A method and system for determining water flow direction based on physical information neural network

Abstract

The present invention discloses a water flow direction determination method and system based on a physical information neural network, which belongs to the field of construction technology. The method and system are based on the motion trajectory data of the magnetic particles to construct a dynamic distribution model of the water flow field in the tunnel, and then construct a multidimensional data set according to the motion trajectory data and the fluid mechanics parameters in the tunnel of the dynamic distribution model of the water flow field in the tunnel. The physical information neural network is trained based on the multidimensional samples and water flow direction labels in the multidimensional data set using an improved parameter optimization algorithm to obtain a water flow direction determination model. Finally, the water flow direction determination model is used to identify the real-time multidimensional data to obtain a water flow direction determination result. The method can effectively increase the accuracy and uniqueness of the water flow direction determination result, can adapt to the changeable water flow conditions in the tunnel, and provide strong protection and timely warning for tunnel construction safety.

Description

Technical Field

[0001] The present invention belongs to the field of construction technology, and in particular relates to a water flow direction determination method and system based on a physical information neural network. Background Art

[0002] In underground construction projects like tunnels and mines, accurately determining the direction of water flow ahead of the tunnel face is crucial for preventing water inrush and flooding. Traditional geophysical exploration methods typically use geological radar, TSP, and advance drilling to determine water flow direction. These methods rely on a single physical field (electromagnetic waves or seismic waves) to invert the location of water bodies. However, due to interference from complex geological conditions (karst and faults), they suffer from low accuracy, poor real-time performance, and high multi-solution requirements. Summary of the Invention

[0003] The present invention provides a water flow direction determination method and system based on a physical information neural network, which are used to solve the problems of low precision, poor real-time performance and strong multi-solution in the prior art.

[0004] In one aspect, the present invention provides a method for determining water flow direction based on a physical information neural network, comprising:

[0005] The high-precision magnetic sensor array arranged at the tunnel face and surrounding areas collects the motion trajectory data of multiple magnetic particles; wherein the magnetic particles are set in the tunnel construction water flow;

[0006] Based on the motion trajectory data of the magnetic particles, a dynamic distribution model of the water flow field in the tunnel is constructed;

[0007] Constructing a multidimensional dataset based on the motion trajectory data and the tunnel fluid mechanics parameters of the tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels;

[0008] Based on the multidimensional samples and water flow direction labels in the multidimensional data set, an improved parameter optimization algorithm is used to train the physical information neural network to obtain a water flow direction determination model;

[0009] Real-time multi-dimensional data is collected, and the water flow direction determination model is used to identify the real-time multi-dimensional data to obtain a water flow direction determination result.

[0010] Furthermore, it also includes: transmitting the water flow direction determination result to a device designated by the staff and transmitting it to a cloud computing center for storage.

[0011] Furthermore, it also includes: when the water flow direction determination result meets the preset warning condition, generating warning information, and transmitting the warning information to the on-site warning equipment and the equipment designated by the staff.

[0012] Furthermore, a multidimensional dataset is constructed based on the motion trajectory data and the tunnel fluid mechanics parameters of the dynamic distribution model of the water flow field in the tunnel, including:

[0013] Based on a preset data sampling frequency, the motion trajectory data and the in-tunnel fluid mechanics parameters of the in-tunnel water flow field dynamic distribution model are collected at a certain fixed length to obtain a multidimensional sample;

[0014] The water flow direction at a future time point after the target time point is collected to determine the water flow direction label corresponding to the multidimensional sample; wherein the target time point is the last data collection time point corresponding to the multidimensional sample.

