CN102799938B - Optimizing method of 9% martensite steel pipeline postweld heat treatment heating width - Google Patents

Abstract

The invention relates to an optimizing method of 9% martensite steel pipeline postweld heat treatment heating width. According to the optimizing method, pipeline postweld heat treatment inner-outer wall temperature difference size data of T groups of pipelines with different sizes under different heating widths, different heat treatment environment temperature and different temperature control conditions are obtained through calculation. The minimum heating widths required by pipeline postweld heat treatment of the pipelines under different heat treatment environment temperatures, different temperature control temperatures and different preset inner-outer wall temperature differences are comprehensively considered, so that an error back propagation-based neural network is established for training and testing the minimum heating widths, pipeline size, heat treatment environment temperature, temperature control temperature and preset inner-outer wall temperature difference are used as inputs, and heating widths are used as outputs. By combining with actually measured data of the pipeline postweld heat treatment, a trained and tested network output threshold is corrected to obtain the optimizing method. According to the optimizing method, the minimum heating widths required by the postweld heat treatment can be rapidly calculated, a heat treatment process can be helped to be guided and optimized, and the heat treatment quality is improved.

Description

Optimization method of heating width for post-weld heat treatment of 9%Cr martensitic steel pipeline

technical field

The invention relates to a method for optimizing the heating width of a 9%Cr martensitic steel pipeline for post-weld heat treatment.

Background technique

9%Cr new martensitic heat-resistant steel mainly includes three new martensitic heat-resistant steels: T/P92, T/P91 and E911, which are widely used in ultra-supercritical boiler main steam pipes, headers and other thick-walled pipes and other components , low weld toughness is a major problem in the installation process of this series of steel pipeline welds. In order to improve the toughness of the weld, local heat treatment must be carried out on the weld. Research at home and abroad shows that the post-weld heat treatment temperature has a great influence on the weld. Treatment, the impact energy of the weld can reach more than 41J. When heating at about 740°C, the constant temperature time must be extended to achieve this index. When the heating temperature is below 730°C, extending the constant temperature time not only has little effect, but the impact energy It is difficult to achieve the toughness index of 41J, and the installation cost will be greatly increased, which seriously affects the construction progress.

At present, on the basis of traditional heat-resistant steel, domestic and foreign have put forward the technical regulations for post-weld heat treatment of pressure-bearing pipelines. The post-weld heat treatment of Cr new martensitic heat-resistant steel is not necessarily applicable, that is, the applicability of existing standards to 9% Cr new martensitic heat-resistant steel needs to be verified.

In addition, there are great disputes on the selection of heating width in domestic and foreign post-weld heat treatment specifications, and the heating width values obtained according to different specifications are very different. This brings difficulties to the on-site heat treatment, and the applicability of the regulations is questionable.

Artificial neural network is a nonlinear science that began to develop rapidly in the late 1980s. The artificial neural network model has strong fault tolerance, learning, adaptive and nonlinear mapping capabilities, and is especially suitable for solving non-linear problems with complex causal relationships. Deterministic reasoning, judgment, recognition and classification problems. At present, the most widely used model in the field of iron and steel metallurgy is the model (BP model) with a multi-layer feed-forward network structure and a reverse error propagation training method.

Contents of the invention

The present invention mainly solves the technical problems existing in the prior art; it provides a method that can not only optimize the heating width of the post-weld heat treatment of 9% Cr new martensitic heat-resistant steel thick-walled pipelines, but also ensure the quality of heat treatment and improve the efficiency of heat treatment is of great significance.

Another purpose of the present invention is to solve the existing problems in the prior art; to provide a method that solves the difference in the selection of heating width for post-weld heat treatment of 9% Cr new martensitic heat-resistant steel pipelines in domestic and foreign heat treatment technical regulations.

