Neural Network-Based Reactor Component Transfer and Positioning Method

Abstract

The present invention discloses a method for transporting and positioning reactor components based on a neural network, including: constructing an original positioning model with a neural network structure based on multi-source sensor data. Establishing a three-dimensional rectangular coordinate system, and planning an optimal transportation path in combination with the component safety requirements, starting position, and target position of the reactor fuel assembly. Training the original positioning model with historical transportation data, and then optimizing the parameters of the original positioning model based on the optimal transportation path to obtain a transportation and positioning model. By inputting the multi-source sensor data obtained by the transportation equipment in real time into the transportation and positioning model, the transportation and positioning model outputs the positioning data for obtaining the reactor fuel assembly in real time. The present invention can assist operators in accurately positioning the reactor fuel assembly in real time in the blind operation area, ensuring the positioning accuracy of the reactor fuel assembly during operation.

Description

Technical Field

[0001] The present invention relates to the technical field of reactor fuel assembly transfer and positioning, and particularly to a method for transferring and positioning reactor assemblies based on a neural network. Background Art

[0002] During the transfer and transportation of reactor fuel assemblies in nuclear power plants, it is a basic operation carried out within the nuclear power plant and a necessary condition for the operation of the nuclear power plant. To improve the transfer efficiency of the assemblies, nuclear power plants generally adopt a method of simultaneously transferring new and spent assemblies in the out-of-reactor process transportation system. However, since the reactor fuel assemblies are all blindly operated during the transfer and use after entering the preheating tank, it is necessary to have a suitable monitoring system to ensure that the reactor fuel assemblies are in the correct position, otherwise it may cause damage to the reactor fuel assemblies.

[0003] Therefore, the prior art highly depends on the operating experience of the operators, and the operators need to continuously pay attention to the monitoring system. When continuously operating the transfer of reactor fuel assemblies for a long time, the operators are prone to distraction or misoperation, which may cause safety accidents in severe cases. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the transfer of reactor fuel assemblies in the prior art, and provide a method for transferring and positioning reactor assemblies based on a neural network. The method constructs an original positioning model through a neural network, and trains the original positioning model with historical transfer data to obtain a transfer positioning model. The running trajectory of the transfer equipment is displayed in real time through the transfer positioning model, realizing high-precision positioning of the reactor fuel assemblies, and automatically correcting the position according to the deviation situation when the reactor fuel assemblies deviate, ensuring that the reactor fuel assemblies are aligned with the reactor core.

[0005] The technical solution of the present invention provides a method for transferring and positioning reactor assemblies based on a neural network, which is characterized by including:

[0006] Constructing an original positioning model of a neural network structure based on multi-source sensor data;

[0007] Establishing a three-dimensional rectangular coordinate system, and planning an optimal transfer path in combination with the component safety requirements, starting position, and target position of the reactor fuel assemblies;

[0008] Training the original positioning model with historical transfer data, and then optimizing the parameters of the original positioning model based on the optimal transfer path to obtain a transfer positioning model;

[0009] By inputting the multi-source sensor data obtained in real time by the transfer device into the transfer positioning model, the transfer positioning model outputs in real time the positioning data of the reactor fuel assembly, and the positioning data includes position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information.

[0010] In one of the alternative technical solutions, it further includes: displaying in real time the running trajectory of the transfer device, and controlling the transfer device to make corresponding positioning corrections if the positioning data meets the risk conditions.

[0011] In one of the alternative technical solutions, the multi-source sensor data includes laser ranging data, height position sensor data, and ultrasonic measurement data;

[0012] The original positioning model is constructed based on a neural network, with multi-source sensor data as the input of the original positioning model, and the positioning data of the reactor fuel assembly is output, including:

[0013] Integrate the laser ranging data, height position sensor data, and ultrasonic measurement data into a multi-source sensor data vector as the input data of the original positioning model;

[0014] Perform normalization processing on the input data to unify the value ranges of different types of sensor data, so as to improve the stability and efficiency of model training;

[0015] Construct an original positioning model with a deep neural network structure including an input layer, multiple hidden layers, and an output layer;

[0016] Receive the normalized multi-source sensor data vector through the input layer;

[0017] Through the hidden layer, perform non-linear transformation and feature extraction on the input data. The neurons of each hidden layer perform weighted summation and activation operations on the input according to the weights and biases, and gradually extract higher-level features;

[0018] The output layer outputs through linear transformation the positioning data of the reactor fuel assembly including position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information.

