A high-precision fiber grating blast furnace temperature detection method and system

Abstract

The application discloses a high-precision fiber grating blast furnace temperature detection method and system, which comprises the following steps: collecting the temperature of a blast furnace cooling wall by using a fiber grating array, and obtaining fiber grating calibration data; establishing a fiber grating temperature compensation model, and compensating the fiber grating calibration data by using a BP neural network model; optimizing the parameters of the BP neural network model by using a differential evolution algorithm until the model converges; and compensating the fiber grating calibration data by using the optimized fiber grating temperature compensation model, and obtaining blast furnace temperature detection data. The application proposes a data conversion layer in the BP neural network to enhance the data information mining capacity of the neural network, and adds a temperature compensation fitting formula in the output layer of the BP neural network to effectively fit the temperature compensation value of the fiber grating and improve the fitting precision of the BP neural network. The application also optimizes the parameters of the BP neural network by using the differential evolution algorithm, and further improves the precision of the model.

Description

Technical Field

[0001] This invention belongs to the field of fiber Bragg gratings, and specifically relates to a high-precision fiber Bragg grating blast furnace temperature detection method and system. Background Technology

[0002] The issue of blast furnace lifespan has increasingly attracted the attention of the steel industry both domestically and internationally. To maintain good blast furnace shape and achieve a long blast furnace lifespan during production, it is essential to constantly monitor and control the erosion of the blast furnace lining, especially under intensified smelting conditions. Cooling walls are crucial equipment for protecting the blast furnace lining. Strengthening temperature monitoring and providing an effective monitoring method is the most fundamental and important aspect of protecting the cooling walls.

[0003] Online monitoring of blast furnace cooling walls requires a large number of temperature sensors to reflect the hot surface condition of the cooling walls. Currently, thermocouples and other electrical sensors are commonly used to measure temperature in the hot monitoring of blast furnace cooling walls, but thermal strain cannot be measured. Using traditional temperature sensors requires drilling numerous thermocouple mounting holes in the cooling walls and blast furnace shell, which inevitably leads to damage and stress concentration in the furnace shell and cooling walls. When using traditional strain gauges to monitor the thermal stress of the cooling walls, the strain gauges often fluctuate due to the harsh on-site environment and strong electromagnetic radiation, resulting in large measurement errors.

[0004] Fiber Bragg gratings (FBGs) are sensing elements developed during the development of fiber optic communication. As a passive sensor, FBGs use wavelength encoding for sensing, offering advantages such as high accuracy and immunity to external signal interference. They have wide applications in temperature monitoring and provide a new technical means for accurate and stable sensing of temperature signals in high-temperature and harsh environments.

[0005] Using fiber Bragg grating sensors to monitor blast furnace cooling wall temperature can avoid the impact of excessive perforations on health monitoring. However, in practice, installation errors are unavoidable. Furthermore, fiber Bragg grating sensors are extremely sensitive to temperature and strain, leading to cross-sensitivity. Additionally, the optical signal attenuates as it propagates within the fiber Bragg grating. These factors directly affect the accuracy and reliability of the measurement data. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by providing a high-precision fiber Bragg grating blast furnace temperature detection method and system.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A high-precision fiber Bragg grating blast furnace temperature detection method, characterized by comprising the following steps:

[0009] Step 1: Use a fiber Bragg grating array to collect the temperature of the blast furnace cooling wall and obtain fiber Bragg grating calibration data;

[0010] Step 2: Establish a fiber Bragg grating temperature compensation model. The fiber Bragg grating temperature compensation model compensates for the fiber Bragg grating calibration data through a BP neural network model. The input layer of the BP neural network model is sequentially equipped with a discrete Fourier transform filter, a discrete inverse Fourier transform filter, and a Gaussian filter.

[0011] Step 3: Use the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges;

[0012] Step 4: Use the fiber grating temperature compensation model optimized in Step 3 to compensate the fiber grating calibration data to obtain blast furnace temperature detection data.