Above-mentioned technical problem of the present invention is mainly solved by following technical scheme:

The method for optimizing the heating width of the 9%Cr martensitic steel pipeline post-weld heat treatment is characterized in that it comprises the following steps:

Step 1, the temperature field calculation module, establishes the calculation model of the temperature field of the hot spot of the upper T group of pipes of different sizes under different heating widths, different heat treatment ambient temperatures, and different temperature control temperatures, and uses finite element analysis software to calculate the post-welding of each group of models The temperature difference between the inner and outer walls of heat treatment (the insulation width is determined according to the electric power standard);

Step 2, the neural network building module, comprehensively consider the minimum heating width required by the pipes of any specification (pipe diameter and wall thickness) under different heat treatment ambient temperatures, different temperature control temperatures, and different preset temperature differences between inner and outer walls. Establish a neural network based on error backpropagation;

Step 3: Prediction model building module, based on the data of T group heating width obtained in step 1, train and test the error-based backpropagation neural network in step 2, and obtain a thick-walled pipe that can predict 9% Cr martensitic heat-resistant steel Prediction model of postweld heat treatment heating width;

Step 4, the model correction module, combined with the experimental measurement data of post-weld heat treatment of 9% Cr martensitic heat-resistant steel thick-walled pipes, corrects the obtained prediction model;

Step 5, heating width optimization module, analyze pipe size (pipe diameter and wall thickness), heat treatment ambient temperature, temperature control temperature, preset temperature difference between inner and outer walls, and input it into the corrected model to obtain the minimum heating width of pipe post-weld heat treatment .

In the method for optimizing the heating width of the post-weld heat treatment of the above-mentioned 9%Cr martensitic steel pipeline, in the step 1, the different size pipelines of the upper T group are established under different heating widths, different heat treatment ambient temperatures, and different temperature control temperatures. The heat treatment temperature field calculation model uses finite element software to calculate the temperature difference between the inner and outer walls of the pipeline post-weld heat treatment under different conditions. The specific method is as follows:

According to the application of 9%Cr new martensitic heat-resistant steel, select the size range of the pipe; according to the domestic and foreign heat treatment technical regulations, calculate the width of the heating zone and the width of the heat preservation for a pipe of a certain specification, select the heating width range, and the heat preservation width according to Electric power standard is selected; according to the temperature control temperature of 9%Cr new martensitic heat-resistant steel and the temperature of heat treatment environment, the range of temperature control temperature and heat treatment environment temperature is selected. Establish a theoretical calculation model for the post-weld heat treatment temperature field of T-group 9%Cr new martensitic heat-resistant steel pipes, and calculate the pipe size (pipe diameter and wall thickness), heating width, temperature control temperature, and heat treatment environment temperature by using finite element software. The influence of the position of the equivalent point is calculated as follows:

Step 1.1, in the finite element software, establish the calculation model of the temperature field of the post-weld heat treatment of 9%Cr new martensitic heat-resistant steel;

Step 1.2, define initial conditions, boundary conditions, and solve;

Step 1.3, after the calculation is completed, check the temperature of the inner wall and the outer wall of the pipe in the post-processor, and calculate the temperature difference between the inner and outer walls.

In the above-mentioned method for optimizing the heating width of the post-weld heat treatment of 9%Cr martensitic steel pipelines, in the step 2, the specific method of establishing a neural network based on error backpropagation is:

Step 2.1, define the input layer and output layer

The pipe size (pipe diameter and wall thickness), the preset temperature difference between the inner and outer walls, the temperature control temperature and the heat treatment environment temperature are selected as input variables, so the number of neurons in the input layer of the network is 5; the minimum required under different conditions The heating width is used as the output of the network model, so the number of neurons in the output layer is 1.

Step 2.2, select the number of hidden layers and hidden layer units: use a single hidden layer, and determine the number of hidden layer nodes as 10.

Step 2.3, determination of other parameters: hidden layer The transfer function of the hidden layer is a unipolar S-type function: f(x)=1/(1+e ^-x ), and the transfer function of the output layer is a linear function: f(x )=x, make the network output any value, the number of training is 1800 times, the error target is 0.5, and the number of selected samples is T, of which N are training samples and TN are testing samples.

