Multivariable fluidized bed temperature control method and system based on RBF neural network

Abstract

The invention belongs to the technical field of industrial automation control, and discloses a multivariable fluidized bed temperature control method and system based on an RBF neural network. Comprising an air intake and exhaust processing module, a fluidized bed body, a heating module, a temperature sensing module and a control module, and comprises the following steps: S1, establishing a nonlinear autoregression moving average model; s2, obtaining a predicted temperature of a future # imgabs0 # step; s3, correcting the predicted temperature; s4, constructing an objective function of the RBF neural network, and performing iterative optimization solution on the input; and S5, outputting the optimized signal to a fluidized bed temperature control system in real time. According to the method, the multivariable coupling relation is accurately modeled through the RBF neural network, and cooperative adjustment of the cold and hot air valve, the heater and the air inlet amount is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of industrial automation control technology and discloses a multivariable fluidized bed temperature control method and system based on RBF neural network, which is suitable for precise temperature control of fluidized bed equipment in the pharmaceutical, food, chemical and other industries. Background Art

[0002] In recent years, an increasing number of pharmaceutical companies have established solid granulation lines, and the use of fluidized beds has become increasingly widespread. Fluidized bed drying technology is currently widely used in various industries, including pharmaceuticals, food chemistry, and environmental technology. Its advantages lie in its intensive heat and mass transfer, targeted temperature control, and compact structure. In most cases, heat input to the fluidized bed is achieved through a combination of convective heating of the fluidized medium and contact heat transfer with the wet material. Therefore, the control of the fluidized bed temperature significantly affects the drying effect and efficiency. However, temperature control faces the following technical difficulties: (1) Large hysteresis and nonlinearity: The drying process involves complex heat and mass transfer, the system dynamic response is slow, and the input-output relationship is highly nonlinear, making it difficult to establish an accurate mathematical model.

[0003] (2) Multiple coupling and strong interference: Factors such as ambient temperature fluctuations, air volume changes, and sensor noise cause significant interference to temperature control.

[0004] (3) Limitations of traditional control methods: Existing technologies mostly use single-input single-output (SISO) PID control, which cannot effectively handle the coupling problem of multiple-input multiple-output (MIMO) systems and has insufficient regulation capabilities for large lag and nonlinear systems, resulting in low control accuracy and slow response speed.

[0005] For example, the publication number CN107765730A discloses a fluidized bed temperature control method and control device, which includes: determining a fluidized bed temperature set value; setting control parameters of a fluidized bed temperature controller and a heat exchange medium temperature controller according to a first parameter group; measuring the actual value of the fluidized bed temperature and the actual value of the heat exchange medium temperature; calculating the output value of the fluidized bed temperature controller based on the current corresponding fluidized bed temperature set value, the current actual value of the fluidized bed temperature, and the control parameters of the current fluidized bed temperature controller; calculating the output value of the heat exchange medium temperature controller based on the calculated output value of the fluidized bed temperature controller, the current actual value of the heat exchange medium temperature, and the control parameters of the current heat exchange medium temperature controller; and controlling the opening of a regulating valve of the temperature control medium so that the output value of the heat exchange medium temperature controller is the calculated output value of the heat exchange medium temperature controller. Although this invention can achieve automatic temperature control during the fluidized bed heating stage to a certain extent, it still belongs to the traditional method of temperature control and cannot solve the above-mentioned technical problems. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides a multivariable fluidized bed temperature control method and system based on RBF neural network.

[0007] In a first aspect, a multivariable fluidized bed temperature control method based on an RBF neural network comprises the following steps: S1, based on historical data, the RBF neural network is trained to establish a multivariable nonlinear autoregressive moving average model of the fluidized bed to obtain a recursive prediction model; S2, according to the recursive prediction model, recursively obtain the future The predicted temperature of the first step; S3, real-time acquisition of the current temperature measured value, dynamic compensation of the predicted temperature, and obtaining the corrected predicted temperature; S4, constructing an objective function with the deviation between the corrected predicted temperature and the set value and the increment of the multi-input control vector as constraints, modifying the RBF neural network with the objective function, and iteratively optimizing the multi-input control vector using the modified RBF neural network; S5, outputs the optimized multi-input control vector to the fluidized bed temperature control system in real time, realizing multi-variable coordinated adjustment and making the actual temperature quickly track the set value.