[0015] Furthermore, based on the multidimensional samples and water flow direction labels in the multidimensional dataset, an improved parameter optimization algorithm is used to train the physical information neural network to obtain a water flow direction determination model, including:

[0016] Initialize the hyperparameters of the physical information neural network and obtain the population for training;

[0017] For any individual in the population, based on the multidimensional samples and water flow direction labels in the multidimensional dataset, obtain the loss function value corresponding to each individual;

[0018] Determine the best individual and the worst individual based on the loss function values ​​corresponding to all individuals;

[0019] For any individual in the population, a population information interaction strategy is used to search the neighborhood of the individual to obtain the individual after the neighborhood search;

[0020] For any individual after the neighborhood search, an extended search is performed on the individual according to the best individual and the worst individual and the overall position collaborative search strategy to obtain the individual after the extended search;

[0021] For any individual after the extended search, an adaptive fine search is performed on the individual according to the optimal individual and an adaptive variable speed search strategy to obtain an individual after the fine search;

[0022] For any individual after fine search, global mutation search is performed on the individual using the surrounding mutation strategy to obtain the individual after global mutation search;

[0023] Repeat the population information interaction strategy, overall position collaborative search strategy, adaptive speed change search strategy and surrounding mutation strategy until the total number of training times reaches the preset maximum number of training times, and then re-obtain the optimal individual and obtain the target individual;

[0024] The hyperparameters in the target individual are used as the final hyperparameters of the physical information neural network to obtain a water flow direction determination model.

[0025] Furthermore, for any individual in the population, a population information interaction strategy is used to perform a neighborhood search on the individual, and the individual after the neighborhood search is obtained as follows:

[0026]

[0027]

[0028]

[0029] in, Indicates the t During the training m Individuals, Indicates the m Individuals after neighborhood search, m =1,2,…,M, where M represents the total number of individuals in the population, represents the central individual, and each dimension of its hyperparameter is the mean of all hyperparameters in the same dimension; represents the population information interaction parameter, Indicates the t The interaction control factor in the training process, represents the first random number between (0,1), represents the preset maximum value of the interaction control factor, It represents the preset minimum value of the interaction control factor, and T represents the preset maximum number of training times.

[0030] Furthermore, for any individual after the neighborhood search, an extended search is performed on the individual based on the best individual and the worst individual and the overall position collaborative search strategy, and the individual after the extended search is obtained as follows:

[0031]

[0032] in, Indicates the t During the training n Individuals after neighborhood search, Indicates the n individuals after extended search, n =1,2,…,M, represents the first extended search control factor, which is a random number uniformly distributed between [0,1]; represents the second extended search control factor, which is a random number uniformly distributed between [0,1]; represents the worst individual, represents the optimal individual, Represents a second random number between (0,1).

[0033] Furthermore, for any individual after the extended search, an adaptive fine search is performed on the individual according to the optimal individual and an adaptive variable speed search strategy, and the individual after the fine search is obtained as follows:

[0034]

[0035]

[0036]

[0037] in, Indicates the t During the training i individuals after extended search, Indicates the i After a detailed search, i =1,2,…,M, represents the fine search step size control factor, Indicates the t +1 chaos factor during training, Represented as an individual Other individuals randomly matched, Indicates the t The chaos factor in the training process, represents the third random number between (0,1), represents the fourth random number between (0,1), represents pi, e represents a natural constant, and sin represents a sine function.

[0038] Furthermore, for any individual after the fine search, a global mutation search is performed on the individual using the surrounding mutation strategy to obtain the individual after the global mutation search, including:

[0039] For any individual after a fine search, the generated parameters around the number of mutations are:

[0040]

[0041] in, Indicates the t During the training i Individuals after a detailed search The corresponding parameter around the number of mutations, represents the rounding function, Represents an individual The corresponding loss function value is, Represents the loss function value corresponding to the optimal individual, Represents the loss function value corresponding to the worst individual, represents the maximum value around the number of mutation parameters, represents the minimum value around the number of mutation parameters;

[0042] Based on the current number of training times, the generated variation range control parameters are:

[0043]

[0044] in, represents the variation range control parameter, Represents the fourth random number between (0,1);

[0045] According to the variation range control parameters, Generate the corresponding global mutation search individual:

[0046]

[0047] in, Represents an individual Generate the corresponding global mutation search individual, represents the first variation control parameter, represents the second variation control parameter, represents the Cauchy variation, represents Gaussian variation;

[0048] Based on the number of variables, Generate multiple global mutation search individuals;

[0049] Based on the individuals after fine search and their corresponding multiple global mutation search individuals, the individual with the smallest loss function value is selected as the individual after global mutation search.