[0019] In one of the alternative technical solutions, the establishment of the three-dimensional rectangular coordinate system includes:

[0020] Taking the most distal vertex angle position of the fuel assembly preheating box as the origin, the transportation channel direction as the X-axis, the horizontal vertical direction as the Y-axis, and the height direction as the Z-axis to determine the spatial position of the fuel assembly.

[0021] In one alternative technical solution, planning an optimal transfer path in combination with the component safety requirements, starting position, target position, starting azimuth angle, and target azimuth angle of the reactor fuel assembly includes:

[0022] In combination with the component safety requirements, interlock and prohibit operations in the X or Y direction where the height in the transfer path does not meet the requirements; comprehensively consider the path length and the risk assessment value of passing through the hazardous area, and construct a path cost function;

[0023] The path length is obtained by calculating the distance between the starting position and the target position in a three-dimensional rectangular coordinate system;

[0024] The risk assessment value is quantified according to the degree of danger;

[0025] Adopt the A* search algorithm. On the basis of considering the component safety requirements and the path cost, start from the starting position, continuously expand adjacent nodes, calculate the path cost of each node, and gradually search to the target position, thereby determining the optimal transfer path.

[0026] In one alternative technical solution, training the original positioning model with historical transfer data includes:

[0027] Collect historical multi-source sensor data during the historical transfer process and the historical positioning data of the corresponding reactor fuel assembly to form a historical transfer data set;

[0028] Divide the historical transfer data set into a training set and a validation set according to a preset ratio for training and performance verification of the original positioning model;

[0029] Adopt the mean square error loss function to calculate the average value of the sum of the squares of the errors between the positioning data predicted by the model and the actual positioning data, and measure the deviation between the predicted value and the true value of the original positioning model through the mean square error loss function to optimize the parameters of the original positioning model;

[0030] Use the stochastic gradient descent optimization algorithm to perform iterative training on the original positioning model on the training set;

[0031] In each iteration, calculate the gradient of the model's weights and biases according to the mean square error loss function, and adjust the weights and biases according to the learning rate to reduce the value of the loss function.

[0032] In one alternative technical solution, optimizing the parameters of the original positioning model based on the optimal transfer path to obtain a transfer positioning model includes:

[0033] Based on the original mean squared error loss function, a deviation term between the predicted path and the optimal transfer path is added. The deviation term is obtained by calculating the differences between the positions and azimuth angles of the fuel assemblies predicted by the model and the corresponding points on the optimal transfer path in a three-dimensional rectangular coordinate system and an azimuth coordinate system.

[0034] A balance coefficient is set to balance the influences of the original loss function and the path deviation term. According to the requirements of the actual transfer task, the magnitude of the balance coefficient is adjusted to determine the emphasis on the accuracy of positioning data and the path compliance during the model optimization process.

[0035] Using optimization algorithms such as stochastic gradient descent, based on the new loss function containing the path deviation term, the parameters of the original positioning model are iteratively updated again to make the path output by the model closer to the optimal transfer path, and a transfer positioning model is obtained.

[0036] In one of the optional technical solutions, inputting the multi-source sensor data obtained by the transfer device in real time into the transfer positioning model, so that the transfer positioning model outputs the positioning data for obtaining the reactor fuel assembly in real time, includes:

[0037] During the operation of the transfer device, laser ranging data, height position sensor data, and ultrasonic measurement data are collected in real time and integrated into a multi-source sensor data vector.

[0038] The multi-source sensor data vector collected in real time is normalized to make its format and value range consistent with the input data during the training of the transfer positioning model, and then input into the transfer positioning model.

[0039] Based on the input real-time data, the transfer positioning model outputs the current position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information of the reactor fuel assembly through forward propagation calculation.

[0040] In one of the optional technical solutions, if the positioning data meets the risk conditions, controlling the transfer device to make corresponding positioning corrections, includes:

[0041] Construct a risk condition judgment function, comprehensively consider the position deviation, azimuth angle deviation, and speed deviation. When the value of the risk condition judgment function exceeds the preset risk threshold, it is determined that the positioning data meets the risk conditions.

[0042] Adopt a PID control algorithm to calculate the correction amounts of the position, azimuth angle, and speed respectively according to the position deviation, azimuth angle deviation, and speed deviation.

[0043] Send the correction amounts to the transfer device to control the transfer device to adjust its own operating state according to the correction amounts, including adjusting the position, azimuth angle, and speed, so as to achieve positioning correction.

[0044] The technical solution of the present invention further provides an electronic device, including a memory, a processor, and an electronic device program on the memory, and the processor executes the electronic device program to implement the steps of any one of the foregoing neural network-based reactor component transfer and positioning methods.

