CN102393884B - Hot continuous rolling electromagnetic induction heating temperature prediction method based on BP (back-propagation) neural network - Google Patents

Abstract

The invention discloses a hot continuous rolling electromagnetic induction heating temperature prediction method based on BP (back-propagation) neural network. The existing prediction method is dependent on manual work, and thus, has the defects of low efficiency and poor reliability. The method disclosed by the invention comprises the following steps: selecting prediction model variables: reasonably selecting input/output variables of a prediction model by utilizing mechanism analysis and prior information; normalizing data to be input; establishing a BP neural network, and training and testing the neural network; and finally, denormalizing the data obtained by the neural network, thereby obtaining the predicted heating temperature. In the invention, the steel billet temperature is predicted according to the operation history data of the electromagnetic induction heater, and has higher prediction precision than the traditional engineering computation method.

Description

Prediction Method of Electromagnetic Induction Heating Temperature in Hot Continuous Rolling Based on BP Neural Network

technical field

The invention belongs to the technical field of automation, and in particular relates to a neural network-based method for predicting the temperature of a billet heated by electromagnetic induction for hot continuous rolling.

Background technique

In the iron and steel industry, the traditional steelmaking, continuous casting, and steel rolling processes are independent production links. However, the modern production mode is gradually transforming into an integrated hot rolling process of "steelmaking-continuous casting-steel rolling". This is the highly intensive automatic hot rolling production line model widely praised at home and abroad. The heating equipment (heating, soaking, holding furnace) used between continuous casting and rolling is the key equipment connecting the continuous casting and rolling production line. From a technical point of view, the intermediate heating equipment needs to supplement the heat during the process of conveying the continuous casting slab to make the temperature of the slab uniform and meet the temperature requirements of continuous rolling, so as to solve the problem of uneven temperature field of the slab. The heating furnace is used as a buffer zone between two parts with different process speeds of the continuous casting machine and the continuous rolling mill. It connects and buffers the logistics between the casting machine and the rolling mill, and coordinates the production of the two.

Due to the many advantages of electromagnetic induction heating technology, such as fast heating speed, controllable power density, no pollution, easy control, and minimal oxidation loss, etc., compared with traditional gas heating furnaces, no open flame heat source is required, and it has zero emissions. Advantages, more suitable for use in hot rolling production lines. Therefore, high-power medium-frequency induction heaters have been gradually applied to hot rolling production lines, see Figure 1.

Since electromagnetic induction heating is a complex nonlinear process with a large lag, it is difficult to establish an accurate mechanism model, and it is difficult to obtain satisfactory results with conventional control methods (such as PID adjustment). Generally, manual experience is used for debugging and control.

In view of the strong nonlinear mapping ability and good fault tolerance of the neural network, it can well approximate the real state data of the system. Therefore, on the basis of a large amount of accumulated induction heating operation data, the neural network method is used to establish a hot rolling electromagnetic induction heating temperature prediction model to predict the billet temperature, which can provide a basis for precise temperature control of electromagnetic induction heating.

Contents of the invention

Aiming at the deficiencies of the prior art, the invention provides a neural network-based method for predicting the temperature of a steel slab heated by electromagnetic induction for hot rolling. The specific steps of this method are:

Step (1) Select predictive model variables.

Using neural network technology to establish a temperature prediction model for hot rolling electromagnetic induction heating slabs, in order to ensure the effectiveness of data-based neural network modeling and avoid the blindness of pure black box modeling, first use mechanism analysis and prior information to select a reasonable prediction model input and output variables.

Based on the mechanism analysis, the billet temperature after electromagnetic induction heating is selected as the output variable of the neural network model, and the main factors affecting the billet temperature are selected as the input variables of the neural network model: ①The temperature of the billet before electromagnetic induction heating; ②The voltage of the induction heater; ③ Induction heater current. The output variable of the neural network model is the billet temperature after electromagnetic induction heating.

For the induction heating process of the steel billet in the induction coil, it starts from the time when the steel billet enters the coil magnetic field until the steel billet leaves the magnetic field. First of all, in the case of ignoring heat conduction and heat radiation, the temperature rise of a certain section of the billet is mainly affected by the voltage and current of the induction heater in the process. Therefore, the voltage and current of the induction heater are sequence variables. Secondly, considering heat conduction and heat radiation, the billet temperature before and after electromagnetic induction heating is also taken as a sequence variable.