[0050] On the other hand, the present invention provides a water flow direction determination system based on a physical information neural network, comprising: a data acquisition module, a model construction module, a multidimensional data set construction module, a data training module, and a real-time data recognition module;

[0051] The data acquisition module is used to collect motion trajectory data of multiple magnetic particles through a high-precision magnetic sensor array arranged at the tunnel face and the surrounding area; wherein the magnetic particles are set in the tunnel construction water flow;

[0052] The model building module is used to build a dynamic distribution model of the water flow field in the tunnel based on the motion trajectory data of the magnetic particles;

[0053] The multidimensional dataset construction module is configured to construct a multidimensional dataset based on the motion trajectory data and the in-tunnel fluid mechanics parameters of the in-tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels;

[0054] The data training module is used to train the physical information neural network based on the multidimensional samples and water flow direction labels in the multidimensional data set using an improved parameter optimization algorithm to obtain a water flow direction determination model;

[0055] The real-time data recognition module is used to collect real-time multi-dimensional data, and use the water flow direction determination model to identify the real-time multi-dimensional data to obtain a water flow direction determination result.

[0056] The present invention provides a water flow direction determination method and system based on a physical information neural network. The method and system construct a dynamic distribution model of the water flow field in a tunnel based on the motion trajectory data of the magnetic particles, and then construct a multidimensional data set based on the fluid mechanics parameters in the tunnel of the motion trajectory data and the dynamic distribution model of the water flow field in the tunnel. The physical information neural network is trained using an improved parameter optimization algorithm based on the multidimensional samples and water flow direction labels in the multidimensional data set to obtain a water flow direction determination model. Finally, the water flow direction determination model is used to identify the real-time multidimensional data to obtain a water flow direction determination result. This method can effectively increase the accuracy and uniqueness of the water flow direction determination result, can adapt to the changeable water flow conditions in the tunnel, and provide strong protection and timely warning for tunnel construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0058] Figure 1 A flow chart of a method for determining water flow direction based on a physical information neural network provided in an embodiment of the present invention.

[0059] Figure 2 A schematic structural diagram of a water flow direction determination system based on a physical information neural network provided in an embodiment of the present invention.

[0060] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0061] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0062] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0063] like Figure 1 As shown, an embodiment of the present invention provides a method for determining water flow direction based on a physical information neural network, comprising:

[0064] S101, collecting motion trajectory data of multiple magnetic particles using a high-precision magnetic sensor array arranged at the tunnel face and surrounding areas; wherein the magnetic particles are set in the tunnel construction water flow;

[0065] To facilitate those skilled in the art to better understand the technical solutions described in the embodiments of the present invention, some of the technologies involved in the embodiments of the present invention are explained. The embodiments of the present invention use magnetic particles with high magnetic response performance to ensure that a traceable motion trajectory is formed under the action of water flow during tunnel construction. In addition, a release channel can be laid out in advance in front of the tunnel face through advance drilling to provide a release path for the magnetic particles. In addition, a high-precision magnetic sensor array can be reasonably arranged in the tunnel face and surrounding areas to capture the motion trajectory of the magnetic particles in real time.

[0066] Optionally, the magnetic particles can be encapsulated in a compression capsule to ensure that they are safely and accurately released into the water flow at a predetermined time, thereby providing a dynamic data source for subsequent water flow monitoring.

[0067] S102, constructing a dynamic distribution model of the water flow field in the tunnel based on the motion trajectory data of the magnetic particles;

[0068] The particle tracking method is a numerical method that studies the characteristics of flow fields by tracking the motion trajectories of particles in a fluid. Its core is to regard the fluid as consisting of many particles and simulate the motion of particles in the flow field by solving the particle motion equations. This method does not require directly solving complex fluid dynamics equations, but instead inverts the velocity, pressure and other parameters of the flow field through the motion trajectories of the particles. The particle tracking method is suitable for various flow field simulations, especially for complex flow fields and unsteady flows. Through prediction-correction algorithms and interpolation methods, the particle tracking method can accurately capture the dynamic changes of the flow field, providing a powerful tool for fluid mechanics research and engineering applications.