[0013] To optimize the above technical solution, the specific measures also include:

[0014] Furthermore, the BP neural network model comprises four layers in sequence: an input layer, a data transformation layer, a hidden layer, and an output layer.

[0015] Furthermore, the data transformation layer is designed as follows:

[0016] The acquired fiber Bragg grating calibration data is transformed at the back end of the input layer of the BP neural network model:

[0017] F = DFT(x) (1)

[0018]

[0019] R = Gfilter(iDFT(S)) (3)

[0020] In the formula, x is the fiber grating calibration data, F is the data after discrete Fourier transform, DFT(·) is the discrete Fourier transform, S is the data after sparsity representation, and f l and f h Let represent the sparsity threshold, sign(·) be the sign function, iDFT(·) be the discrete inverse Fourier transform, Gfilter(·) be the Gaussian filter, and R be the transformed fiber Bragg grating saliency data. The following compensatory fitting is performed on the fiber Bragg grating saliency data:

[0021] y = aR + b + cx (4)

[0022]

[0023]

[0024] In the formula, y is the fitted temperature value, N represents the number of fiber grating calibration data, and R nThis represents the saliency data of the nth fiber Bragg grating, where n = 1, 2, ..., N, y n denoted by , represents the actual temperature value corresponding to the saliency data of the nth fiber grating, and c represents the calibration data bias.

[0025] Furthermore, the design of the hidden layer and the output layer is as follows:

[0026] The hidden layer is shown in the following formula:

[0027] q j =f1(∑ i v ij x i +θ j (7)

[0028] In the formula, x i v represents the input of the i-th neuron node in the input layer. ij θ represents the weights from the i-th neuron in the input layer to the j-th neuron in the hidden layer. j f1 is the threshold value of the hidden layer nodes in the network, f1 is the transfer function of the hidden layer in the network, and q is the threshold value of the hidden layer nodes in the network. j It is the output of the j-th node in the hidden layer;

[0029] The output layer is shown in the following formula:

[0030] y k =f2(∑ j w jk q j +φ k )+y (8)

[0031] In the formula, w jk φ represents the weights from the j-th neuron in the hidden layer to the k-th neuron in the output layer. k f2 is the threshold value of the network output layer nodes, f2 is the transfer function of the network output layer, and y is the threshold value of the network output layer nodes. k It is the output of the kth node in the output layer.

[0032] Furthermore, the training process of the BP neural network model is as follows:

[0033] The neural network is iterated through forward and backward propagation to update the parameters in the neural network. When the loss function is less than the set threshold or the initial set number of iterations is reached, the training of the BP neural network model ends and the trained fiber optic grating temperature compensation model is obtained.

[0034] The loss function of the neural network is shown in the following formula:

[0035]

[0036] In the formula, O kIt is the actual value of the fiber grating calibration data corresponding to the output layer of the neural network, and E is the loss function of the neural network;

[0037] The weight changes in a neural network satisfy the following equation:

[0038]

[0039]

[0040] In the formula, Δv ij and Δw jk These represent the changes in weights of the hidden layer and the output layer, respectively, and η is the learning rate of the neural network.

[0041] Furthermore, the parameter optimization of the BP neural network model using the differential evolution algorithm includes the following steps:

[0042] Step 3.1: The population consists of NP individuals z with D-dimensional vectors. p The population is composed of NP, where NP is the population size, D represents the total number of weights and thresholds in the neural network, and p∈[1,2,...,NP], then the population is represented as Pop={z1,z2,...,z...} NP}, individual z p The upper and lower boundaries of the d-th dimension are respectively and The population is initialized using the following formula:

[0043]

[0044] In the formula, z p,d Let be the d-th dimension variable of the p-th individual, where d = 1, 2, ..., D, and rand. d (0, 1) is a random number that follows a uniform distribution over the interval;