[0045] Adopting the above technical solution, the following beneficial effects are achieved:

[0046] The neural network-based reactor component transfer and positioning method provided by the present invention constructs an original positioning model through a neural network, and trains the original positioning model with historical transfer data to obtain a transfer and positioning model. The operation trajectory of the transfer device is displayed in real time through the transfer and positioning model, realizing high-precision positioning of the reactor fuel assembly. When the reactor fuel assembly deviates, the position is automatically corrected according to the deviation situation to ensure that the reactor fuel assembly is aligned with the operation position, thereby reducing the demand for the operation experience of the operator during the transfer process. After the reactor fuel assembly enters the preheating tank, during the blind operation of the operator, it is ensured that the reactor fuel assembly is in the correct position and the azimuth angle is correct, avoiding the burning of the reactor fuel assembly or jamming when it enters the reactor core. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Referring to the drawings, the disclosure of the present invention will become more readily understood. It should be understood that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In the figures:

[0048] Figure 1 is a working flow chart of the neural network-based reactor component transfer and positioning method provided by an embodiment of the present invention;

[0049] Figure 2 is a schematic structural diagram of the electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The following further describes the specific embodiments of the present invention with reference to the drawings.

[0051] It should be noted that the reactor fuel assembly transfer control method provided by the embodiments of the present invention is applied to a reactor fuel assembly transfer system. By using transfer equipment, various data of the reactor fuel assembly are obtained. For example, when the transport device is a transfer cart, multiple sensors can be set on the transfer cart to measure laser ranging data, height position sensor data, and ultrasonic measurement data. During the process of transferring the reactor fuel assembly from the fuel storage container in the fuel depot to the preheating box, the code, position change, and various data of the reactor fuel assembly are recorded. The transport container on the transfer cart is docked with the preheating box, and the positioning component of the preheating box assembly is used to position and insert it into the specified position of the transport container. After ensuring that the azimuth angle is correct, the component code and position are recorded. Then the reactor fuel assembly is transported to the reactor for reaction. During the transfer process, the reactor fuel assembly transfer control method provided by the present invention can ensure the safety of the reactor fuel assembly on the premise of transferring the reactor fuel assembly as soon as possible.

[0052] As Figure 1 shown in a method for positioning the transfer of a reactor assembly based on a neural network provided by an embodiment of the present invention, which is characterized by including:

[0053] Step S101: Construct an original positioning model of the neural network structure based on multi-source sensor data.

[0054] Step S102: Establish a three-dimensional rectangular coordinate system and an azimuth coordinate system, and plan an optimal transfer path in combination with the component safety requirements, starting position, and target position of the reactor fuel assembly.

[0055] Step S103: Train the original positioning model with historical transfer data, and then optimize the parameters of the original positioning model based on the optimal transfer path to obtain a transfer positioning model.

[0056] Step S104: Input the multi-source sensor data obtained in real time by the transfer equipment into the transfer positioning model, so that the transfer positioning model outputs in real time the positioning data of the reactor fuel assembly. The positioning data includes position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information.

[0057] Among them, step S101 realizes the deep fusion processing of multi-source heterogeneous sensor data by constructing an original positioning model based on a neural network. After the multi-source sensor data undergoes model feature extraction and non-linear mapping, the three-dimensional spatial coordinates and attitude information of the fuel assembly with high precision are output. Adaptive compensation through deep learning algorithms lays a reliable data foundation for subsequent path planning. Step S102 constructs a standardized kinematic model for the transfer of the fuel assembly by establishing a spatial representation system of a three-dimensional rectangular coordinate system, which can generate an optimal transfer path that takes into account both safety and efficiency. Step S103 uses historical transfer data to perform transfer learning training on the original positioning model. The historical transfer data can include sensor readings and verified pose data under typical working conditions, and then focuses on optimizing the feature weight distribution of the deep neural network. Step S104 inputs data such as laser ranging data, height position sensor data, and ultrasonic measurement data in the multi-source sensor data obtained by the transfer device in real time into the transfer positioning model, enabling the model to output the positioning data of the reactor fuel assembly in real time, realizing the real-time application of the positioning model, ensuring that the system can dynamically perceive the position and attitude of the fuel assembly, and the system can monitor the transfer process in real time, providing data support for subsequent risk judgment and correction operations.