[0069] Therefore, embodiments of the present invention utilize particle tracking to construct a dynamic distribution model of the water flow field within a tunnel. This effectively captures tunnel fluid dynamics parameters (such as water velocity and pressure distribution), facilitating the subsequent construction of a multidimensional dataset. However, it is worth noting that constructing a dynamic distribution model of the water flow field within a tunnel is a relatively common technique in the field, and existing technologies can be readily employed. This embodiment of the present invention will not be further elaborated upon.

[0070] S103: constructing a multidimensional dataset based on the motion trajectory data and the tunnel fluid mechanics parameters of the tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels;

[0071] In an embodiment of the present invention, a multidimensional dataset is constructed based on the motion trajectory data and the tunnel fluid mechanics parameters of the dynamic distribution model of the water flow field in the tunnel, including:

[0072] Based on a preset data sampling frequency, the motion trajectory data and the in-tunnel fluid mechanics parameters of the in-tunnel water flow field dynamic distribution model are collected at a certain fixed length to obtain a multidimensional sample; the water flow direction at a future time point after the target time point is collected to determine the water flow direction label corresponding to the multidimensional sample; wherein the target time point is the last data collection time point corresponding to the multidimensional sample.

[0073] For example, the motion trajectory data at NP time points and the fluid mechanics parameters in the tunnel of the dynamic distribution model of the water flow field in the tunnel can be collected, and then the collected data can be constructed into a data matrix to obtain a multidimensional sample. Then, the water flow direction at the NP+1th time point is collected to obtain the water flow direction label corresponding to the multidimensional sample.

[0074] S104, based on the multidimensional samples and water flow direction labels in the multidimensional data set, using an improved parameter optimization algorithm to train the physical information neural network to obtain a water flow direction determination model;

[0075] Physics-Informed Neural Networks (PINNs) are advanced machine learning models that combine deep learning with physical modeling. Their core concept is to embed physical laws (usually expressed as partial differential equations) into the neural network's loss function, thereby leveraging both data information and physical constraints during training. PINNs not only rely on data-driven approaches but also penalize the residuals of partial differential equations to ensure that the model's predictions conform to physical laws. This approach is particularly suitable for data-scarce or complex physical scenarios, such as fluid mechanics, materials science, and engineering optimization. It can effectively solve partial differential equations and provide highly accurate predictions. Therefore, the embodiments of the present invention use a physics-informed neural network to construct a water flow direction determination model, effectively improving the accuracy of water flow direction predictions.

[0076] S105 , collecting real-time multi-dimensional data, and using the water flow direction determination model to identify the real-time multi-dimensional data to obtain a water flow direction determination result.

[0077] The magnetic particles provided in the embodiment of the present invention can provide high-resolution trajectory information, and combined with the physical information neural network, can accurately determine the direction of water flow, so that the system can monitor the dynamics of water flow in real time during construction, adapt to the changing water flow conditions in the tunnel, and provide strong protection and timely warning for tunnel construction safety.

[0078] In an embodiment of the present invention, the water flow direction determination result is also transmitted to a device designated by the staff and to a cloud computing center for storage, so that the staff can make timely judgments on the construction site conditions and facilitate tracing.

[0079] In an embodiment of the present invention, the method further includes: when the water flow direction determination result meets a preset warning condition, generating warning information, and transmitting the warning information to an on-site warning device and a device designated by a staff member.

[0080] For example, it can be set to a fixed direction. When the angle between the water flow direction determination result and the preset fixed direction is less than a certain threshold, it is determined that the water flow direction determination result meets the preset warning condition.

[0081] In an embodiment of the present invention, based on the multidimensional samples and water flow direction labels in the multidimensional dataset, an improved parameter optimization algorithm is used to train the physical information neural network to obtain a water flow direction determination model, including:

[0082] Initialize the hyperparameters of the physical information neural network and obtain the population for training;

[0083] For example, the hyperparameters of the physical information neural network can be initialized by randomly initializing between upper and lower limits, and then the initialized hyperparameters can be encoded into vectors, so that individuals can be obtained, and multiple different individuals can be repeatedly obtained to obtain a population for training.

