CN102622000B - Fuzzy neural network-based slurry supply system flow control method - Google Patents

Abstract

The invention discloses a fuzzy neural network-based slurry supply system flow control method. According to the method, a programmable logic controller (PLC) serves as a controller; the rotating speed of a paper machine is obtained by an encoder; and flow change parameters which need to be controlled are calculated through a fuzzy neural network by combining the changes of the speed and the concentration of the current paper machine by the PLC according to parameters designed by the system. By the method, the stable work of the slurry supply system can be guaranteed, so that the quality stability of paper is guaranteed.

Description

A Flow Control Method of Slurry Supply System Based on Fuzzy Neural Network

technical field

The invention relates to a flow control method of a slurry supply system based on a fuzzy neural network.

Background technique

The characteristics and development trend of the slurry supply system in the 21st century are that the process strives for simplicity and efficiency, and the system control is sensitive. With the development of the times, the progress of society, and the continuous improvement of manufacturing technology and production technology, high-tech products and new technologies have been introduced into the field of papermaking. The future paper machines will develop in the direction of large-scale, high-speed, high-efficiency, and low-cost products. The pulp supply system tries to increase the pulp concentration on the net, reduce the return flow of white water, shorten the process, reduce the volume of efficient degassing device, and reduce the consumption of unit products, which will be the inevitable trend of the future development of pulp supply system for papermaking. Main functions of paper machine pulp supply system 1. Pulp and chemicals are uniformly mixed with white water in various proportions in the pulp supply system. 2. Effectively remove impurities brought into the system by auxiliary materials such as slurry, fillers, chemicals, etc., and improve the cleanliness of slurry on the Internet. Can effectively remove the air in the system. 3. After the slurry passes through the system, it can be well dispersed and the floc such as fiber bundles can be removed. 4. Ensure the stability of the concentration, flow and pressure of the slurry entering the headbox and reduce the pulse caused by the performance of the equipment (pump, screen) as much as possible. 5. Shorten as much as possible the time required for the system to stabilize after changing varieties and adjusting processes. If the system is stable, the flow control is particularly important.

Fuzzy control is a nonlinear control based on fuzzy set theory, fuzzy language variables and fuzzy reasoning. It has become an important and effective form of intelligent control. The fundamental difference between fuzzy control and classical control is that it does not need to establish a precise mathematical model of the controlled object, but completely relies on human experience and knowledge, summarizes and formally describes the experience of technicians, and expresses it into a set of qualitative Conditional statements and imprecise decision rules, and then use fuzzy sets as a tool to quantify them. Through fuzzy logic and approximate reasoning methods, human knowledge and experience can be turned into a control model acceptable to computers, so that computers can replace people. Take control.

The purpose of the control system is to determine the appropriate input control quantity, so that the actual output of the system is close to the desired output. The neural network adjusts the connection weight of the neural network through the error between the actual output of the system and the expected output, that is, the process of letting the neural network learn until the error tends to zero is the basic idea of the neural network to achieve direct control.

Both neural network and fuzzy set theory are methods between the symbolic reasoning of traditional artificial intelligence and the numerical calculation of traditional control theory. Both have some basic characteristics in common, and they can be considered as complementary. Fuzzy control utilizes the prior knowledge of domain experts to perform approximate processing, but it lacks the ability of online learning or self-adjustment for nonlinear systems with time-varying parameters in practical engineering applications. How to automatically generate or adjust the membership function or adjust the fuzzy rules is a very complicated problem. Neural network is a typical black-box learning mode. After learning, the input/output relationship obtained by neural network cannot be expressed in an easily acceptable way; although neural network has parallel computing, distributed information storage, fault tolerance It has a series of advantages such as strong and adaptive learning functions, but it is not suitable for expressing rule-based knowledge. Therefore, when the neural network is trained, because it cannot make good use of the existing empirical knowledge, it often can only set the initial weight to zero or a random number, thereby increasing the training time of the network or falling into an unrequired local extremum. How to combine the two advantages of fuzzy theory's easy knowledge expression and neural network's strong self-learning ability, learn from each other's strengths, and improve the learning ability and expression ability of the entire system is a problem that the entire control engineering field needs to solve.