[0058] In summary, the method for positioning the transfer of reactor components based on a neural network provided by the embodiment of the present invention constructs an original positioning model through a neural network, and trains the original positioning model through historical transfer data to obtain a transfer positioning model. The operation trajectory of the transfer device is displayed in real time through the transfer positioning model, realizing the high-precision positioning of the reactor fuel assembly, and automatically correcting the position according to the deviation when the reactor fuel assembly deviates, ensuring that the reactor fuel assembly is aligned with the reactor core, thereby reducing the requirement for the operating experience of the operator during the transfer process of the reactor fuel assembly. After the reactor fuel assembly enters the preheating box, during the blind operation of the operator, it is ensured that the reactor fuel assembly is in the correct position and the azimuth angle is correct, avoiding the burning of the reactor fuel assembly or jamming when it enters the reactor core.

[0059] In one embodiment, the multi-source sensor data includes laser ranging data, height position sensor data, and ultrasonic measurement data.

[0060] Further, step S101 includes the following sub-steps:

[0061] Integrate the laser ranging data, height position sensor data, and ultrasonic measurement data into a multi-source sensor data vector as the input data of the original positioning model;

[0062] Perform normalization processing on the input data to unify the value ranges of different types of sensor data, so as to improve the stability and efficiency of model training;

[0063] Construct an original positioning model of a deep neural network structure including an input layer, multiple hidden layers, and an output layer;

[0064] Receive the normalized multi-source sensor data vector through the input layer;

[0065] Through the hidden layer, perform non-linear transformation and feature extraction on the input data. The neurons in each hidden layer perform weighted summation and activation operations on the input according to the weights and biases, gradually extracting higher-level features;

[0066] The output layer outputs the positioning data of the reactor fuel assembly through linear transformation, including position information, velocity information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information.

[0067] In this embodiment, an original positioning model is constructed using a deep neural network (DNN). The network includes an input layer, several hidden layers, and an output layer. Let the number of hidden layers be L, the number of neurons in the input layer be n, corresponding to the dimension of the multi-source sensor data, and the number of neurons in the output layer be 3, corresponding to the position information p, velocity information v, and height position information θ. T that appears below represents the transpose of a matrix or vector.

[0068] The multi-source sensor data sources as input are:

[0069] x = [x laser , x angle , x ultra T

[0070] where x laser is the laser ranging data, x angle is the height position sensor data, and x ultra is the ultrasonic measurement data.

[0071] The input of the l-th hidden layer (l = 1, 2,..., L) is a (l-1) , and when it is the input layer, a (0) = x, the weight matrix is W (l) , and the bias vector is b (l) , then the weighted input z (l) is:

[0072] z (l) = W (l) a (l-1) + b (l)

[0073] After passing through the activation function σ, such as the ReLU function σ(z) = max(0, z), the activation output a (l) is obtained;

[0074] The output of the output layer y = [p, v, θ]​T is:

[0075] y = W (l+1) a (L) + b (L+1)

[0076] Combined, the original positioning model is:

[0077] y = f(x; W, b)

[0078] where W = {W (1) , …, W (L+1)}, b = {b (1) , …, b (L+1)}, both W and b are model parameters,

[0079] The current azimuth angle situation of the reactor fuel assembly can be obtained through the rotation angle information relative to the original position. Specifically, the starting azimuth angle, target azimuth angle, and azimuth angle deviation can be obtained. Therefore, when planning the optimal conversion path, the starting azimuth angle and target azimuth angle of the reactor fuel assembly also need to be obtained.

[0080] In one embodiment, step S102 includes the following sub-steps:

[0081] Taking a specific fixed point in the transfer area as the origin, determine the positive directions of the three-dimensional rectangular coordinate system to represent the position of the reactor fuel assembly;

[0082] Taking the true north or a specific reference direction as the benchmark, establish an azimuth coordinate system to determine the azimuth angle of the fuel assembly.

[0083] Combined with the component safety requirements, interlock and prohibit operations in the X or Y direction where the height in the transfer path does not meet the requirements; comprehensively consider the path length, azimuth angle deviation, and the risk assessment value of passing through the dangerous area, and construct a path cost function;

[0084] The path length is obtained by calculating the distance between the starting position and the target position in the three-dimensional rectangular coordinate system;

[0085] The azimuth angle deviation is the difference between the starting azimuth angle and the target azimuth angle; the risk assessment value is quantified according to the area and danger level of the path passing through the dangerous area;

[0086] Adopt the A* search algorithm. On the basis of considering the component safety requirements and the path cost, start from the starting position, continuously expand adjacent nodes, calculate the path cost of each node, and gradually search for the target position to determine the optimal transfer path.