Step (2) Data normalization processing.

The input data in the training sample contains three items, and the order of magnitude differs greatly. In order to ensure the equal status of each factor and speed up the convergence speed, the data is normalized and converted into a value in the range of [0, 1] .

in is the maximum value in the input data, is the minimum value in the input data. for the input data, Normalizes the processed value for the input data.

Step (3) Build the BP neural network framework.

Call the newff function in the MatlabR2009a neural network toolbox to establish a BP neural network, Net= newff (PR, , , BTF, BLF, PF ); Net is the BP neural network framework, PR is a value range determined by the largest and smallest elements in the input matrix, is the number of neurons in the i-th layer, is the transfer function of the i-th layer, , is the total number of layers of the neural network, BTF is the training function of the BP neural network, BLF is the weight and bias value, and PF is the network performance function.

Step (4) training BP neural network. The specific method is:

a. Initialize the BP neural network, use the value generated by the random function to assign the weight and bias value, and then call the init function to initialize the BP neural network.

b. Set the number of network training times and the training target error.

c. Set the training data as the input matrix P, set the target value as the matrix T, and call the train function in the MatlabR2009a neural network toolbox to perform data training on the BP neural network Net until convergence, Net = train (Net, P, T).

Step (5) Test the BP neural network.

Test the trained BP neural network, use historical data to form the network test matrix P_test for electromagnetic induction heating temperature prediction, directly call the sim function in the MatlabR2009a neural network toolbox, D=sim(Net, P_test), and test the matrix Carry out simulation, where D is the objective function.

Step (6) Data denormalization processing.

According to the formula Perform denormalization processing, where is the billet temperature after inverse normalization processing, is the billet temperature obtained from the simulation test, is the maximum billet temperature, is the minimum temperature of the billet.

The present invention will use the computer's simulation computing ability to build an error backpropagation (BP) neural network, automatically acquire knowledge from the operation history data of the induction heater, and gradually combine the new knowledge into its mapping function, thereby realizing nonlinearity The approximation of the function can accurately predict the temperature of the billet heated by the inductor.

The beneficial effect of the inventive method:

1. The BP neural network has the ability to approximate the nonlinear mapping function, so the use of electromagnetic induction heater operation history data to predict the billet temperature is higher than the prediction accuracy obtained by traditional engineering calculation methods.

2. The temperature prediction of hot rolling electromagnetic induction billet based on BP neural network avoids solving the traditional finite element model of electromagnetic field and temperature field coupling based on induction heating, heat conduction and heat radiation theory, and also avoids solving the finite element model by computer For the strict requirements of basic data such as boundary conditions, only a large amount of electromagnetic induction heating historical operation data is needed, which is more practical.

3. Increase the length of BP neural network training data and the dimension of the input matrix, improve the calculation speed of the computer, and improve the prediction accuracy.

Description of drawings

Figure 1 is a schematic diagram of induction heating for hot rolling.

Detailed ways

Taking the billet induction heating of the hot rolling production line of a steel factory as an example, the specific implementation of the modeling of the billet temperature prediction model for hot rolling electromagnetic induction heating based on BP neural network is carried out.

Step (1) Prediction model variable selection

Mechanism analysis. The output of the model is the heated slab temperature. According to the law of energy conservation, the energy after billet heating = the energy before billet heating + the energy absorbed by the billet, and the energy absorbed by the billet comes from the induction heater. From the PLC control system of the electromagnetic induction heater, it can be known that the voltage and current of the induction heater are measured. If the time for the billet to pass through the induction heater is known, then theoretically the product of the voltage, current and time of the induction heater is the amount absorbed by the billet. energy. Based on the above mechanism analysis, it can be determined that the input of the neural network model is three items: the temperature of the billet before heating, the voltage and current of the inductor.