Since the width of the paper produced by the paper machine is fixed, as long as the flow and concentration of the pulp provided by the pulp supply system are known, and the speed of the paper machine is known, the weight of the paper can be calculated.

L is the width of the paper; Q: water flow; P: the concentration of the pulp; V: the operating speed of the paper machine.

The weight of paper per square meter without considering the ash content is:

If you know the weight of the ash: M _ash , you can find the weight M _paper of the paper. The formula is:

Among them: M _ash = M _paper × ash coefficient

When the pulp concentration changes, in order to keep the thickness of the paper constant, that is, to keep the weight of the paper constant, the flow Q must be controlled to keep the weight of the paper constant. Flow control becomes an important purpose for the stability of the slurry supply system.

Contents of the invention

The purpose of the present invention is to solve the above problems, to provide a method for controlling the flow of pulp supply system based on fuzzy neural network, which is used to control the flow of pulp when the concentration changes. After obtaining the speed of the paper machine, the PLC calculates the change parameters of the flow that should be controlled through the fuzzy neural network according to the parameters of the system design, combined with the current speed and concentration changes of the paper machine.

To achieve the above object, the present invention adopts the following technical solutions:

A flow control method of the pulp supply system based on fuzzy neural network, which uses PLC as the controller to obtain the speed of the paper machine through the encoder, PLC according to the parameters of the system design, combined with the current speed and concentration changes of the paper machine, Calculate the change parameters of the traffic that should be controlled through the fuzzy neural network. The specific process is:

When the slurry supply system is running, record the current concentration value C and the current flow value Q _before , when the system concentration is changed _to C, use _the formula And M _paper = QC + M _{ash content After} calculating the corresponding pulp flow Q under the current concentration, the change of flow ΔQ = _after Q - _before Q; if ΔQ is greater than zero, it means that the concentration has decreased, and the flow needs to be increased. If ΔQ is less than zero, it means that the concentration Increased, the corresponding flow should decrease; flow control is achieved by adjusting the acceleration and deceleration of the motor, using a method of stably controlling the acceleration and deceleration by adjusting the acceleration time, the design of the acceleration and deceleration per unit time, only need to know how much The required speed value can be reached within a certain time, that is, the longer the switch is turned on, the longer the acceleration or deceleration time will be. Value, several sets of actual values are obtained through on-site testing, and the proportional value Y is calculated by ΔQ based on the fuzzy neural network. Multiplying ΔQ by the proportional value is the time of this speed up and down, and then the time interval of the valve action is input with the touch screen, which is the pulse width. .

The input of the fuzzy neural network is two variables X1 (concentration change value ΔC), X2 ( _before the current concentration value C), and the output variable is the proportional value Y of the flow rate, and each input factor is divided into: NM, NS, ZO , PS, PM these five fuzzy states; the fuzzy neural network is used for learning, and online learning is selected during learning, and the learning speed remains unchanged during online learning. When the control object is determined, the index value is determined based on experience, and the specific learning steps are:

① The initial values of θji, σji, ωi and η are randomly selected between [0,1], and the value of η is a constant value, which is determined based on experience;

②Calculate the ideal values of θji(k+1), σji(k+1), and ωi(k+1) according to the fuzzy neural algorithm;

③ Calculate E according to the fuzzy neural algorithm. If E≤0.002, the iteration ends; otherwise, set θji(k+1), σji(k+1), ωi(k+1) as initial values and return to ②.

The weight is calculated through the cooperation of the upper computer and the lower computer; the upper computer establishes a fuzzy neural network, the lower computer collects data, and the lower computer continuously collects data and sends it to the upper computer to continuously modify the network until the network meets the requirements.