[0087] Among them, the A* algorithm can consider the time complexity and space complexity of the path. The time complexity can reach exponential complexity in the worst case theoretically, but in practical applications, due to the role of the heuristic function, the optimal solution can usually be found in a short time. The space complexity is similar to the time complexity. In the worst case, it may be necessary to store an exponential number of nodes, but a good heuristic function can greatly reduce the number of nodes to be stored.

[0088] In this embodiment, a three-dimensional rectangular coordinate system x and an azimuth coordinate system θ are established. Let the starting position of the reactor fuel assembly be (x δ , θ δ ), and the target position be (x t , θ t ). The component safety requirements can be represented by a safety region matrix S, where the element S ij represents whether the coordinate (i, j) is a safe region, 1 for safe and 0 for dangerous.

[0089] Define the path cost function C as:

[0090] C = ω1·|x t - x path | + ω2·|θ t - θ path | + ω3·R safety

[0091] Among them, x path and θ path are the position and direction angle of a certain point on the path, R safety is the risk valuation of the path passing through the dangerous area, which can be calculated according to the safety region matrix S. ω1, ω2, and ω3 are weight coefficients, and ω1 + ω2 + ω3 = 1.

[0092] The calculation method of the matrix S is:

[0093] S ij = α·S static (i,j) + (1 - α)·D(i,j)

[0094] Among them, α is a static weight coefficient, S static (i,j) is a reference safety static matrix (set according to industry safety specifications), and D(i,j) is a reference safety dynamic matrix (usually a diagonal matrix). It is necessary to first obtain the corresponding coordinate point (x path according to x n , y n ), and then calculate R safety according to the safety region matrix S. The specific calculation method is as follows:

[0095]

[0096] Wherein, N is the total number of sampling points on the path.

[0097] Through the above method, the path cost can be calculated, and then the optimal transfer path can be determined.

[0098] In one embodiment, step S103 includes the following sub-steps:

[0099] Collect historical multi-source sensor data and corresponding historical positioning data of reactor fuel assemblies during historical transfer processes to form a historical transfer data set;

[0100] Divide the historical transfer data set into a training set and a validation set according to a preset ratio for training and performance verification of the original positioning model;

[0101] Adopt the mean square error loss function to calculate the average value of the sum of squared errors between the positioning data predicted by the model and the actual positioning data. Measure the deviation degree between the predicted value and the true value of the original positioning model through the mean square error loss function to optimize the parameters of the original positioning model;

[0102] Use the stochastic gradient descent optimization algorithm to iteratively train the original positioning model on the training set;

[0103] In each iteration, calculate the gradients of the model's weights and biases according to the mean square error loss function, and adjust the weights and biases according to the learning rate to reduce the value of the loss function.

[0104] Furthermore, step S103 also includes the following sub-steps:

[0105] On the basis of the original mean square error loss function, add a deviation term between the predicted path and the optimal transfer path. The deviation term is obtained by calculating the differences between the predicted positions and azimuth angles of the fuel assemblies in the three-dimensional rectangular coordinate system and the corresponding points on the optimal transfer path;

[0106] Set a balance coefficient to balance the influences of the original loss function and the path deviation term, and adjust the magnitude of the balance coefficient according to the requirements of the actual transfer task to determine the emphasis on the accuracy of positioning data and the path compliance during the model optimization process;

[0107] Use optimization algorithms such as stochastic gradient descent, and based on the new loss function including the path deviation term, iteratively update the parameters of the original positioning model again to make the path output by the model closer to the optimal transfer path, and obtain the transfer positioning model.

[0108] In this embodiment, historical transfer data is used to train the original positioning model, where N is the number of samples. Adopt the mean square error loss function J1:

[0109]

[0110] Among them, y i = [p i , v i , θ i T is the true positioning data, is the positioning data predicted by the model. Then, optimization algorithms such as Stochastic Gradient Descent (SGD) are used to update the model parameters.

[0111] Add a path constraint term J2 to the loss function to further optimize the model based on the optimal transfer path:

[0112] J = J1 + λJ2

[0113] Among them, λ is the balance coefficient, and J2 is:

[0114]

[0115] Among them, x pathi and θ pathi are respectively the position and the direction angle of the i-th point on the optimal transfer path, and are respectively the position and the direction angle of the corresponding point on the model-predicted path.

[0116] Use the optimization algorithm to update the model parameters again, and the transfer positioning model can be obtained according to the original positioning model.

