue gas purification system of the incinerator to the greatest extent, meeting both environmental protection goals and cost goals. Compared with the traditional method of using CEMS equipment to monitor the emission concentration of flue gas pollutants and manually adjusting the operation of the flue gas purification system equipment based on the monitoring data, the method proposed in the present invention has the advantages of fast, efficient, accurate, collaborative, etc., can avoid the hysteresis caused by CEMS equipment, and greatly save time and labor costs. In addition, the waste incinerator can also greatly improve the intelligence level and cost-effectiveness of the operation of the incinerator by carrying the collaborative prediction and intelligent control system for multiple pollutants in waste incineration flue gas proposed in the present invention, and quickly and conveniently realize the control of various pollutants in the incineration flue gas.

[0140] Obviously, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art may make various subsequent applications, supplements, modifications and variations to the present invention without departing from the spirit and scope of the present invention. If various applications, supplements, modifications and variations based on the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include such applications, supplements, modifications and variations.

Claims

1. A method for collaborative prediction and intelligent control of multiple pollutants in waste incineration flue gas, characterized in that: The method comprises the following steps: (1) Collect all historical data of the operating parameters and flue gas pollutant emission concentrations of the waste incinerator at set time intervals within the same time period; Among them, the incinerator operation parameter data comes from the storage database of the waste incinerator distributed control system (DCS); flue gas pollutants include four categories of hydrogen chloride (HCl), sulfur dioxide (SO2), nitrogen oxides (NOx) and particulate matter (PM), and their respective emission concentration data are all from the storage database of the flue gas continuous emission monitoring system (CEMS); (2) Calculate the Pearson correlation coefficient between the incinerator operating parameters and flue gas pollutant emission concentrations in the collected samples obtained in step (1); screen all operating parameters in the collected samples based on the Pearson correlation coefficient with the emission concentration data of any one of the four types of flue gas pollutants, and retain the screening results as the input features required for collaborative prediction training; the four types of flue gas pollutant emission concentration data corresponding to the samples at the same time are used as data labels corresponding to the input features of the samples at that time; (3) For the data selected in step (2), perform mean downsampling with a larger time span; then select the incinerator operation parameter data of a portion of the samples, perform time series processing on them, and use them as input data for training; A multi-pollutant collaborative prediction model based on the LSTM layer structure was constructed, and the incinerator operation parameter data of the set time period was input to train the model. The collaborative prediction results of the four types of flue gas pollutant concentrations after the set time period were used as the output of the model. (4) The flue gas purification system uses corresponding absorbents to absorb and treat the four types of flue gas pollutants. A multi-objective optimization function F(x) is constructed with the goal of controlling the amount of each absorbent. The function takes into account two types of objective functions: a cost index function of the amount of absorbent used and an environmental index function that considers the amount of pollutant emissions. The final optimization goal is to minimize the multi-objective optimization function F(x). The input variables of the cost index function and the environmental index function are both the flue gas multi-pollutant collaborative prediction concentration output by the multi-pollutant collaborative prediction model. (5) According to the actual dosage range of each absorbent in engineering applications and the emission concentration limit standards of each pollutant, set the constraints of the multi-objective optimization function F(x); use the multi-objective optimization algorithm to solve the function and calculate the optimal dosage data of the absorbent corresponding to each of the four types of flue gas pollutants; (6) The calculation result of step (5) is input into the DCS system of the waste incinerator, and the optimal dosage of each absorbent is converted into an actual analog control signal, which is then sent to the delivery valve of each absorbent in the flue gas purification system; the feedback adjustment of the dosage of each absorbent is achieved by controlling the valve opening, thereby reducing the dosage of the absorbent while ensuring that the flue gas pollutants meet the environmental emission standards, thereby achieving the cost-effectiveness goal of the incinerator flue gas purification system.