[0045] Step 3.2: Perform differential mutation on the population to generate a mutation vector. The mutation strategy formula is shown below:

[0046] U p =z r1 +F*(z r2 -z r3 (13)

[0047] In the formula, r1, r2, r3 ∈ {1, 2, ..., NP} are distinct integers, and z r1 It is the r1th individual, z r2 It is the r2th individual, z r3 It is the r-3rd individual, F is the scaling factor, F is used to control the size of the difference vector, U p It is a mutation vector, U p={v p，1 v p，2 , ..., v p，D}, v p，d U p The d-th dimension variable;

[0048] Step 3.3: Place z p，d and v p，d Perform a binomial crossover operation according to the following formula to generate the final mutation vector:

[0049]

[0050] In the formula, d rand It is an integer randomly selected from the set {1, 2, ..., D} to ensure that v p，d At least one dimension of information is preserved, and the crossover probability CR is a constant between the intervals (0, 1);

[0051] Step 3.4: Compare the mutation vectors U p With the target vector z p The objective function is minimized. If the mutated vector has a better objective function, then the objective vector is replaced by the mutated vector. The selection operation uses the following formula:

[0052]

[0053] In the formula, E(U) p ) represents the mutation vector U p The objective function, E(z) p ) represents the target vector z p The objective function is E, and the loss function is E for the neural network.

[0054] This invention also proposes a high-precision fiber Bragg grating blast furnace temperature detection system, characterized in that it includes: a fiber Bragg grating array, a photodetector, a signal conditioning circuit, a data acquisition card, and a host computer;

[0055] The fiber Bragg grating array collects the temperature of the blast furnace cooling wall to obtain fiber Bragg grating calibration data; the photodetector converts the optical signal of the fiber Bragg grating array into an electrical signal, and after the electrical signal passes through the signal conditioning circuit, the digital fiber Bragg grating calibration data is acquired by the data acquisition card and sent to the host computer.

[0056] The host computer includes a modeling unit, a calculation unit, and a compensation unit:

[0057] The modeling unit is used to establish a fiber optic grating temperature compensation model. The fiber optic grating temperature compensation model compensates for the fiber optic grating calibration data through a BP neural network model. The input layer of the BP neural network model is sequentially equipped with a discrete Fourier transform filter, a discrete inverse Fourier transform filter, and a Gaussian filter.

[0058] The computing unit uses the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges.

[0059] The compensation unit uses a fiber optic grating temperature compensation model optimized by the computing unit to compensate for the fiber optic grating calibration data, thereby obtaining blast furnace temperature detection data.

[0060] Furthermore, the BP neural network model comprises four layers in sequence: an input layer, a data transformation layer, a hidden layer, and an output layer.

[0061] The beneficial effects of this invention are:

[0062] 1. This invention proposes a data transformation layer in a BP neural network through discrete Fourier transform, discrete inverse Fourier transform, and Gaussian filter to enhance the data information mining capability of the neural network;

[0063] 2. This invention incorporates a temperature compensation fitting formula for fiber Bragg grating calibration data into the output layer of the BP neural network, which effectively fits the temperature compensation value of the fiber Bragg grating and improves the fitting accuracy of the BP neural network.

[0064] 3. This invention utilizes the differential evolution algorithm to optimize the parameters of the BP neural network model, further improving the model's accuracy. The improved BP neural network enables high-precision temperature detection in fiber Bragg grating blast furnaces. Attached Figure Description

[0065] Figure 1 This is a flowchart of a high-precision fiber Bragg grating blast furnace temperature detection method proposed in this invention.

[0066] Figure 2 This is a hardware block diagram of a high-precision fiber Bragg grating blast furnace temperature detection system proposed in this invention. Detailed Implementation

[0067] The invention will now be described in further detail with reference to the accompanying drawings.