Before this modeling, the sampling data are preprocessed first, and the data of each sample (a steel billet induction heating process) is integrated. Among them, the temperature of the billet before heating and the temperature after heating are integrated into 100 values, according to the induction The length of the device and the billet, the voltage and current are integrated into 152 values. According to the position correlation analysis, the billet temperature at the nth point corresponds to the current and voltage values of the n+25th to n+46th inductors, and the temperature of the billet at the nth point before heating and its corresponding voltage and current values are taken as Input the matrix, and the temperature of the billet at the nth point after heating is used as the output matrix. In this way, each sample can form 100 pairs of input and output matrices.

Let the A matrix be the temperature of the billet before heating, the B matrix be the voltage of the induction heater, the C matrix be the current of the induction heater, and D be the temperature of the billet after heating. Then enter the matrix: P= , the specific arrangement is:

P = ,in is the temperature of the billet before heating at the nth moment,

is the voltage of the induction heater at the nth moment,

is the current of the induction heater at the nth moment, so that each column of the matrix P constitutes the input matrix.

Corresponding output matrix: D=

Step (2) data normalization processing

Using the normalization formula Normalize the input and output matrices separately.

For the input matrix P, the specific method is to normalize the three matrices A, B, and C respectively. Take the A matrix as an example:

A=

The A matrix is normalized according to the normalization formula to get . Then perform similar processing on the B and C matrices in turn, and the normalized input matrix is: = .

For the output matrix D= , is the temperature of the billet after heating, is the minimum value of the temperature after the billet is heated, is the maximum value of the temperature of the billet after heating, is the temperature of the billet after heating after the normalization treatment. The normalized input matrix is .

Step (3) Construct BP neural network

Build the BP neural network framework, call the newff function in the MatlabR2009a function library,

Net=newff(threshold,[5,1],'tansig','purelin',trainlm)

Where Threshold is a 45*2 matrix defining the minimum and maximum values of 45 input and output vectors; [5,1] means that there are 5 neurons in the first layer and 1 neuron in the second layer; tansig is the input layer Transfer function; purelin is the transfer function of the output layer; trainlm is the training function based on the l-m algorithm.

Step (4) training BP neural network

a. Initialize the network

net.initFcn is used to determine the initialization function of the entire network. The parameter net.layer{i}.initFcn is used to determine the initialization function of each layer. The initwb function initializes the weight matrix and bias according to each layer's own initialization parameters (net.inputWeights{i,j}.initFcn). The initialization weight is usually set to rans. The specific method is as follows:

net.layers{1}.initFcn = 'initwb';

net.inputWeights{1,1}.initFcn = 'rands';

net.layerWeights{2,1}.initFcn = 'rands';

net.biases{1,1}.initFcn = 'rands';

net.biases{2,1}.initFcn = 'rands';

net = init(net);

net.IW{1,1} is the weight matrix from the input layer to the hidden layer

net.LW{2,1} is the weight matrix between the hidden layer and the output layer; net.b{1,1} is the threshold vector of the hidden layer, net.b{2,1} is the valve of the output node value;

b. Set the number of network training times, training target error and the number of steps for display

net.trainParam.epochs=2000;

net.trainParam.goal=0.002;

net.trainParam.show=50;

Set the number of network training times to 2000 steps, the training target error to 0.0002, and display the number of training steps as 50 steps.

c. Using the input matrix and the target matrix is set to , by calling the train function, net=train(net, , ) to train the temperature prediction network after billet heating until convergence.

Step (5) Network test

The historical data used for testing is composed of the matrix p_test for billet temperature network testing according to the input matrix format in step (1), and then normalized according to step (2), the normalized test matrix is . Call the sim() function in the Matlab toolbox to simulate the trained network. The calling program code is: D=sim(net, ); the D matrix is the predicted value of the forebay water level of the sewage pumping station.

Step (6) denormalization processing

According to the formula Perform denormalization processing, where is the final slab temperature after denormalization, is the billet temperature obtained from the simulation test, is the maximum billet temperature, is the minimum value of billet temperature. After denormalization, the billet temperature is , that is, the billet temperature obtained by the test is .