[0084] For any individual in the population, based on the multidimensional samples and water flow direction labels in the multidimensional dataset, obtain the loss function value corresponding to each individual;

[0085] For example, after applying the hyperparameters contained in the individual to the physical information neural network, the multidimensional samples can be used as input and the water flow direction label as the expected output to obtain the loss function value corresponding to the individual.

[0086] According to the loss function values ​​corresponding to all individuals, the optimal individual and the worst individual are determined, that is, the individual with the smallest loss function value is the optimal individual, and the individual with the largest loss function value is the worst individual.

[0087] For any individual in the population, a population information interaction strategy is used to search the neighborhood of the individual to obtain the individual after the neighborhood search;

[0088] For any individual after the neighborhood search, an extended search is performed on the individual according to the best individual and the worst individual and the overall position collaborative search strategy to obtain the individual after the extended search;

[0089] For any individual after the extended search, an adaptive fine search is performed on the individual according to the optimal individual and an adaptive variable speed search strategy to obtain an individual after the fine search;

[0090] For any individual after fine search, global mutation search is performed on the individual using the surrounding mutation strategy to obtain the individual after global mutation search;

[0091] Repeat the population information interaction strategy, overall position collaborative search strategy, adaptive speed change search strategy and surrounding mutation strategy until the total number of training times reaches the preset maximum number of training times, and then re-obtain the optimal individual and obtain the target individual;

[0092] The hyperparameters in the target individual are used as the final hyperparameters of the physical information neural network to obtain a water flow direction determination model.

[0093] Since the existing technology generally uses gradient descent or manual adjustment methods to train physical information neural networks, which often leads to incomplete training and makes the physical information neural network after training unable to accurately determine the direction of water flow. Therefore, an embodiment of the present invention provides a new hyperparameter adjustment method to improve the training effect of the physical information neural network and enhance the accuracy of water flow direction determination.

[0094] In the embodiment of the present invention, for any individual in the population, a population information interaction strategy is used to perform a neighborhood search on the individual, and the individual obtained after the neighborhood search is:

[0095]

[0096]

[0097]

[0098] in, Indicates the t During the training m Individuals, Indicates the m Individuals after neighborhood search, m =1,2,…,M, where M represents the total number of individuals in the population, represents the central individual, and each dimension of its hyperparameter is the mean of all hyperparameters in the same dimension; represents the population information interaction parameter, Indicates the t The interaction control factor in the training process, represents the first random number between (0,1), Indicates the preset maximum value of the interaction control factor (such as 1.5), represents the preset minimum value of the interaction control factor (such as 0.0001), and T represents the preset maximum number of training times.

[0099] The population information interaction strategy provided by the present invention enables individuals to adaptively learn the average position of the entire population, while simultaneously improving the training speed of individuals with poor positions while achieving neighborhood search. The introduction of an interaction control factor also allows for more refined search in the later stages of the algorithm.

[0100] In the embodiment of the present invention, for any individual after the neighborhood search, an extended search is performed on the individual based on the best individual and the worst individual and using the overall position collaborative search strategy, and the individual after the extended search is obtained as follows:

[0101]

[0102] in, Indicates the t During the training n Individuals after neighborhood search, Indicates the n individuals after extended search, n =1,2,…,M, represents the first extended search control factor, which is a random number uniformly distributed between [0,1]; represents the second extended search control factor, which is a random number uniformly distributed between [0,1]; represents the worst individual, represents the optimal individual, Represents a second random number between (0,1).

[0103] The overall position collaborative search strategy provided by the embodiment of the present invention can enable the entire population to perform collaborative search in a certain manner, and stay away from the worst individual and close to the best individual during the search process, which can effectively improve the ability to find the global optimal solution.