[0068] In one embodiment of the present invention, a high-precision fiber Bragg grating blast furnace temperature detection method is proposed, such as... Figure 1 As shown, the method specifically includes the following steps:

[0069] Step 1: Use a fiber Bragg grating array to collect the temperature of the blast furnace cooling wall and obtain fiber Bragg grating calibration data.

[0070] Step 2: Design a fiber optic grating temperature compensation model and establish a four-layer BP neural network model to compensate for the temperature detected by the fiber optic grating. At the same time, add discrete Fourier transform, discrete inverse Fourier transform and Gaussian filter to the back end of the input layer of the BP neural network.

[0071] The process of adding Discrete Fourier Transform, Discrete Inverse Fourier Transform, and Gaussian filter to the back end of the input layer of a BP neural network can be represented as follows:

[0072] The collected calibration data is transformed at the back end of the input layer of the BP neural network:

[0073] F = DFT(x) (1)

[0074]

[0075] R = Gfilter(iDFT(S)) (3)

[0076] Where x is the calibration data acquired by the fiber Bragg grating, F is the data after the Discrete Fourier Transform (DFT), DFT(·) is the Discrete Fourier Transform, S is the data after sparsity representation, and f l f h Here, represents the sparsity threshold, sign(·) is the sign function, iDFT(·) is the discrete inverse Fourier transform, Gfilter(·) is a Gaussian filter, and R is the transformed fiber Bragg grating saliency data. Compensatory fitting is then performed on the fiber Bragg grating saliency data.

[0077] y = aR + b + cx (4)

[0078]

[0079]

[0080] In the formula, y is the fitted temperature value, N represents the number of fiber grating calibration data, and R n This represents the saliency data of the nth fiber Bragg grating, where n = 1, 2, ..., N, y n This represents the actual temperature value corresponding to the nth saliency data point of the fiber Bragg grating, and c represents the calibration data bias. Since the actual value and the acquired value are not linearly related, further compensation of the fitted curve is required through a neural network to obtain a high-precision fiber Bragg grating temperature compensation model.

[0081] The fiber Bragg grating temperature compensation model mainly consists of a four-layer backpropagation (BP) neural network model. Its structure mainly includes: an input layer, a data conversion layer, a hidden layer, and an output layer. The input layer consists of data collected by the fiber Bragg grating. The data conversion layer consists of equations 1 to 6. The hidden layer is shown in the following equation:

[0082] q j =f1(∑ i v ij x i +θ j (7)

[0083] In the formula, x i v represents the input of the i-th neuron node in the input layer (i.e., the temperature value y output by the data conversion layer). ij θ represents the weights from the i-th neuron in the input layer to the j-th neuron in the hidden layer. j f1 is the threshold value of the hidden layer nodes in the network, f1 is the transfer function of the hidden layer in the network, and q is the threshold value of the hidden layer nodes in the network. j It is the output of the hidden layer.

[0084] The output layer is shown in the following formula:

[0085] y k =f2(∑ j w jk q j +φ k )+y (8)

[0086] In the formula, w jk φ represents the weights from the j-th neuron in the hidden layer to the k-th neuron in the output layer. k f2 is the threshold value of the network output layer nodes, f2 is the transfer function of the network output layer, and y is the threshold value of the network output layer nodes. k It is the output of the kth node in the output layer.

[0087] The loss function of a BP neural network is shown in the following equation:

[0088]

[0089] O k It is the output layer y of the neural network k The corresponding actual value of the calibration data, where E is the loss function of the neural network.

[0090] The neural network first performs forward propagation, where training samples are passed through the network to obtain output values. Then, error propagation occurs in the backpropagation layer, where the error E obtained from the forward propagation is calculated. This error is then fed back to the input layer, simultaneously updating the weights and thresholds in the network. The changes in weights must satisfy the following formula:

[0091]

[0092]

[0093] In the formula, η is the learning rate of the neural network.

