Sensor array multi-objective optimization method for magnetic field signal denoising

Abstract

The invention discloses a sensor array multi-objective optimization method for magnetic field signal denoising, and the method comprises the implementation steps: building a magnetic field interference model in a constraint mode, carrying out the transient analysis, and obtaining the time sequence data of magnetic field signals collected by all sensors; setting an optimization index, taking the optimization index as a multi-target fitness function, optimizing the multi-target fitness function, and obtaining an optimal auxiliary sensor layout scheme under the condition that the fitness function value is maximum; establishing a main sensor and an optimal auxiliary sensor array based on the optimal auxiliary sensor layout scheme; and in an interference source influence environment, processing a noise signal of the optimal auxiliary sensor array based on the trained deep learning network model to obtain a denoised magnetic field signal of the main sensor, reconstructing a main position noise through sensor layout optimization and auxiliary sensor array noise, and obtaining an optimal auxiliary sensor array. Magnetic field signal denoising under the condition that an object is disturbed is realized, and the precision of magnetic field testing is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic field signals, and specifically to a multi-objective optimization method for a sensor array for magnetic field signal denoising. Background Technique

[0002] Magnetic field measurement technology is a core means in fields such as power equipment monitoring, biomedical imaging, and industrial non-destructive testing. Its accuracy directly affects the accuracy of fault diagnosis and the reliability of signal analysis. However, in practical applications, the target magnetic field is often interfered by complex environmental noises, such as switch surges in the power system, electromagnetic radiation from adjacent devices, or environmental stray magnetic fields in biomedical scenarios. These noises often overlap with the target signal in the frequency domain and have non-stationary and non-linear characteristics, posing severe challenges to traditional denoising methods.

[0003] In the prior art, single-sensor filtering schemes (such as adaptive filtering, wavelet threshold denoising) can suppress some noises, but it is difficult to distinguish signal and interference components in the same frequency band. Especially when the noise intensity is comparable to the target signal, signal distortion is likely to occur. To solve the spatial noise correlation, fixed multi-sensor arrays (such as uniform or regular distributions) are used for spatial filtering, but their rigid layouts lead to two key defects: First, the sensor positions cannot adapt to the spatial distribution changes of interference sources. When the interference source moves or new ones are added, the noise suppression efficiency of the array drops significantly. Second, the fixed layout does not optimize the signal correlation between sensors. Some sensors may collect low-correlation noises, introducing redundant calculations and error propagation instead.

[0004] On the other hand, deep learning-based denoising methods (such as CNN, LSTM) show the potential to handle non-linear noises through end-to-end modeling, but their performance highly depends on the spatial representativeness of the input data. The prior art usually directly uses a fixed sensor array as the network input without combining layout optimization, resulting in the network having to learn the spatial redundant features of the noise additionally, increasing the model complexity and the risk of overfitting. In addition, traditional linear denoising assumptions (such as principal component analysis) are difficult to describe the complex coupling relationship between interference noises and the target magnetic field, further limiting the denoising accuracy. Summary of the Invention

[0005] The technical solution of the present invention: Provide a novel denoising architecture that combines sensor dynamic optimization and deep learning. Screen high-correlation auxiliary sensors through multi-objective metrics, and use the non-linear mapping ability to decouple noises and signals, thereby enhancing the robustness of magnetic field measurement in complex electromagnetic environments and breaking through the technical bottlenecks of traditional methods in adaptive layout and non-linear denoising.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A multi-objective optimization method for a sensor array for magnetic field signal denoising, comprising the following steps: S1: Establish a magnetic field interference model under constrained modes and perform transient analysis to obtain the time series data of magnetic field signals collected by all sensors; S2: Set optimization indices based on the magnetic field signal data, use the optimization indices as a multi-objective fitness function, optimize the multi-objective fitness function, and obtain the optimal layout scheme of auxiliary sensors when the fitness function value reaches the maximum; S3: Based on the optimal layout scheme of auxiliary sensors, establish a main sensor and an optimal auxiliary sensor array; S4: Under the influence of the interference source environment, process the noise signals of the optimal auxiliary sensor array based on the trained deep learning network model to obtain the denoised magnetic field signals of the main sensor.