The specific process is as follows:

Lower computer operation during the learning process:

(1) Initialization

Initialize the sample value (including two input values: the concentration change value, the current concentration value and an output value: the flow change value) and prepare for the subsequent transmission of the sample value. The learning stage continuously writes the sample value to the storage address, and at the same time Also continuously send the sample value to the host computer;

(2) Accept the request

Accept the command request from the upper computer, decide whether the lower computer sends sample data to the upper computer, if it is to execute step (3), otherwise stop;

(3) Send data

Send the sample data to the upper computer through the communication of the lower computer, end the transmission of the sample after sending, start a new acceptance, and wait for the upper computer to intercede the data transmission signal;

The specific process of the upper computer during the learning process is as follows:

(1) Initialization

First randomly select the initial values of θji, σji, ωi, and η within [0, 1], and then assign values to the constants used in the learning process, which are also assigned to the storage unit; finally, request the lower computer to transmit sample data;

(2) Initial value calculation

1) Since data transmission is requested during initialization, the data is first obtained through the communication program of the upper computer, and after receiving the sample data, the lower computer is notified not to transmit data;

2) Then use the initialized first set of weights to calculate the output value when the first set of sample values are input, the difference between the output value and the expected output value, and the data to be used for subsequent calculations;

3) Calculation of weight and performance index E value

Calculate weight and E value on the basis of above-mentioned 2) step;

4) E value judgment Compare the calculated E value with 0.002; if E ≤ 0.002, it means that the calculated function variables and weights have reached the expected goal, and the learning process is over; at the end, the switch value of the external device is triggered at the same time, and the external device is used Read these calculation results; on the contrary, you need to continue the learning process, and assign the function variables and weights calculated in step (3) that do not meet the performance indicators to the address required to recalculate the t value in the next step, and put the requested data The flag is set and sent to the lower computer to prepare for the calculation of the new t value;

In the present invention, the program design function of the upper computer in the on-site work is consistent with the learning stage, and the main difference is that the sample values initialized in the learning stage and that need to be transmitted to the lower computer become the values collected on-site by external devices. In the initialization of the PLC program, assign the collection value from the address of the external device to the data area of the sending area. Because the collection value changes within a certain period, it is real-time. Therefore, there is no need for an address pointer to synchronize the work of the two. The lower computer program is realized, and the specific steps of the lower computer in the field work process: (1) Initialization. The initialization of the lower computer firstly assigns the values of θji, σji, ωi and η trained in the learning process to the storage unit; secondly, assigns the constants used in the subsequent t value calculation process to the storage unit as well ; Finally, send the requested data to the host computer. (2) Receive the collection value. First receive the collected value from the host computer, and then assign the collected value to the storage address where the t value operation will be performed. At the same time, the upper computer is required to stop and continue to transmit the collected values to the lower computer. (3) Output t value calculation. Use the collected data provided in the previous step, the weights in the initialization step and the fuzzy neural network algorithm to assign the calculated value to the storage address of the external output device, and control whether the requested data is sent according to the site situation. (4) Data transmission judgment. If it needs to be sent, skip to the second step. Otherwise end.

The beneficial effect of the present invention is: utilize the characteristic of fuzzy neural network

Description of drawings

Fig. 1 is a control flowchart;

Fig. 2 is the structure of fuzzy neural network;

Fig. 3 is the flow chart of upper computer;

Figure 4 is a flow chart of the lower computer.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

A flow control method of the pulp supply system based on fuzzy neural network, which uses PLC as the controller to obtain the speed of the paper machine through the encoder, PLC according to the parameters of the system design, combined with the current speed and concentration changes of the paper machine, Calculate the change parameters of the traffic that should be controlled through the fuzzy neural network. The specific process is:

When the slurry supply system is running, record the current concentration value C and the flow value Q _before , when the system concentration is changed _and set _to C, use the formula And M _paper = QC + M _{ash content After} calculating the corresponding pulp flow Q under the current concentration, the change of flow ΔQ = _after Q - _before Q; if ΔQ is greater than zero, it means that the concentration has decreased, and the flow needs to be increased. If ΔQ is less than zero, it means that the concentration Increased, the corresponding flow should be reduced; flow control is achieved by adjusting the acceleration and deceleration of the motor, using a method of stably controlling the acceleration and deceleration by adjusting the acceleration time, the design of the acceleration and deceleration per unit time, only need to know The required speed value can be reached within a certain amount of time, that is, the longer the switch is turned on, the longer the acceleration or deceleration time will be. Several sets of actual values are obtained through on-site testing, and the proportional value Y is calculated by ΔQ based on the fuzzy neural network. Multiplying ΔQ by the proportional value is the time for this speed up and down, and then the time interval of the valve action is input with the touch screen, which is the pulse. width.

The input of the fuzzy neural network is two variables X1 (concentration change value ΔC), X2 ( _before the current concentration value C), and the output variable is the proportional value Y of the flow rate, and each input factor is divided into: NM, NS, ZO , PS, PM these five fuzzy states; the fuzzy neural network is used for learning, and online learning is selected during learning, and the learning speed remains unchanged during online learning. When the control object is determined, the index value is determined based on experience, and the specific learning steps are:

① The initial values of θji, σji, ωi and η are randomly selected between [0,1], and the value of η is a constant value, which is determined based on experience;

②Calculate the ideal values of θji(k+1), σji(k+1), and ωi(k+1) according to the fuzzy neural algorithm;

③ Calculate E according to the fuzzy neural algorithm. If E≤0.002, the iteration ends; otherwise, set θji(k+1), σji(k+1), ωi(k+1) as initial values and return to ②.

The weight is calculated through the cooperation of the upper computer and the lower computer; the upper computer establishes a fuzzy neural network, the lower computer collects data, and the lower computer continuously collects data and sends it to the upper computer to continuously modify the network until the network meets the requirements.

The specific process is as follows:

Lower computer operation during the learning process:

(1) Initialization

Initialize the sample value (including two input values: the concentration change value, the current concentration value and an output value: the flow change value) and prepare for the subsequent transmission of the sample value. The learning stage continuously writes the sample value to the storage address, and at the same time Also continuously send the sample value to the host computer;

(2) Accept the request

Accept the command request from the upper computer, decide whether the lower computer sends sample data to the upper computer, if it is to execute step (3), otherwise stop;

(3) Send data

Send the sample data to the upper computer through the communication of the lower computer, end the transmission of the sample after sending, start a new acceptance, and wait for the upper computer to intercede the data transmission signal;

The specific process of the upper computer during the learning process is as follows:

(1) Initialization

First randomly select the initial values of θji, σji, ωi, and η within [0, 1], and then assign values to the constants used in the learning process, which are also assigned to the storage unit; finally, request the lower computer to transmit sample data;

(2) Initial value calculation

1) Since data transmission is requested during initialization, the data is first obtained through the communication program of the upper computer, and after receiving the sample data, the lower computer is notified not to transmit data;

2) Then use the initialized first set of weights to calculate the output value when the first set of sample values are input, the difference between the output value and the expected output value, and the data to be used for subsequent calculations;

3) Calculation of weight and performance index E value

Calculate weight and E value on the basis of above-mentioned 2) step;

4) E value judgment Compare the calculated E value with 0.002; if E ≤ 0.002, it means that the calculated function variables and weights have reached the expected goal, and the learning process is over; at the end, the switch value of the external device is triggered at the same time, and the external device is used Read these calculation results; on the contrary, you need to continue the learning process, and assign the function variables and weights calculated in step (3) that do not meet the performance indicators to the address required to recalculate the t value in the next step, and put the requested data The flag is set and sent to the lower computer to prepare for the calculation of the new t value;

5) Calculation of t value during the learning process

Since the request data flag has been set, firstly take the sample through the communication program, tell the host computer not to transmit the sample value after taking the sample value, and then calculate the new t value in order to calculate the new function variable, weight and E value; when the learning phase is over, collect data on site, establish a database and use the collected data as input, use the trained weight and fuzzy neural network algorithm to obtain the control value required by the control object.