Claims (1)

1. the hot continuous rolling electromagnetic induction heating temperature prediction method based on BP neural network, it is characterized in that the method may further comprise the steps: Step 1. Select the predictive model variables, specifically using mechanism analysis and prior information to select the input and output variables of the predictive model; The specific process of mechanism analysis is: the output of the model is the temperature of the billet after heating; according to the law of energy conservation, the energy after the billet is heated = the energy before the billet is heated + the energy absorbed by the billet, where the energy absorbed by the billet comes from the induction heater; The PLC control system of the electromagnetic induction heater can know the voltage and current of each measured induction heater. If the time for the billet to pass through the induction heater is known, then theoretically the product of the voltage, current and time of the induction heater is the energy absorbed by the billet. ; Based on the above-mentioned mechanism analysis, it is determined that the input of the neural network model is three items: the billet temperature before electromagnetic induction heating, the voltage of the induction heater and the current of the induction heater; Before this modeling, preprocess each sampled data and integrate each sample data. Among them, the temperature before heating and the temperature after heating of the billet are integrated into 100 values. According to the length of the sensor and the billet, the The voltage and current are integrated into 152 values; according to the position correlation analysis, the billet temperature at the nth point corresponds to the current and voltage values of the sensors at the n+25th to n+46th, and the temperature of the billet at the nth point before heating is taken now And its corresponding voltage and current values, as the input matrix, and the temperature of the heated billet at the nth point as the output matrix, so that each sample constitutes 100 pairs of input and output matrices; Let the A matrix be the billet temperature before heating, the B matrix be the voltage of the induction heater, the C matrix be the current of the induction heater, and D be the temperature of the billet after heating; then input the matrix: P=[A B C] ^T , the specific arrangement The way is: $P=\left[\begin{array}{cccc}{a}_1& a_2& ...& a_{100}\\ b_{26}& b_{27}& ...& b_{125}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ b_{47}& b_{48}& ...& b_{146}\\ c_{26}& c_{27}& ...& c_{125}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ c_{47}& c_{48}& ...& c_{146}\end{array}\right],$ Where a _n is the billet temperature before heating at the nth moment, b _n is the voltage of the induction heater at the nth moment, c _n is the current of the induction heater at the nth moment, so that each column of the matrix P constitutes the input matrix; Corresponding output matrix: D=[d ₁ d ₂ ... d ₁₀₀ ]; Step 2. Data normalization processing, specifically: The input data in the training sample contains three items, and the order of magnitude differs greatly. In order to ensure the equal status of each factor and speed up the convergence speed, the data is normalized and converted into a value in the range of [0,1] $\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; min</span>}}}$ Where x _max is the maximum value in the input data, x _min is the minimum value in the input data, x is the input data, Normalize the processed value for the input data; Step 3. Build the BP neural network framework, specifically: Call the newff function in the MatlabR2009a neural network toolbox to establish a BP neural network, Net=newff(PR,[s ₁ ,s ₂ ,…,s _i ],{TF ₁ ,TF ₂ ,…,TF _i },BTF,BLF ,PF); Net is the BP neural network framework, PR is a value range determined by the largest element and the smallest element in the input matrix, s _i is the number of neurons in the i-th layer, and TF _i is the transfer function of the i-th layer , 1≤i≤N ₁ , N ₁ is the total number of layers of the neural network, BTF is the training function of the BP neural network, BLF is the weight and bias value, and PF is the network performance function; Step 4. Train the BP neural network, the specific method is: a. Initialize the BP neural network, use the value generated by the random function to assign weights and bias values, and then call the init function to initialize the BP neural network; b. Set the number of network training times and the training target error; c, set training data as input matrix P, set target value as matrix T, call the train function in the MatlabR2009a neural network toolbox to carry out data training to BP neural network Net until convergence, Net=train(Net,P,T); Step 5. Test the BP neural network, specifically: Test the trained BP neural network, use historical data to form the network test matrix P_test for electromagnetic induction heating temperature prediction, directly call the sim function in the MatlabR2009a neural network toolbox, D=sim(Net, P_test), and test the matrix Perform simulation, where D is the objective function; Step 6. Data denormalization processing, specifically: According to the formula Perform denormalization processing, where x' is the billet temperature after denormalization processing, is the steel slab temperature obtained from the simulation test, x′ _max is the highest temperature of the slab, and x′ _min is the minimum temperature of the slab.