[0007] Further, the magnetic field interference model under constrained modes in step 1 is established through finite element analysis software, and the process is as follows: Use finite element analysis software to build a magnetic field interference model, and the magnetic field interference model includes an interference source and a disturbed plane. N + 1 sensors for detecting magnetic field signals are arranged on the surface of the disturbed plane, where 1 sensor is the main sensor located at the main position of the disturbed plane, and the remaining N sensors are candidate auxiliary sensors evenly distributed radially on the disturbed plane.

[0008] Further, the process of obtaining the magnetic field signal data collected by all sensors in step 1 is as follows: In the finite element analysis software, define the material properties of the established magnetic field interference model, and perform mesh division. Based on the selected element type and mesh density, apply fixed constraints and define the current load that changes with time, and perform transient analysis to obtain the magnetic field signal data in three directions collected by all sensors.

[0009] Further, the optimization indices in step 2 include optimization index 1 and optimization index 2. Among them, optimization index 1 is the comprehensive dynamic time warping distance between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor, and optimization index 2 is the comprehensive Chatterjee correlation coefficient between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor; The optimal layout scheme of auxiliary sensors refers to the optimal layout scheme of auxiliary sensors selected from N candidate auxiliary sensor layouts; The genetic algorithm with adaptive double structure coding is used to optimize the multi-objective fitness function.

[0010] Further, the process of optimizing the multi-objective fitness function using the genetic algorithm with adaptive dual-structure encoding is as follows: S21: Initialize the population size, maximum number of iterations, number of candidate auxiliary sensors, number of selected auxiliary sensors, mutation selection probability, crossover selection probability, optimization index weight ratio, number of elite individuals, and stagnation threshold iteration number; S22: Construct the individual of the dual-structure encoding population. The dual-structure of the individual chromosome consists of two rows: position encoding and confirmation code. The position encoding indicates the selectable layout positions of the auxiliary sensors, and the confirmation code indicates whether the position is selected; S23: Set the multi-objective fitness function; S24: Optimize the multi-objective fitness function to obtain the optimal layout scheme of the auxiliary sensors when the fitness function value is maximized.

[0011] Further, in S23: The specific process of setting the multi-objective fitness function is as follows: S231: Take Optimization Index 1 and Optimization Index 2 as the multi-objective fitness function, where Optimization Index 1 is negative and Optimization Index 2 is positive;

[0012]

[0013] S232: Normalize the calculated Optimization Index 1 and Optimization Index 2, and weight them according to the optimization index weight ratio to obtain the comprehensive fitness value of a single chromosome. The formula expression is:

[0014] Integrated_Fitness = [weight index_1 , weight index_2 × [Norm Index_1 , Norm Index_2 T ;

[0015] where weight index_1 and weight index_2 respectively represent the proportion of the optimization index weights, and Norm Index_1 and Norm Index_2 respectively represent the normalization results of Optimization Index 1 and Optimization Index 2 within the population.

[0016] Further, in S24: Optimize the multi-objective fitness function to obtain the optimal layout scheme of the auxiliary sensors when the fitness function value is maximized. The specific process is as follows:

[0017] S241: Calculate the comprehensive fitness values of all chromosomes in the initial population. Through roulette wheel selection, partially matched crossover, and inverse mutation, and using the elitist strategy, retain the number of elite individuals of the chromosomes with the highest fitness values to generate a new population;

[0018] ​S242: Propose a non - linear adaptive adjustment mechanism by real - time quantifying the characteristics of population distribution;

[0019] By statistically analyzing the standard deviation and range distribution of population fitness, construct a composite diversity evaluation index, and use the Sigmoid function to achieve the smooth dynamic adjustment of the crossover probability P c and the mutation probability P m , and the formula is as follows:

[0020]

[0021] Among them, set the comprehensive fitness value of the current population as There is:

[0022]

[0023] P c is the crossover probability, P m is the mutation probability, D is the composite diversity index, e is the natural constant, P size represents the population size, f p represents the comprehensive fitness value of the p - th chromosome in the population, f max represents the maximum comprehensive fitness in the population, f min represents the minimum comprehensive fitness in the population, μ is the average fitness, σ is the fitness standard deviation, and Δ is the range normalization factor;

[0024] S243: In S241, add the non - linear adaptive adjustment mechanism and repeat this step until reaching the maximum number of iterations or until the maximum - fitness chromosome in the population does not change after the number of iterations of the stagnation threshold, obtain the large - fitness chromosome, and deconstruct the double - structure encoding to obtain the optimal auxiliary sensor layout.