Claims (1)

1. A flow control method for the pulp supply system based on fuzzy neural network. The speed of the paper machine is obtained through the encoder, and the current speed of the paper machine and the concentration change of the system are combined to calculate the flow rate that should be controlled through the fuzzy neural network. Change parameters, which are characterized by: _before the current concentration value C and current flow value Q are recorded _when the pulp supply system is running, _when the system concentration changes to C, without considering the ash content, the weight of paper per square meter is: In order to keep the weight of the paper constant, the formula is used: _After calculating the corresponding pulp flow Q _after the concentration C; in the case of considering the ash content, use the formula: M _paper = Q _{after C} + M _ash , _{after calculating the corresponding pulp flow Q after the} concentration C, among them, M _ash = M _paper ×Ash coefficient; flow change ΔQ=Q _post -Q _front ; among them, L: paper width; V: paper machine operating speed; when the pulp concentration changes, control the flow to keep the weight of the paper unchanged; The above method includes the following steps: first, initialize the parameters, and then determine whether the concentration has changed, and if so, determine whether △Q is greater than zero, if △Q is greater than zero, it means that the concentration has decreased, and the flow needs to be increased, then the motor is accelerated; if △Q is less than zero It shows that when the concentration increases, the corresponding flow should decrease, and the motor will decelerate; the flow control is realized by adjusting the acceleration and deceleration of the motor, and the method of stably controlling the acceleration and deceleration by adjusting the acceleration and deceleration time is adopted, and the acceleration and deceleration per unit time is designed. Decrement, you only need to know how much time you can reach the required speed value, that is, the longer the switch is turned on, the longer the acceleration or deceleration time, the acceleration and deceleration time multiplied by the acceleration and deceleration per unit time is the motor The values that need to be added and subtracted, the actual values of several groups of pulp concentration and pulp flow are obtained through field tests, the change of flow △Q is obtained through the above calculation, and the proportional value Y is calculated based on the fuzzy neural network, and △Q is multiplied by the proportional value. The time of this speed up and down, and then use the touch screen to input the time interval of the valve action, which is the pulse width; The input of the fuzzy neural network is two variables X1 and X2: X1 represents the concentration change value △C, X2 represents the _current concentration value C, the output variable is the proportional value Y, and each input factor is divided into: NM, NS, The five fuzzy states of ZO, PS, and PM; the fuzzy neural network is used for learning, and online learning is selected during learning. The learning speed remains unchanged during online learning. The termination condition of online learning is that the performance index E is less than or equal to a certain value. The specific learning steps are : ① The initial values of θji, σji, ωi and η are randomly selected between [0,1], θji, σji, ωi are the weights of the above-mentioned fuzzy neural network, and the value of η is a constant value, which is determined according to experience; ②Calculate the ideal values of θji(k+1), σji(k+1), and ωi(k+1) according to the fuzzy neural algorithm; ③ Calculate E according to the fuzzy neural algorithm, if E≤0.002, the iteration ends; otherwise, set θji(k+1), σji(k+1), ωi(k+1) as initial values and return to ②; The above weights are calculated through the cooperation of the upper computer and the lower computer: the upper computer establishes a fuzzy neural network, the lower computer collects data, and the lower computer continuously collects data and sends it to the upper computer to continuously modify the above network until the above network meets the requirements ; Lower computer operation during the learning process: (1) Initialization: Initialize the sample value, which includes the above two input values and an output value, and prepare for the subsequent transmission of the sample value. The learning stage continuously writes the sample value into the storage address, and also continuously writes the sample value The value is sent to the host computer; (2) Accept the request: accept the command request from the upper computer, decide whether the lower computer sends sample data to the upper computer, if it is to execute step (3), otherwise stop; (3) Send data: send the sample data to the upper computer through the communication of the lower computer, stop transmitting the sample after sending, start a new acceptance, and wait for the upper computer to request a data transmission signal; During the learning process, the upper computer performs initialization, initial value calculation, weight and performance index E value calculation, E value judgment and compares the calculated E value with 0.002.