[0117] In one of the embodiments, step S104 includes the following sub-steps:

[0118] During the operation of the transfer device, laser ranging data, height position sensor data, and ultrasonic measurement data are collected in real time and integrated into a multi-source sensor data vector;

[0119] Normalize the multi-source sensor data vector collected in real time to make its input data format and value range consistent with those during the training of the transfer positioning model, and then input it into the transfer positioning model;

[0120] According to the input real-time data, the transfer positioning model outputs the current position information, speed information, and height position information of the reactor fuel assembly through forward propagation calculation.

[0121] In this embodiment, the multi-source sensor data x real obtained by the transfer device in real time is input into the trained transfer positioning model, and the model outputs the positioning data y real = [p real , v real , θ real ​​T 。

[0122] To avoid uncertainty in the predicted positioning data, in this embodiment, the risk of the real-time predicted positioning data is further judged according to the running trajectory of the transfer device, and when the positioning data meets the risk conditions, the transfer device is controlled to make corresponding positioning corrections.

[0123] Preferably, it further includes step S105: real-time display of the running trajectory of the transfer device, and when the positioning data meets the risk conditions, control the transfer device to make corresponding positioning corrections.

[0124] Step S105 enables the operator to visually observe the position and attitude of the reactor fuel assembly by real-time displaying the running trajectory of the transfer device, and when the positioning data meets the risk conditions, control the transfer device to make corresponding positioning corrections, realizing visual monitoring and risk warning of the transfer process, ensuring that the system can timely detect and correct potential problems, enabling the system to dynamically adjust operations during the transfer process, no longer overly relying on manual monitoring and operation by the operator, thereby ensuring the safety and reliability of the continuous running transfer process.

[0125] In one of the embodiments, step S105 includes the following sub-steps:

[0126] Construct a risk condition judgment function, comprehensively considering position deviation, azimuth deviation, and speed deviation. When the value of the risk condition judgment function exceeds the preset risk threshold, it is determined that the positioning data meets the risk conditions;

[0127] Adopt the PID control algorithm to calculate the correction amounts of position, azimuth, and speed respectively according to the position deviation, azimuth deviation, and speed deviation;

[0128] Send the correction amounts to the transfer device, and control the transfer device to adjust its own running state according to the correction amounts, including adjusting position, azimuth, and speed, so as to achieve positioning correction.

[0129] In this embodiment, the risk condition judgment function is defined as R(y real ), and when R(y real ) meets the preset risk threshold, it is considered that the positioning data meets the risk conditions.

[0130] R(y real ) = β1·|x t - p real | + β2·|θ t - θ real | + β3·Δy

[0131] Wherein, β1, β2, and β3 are weight coefficients, and Δy is the correction amount.

[0132] Set the risk threshold to τ. When R(y real ) > τ, control the transfer device to make corresponding positioning corrections. The correction amount Δy = [Δp, Δv, Δθ] T can be calculated through feedback control algorithms such as PID control. The specific calculation method is as follows:

[0133]

[0134] where v ref is the reference speed, and K pn , K in and K dn are the proportional coefficient, integral coefficient, and differential coefficient of the PID controller. The transfer device adjusts its own state according to the correction amount Δy to achieve real-time positioning correction and attitude correction.

[0135] During the entire transfer process of the reactor fuel assembly, assign a unique identifier to each reactor fuel assembly, and record its position, azimuth angle, and operation node information in real time to ensure that each reactor fuel assembly can be correctly distinguished and recorded in the case of multiple reactor fuel assemblies being transferred simultaneously, and perform risk monitoring and position correction for each reactor fuel assembly respectively to ensure the safety and efficiency of the overall system.

[0136] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0137] As Figure 2 shown, it is a schematic hardware structure diagram of an electronic device of the present invention, including a memory 202, a processor 201, and an electronic device program on the memory 202. The processor 201 executes the electronic device program to implement the steps of the reactor component transfer positioning method based on a neural network in any of the above embodiments.

[0138] Figure 2 Here, a processor 201 is taken as an example.

[0139] The electronic device may further include: an input device 203 and a display device 204.

[0140] The processor 201, the memory 202, the input device 203, and the display device 204 may be connected through a bus or other means. In the figure, it is taken as an example of being connected through a bus.

[0141] The memory 202, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, executable programs for non-volatile electronic devices, and modules, such as program instructions / modules corresponding to the method for transporting and positioning reactor components based on a neural network in the embodiments of the present application. The processor 201 executes various functional applications and data processing by running the non-volatile software programs, instructions, and modules stored in the memory 202, that is, to implement the method for transporting and positioning reactor components based on a neural network in the above embodiments.