[0008] Optionally, in step S1, the fluidized bed multivariate nonlinear autoregressive moving average model is specifically: , Where n is the system output order, u is the control signal of the multi-input control vector, k is the discrete time sequence number of the current sampling moment, m is the system input order, d is the number of delay steps, and y(k) is the output of the fluidized bed temperature control system; and, , For input Dimensions; The control signals of the fluidized bed temperature control system include the opening degree signals of the cold and hot air valves and the heater opening degree signals.

[0009] Optionally, in step S3, the current temperature measured value is collected in real time, and the predicted temperature is dynamically compensated to obtain a corrected predicted temperature, specifically: 1) According to the recursive prediction model in step S1, the predicted temperature at the current moment is predicted; 2) Calculate the difference between the predicted temperature at the current moment and the actual temperature at the current moment to obtain the prediction error at the current moment; 3) The prediction error at the current moment is compared with the future prediction error in step S2. The predicted temperature of the first step is dynamically compensated to obtain the corrected predicted temperature.

[0010] Optionally, in step S4, the objective function Specifically: , in, To set the temperature, To correct the predicted temperature, Optionally, in step S5, the multi-input control vector includes the opening of the hot and cold air valves, the heater power and the air intake volume.

[0011] Optionally, the method is implemented by a PLC controller, the programming language of the control algorithm is SCL language, and the sampling period is set to 10ms-1s.

[0012] In a second aspect, a multivariable fluidized bed temperature control system based on an RBF neural network includes: Inlet and exhaust air processing module, used to regulate the inlet and exhaust air; fluidized bed; Heating module, used to adjust the heater opening and the hot and cold air valve opening; Temperature sensor module, real-time detection of inlet air temperature; The control module controls the RBF neural network algorithm to perform optimization iterations, and displays and executes the control results.

[0013] Optionally, the control module includes: The actuator controls other modules of the fluidized bed temperature control system based on the multi-input control vector signal iterated by the RBF neural network algorithm optimization; The display device is used to display data from various links of the fluidized bed temperature control system.

[0014] The present invention has the following advantages: (1) Multivariable decoupling control: The multivariable coupling relationship is accurately modeled through the RBF neural network to achieve coordinated adjustment of the hot and cold air valves, heaters, and air intake volume.

[0015] (2) High temperature control accuracy and fast response: The recursive prediction and error compensation mechanism narrows the temperature fluctuation range and greatly reduces the response time.

[0016] (3) Strong anti-interference capability: The MIMO control strategy can maintain stability even when some sensors fail. When facing a single sensor or actuator failure, it maintains stable operation through other normal input and output signals, thereby improving the overall reliability and stability of the system.

[0017] (4) Scalability: The system supports modular design and can be adapted to fluidized bed equipment of different specifications and meet diverse process requirements by adjusting network parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the solution in this application, the following will describe the methods required in the embodiment of this application. A brief introduction to the drawings is given. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 This is a flow chart of fluidized bed temperature control in an embodiment of the present invention; Figure 2 A schematic diagram of a neural network model in an embodiment of the present invention; Figure 3 2 is a system structure diagram in an embodiment of the present invention. DETAILED DESCRIPTION Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those in the technical field of this application. The meanings generally understood by technicians are the same; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" in the specification and claims of the present application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0020] This embodiment provides a multivariable fluidized bed temperature control method and system based on RBF neural network.