In the above-mentioned method for optimizing the heating width of the post-weld heat treatment of 9%Cr martensitic steel pipelines, in the step 2, the neural network based on error backpropagation includes an input layer, an intermediate layer and an output layer, and the input layer has 5 neurons, the middle layer has 10 neurons, and the output layer has 1 neuron; the transfer function of the middle layer of the prediction model is a unipolar S-type function, and the transfer function of the output layer is a linear function, so that the network output Any value; step 1 obtains the T heating width, and the specific steps of training and testing based on the error backpropagation neural network in step 2 are as follows:

Step 3.1, set weights, thresholds and training times, and initialize weights and thresholds, randomly pick T-N group samples from T group samples as training samples, and N group samples as test samples, and input T-N group training samples, The sample is the size of the heating width of the T group obtained in step 1 and the influencing factors of the minimum heating width of the post-weld heat treatment of the T group 9%Cr martensitic heat-resistant steel pipeline;

Step 3.2, calculate the network output, obtain the weights and thresholds of each layer in the backpropagation neural network, and calculate the weights and threshold correction factors of each layer in the backpropagation neural network, according to the TN group A obtained in step 1 ₁ temperature calculation value and network output calculation network output error, the network output error is the calculation value of the TN group heating width obtained in step 1 and the comparison difference of the network output calculated in this step;

Step 3.3, judge whether the maximum number of training times is reached, and choose to perform the following steps according to whether the maximum number of training times is reached:

Choose to execute step 1. If the maximum number of training times has not been reached, judge whether the network output error is less than the expected error in step 3.2. If it is less than the expected error, the training ends, and save the weights of each layer in the backpropagation neural network in step 3.2. and the threshold to obtain the undetermined prediction model; if it is greater than the expected error, correct the weights of each layer in the backpropagation neural network and the threshold and then repeat step 3.2. Wherein the correction factor adopts the correction factor calculated in step 3.2;

Choose to execute step 2. If the maximum number of training times is reached, the backpropagation neural network cannot converge within the given number of training times, and the training ends;

Step 3.4: Input N groups of test samples one by one to select and execute the undetermined prediction model in step 1. If the prediction error is lower than the specified level, it indicates that the undetermined prediction model can be used to predict the post-weld heat treatment of 9% Cr martensitic heat-resistant steel pipes The required minimum heating width, that is, the undetermined prediction model is the prediction model obtained in step 3; otherwise, the undetermined prediction model does not conform, and the whole step ends.

In the above-mentioned method for optimizing the heating width of post-weld heat treatment of 9%Cr martensitic steel pipes, in the step 4, the data and model of the post-weld heat treatment experiments of 9%Cr new-type martensitic heat-resistant steel thick-walled pipes are measured The calculated values are compared and analyzed, and the output threshold of the model is corrected.

Therefore, the present invention has the following advantages: 1. The method for optimizing the post-weld heat treatment heating width of 9% Cr new type martensitic heat-resistant steel thick-walled pipeline has very important significance for ensuring heat treatment quality and improving heat treatment efficiency; 2. Solving The differences in the selection of heating width for post-weld heat treatment of 9%Cr new martensitic heat-resistant steel pipes in domestic and foreign heat treatment technical regulations are discussed.

Description of drawings

Fig. 1 is the BP neural network model diagram used in the present invention.

Fig. 2 is a flow chart of BP neural network training in the present invention.

Fig. 3 is a BP neural network training error diagram in the present invention.

Detailed ways

The technical solutions of the present invention will be further specifically described below through the embodiments and in conjunction with the accompanying drawings.