2. The method according to claim 1, characterized in that In the step (2), operating parameters whose absolute value of the Pearson correlation coefficient with any type of pollutant emission concentration data is greater than 0.3 are selected, and these operating parameters are used as input features of the collaborative prediction model, and the remaining operating parameters are discarded.

3. The method according to claim 1, characterized in that The time interval for collecting data in step (1) is 1 second, and the data screened in step (2) is downsampled to a 5-minute mean value; for the prediction target of each sample after downsampling, the incinerator operation parameter data of the 12 samples before the sample are used as the input feature data of the multi-pollutant collaborative prediction model.

4. The method according to claim 1, characterized in that: In step (3), the multi-pollutant collaborative prediction model has 2 LSTM layers and 1 Dense layer; the number of neurons in the Dense layer is 4, the input time step is 12, the output prediction step is 1, the optimizer is adam, the loss function is mse, and the maximum number of iterations is set to 100 times; the number of neurons in each LSTM layer, the type of activation function and the learning rate are used as the hyperparameters of the model, and the optimal parameters are determined by the grid search method.

5. The method according to claim 1, characterized in that In step (4), the absorbents corresponding to the absorption and treatment of the four types of pollutants, namely HCl, SO2, NOx and PM, are slaked lime, sodium hydroxide, ammonia water and activated carbon, respectively.

6. The method according to claim 1, characterized in that In step (4), the expression of the multi-objective optimization function F(x) is as follows: min F(x) = [f1(x), f2(x), f3(x), f4(x), g1(x), g2(x), g3(x), g4(x)] Among them, f(x) is the cost index function, g(x) is the environmental index function, and the subscripts 1, 2, 3, and 4 refer to the four types of flue gas pollutants, namely HCl, SO2, NOx, and PM, respectively; The calculation formula of the cost index function f(x) is as follows: ； Where, f(x) is the amount of absorbent; C in is the concentration of flue gas pollutants at the inlet of the flue gas purification system; C out is the flue gas pollutant synergistic predicted concentration; V is the flue gas flow rate; M Abs is the molecular molar mass of the main reactive component of the absorbent; M P is the molecular molar mass of the main component of a certain type of pollutant; η is the actual pollutant removal efficiency; The calculation formula of the environmental protection index function g(x) is as follows: ； Where g(x) is the pollutant concentration; f is the synergistic prediction model of flue gas pollutants; Q is the dosage of the controllable variable absorbent.

7. The method according to claim 1, characterized in that In step (5), the multi-objective optimization algorithm is a particle swarm optimization algorithm, and its algorithm model is constructed by calling the pso function in the pyswarm library. The number of particles is set to 10 and the maximum number of iterations is 10.

8. The method according to claim 1, characterized in that In step (6), the analog control signal is a current signal of 4 to 20 mA; when the absorbent dosage is controlled to be 0%, the valve is fully closed, corresponding to a current signal of 4 mA; when the absorbent dosage is controlled to be 100%, the valve is fully opened, corresponding to a current signal of 20 mA; in the process of sending the control signal, Modbus is used as the data communication protocol.

9. A multi-pollutant collaborative prediction and intelligent control system for waste incineration flue gas, characterized in that: The control system includes the following modules: Waste incinerator distributed control system module: used to collect and store real-time operating parameter data of the waste incinerator; Flue gas emission continuous monitoring system module: located at the end of the flue of the waste incinerator, it is used to collect and store real-time emission concentration data of four types of pollutants, namely HCl, SO2, NOx and PM, in the flue gas of waste incineration; Multi-pollutant collaborative prediction module: used to perform the operations described in steps (1) to (5) of claim 1, realize data processing, model training, use the model to output the collaborative prediction concentration of multiple pollutants in flue gas and calculate the optimal dosage of absorbent; Multi-pollutant intelligent control module: used to perform the operation described in step (6) of claim 1, send control signals to the openings of various absorbent delivery valves in the incinerator flue gas purification system, and reduce the flue gas purification cost of the incinerator system.

Images