[0094] The neural network is iterated through forward and backward propagation to update the parameters in the network. When the loss function E is less than the set threshold or the initially set number of iterations is reached, the model training ends and a trained fiber optic grating temperature compensation model is obtained.

[0095] Step 3: Fine-tune the fiber grating temperature compensation model: Use the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges.

[0096] Step 3.1: Initialize the population: The population consists of NP D-dimensional vector individuals z. p Composition, denoted by z p Let p be a D-dimensional variable, where D represents the total number of weights and thresholds in the neural network, and p∈[1,2,...,NP]. Then a population can be represented as Pop={z1,z2,...,z...} NP} where NP is the population size, set to 8D, and assuming individual z p The upper and lower boundaries of the d-th dimension are respectively and The population is initialized using the following formula:

[0097]

[0098] In the formula, z p,d Let be the d-th dimension variable of the p-th individual, where d = 1, 2, ..., D, and rand. d (0, 1) is a random number that follows a uniform distribution over the interval.

[0099] Step 3.2: Mutation Operation: Perform differential mutation operation on the population to generate a mutation vector. The mutation strategy formula is shown below:

[0100] U p =z r1 +F*(z r2 -z r3 (13)

[0101] In the formula, r1, r2, r3 ∈ {1, 2, ..., NP} are distinct integers, and are different from the current target vector index p. r1 It is the r1th individual, z r2 It is the r2th individual, z r3 It is the r-3rd individual, F is the scaling factor, F is used to control the size of the difference vector, U p It is a mutation vector, U p ={v p，1 v p，2 , ..., v p，D}, v p，d U pThe d-th dimension variable;

[0102] Step 3.3: Crossover operation: The target vector z... p，d and mutation vector v p,d The final test vector is generated by binomial crossover according to the following formula:

[0103]

[0104] In the formula, d rand It is an integer randomly selected from the set {1, 2, ..., D} to ensure that the mutation vector v p,d At least one dimension of information is retained, and the crossover probability CR is a constant between the intervals (0, 1).

[0105] Step 3.4: Selection operation: For the mutation vector U p With the target vector z p The objective functions are compared, and if the mutated vector has a better objective function, the objective vector is replaced with the mutated vector. The selection operation, based on minimizing the objective function, can be expressed by the following formula:

[0106]

[0107] In the formula, E(U) p ) represents the mutation vector U p The objective function, E(z) p ) represents the target vector z p The objective function is E, where E is the neural network loss function in Equation 9.

[0108] Step 4: Online application of the model: The fiber grating temperature compensation model trained in Steps 1 to 3 is applied online to achieve high-precision detection of blast furnace temperature.

[0109] In another embodiment of the present invention, a high-precision fiber Bragg grating blast furnace temperature detection system is proposed, such as... Figure 2 As shown, the system includes: a fiber optic grating array, a photodetector, a signal conditioning circuit, a data acquisition card, and a host computer.

[0110] The fiber Bragg grating array collects the temperature of the blast furnace cooling wall to obtain fiber Bragg grating calibration data; the photodetector converts the optical signal of the fiber Bragg grating array into an electrical signal, and after the electrical signal passes through the signal conditioning circuit, the digital fiber Bragg grating calibration data is acquired by the data acquisition card and sent to the host computer.

[0111] The host computer includes a modeling unit, a calculation unit, and a compensation unit:

[0112] The modeling unit is used to establish a fiber optic grating temperature compensation model. The fiber optic grating temperature compensation model compensates for the fiber optic grating calibration data through a BP neural network model. The BP neural network model consists of four layers in sequence: an input layer, a data conversion layer, a hidden layer, and an output layer. The input layer is backed by a discrete Fourier transform filter, a discrete inverse Fourier transform filter, and a Gaussian filter.

[0113] The computational unit uses the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges.

[0114] The compensation unit uses the fiber grating temperature compensation model optimized by the computing unit to compensate the fiber grating calibration data to obtain blast furnace temperature detection data.