The method for optimizing the heating width of the 9%Cr martensitic steel pipe post-weld heat treatment of the present invention comprises the following steps:

Step 1, the temperature field calculation module, establishes the calculation model of the temperature field of the hot spot of the upper T group of pipes of different sizes under different heating widths, different heat treatment ambient temperatures, and different temperature control temperatures, and uses finite element analysis software to calculate the post-welding of each group of models The temperature difference between the inner and outer walls of the heat treatment (the insulation width is determined according to the electric power standard), the specific method is:

According to the application of 9%Cr new martensitic heat-resistant steel, select the size range of the pipe; according to the domestic and foreign heat treatment technical regulations, calculate the width of the heating zone and the width of the heat preservation for a pipe of a certain specification, select the heating width range, and the heat preservation width according to Electric power standard is selected; according to the temperature control temperature of 9%Cr new martensitic heat-resistant steel and the temperature of heat treatment environment, the range of temperature control temperature and heat treatment environment temperature is selected. Establish a theoretical calculation model for the post-weld heat treatment temperature field of T-group 9%Cr new martensitic heat-resistant steel pipes, and calculate the pipe size (pipe diameter and wall thickness), heating width, temperature control temperature, and heat treatment environment temperature by using finite element software. The influence of the position of the equivalent point is calculated as follows:

Step 1.1, in the finite element software, establish the calculation model of the temperature field of the post-weld heat treatment of 9%Cr new martensitic heat-resistant steel;

Step 1.2, define initial conditions, boundary conditions, and solve;

Step 1.3, after the calculation is completed, check the temperature of the inner wall and the outer wall of the pipe in the post-processor, and calculate the temperature difference between the inner and outer walls.

Step 2, the neural network building module, comprehensively consider the minimum heating width required by the pipes of any specification (pipe diameter and wall thickness) under different heat treatment ambient temperatures, different temperature control temperatures, and different preset temperature differences between inner and outer walls. Establish a neural network based on error backpropagation, the specific method is:

1) Design of input layer and output layer

The pipe size (pipe diameter and wall thickness), preset temperature difference between inner and outer walls, temperature control temperature, and heat treatment ambient temperature are selected as input variables, so the number of neurons in the input layer of the network is 5; The required minimum heating width is used as the output of the network model, so the number of neurons in the output layer is 1.

2) Selection of the number of hidden layers and the number of hidden layer units

In 1989, Robert Hecht-Nielson proved that a continuous function in any closed interval can be approximated by a hidden layer BP network. Because a 3-layer BP network can complete any continuous mapping from n-dimensional to m-dimensional, this model uses a single hidden layer, and the selection of the number of nodes in the hidden layer is a relatively complicated problem. Try, and finally determine that the number of hidden layer nodes is 10.

3) Determination of other parameters

Hidden layer The transfer function of the hidden layer is a unipolar S-type function: f(x)=1/(1+e ^-x ), and the transfer function of the output layer is a linear function: f(x)=x, so that the network can output any value, the number of training times is 1800, the error target is 1, and the number of selected samples is T, of which N are training samples and TN are testing samples.

In this step, the neural network based on error backpropagation includes an input layer, an intermediate layer and an output layer, the input layer has 5 neurons, the intermediate layer has 10 neurons, and the output layer has 1 neuron; The transfer function of the middle layer of the prediction model is a unipolar S-type function, and the transfer function of the output layer is a linear function, so that the network can output any value. The structure diagram is shown in Figure 1.

Step 3: Prediction model building module, based on the data of T group heating width obtained in step 1, train and test the error-based backpropagation neural network in step 2, and obtain a thick-walled pipe that can predict 9% Cr martensitic heat-resistant steel The prediction model of the heating width of post-weld heat treatment; for the T group of heating width data obtained in step 1, the specific steps of training and testing based on the error backpropagation neural network in step 2 are as follows:

Step 3.1, set weights, thresholds and training times, and initialize weights and thresholds, randomly pick T-N group samples from T group samples as training samples, and N group samples as test samples, and input T-N group training samples, The sample is the size of the heating width of the T group obtained in step 1 and the influencing factors of the minimum heating width of the post-weld heat treatment of the T group 9%Cr martensitic heat-resistant steel pipeline;