[0104] In the embodiment of the present invention, for any individual after the extended search, an adaptive fine search is performed on the individual based on the optimal individual and an adaptive variable speed search strategy, and the individual after the fine search is obtained as follows:

[0105]

[0106]

[0107]

[0108] in, Indicates the t During the training i individuals after extended search, Indicates the i After a detailed search, i =1,2,…,M, represents the fine search step size control factor, Indicates the t +1 chaos factor during training, Represented as an individual Other individuals randomly matched, Indicates the t The chaos factor in the training process (the chaos factor can be set to a constant between (0, 1) at the initial moment, such as 0.78), represents the third random number between (0,1), represents the fourth random number between (0,1), represents pi, e represents a natural constant, and sin represents a sine function.

[0109] The adaptive variable-speed search strategy provided by the present invention effectively learns the location information of optimal individuals and random individuals, enhancing the efficiency of searching unfamiliar areas between individuals, thereby improving search accuracy and the ability to find the global optimal solution. Furthermore, a chaotic factor and a fine search step-size control factor are provided to provide a larger search step size in the early stages of the algorithm to ensure algorithm training speed. As the algorithm progresses, the search step size gradually decreases, and the search step size exhibits chaotic characteristics, which further enhances the ability to find more optimal solutions around individuals.

[0110] In the embodiment of the present invention, for any individual after fine search, a global mutation search is performed on the individual using a surrounding mutation strategy to obtain an individual after global mutation search, including:

[0111] For any individual after a fine search, the generated parameters around the number of mutations are:

[0112]

[0113] in, Indicates the t During the training i Individuals after a detailed search The corresponding parameter around the number of mutations, represents the rounding function, Represents an individual The corresponding loss function value is, Represents the loss function value corresponding to the optimal individual, Represents the loss function value corresponding to the worst individual, represents the maximum value around the number of mutation parameters, represents the minimum value around the number of mutation parameters;

[0114] Based on the current number of training times, the generated variation range control parameters are:

[0115]

[0116] in, represents the variation range control parameter, Represents the fourth random number between (0,1);

[0117] According to the variation range control parameters, Generate the corresponding global mutation search individual:

[0118]

[0119] in, Represents an individual Generate the corresponding global mutation search individual, represents the first variation control parameter, represents the second variation control parameter, represents the Cauchy variation, represents Gaussian variation;

[0120] Based on the number of variables, Generate multiple global mutation search individuals;

[0121] Based on the individuals after fine search and their corresponding multiple global mutation search individuals, the individual with the smallest loss function value is selected as the individual after global mutation search.

[0122] The surrounding mutation strategy provided by the embodiment of the present invention can enable the individual to quickly change its position with itself as the center, thereby jumping out of the local optimal solution, and the adaptive optimization criterion can effectively ensure the training speed of the algorithm.

[0123] In summary, the mutual coordination of several strategies can provide better training effects, enabling the water flow direction determination model to accurately determine the water flow direction, thereby improving the accuracy of the determination.

[0124] The present invention provides a water flow direction determination method based on a physical information neural network. The method constructs a dynamic distribution model of the water flow field in a tunnel based on the motion trajectory data of the magnetic particles, and then constructs a multidimensional data set based on the fluid mechanics parameters in the tunnel of the motion trajectory data and the dynamic distribution model of the water flow field in the tunnel. The physical information neural network is trained using an improved parameter optimization algorithm based on the multidimensional samples and water flow direction labels in the multidimensional data set to obtain a water flow direction determination model. Finally, the water flow direction determination model is used to identify the real-time multidimensional data to obtain a water flow direction determination result. This method can effectively increase the accuracy and uniqueness of the water flow direction determination result, adapt to the changeable water flow conditions in the tunnel, and provide strong protection and timely warning for tunnel construction safety.

[0125] like Figure 2 As shown, an embodiment of the present invention provides a water flow direction determination system based on a physical information neural network, comprising: a data acquisition module 201, a model construction module 202, a multidimensional data set construction module 203, a data training module 204, and a real-time data recognition module 205;

[0126] The data acquisition module 201 is used to collect motion trajectory data of multiple magnetic particles through a high-precision magnetic sensor array arranged at the tunnel face and the surrounding area; wherein the magnetic particles are set in the tunnel construction water flow;

[0127] The model building module 202 is used to build a dynamic distribution model of the water flow field in the tunnel based on the motion trajectory data of the magnetic particles;

[0128] The multidimensional dataset construction module 203 is configured to construct a multidimensional dataset based on the motion trajectory data and the tunnel fluid mechanics parameters of the tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels;

[0129] The data training module 204 is configured to train the physical information neural network using an improved parameter optimization algorithm based on the multidimensional samples and water flow direction labels in the multidimensional dataset to obtain a water flow direction determination model;

[0130] The real-time data identification module 205 is used to collect real-time multi-dimensional data, and use the water flow direction determination model to identify the real-time multi-dimensional data to obtain a water flow direction determination result.