[0115] The specific working mode of each unit is the same as that of the high-precision fiber optic grating blast furnace temperature detection method in the first embodiment.

[0116] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A high-precision fiber Bragg grating blast furnace temperature detection method, characterized in that, Includes the following steps: Step 1: Use a fiber Bragg grating array to collect the temperature of the blast furnace cooling wall and obtain fiber Bragg grating calibration data; Step 2: Establish a fiber Bragg grating temperature compensation model. The fiber Bragg grating temperature compensation model compensates for the fiber Bragg grating calibration data through a BP neural network model. The input layer of the BP neural network model is sequentially equipped with a discrete Fourier transform filter, a discrete inverse Fourier transform filter, and a Gaussian filter. The BP neural network model consists of four layers: an input layer, a data conversion layer, a hidden layer, and an output layer. The data transformation layer is designed as follows: The acquired fiber Bragg grating calibration data is transformed at the back end of the input layer of the BP neural network model: （1） （2） （3） In the formula, x It is fiber grating calibration data. F It is the data after the Discrete Fourier Transform. DFT (•) is the Discrete Fourier Transform. S It is the data after sparse representation. f l and f h Represents the sparsity threshold. sign (•) is used to find the sign function. iDFT (•) is the discrete inverse Fourier transform. Gfilter (•) is a Gaussian filter. The transformed fiber Bragg grating saliency data are used to perform the following compensatory fitting: （4） （5） （6） In the formula, The fitted temperature value, N This indicates the number of fiber Bragg grating calibration data. Indicates the first n Significance data for each fiber Bragg grating n =1,2,…, N , y n Indicates the first n The actual temperature value corresponding to each fiber Bragg grating significance data point. c Indicates calibration data bias; The design of the hidden layer and the output layer is as follows: The hidden layer is shown in the following formula: （7） In the formula, Indicates the input layer's first... i The input of each neuron node, v ij Indicates the input layer's first... i The first neuron node is connected to the hidden layer. j Each node's weight, θ j It is the threshold of hidden layer nodes in the network. f 1 is the hidden layer transfer function of the network. q j It is the hidden layer. j The output of each node; The output layer is shown in the following formula: （8） In the formula, w jk Indicates the hidden layer number 1 j The nth neuron node is connected to the output layer. k Each node's weight, It is the threshold of the network output layer nodes. f 2 is the network output layer transfer function. It is the output layer. k The output of each node; Step 3: Use the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges; Step 4: Use the fiber grating temperature compensation model optimized in Step 3 to compensate the fiber grating calibration data to obtain blast furnace temperature detection data.

2. The high-precision fiber Bragg grating blast furnace temperature detection method as described in claim 1, characterized in that: The training process of the BP neural network model is as follows: The neural network is iterated through forward and backward propagation to update the parameters in the neural network. When the loss function is less than the set threshold or the initial set number of iterations is reached, the training of the BP neural network model ends and the trained fiber optic grating temperature compensation model is obtained. The loss function of the neural network is shown in the following formula: （9） In the formula, O k These are the actual values ​​of the fiber Bragg grating calibration data corresponding to the output layer of the neural network. It is the loss function of the neural network; The weight changes in a neural network satisfy the following equation: （10） （11） In the formula, and These represent the changes in weights of the hidden layer and the output layer, respectively. It is the learning rate of the neural network.