Step 3.2, calculate the network output, obtain the weights and thresholds of each layer in the backpropagation neural network, and calculate the weights and threshold correction factors of each layer in the backpropagation neural network, according to the TN group A obtained in step 1 ₁ temperature calculation value and network output calculation network output error, the network output error is the calculation value of the TN group heating width obtained in step 1 and the comparison difference of the network output calculated in this step;

Step 3.3, judge whether the maximum number of training times is reached, and choose to perform the following steps according to whether the maximum number of training times is reached:

Choose to execute step 1. If the maximum number of training times has not been reached, judge whether the network output error is less than the expected error in step 3.2. If it is less than the expected error, the training ends, and save the weights of each layer in the backpropagation neural network in step 3.2. and the threshold to obtain the undetermined prediction model; if it is greater than the expected error, correct the weights of each layer in the backpropagation neural network and the threshold and then repeat step 3.2. Wherein the correction factor adopts the correction factor calculated in step 3.2;

Choose to execute step 2. If the maximum number of training times is reached, the backpropagation neural network cannot converge within the given number of training times, and the training ends;

Step 3.4: Input N groups of test samples one by one to select and execute the undetermined prediction model in step 1. If the prediction error is lower than the specified level, it indicates that the undetermined prediction model can be used to predict the post-weld heat treatment of 9% Cr martensitic heat-resistant steel pipes The required minimum heating width, that is, the undetermined prediction model is the prediction model obtained in step 3; otherwise, the undetermined prediction model does not conform, and the whole step ends.

In this embodiment, training and testing refer to 3600 groups of 9%Cr new martensitic heat-resistant steel pipeline post-weld heat treatment heating width data obtained under 3650 groups calculated by the finite element software above under different conditions as training samples. The model was trained, and the remaining 50 sets of post-weld heat treatment heating width data of 9% Cr new martensitic heat-resistant steel pipes under different conditions were used as test samples to test the trained BP network. The network model network is trained using the error back propagation algorithm. The training process is shown in Figure 2. After repeated training, when the output error of the neural network reaches 0.5mm, the training can be stopped. The training error diagram is shown in Figure 3. When the prediction error of the neural network for 50 groups of test samples is lower than the specified level, it shows that the network model can be used to predict the heating width of the post-weld heat treatment of the 9%Cr new martensitic heat-resistant steel thick-walled pipeline.

Step 4, the model correction module, combined with the experimental measurement data of post-weld heat treatment of 9% Cr martensitic heat-resistant steel thick-walled pipes, corrects the obtained prediction model, and corrects the threshold value of the model output layer;

Step 5, heating width optimization module, analyze pipe size (pipe diameter and wall thickness), heat treatment ambient temperature, temperature control temperature, preset temperature difference between inner and outer walls, and input it into the corrected model to obtain the minimum heating width of pipe post-weld heat treatment .

In the present invention, the pipe size (pipe diameter and wall thickness), preset temperature difference between inner and outer walls, heat treatment ambient temperature and temperature control temperature are selected as input parameters, and the scope of application is as follows:

Pipe inner diameter (radius): 100mm-500mm;

Pipe wall thickness: 30mm-140mm;

Preset temperature difference between inner and outer walls: 0°C-50°C;

Heat treatment ambient temperature: -10°C-30°C;

Temperature control temperature: 750°C-780°C.

Example:

The BP neural network optimization method involved in the present invention is compared with the measured temperature difference between the inner and outer walls of the pipeline:

Analyze and record the three specifications of 9%Cr martensitic heat-resistant steel pipes shown in Table 1. Pipe size (pipe diameter and wall thickness), heat treatment ambient temperature, temperature control temperature and preset temperature difference between inner and outer walls, and each influencing factor Input the value of the value into the model for calculation, and the minimum heating width of the post-weld heat treatment of the 9% Cr martensitic heat-resistant steel pipe under this condition can be quickly calculated. In addition, the accuracy of the model is verified by experiments. In this example, the results obtained by the present invention and the measured results are shown in Table 2 below.