[0131] The embodiment of the present invention provides a water flow direction determination system based on physical information neural network, which can execute the above-mentioned method and technical solution. Its principles and beneficial effects are similar and will not be repeated here.

[0132] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and variations can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for determining water flow direction based on physical information neural network, characterized in that: include: The high-precision magnetic sensor array arranged at the tunnel face and surrounding areas collects the motion trajectory data of multiple magnetic particles; wherein the magnetic particles are set in the tunnel construction water flow; Based on the motion trajectory data of the magnetic particles, a dynamic distribution model of the water flow field in the tunnel is constructed; Constructing a multidimensional dataset based on the motion trajectory data and the tunnel fluid mechanics parameters of the tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels; Based on the multidimensional samples and water flow direction labels in the multidimensional data set, an improved parameter optimization algorithm is used to train the physical information neural network to obtain a water flow direction determination model; Collecting real-time multi-dimensional data, and using the water flow direction determination model to identify the real-time multi-dimensional data to obtain a water flow direction determination result; Based on the multidimensional samples and water flow direction labels in the multidimensional dataset, an improved parameter optimization algorithm is used to train the physical information neural network to obtain a water flow direction determination model, including: Initialize the hyperparameters of the physical information neural network and obtain the population for training; For any individual in the population, based on the multidimensional samples and water flow direction labels in the multidimensional dataset, obtain the loss function value corresponding to each individual; Determine the best individual and the worst individual based on the loss function values ​​corresponding to all individuals; For any individual in the population, a population information interaction strategy is used to search the neighborhood of the individual to obtain the individual after the neighborhood search; For any individual after the neighborhood search, an extended search is performed on the individual according to the best individual and the worst individual and the overall position collaborative search strategy to obtain the individual after the extended search; For any individual after the extended search, an adaptive fine search is performed on the individual according to the optimal individual and an adaptive variable speed search strategy to obtain an individual after the fine search; For any individual after fine search, global mutation search is performed on the individual using the surrounding mutation strategy to obtain the individual after global mutation search; Repeat the population information interaction strategy, overall position collaborative search strategy, adaptive speed change search strategy and surrounding mutation strategy until the total number of training times reaches the preset maximum number of training times, and then re-obtain the optimal individual and obtain the target individual; The hyperparameters in the target individual are used as the final hyperparameters of the physical information neural network to obtain a water flow direction determination model.

2. The water flow direction determination method based on physical information neural network according to claim 1 is characterized in that: Also includes: The water flow direction determination result is transmitted to the equipment designated by the staff and to the cloud computing center for storage.

3. The water flow direction determination method based on physical information neural network according to claim 1 is characterized in that: Also includes: When the water flow direction determination result meets the preset warning condition, a warning message is generated and transmitted to the on-site warning device and the device designated by the staff.

4. The water flow direction determination method based on physical information neural network according to claim 1 is characterized in that: Constructing a multidimensional data set based on the motion trajectory data and the tunnel fluid mechanics parameters of the tunnel water flow field dynamic distribution model, including: Based on a preset data sampling frequency, the motion trajectory data and the in-tunnel fluid mechanics parameters of the in-tunnel water flow field dynamic distribution model are collected at a certain fixed length to obtain a multidimensional sample; The water flow direction at a future time point after the target time point is collected to determine the water flow direction label corresponding to the multidimensional sample; wherein the target time point is the last data collection time point corresponding to the multidimensional sample.