[0021] In the first aspect, a multivariable fluidized bed temperature control method based on RBF neural network is provided. Figure 1 As shown, the following steps are included: S1, based on historical data, the RBF neural network is trained to establish a multivariable nonlinear autoregressive moving average model of the fluidized bed to obtain a recursive prediction model; S2, according to the recursive prediction model, recursively obtain the future The predicted temperature of the first step; S3, real-time acquisition of the current temperature measured value, dynamic compensation of the predicted temperature, and obtaining the corrected predicted temperature; S4, constructing an objective function with the deviation between the corrected predicted temperature and the set value and the increment of the multi-input control vector as constraints, modifying the RBF neural network with the objective function, and iteratively optimizing the multi-input control vector using the modified RBF neural network; S5, outputs the optimized multi-input control vector to the fluidized bed temperature control system in real time, realizing multi-variable coordinated adjustment and making the actual temperature quickly track the set value.

[0022] Optionally, in step S1, the fluidized bed multivariate nonlinear autoregressive moving average model is specifically:

[0023] (1), Where n is the system output order, u is the control signal of the multi-input control vector, k is the discrete time sequence number of the current sampling moment, m is the system input order, d is the number of delay steps, and y(k) is the output of the fluidized bed temperature control system; and, , For input Dimensions; The control signals of the fluidized bed temperature control system include the opening degree signals of the cold and hot air valves and the heater opening degree signals.

[0024] Optionally, according to formula (1), the recursive prediction model in step S1 is: (2), in, For the The predicted output of the actuator at the moment; According to formula (2), the future Predicted temperature of the first step , specifically (3).

[0025] Optionally, in step S3, the current temperature measured value is collected in real time, and the predicted temperature is dynamically compensated to obtain a corrected predicted temperature, specifically: 1) According to the recursive prediction model in step S1, predict the predicted temperature at the current moment ; 2) The predicted temperature at the current moment The actual temperature at the current moment Perform difference calculation to get the prediction error at the current moment , The error compensation model is: (4); 3) The prediction error at the current moment For the future in step S2 Predicted temperature of the first step Perform dynamic compensation to obtain the corrected predicted temperature, that is, Combining equations (3) and (4), the corrected predicted temperature in step S3 can be obtained: ,Right now (5) Where h is the number of neurons in the hidden layer.

[0026] Optionally, in step S4, an objective function is constructed with the deviation between the predicted temperature and the set value and the multi-input control vector increment as constraints. It is necessary to calculate the of , so that the corrected temperature prediction output value As close to the set temperature as possible ,In the control process, it is often not desirable for the actuator to control the increment changes too quickly, so it is used as a constraint to obtain The objective function at the moment , specifically (6) In order to get Optimal control value, considering the objective function , using the gradient descent method, let ,get (7) According to formula (7), combined with formula (3) and formula (4), since Output before time and predicted output and It doesn't matter, so we can get (8) According to formula (2), we can know that hour, Only in When the control signal Therefore, the modified neural network prediction model is as follows: Figure 2 As shown, the neural network model contains one hidden layer, a single input neuron and multiple output neurons, so (9) Among them, h is the number of neurons in the hidden layer, The dimension of the actuator output signal, the control signal Including the actuator opening of the fluidized bed temperature control system, is the delay step number, , , is the radial basis vector of the RBF network structure, For the The center vector of each node, For the hidden neurons to the The weight value between the outputs of the actuators is obtained as follows: Optional, such as Figure 2 As shown, , For input The dimension of , let the radial basis vector of RBF neural network structure be , the base width vector is , the weight vector is ,in is the number of neurons in the hidden layer, wherein in step S4, the hidden layer activation function of the RBF neural network is a Gaussian function, and the Gaussian function is: , (10) in, is the radial basis vector of the RBF network structure, The center vector of the node is , ,in For input Dimensions; Will Time prediction output Expressed as (10) Then get (11) because The control signal at the moment, that is, the actuator opening The value is unknown, so take the initial value first , using an iterative algorithm to find the optimal In practice, due to the limitation of sampling period, a maximum number of iterations should be specified, so that even if the optimal solution is not reached in the end, a more satisfactory solution can still be obtained along the direction of the optimal solution.

[0027] Optionally, in step S5, the multiple input control vectors include the opening of the hot and cold air valves, heater power and air intake volume.