Table 1 Parameters of post-weld heat treatment for 9%Cr martensitic heat-resistant steel pipes

Pipe specification/mm Heating width/mm Insulation width/mm ambient temperature/ Temperature control temperature/ Preset temperature difference between inner and outer walls/ ID296*65 510 750 13 755 31 ID430*90 937 1137 10 756 33 ID288*110 866 1066 15 765 32

Table 2 adopts the method of the present invention to compare with measured data

Pipe specification/mm The method of the present invention/ Measured value/ error/ ID296*65 515 510 5 ID430*90 930 937 -7 ID288*110 870 866 3

The calculation results show that the calculated data obtained by using the method for optimizing the heating width of the post-weld heat treatment of the 9% Cr new-type martensitic heat-resistant steel thick-walled pipeline proposed by the present invention is relatively consistent with the experimental data, and the absolute value of the heating width error is less than 10mm. Compared with the experimental method, it obviously has many advantages. In addition to the convenient and rapid calculation of the heating width of the post-weld heat treatment of 9% Cr new martensitic heat-resistant steel thick-walled pipes, and the optimization of the heat treatment process, it can also solve the problem of differences in heat treatment technical regulations at home and abroad. .

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention belongs can make various modifications or supplements to the described specific embodiments or adopt similar methods to replace them, but they will not deviate from the spirit of the present invention or go beyond the definition of the appended claims range.

Claims (4)

1. The optimization method of 1.9%Cr martensite steel pipeline post weld heat treatment width of heating, is characterized in that, comprise the following steps:

   Step 1, calculate the post weld heat treatment calculation model for temperature field of T group different size pipeline under different heating width, different heat treatment environment temperature, different control temperature by Temperature calculating module, adopt finite element analysis software to calculate the post weld heat treatment inside and outside wall temperature difference of each group model;

   Step 2, by neural network module in conjunction with any specification pipeline different heat treatment environment temperature, different control temperature and different preset inside and outside wall temperature difference condition under, width of heating minimum needed for pipeline; Set up based on error backward propagation method;

   Step 3, forecast model sets up module, the data obtaining T group width of heating for step 1 carry out training and testing in step 2 based on error backward propagation method, obtain the forecast model that can be predicted 9%Cr martensite heat-resistant steel posted sides pipeline post weld heat treatment width of heating;

   Step 4, Modifying model module, revises in conjunction with the forecast model of 9%Cr martensite heat-resistant steel posted sides pipeline post weld heat treatment measured data of experiment to gained;

   Step 5, width of heating optimizes module, and analysis conduit size, heat treatment environment temperature, control temperature, the default inside and outside wall temperature difference, be input to the minimum width of heating that revised model can obtain pipeline post weld heat treatment;

   In described step 1, calculate the heat place calculation model for temperature field of T group different size pipeline under different heating width, different heat treatment environment temperature, different control temperature, pipeline post weld heat treatment inside and outside wall temperature extent under utilization finite element software calculating different condition, concrete grammar is:

   According to the applicable cases of 9%Cr martensite steel pipeline, choose line size scope; According to domestic and international heat treatment technics code, pipeline is calculated to the size of heating tape width, insulation width, choose width of heating scope, insulation width is chosen according to electric power standard; According to control temperature and the heat treatment environment temperature conditions of 9%Cr martensite steel pipeline, select the scope of control temperature and heat treatment environment temperature, set up T group 9%Cr martensite steel pipeline post weld heat treatment temperature field theoretical calculation model, by using finite element software to calculate line size, width of heating, control temperature and heat treatment environment temperature to the impact of equivalent point position, computing method are as follows:

   Step 1.1, in finite element software, sets up 9%Cr martensite steel pipeline post weld heat treatment calculation model for temperature field;

   Step 1.2, definition starting condition, boundary condition, solve;