5. The water flow direction determination method based on physical information neural network according to claim 1 is characterized in that: For any individual in the population, the population information interaction strategy is used to perform neighborhood search on the individual, and the individual after the neighborhood search is obtained as follows: in, Indicates the t During the training m Individuals, Indicates the m Individuals after neighborhood search, m =1,2,…,M, where M represents the total number of individuals in the population, represents the central individual, and each dimension of its hyperparameter is the mean of all hyperparameters in the same dimension; represents the population information interaction parameter, Indicates the t The interaction control factor in the training process, represents the first random number between (0,1), represents the preset maximum value of the interaction control factor, It represents the preset minimum value of the interaction control factor, and T represents the preset maximum number of training times.

6. The method for determining water flow direction based on physical information neural network according to claim 5, characterized in that: For any individual after the neighborhood search, the individuals are extended according to the best individual and the worst individual and the overall position collaborative search strategy. The individuals after the extended search are: in, Indicates the t During the training n Individuals after neighborhood search, Indicates the n individuals after extended search, n =1,2,…,M, represents the first extended search control factor, which is a random number uniformly distributed between [0,1]; represents the second extended search control factor, which is a random number uniformly distributed between [0,1]; represents the worst individual, represents the optimal individual, Represents a second random number between (0,1).

7. The water flow direction determination method based on physical information neural network according to claim 6 is characterized in that: For any individual after the extended search, an adaptive fine search is performed on the individual according to the optimal individual and an adaptive variable speed search strategy. The individual after the fine search is: in, Indicates the t During the training i individuals after extended search, Indicates the i After a detailed search, i =1,2,…,M, represents the fine search step size control factor, Indicates the t +1 chaos factor during training, Represented as an individual Other individuals randomly matched, Indicates the t The chaos factor in the training process, represents the third random number between (0,1), represents the fourth random number between (0,1), represents pi, e represents a natural constant, and sin represents a sine function.

8. The method for determining water flow direction based on physical information neural network according to claim 7, characterized in that: For any individual after fine search, a global mutation search is performed on the individual using the surrounding mutation strategy to obtain the individual after global mutation search, including: For any individual after a fine search, the generated parameters around the number of mutations are: in, Indicates the t During the training i Individuals after a detailed search The corresponding parameter around the number of mutations, represents the rounding function, Represents an individual The corresponding loss function value is, Represents the loss function value corresponding to the optimal individual, represents the loss function value corresponding to the worst individual, represents the maximum value around the number of mutation parameters, represents the minimum value around the number of mutation parameters; Based on the current number of training times, the generated variation range control parameters are: in, represents the variation range control parameter, Represents the fourth random number between (0,1); According to the variation range control parameters, Generate the corresponding global mutation search individual: in, Represents an individual Generate the corresponding global mutation search individual, represents the first variation control parameter, represents the second variation control parameter, represents the Cauchy variation, represents Gaussian variation; Based on the number of variables, Generate multiple global mutation search individuals; Based on the individuals after fine search and their corresponding multiple global mutation search individuals, the individual with the smallest loss function value is selected as the individual after global mutation search.

9. A water flow direction determination system based on a physical information neural network, which is capable of executing the water flow direction determination method based on a physical information neural network according to any one of claims 1 to 8, and is characterized in that: include: Data acquisition module, model building module, multidimensional data set building module, data training module and real-time data recognition module; The data acquisition module is used to collect motion trajectory data of multiple magnetic particles through a high-precision magnetic sensor array arranged at the tunnel face and the surrounding area; wherein the magnetic particles are set in the tunnel construction water flow; The model building module is used to build a dynamic distribution model of the water flow field in the tunnel based on the motion trajectory data of the magnetic particles; The multidimensional dataset construction module is configured to construct a multidimensional dataset based on the motion trajectory data and the in-tunnel fluid mechanics parameters of the in-tunnel water flow field dynamic distribution model; wherein the multidimensional dataset includes multidimensional samples and water flow direction labels; The data training module is used to train the physical information neural network based on the multidimensional samples and water flow direction labels in the multidimensional data set using an improved parameter optimization algorithm to obtain a water flow direction determination model; The real-time data recognition module is used to collect real-time multi-dimensional data, and use the water flow direction determination model to identify the real-time multi-dimensional data to obtain a water flow direction determination result.

Images