[0028] Optionally, the method is implemented by a PLC controller, the programming language of the control algorithm is SCL language, and the sampling period is set to 10ms-1s.

[0029] In a second aspect, a multivariable fluidized bed temperature control system based on an RBF neural network includes: Inlet and exhaust air processing module, used to regulate the inlet and exhaust air; fluidized bed; Heating module, used to adjust the heater opening and the hot and cold air valve opening; Temperature sensor module, real-time detection of inlet air temperature; The control module controls the RBF neural network algorithm to perform optimization iterations, and displays and executes the control results.

[0030] Optionally, the control module includes: The actuator controls other modules of the fluidized bed temperature control system based on the multi-input control vector signal iterated by the RBF neural network algorithm optimization; The display device is used to display data from various links of the fluidized bed temperature control system. Example 1

[0031] This embodiment aims at fluidized bed temperature control and designs a multivariable fluidized bed temperature control method and system based on S71200 PLC controller and RBF neural network. The system uses SCL language to program the RBF neural network intelligent fluidized bed multi-input temperature control algorithm.

[0032] like Figure 3 As shown, the system consists of an air handling module, a heating module, a fluidized bed, a temperature sensing module, and a control module. The air handling module primarily includes the air handling motor; the heating module includes an electric heater and hot and cold air valves; the temperature sensing module includes a PT100 thermometer; and the control module includes an S7 1500 PLC and a human-machine interface (HMI). In this embodiment, the air handling motor has an 18.5 kW power output and a maximum speed of 2900 rpm. A 60 kW electric heater with a maximum temperature of 200°C is installed inside the air handling motor to heat the incoming air. The heater's opening can be adjusted from 0-100% by an electric regulator, which can be controlled via the PLC's AO module using a 0-20 mA analog signal. The temperature control system also uses hot and cold air valves to control the incoming air volume, further enabling precise control of the incoming air temperature. The valve's opening is controlled via the PLC's AO module using a 0-10 V analog signal. The actual inlet air temperature is measured by a PT100 thermometer with a range of 0-200°C. The temperature is converted by the control module and transmitted to the PLC. The air is filtered by the heating module and then enters the fluidized bed, where it lifts the material and causes it to boil. The S7 1500 processor performs both offline and online calculations of the RBF neural network. A touch screen serves as the human-machine interface, enabling data input, command control, and parameter display and data storage.

[0033] The specific training method of the present invention is: 1) Train the RBF neural network based on historical data and establish a nonlinear autoregressive moving average model of multiple variables in the fluidized bed (1) in The inlet air temperature value actually measured by the PT100 of the fluidized bed temperature control system, the system output order , The control signals for the fluidized bed temperature control system include the heater opening and the hot and cold air valves, i.e. , dimension , system input order , delay steps , where the control system order and delay steps need to be selected according to actual conditions. The higher the order, the greater the amount of calculation required.

[0034] In practical applications, the sampling period is usually 100ms, and the appropriate maximum number of iterations is 200. In this way, even if the optimal solution is not reached in the end, a satisfactory solution is still obtained along the direction of the optimal solution.

[0035] Fixed the inlet air speed and temperature setting value, measured the corresponding heater opening value and hot and cold air valve opening value without control, and the actual temperature output value, and obtained the measurement data as shown in Table 1 below.

[0036]

[0037] Table 1 Consider the input of this control module The dimension is 2, the number of hidden layer neurons , there are the following Figure 2 The neural network structure shown needs to be trained, that is, the open-loop input and output signals of the controlled module (data in Table 1) are obtained by pre-measurement and used as samples to repeatedly train the neural network model, and finally a recursive prediction model is obtained.

[0038] 2) Set the control amount The initial value of .

[0039] 3) At this moment, the system inlet air temperature is measured , using the predicted value , calculate the model bias correction .

[0040] 4) Take , Will Input the recursive multi-step prediction model formula (3) and get its The predicted value of the inlet air temperature at time t is

[0041] 5) Obtained from steps 3) and 4) and According to formula (5), the error compensation predicted temperature is obtained

[0042] 6) Reuse the formula to find new .