   Step 1.3, after having calculated, checks inner-walls of duct temperature and outside wall temperature in preprocessor, calculates inside and outside wall temperature extent.
2. 2. the optimization method of 9%Cr martensite steel pipeline post weld heat treatment width of heating according to claim 1, is characterized in that, in described step 2, the concrete grammar set up based on error backward propagation method is:

   Step 2.1, definition input layer and output layer:

   Choose caliber, wall thickness, preset the numerical value of the inside and outside wall temperature difference, control temperature and heat treatment environment temperature as input variable, therefore the neuron number of this network input layer is 5; Minimum width of heating required under different condition is as the output of network model, and therefore output layer neuron number is 1;

   Step 2.2, selects hidden layer number and Hidden unit number: adopt single hidden layer, and determine that the number of hidden nodes is 10;

   Step 2.3, the determination of other parameters: the transport function of hidden layer is unipolarity S type function: f (x)=1/ (1+e ^-x), the transport function of output layer is linear function: f (x)=x, makes network export any value, and frequency of training is 1800 times, and error target is 0.5, and selection sample number is T, wherein N number of test sample book, T-N training sample.
3. 3. the optimization method of 9%Cr martensite steel pipeline post weld heat treatment width of heating according to claim 1, it is characterized in that, in described step 2, an input layer, a middle layer and an output layer is comprised based on error backward propagation method, input layer has 5 neurons, there are 10 neurons in middle layer, and output layer has 1 neuron; The transport function in the middle layer of described forecast model is unipolarity S type function, and the transport function of output layer is linear function, makes network export any value; T width of heating is obtained to step 1 as follows to carrying out the concrete steps of training and testing based on error backward propagation method in step 2:

   Step 3.1, setting weights and threshold and frequency of training, and initialization is carried out to weights and threshold, win T-N group sample in T group sample at random as training sample, N group sample is as test sample book, input T-N group training sample, described sample is the size of the T group width of heating obtained in step 1 and the influence factor of the minimum width of heating of T group 9%Cr martensite heat-resistant steel pipeline post weld heat treatment;

   Step 3.2, computational grid exports, and obtains weights and the threshold value of each layer in reverse transmittance nerve network, and calculates the weights of each layer and the modifying factor of threshold value in reverse transmittance nerve network, according to the T-N group A obtained in step 1 ₁temperature calculations and network export computational grid output error, and described network output error is the comparison difference of the calculated value of the T-N group width of heating obtained in step 1 and the network output of this step calculating;

   Step 3.3, judges whether to reach maximum frequency of training, and according to whether reaching maximum frequency of training selection execution following steps:

   Select to perform step 1, if not yet reach maximum frequency of training, judge whether network output error is less than anticipation error in step 3.2, if be less than anticipation error, then train end, preserve weights and the threshold value of each layer in reverse transmittance nerve network in step 3.2 simultaneously, obtain forecast model undetermined; If be greater than anticipation error, after revising the weights of each layer in reverse transmittance nerve network and threshold value step repeat 3.2. wherein modifying factor adopt the modifying factor calculated in step 3.2;

   Select to perform step 2, if reach maximum frequency of training, then this reverse transmittance nerve network can not be restrained in given frequency of training, and training terminates;

   Step 3.4, by the forecast model undetermined in N group test sample book one by one input selection execution step 1, if predicated error is lower than showing during prescribed level that this forecast model undetermined can be used in predicting the minimum width of heating needed for the post weld heat treatment of 9%Cr martensite heat-resistant steel pipeline, namely namely this forecast model undetermined is the forecast model obtained in step 3; Otherwise this forecast model undetermined does not meet, and terminates whole step.
4. 4. the optimization method of 9%Cr martensite steel pipeline post weld heat treatment width of heating according to claim 1, it is characterized in that, in described step 4, the data of 9%Cr martensite steel pipeline post weld heat treatment experiment measuring and model calculation value are analyzed, and correction model exports threshold value.