[0142] The memory 202 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the method for transporting and positioning reactor components based on a neural network, etc. In addition, the memory 202 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 202 may optionally include a memory remotely provided with respect to the processor 201, and these remote memories can be connected to the device that executes the method for transporting and positioning reactor components based on a neural network through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0143] The input device 203 can receive input user clicks and generate signal inputs related to user settings and function controls of the method for transporting and positioning reactor components based on a neural network. The display device 204 may include a display screen and other display devices.

[0144] When the one or more modules are stored in the memory 202 and are run by the one or more processors 201, they execute the method for transporting and positioning reactor components based on a neural network in any of the above method embodiments.

[0145] When the electronic device disclosed in the present invention is running, it can execute all steps of the above method for transporting and positioning reactor components based on a neural network, construct an original positioning model through a neural network, train the original positioning model with historical transportation data to obtain a transportation and positioning model, and display the running trajectory of the transportation device in real time through the transportation and positioning model, so as to achieve high-precision positioning of reactor fuel components, and automatically correct the position according to the deviation situation when the reactor fuel components deviate, ensuring that the reactor fuel components are aligned with the reactor core.

[0146] An embodiment of the present invention provides a computer-readable storage medium that stores an electronic device program / instruction, and when the electronic device program / instruction is executed by the processor 201, it implements all steps of the method for transporting and positioning reactor components based on a neural network as described above.

[0147] In the context of the present disclosure, a storage medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The storage medium may be a machine-readable signal medium or a machine-readable storage medium. Optionally, the storage medium may be a non-transitory electronic device-readable storage medium. For example, the non-transitory electronic device-readable storage medium may be a ROM, a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0148] An embodiment of the present invention provides an electronic device program product, including an electronic device program / instructions, which when executed by a processor, implement the steps of the method for transporting and positioning reactor components based on a neural network as described above.

[0149] By running the above-mentioned electronic device program product, all the steps of the method for transporting and positioning reactor components based on a neural network as described above can be executed. An original positioning model is constructed through a neural network, and the original positioning model is trained with historical transportation data to obtain a transportation and positioning model. The running trajectory of the transportation device is displayed in real time through the transportation and positioning model, achieving high-precision positioning of the reactor fuel assembly, and automatically correcting the position according to the deviation situation when the reactor fuel assembly deviates, ensuring that the reactor fuel assembly is aligned with the reactor core.

[0150] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A method for transporting and positioning reactor components based on a neural network, characterized in that, Including: An original positioning model that constructs a neural network structure based on multi-source sensor data; Establish a three-dimensional rectangular coordinate system, and plan an optimal transfer path in combination with the component safety requirements, starting position, and target position of the reactor fuel assembly; Train the original positioning model with historical transfer data, and then optimize the parameters of the original positioning model based on the optimal transfer path to obtain a transfer positioning model; By inputting the multi-source sensor data obtained by the transfer device in real time into the transfer positioning model, the transfer positioning model outputs the positioning data of the reactor fuel assembly in real time, and the positioning data includes position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information.

2. The method for transporting and positioning a reactor component based on a neural network according to claim 1, wherein Also including: Real-time display of the operation trajectory of the transfer device, and control the transfer device to make corresponding positioning corrections when the positioning data meets the risk conditions.

3. The method for transporting and positioning a reactor component based on a neural network according to claim 2, wherein The multi-source sensor data includes laser ranging data, height position sensor data, and ultrasonic measurement data; The original positioning model constructed based on the neural network uses multi-source sensor data as the input of the original positioning model and outputs the positioning data of the reactor fuel assembly, including: Integrate the laser ranging data, height position sensor data, and ultrasonic measurement data into a multi-source sensor data vector as the input data of the original positioning model; Perform normalization processing on the input data to unify the value ranges of different types of sensor data to improve the stability and efficiency of model training; Construct an original positioning model with a deep neural network structure including an input layer, multiple hidden layers, and an output layer; Receive the normalized multi-source sensor data vector through the input layer; The hidden layer performs nonlinear transformation and feature extraction on the input data. The neurons of each hidden layer perform weighted summation and activation operations on the input according to the weights and biases, and gradually extract higher-level features; The output layer outputs the positioning data of the reactor fuel assembly including position information, speed information, acceleration information, temperature information, rotation angle information relative to the original position, and height position information through linear transformation.