3. The high-precision fiber Bragg grating blast furnace temperature detection method as described in claim 1, characterized in that: The method of optimizing the parameters of a BP neural network model using the differential evolution algorithm includes the following steps: Step 3.1: Population composition NP indivual D dimensional vector individuals Composition, in which NP It refers to population size. D This represents the total number of weights and thresholds in the neural network. The population is represented as ,individual The d The upper and lower boundaries of the dimension are respectively and The population is initialized using the following formula: （12） In the formula, It is the first p The first individual d dimensional variables, , It is a random number that follows a uniform distribution over an interval; Step 3.2: Perform differential mutation on the population to generate a mutation vector. The mutation strategy formula is shown below: （13） In the formula, They are all distinct integers. It is the first r One individual, It is the first r Two individuals, It is the first r 3 individuals, F It is a scaling factor. F Used to control the size of the difference vector. It is a mutation vector. , express The d Dimensional variables; Step 3.3: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require and Perform a binomial crossover operation according to the following formula to generate the final mutation vector: （14） In the formula, d rand From a set A randomly selected integer is used to guarantee... At least one dimension of information is retained, and the crossover probability CR is a constant between the intervals (0,1); Step 3.4: Compare mutation vectors With the target vector The objective function is minimized. If the mutated vector has a better objective function, then the objective vector is replaced by the mutated vector. The selection operation uses the following formula: （15） In the formula, Represents the mutation vector The objective function, Represents the target vector The objective function, This is the loss function for the neural network.

4. A high-precision fiber Bragg grating blast furnace temperature detection system, characterized in that, include: Fiber Bragg grating array, photodetector, signal conditioning circuit, data acquisition card and host computer; The fiber Bragg grating array collects the temperature of the blast furnace cooling wall to obtain fiber Bragg grating calibration data; the photodetector converts the optical signal of the fiber Bragg grating array into an electrical signal, and after the electrical signal passes through the signal conditioning circuit, the digital fiber Bragg grating calibration data is acquired by the data acquisition card and sent to the host computer. The host computer includes a modeling unit, a calculation unit, and a compensation unit: The modeling unit is used to establish a fiber optic grating temperature compensation model. The fiber optic grating temperature compensation model compensates for the fiber optic grating calibration data through a BP neural network model. The input layer of the BP neural network model is sequentially equipped with a discrete Fourier transform filter, a discrete inverse Fourier transform filter, and a Gaussian filter. The BP neural network model consists of four layers in sequence: an input layer, a data conversion layer, a hidden layer, and an output layer. The data transformation layer is designed as follows: The acquired fiber Bragg grating calibration data is transformed at the back end of the input layer of the BP neural network model: （1） （2） （3） In the formula, x It is fiber grating calibration data. F It is the data after the Discrete Fourier Transform. DFT (•) is the Discrete Fourier Transform. S It is the data after sparse representation. f l and f h Represents the sparsity threshold. sign (•) is used to find the sign function. iDFT (•) is the discrete inverse Fourier transform. Gfilter (•) is a Gaussian filter. The transformed fiber Bragg grating saliency data are used to perform the following compensatory fitting: （4） （5） （6） In the formula, The fitted temperature value, N This indicates the number of fiber Bragg grating calibration data. Indicates the first n Significance data for each fiber Bragg grating n =1,2,…, N , y n Indicates the first n The actual temperature value corresponding to each fiber Bragg grating significance data point. c Indicates calibration data bias; The design of the hidden layer and the output layer is as follows: The hidden layer is shown in the following formula: （7） In the formula, Indicates the input layer's first... i The input of each neuron node, v ij Indicates the input layer's first... i The first neuron node is connected to the hidden layer. j Each node's weight, θ j It is the threshold of hidden layer nodes in the network. f 1 is the hidden layer transfer function of the network. q j It is the hidden layer. j The output of each node; The output layer is shown in the following formula: （8） In the formula, w jk Indicates the hidden layer number 1 j The nth neuron node is connected to the output layer. k Each node's weight, It is the threshold of the network output layer nodes. f 2 is the network output layer transfer function. It is the output layer. k The output of each node; The computing unit uses the differential evolution algorithm to optimize the parameters of the BP neural network model until the model converges. The compensation unit uses a fiber optic grating temperature compensation model optimized by the computing unit to compensate for the fiber optic grating calibration data, thereby obtaining blast furnace temperature detection data.

Images