[0043] 7) Judgment Whether the requirements are met, that is, whether the absolute value of the error between the predicted value and the actual value is less than the set value ,set up , or if the number of iterations exceeds , then go to step 8); otherwise go to step 4), using the new Value calculation.

[0044] 8) The result obtained in step 7) The actual air inlet temperature is generated by sending it to the control module and recursive prediction model. Temperature predicted by the model .

[0045] 9) At the next moment, use the real-time actual inlet air temperature Go back to step 3) and calculate the control signal for the next step The control signal is output to the fluidized bed temperature control system in real time to achieve multi-variable coordinated adjustment, so that the actual temperature can quickly track the set value.

Claims

1. A multivariable fluidized bed temperature control method based on RBF neural network, characterized in that: The following steps are involved: S1, based on historical data, the RBF neural network is trained to establish a multivariable nonlinear autoregressive moving average model of the fluidized bed to obtain a recursive prediction model; S2, according to the recursive prediction model, recursively obtain the future The predicted temperature of the first step; S3, real-time acquisition of the current temperature measured value, dynamic compensation of the predicted temperature, and obtaining the corrected predicted temperature; S4, constructing an objective function with the deviation between the corrected predicted temperature and the set value and the increment of the multi-input control vector as constraints, modifying the RBF neural network with the objective function, and iteratively optimizing the multi-input control vector using the modified RBF neural network; S5, outputs the optimized multi-input control vector to the fluidized bed temperature control system in real time, realizing multi-variable coordinated adjustment and making the actual temperature quickly track the set value.

2. The method for controlling temperature of a fluidized bed based on a multivariable RBF neural network according to claim 1, characterized in that: In step S1, the nonlinear autoregressive moving average model of the fluidized bed multivariate is specifically: ， Where n is the system output order, u is the control signal of the multi-input control vector, k is the discrete time sequence number of the current sampling moment, m is the system input order, d is the number of delay steps, and y(k) is the output of the fluidized bed temperature control system; and, , For input Dimensions; The control signals of the fluidized bed temperature control system include the opening degree signals of the cold and hot air valves and the heater opening degree signals.

3. The multivariable fluidized bed temperature control method based on RBF neural network according to claim 1, characterized in that: In step S3, the current temperature measured value is collected in real time, and the predicted temperature is dynamically compensated to obtain the corrected predicted temperature, specifically: 1) According to the recursive prediction model in step S1, the predicted temperature at the current moment is predicted; 2) Calculate the difference between the predicted temperature at the current moment and the actual temperature at the current moment to obtain the prediction error at the current moment; 3) The prediction error at the current moment is compared with the future prediction error in step S2. The predicted temperature of the first step is dynamically compensated to obtain the corrected predicted temperature.

4. The method for controlling temperature of a fluidized bed based on a multivariable RBF neural network according to claim 1, characterized in that: In step S4, the objective function is specifically: ， in, is the objective function, To set the temperature, To correct the predicted temperature, Increment for the multi-input control vector.

5. The method for controlling temperature of a fluidized bed based on a multivariable RBF neural network according to claim 1, characterized in that: In step S5, the multiple input control vectors include the opening of the hot and cold air valves, the heater power and the air intake volume.

6. A multivariable fluidized bed temperature control system based on RBF neural network, characterized in that: include: Inlet and exhaust air processing module, used to regulate the inlet and exhaust air; fluidized bed; Heating module, used to adjust the heater opening and the hot and cold air valve opening; Temperature sensor module, real-time detection of inlet air temperature; The control module controls the RBF neural network algorithm to perform optimization iterations, and displays and executes the control results.

7. A multivariable fluidized bed temperature control system based on RBF neural network according to claim 6, characterized in that: The control module includes: The actuator controls other modules of the fluidized bed temperature control system based on the multi-input control vector signal iterated by the RBF neural network algorithm optimization; The display device is used to display data from various links of the fluidized bed temperature control system.

Images