4. The method for transporting and positioning a reactor component based on a neural network according to claim 1, characterized in that, The establishment of the three-dimensional rectangular coordinate system includes: Taking the most distal vertex position of the fuel assembly preheating box as the origin, the transportation channel direction as the X-axis, the horizontal vertical direction as the Y-axis, and the height direction as the Z-axis to determine the spatial position of the fuel assembly.

5. The method for transferring and positioning a reactor component based on a neural network according to claim 4, wherein, The combination of the component safety requirements, starting position, and target position of the reactor fuel assembly to plan the optimal transfer path includes: Combined with the component safety requirements, interlock the operations in the X or Y direction when the height in the transfer path does not meet the requirements; Comprehensively consider the path length and the risk assessment value of passing through the dangerous area, and construct a path cost function; The path length is obtained by calculating the distance between the starting position and the target position in the three-dimensional rectangular coordinate system; The risk assessment value is quantified according to the degree of danger; Adopt the A* search algorithm. On the basis of considering the component safety requirements and the path cost, start from the starting position, continuously expand adjacent nodes, calculate the path cost of each node, and gradually search for the target position to determine the optimal transfer path.

6. The method for transferring and positioning a reactor component based on a neural network according to claim 1, wherein, The training of the original positioning model with historical transfer data includes: Collect historical multi-source sensor data during the historical transfer process and the corresponding historical positioning data of the reactor fuel assemblies to form a historical transfer data set; Divide the historical transfer data set into a training set and a validation set according to a preset ratio for the training and performance verification of the original positioning model; Adopt the mean square error loss function to calculate the average of the sum of the squares of the errors between the positioning data predicted by the model and the actual positioning data, and measure the deviation degree between the predicted value and the true value of the original positioning model through the mean square error loss function to optimize the parameters of the original positioning model; Use the stochastic gradient descent optimization algorithm to iteratively train the original positioning model on the training set; In each iteration, calculate the gradients of the weights and biases of the model according to the mean square error loss function, and adjust the weights and biases according to the learning rate to reduce the value of the loss function.

7. The method for transferring and positioning a reactor component based on a neural network according to claim 6, wherein Optimizing the parameters of the original positioning model based on the optimal transfer path to obtain a transfer positioning model includes: On the basis of the original mean square error loss function, add a deviation term between the predicted path and the optimal transfer path, and the deviation term is obtained by calculating the differences between the predicted positions and azimuth angles of the fuel assemblies in the three-dimensional rectangular coordinate system and the azimuth coordinate system and the corresponding points on the optimal transfer path; Set a balance coefficient to balance the influences of the original loss function and the path deviation term, and adjust the size of the balance coefficient according to the requirements of the actual transfer task to determine the emphasis on the accuracy of the positioning data and the path compliance during the model optimization process; Use the stochastic gradient descent optimization algorithm to iteratively update the parameters of the original positioning model again based on the new loss function containing the path deviation term, so that the path output by the model is closer to the optimal transfer path to obtain the transfer positioning model.

8. The method for transporting and positioning reactor components based on a neural network according to claim 1, characterized in that, Inputting the multi-source sensor data obtained by the transfer device in real time into the transfer positioning model, so that the transfer positioning model outputs the positioning data of the reactor fuel assembly in real time, including: During the operation of the transfer device, laser ranging data, height position sensor data, and ultrasonic measurement data are collected in real time and integrated into a multi-source sensor data vector; Normalize the multi-source sensor data vector collected in real time to make it consistent with the input data format and value range during the training of the transfer positioning model, and then input it into the transfer positioning model; According to the input real-time data, the transfer positioning model outputs the current position information, speed information, acceleration information, rotation angle information relative to the original position, and height position information of the reactor fuel assembly through forward propagation calculation.

9. The method for transporting and positioning a reactor component based on a neural network according to claim 2, wherein If the positioning data meets the risk conditions, control the transfer device to make corresponding positioning corrections, including: Construct a risk condition judgment function, comprehensively consider the position deviation, azimuth deviation, and speed deviation, and when the value of the risk condition judgment function exceeds the preset risk threshold, it is determined that the positioning data meets the risk conditions; Adopt the PID control algorithm to calculate the correction amounts of the position, azimuth, and speed respectively according to the position deviation, azimuth deviation, and speed deviation; Send the correction amounts to the transfer device to control the transfer device to adjust its own operating state according to the correction amounts, including adjusting the position, azimuth, and speed, so as to achieve positioning correction.

10. An electronic device, comprising a memory, a processor, and an electronic device program on the memory, characterized in that, The processor executes the electronic device program to implement the steps of the neural network-based reactor component transfer and positioning method according to any one of claims 1-9.

Images