[0025] Furthermore, the main sensor layout in the optimal auxiliary sensor layout scheme is close to the detected position. The optimal auxiliary sensor layout is obtained by comprehensively optimizing using a multi - objective fitness function, and this optimal auxiliary sensor layout changes with the position and type of the interference source; the interference source affecting the environment means that the current - carrying conductor is not energized and only the interference source generates a magnetic field.

[0026] Furthermore, during the training process of the deep - learning network model, use the noise signal corresponding to the optimal auxiliary sensor in the noise signal of the optimal auxiliary sensor array as the input, and the noise signal corresponding to the main sensor as the output.

[0027] Furthermore, the noise signal of the optimal auxiliary sensor array is processed based on the trained deep learning network model to obtain the denoised magnetic field signal of the main sensor. The process is as follows: The trained deep learning network model is used to reconstruct the noise signal of the optimal auxiliary sensor array and output the noise signal at the main position; the disturbed signal obtained by the main sensor is subtracted from the noise signal at the main position to obtain the denoised magnetic field signal.

[0028] The present invention has the following beneficial effects:

[0029] (1) Multi-modal optimization to enhance anti-interference ability: Through the multi-objective optimization that comprehensively considers the dynamic time warping (DTW) distance and the Chatterjee correlation coefficient, while taking into account the time-domain alignment characteristics and non-linear correlation of the sensor signals, the robustness of the auxiliary sensor layout is significantly improved. Compared with the traditional single-index optimization method, it can more accurately screen out the auxiliary nodes with the strongest correlation with the noise of the main sensor, laying a data foundation for subsequent denoising.

[0030] (2) Deep learning-driven high-precision noise separation: Using a deep learning network to model the main-auxiliary sensor noise can adaptively capture non-linear interference patterns in complex environments, breaking through the limitation of traditional filtering algorithms that rely on prior noise models. The network trained with a large amount of data can dynamically reconstruct the noise at the main position, especially performing excellently in transient mutation or non-stationary interference scenarios, and significantly improving the denoising accuracy.

[0031] (3) Full-process closed-loop optimization design: Forming a closed-loop system from finite element modeling, multi-objective optimization layout to online noise cancellation, combining the dual advantages of physical mechanism and data-driven. The finite element model provides theoretical constraints for sensor layout, reducing experimental costs; data-driven optimization dynamically adapts to actual working conditions to ensure the universality of the scheme in complex electromagnetic environments.

[0032] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. Description of the Drawings

[0033] Figure 1 It is a flow chart of a multi-objective optimization method for a sensor array for magnetic field signal denoising according to the present invention;

[0034] Figure 2 It is a flow chart for implementing the method of the present invention;

[0035] Figure 3 It is the overall technical roadmap of this method;

[0036] Figure 4 It is a physical structure simulation diagram of this method for electromagnetic interference;

[0037] Figure 5It is the simulation diagram of the main - auxiliary sensor layout;

[0038] Figure 6 It is the simulation result diagram under electromagnetic interference;

[0039] Figure 7 It is the position coding diagram of the auxiliary sensors;

[0040] Figure 8 It is the curve diagram of each index in the optimization process of the standard GA algorithm and the new GA algorithm when the number of auxiliary sensors is 4;

[0041] Figure 9 It is the comprehensive comparison diagram of the optimization process of the standard GA algorithm and the new GA algorithm when the number of auxiliary sensors is 4;

[0042] Figure 10 It is the layout result diagram of the optimized auxiliary sensors;

[0043] Figure 11 It is the signal waveform of the main sensor and the reconstructed noise waveform diagram when the current - carrying conductor is energized under electromagnetic interference;

[0044] Figure 12 It is the comparison diagram of the denoised signal waveform and the ideal waveform obtained by implementing the method of the present invention; Figure 13 It is the error comparison diagram of the denoised signal obtained by implementing the method of the present invention compared with the case of randomly arranging the auxiliary sensors. Detailed implementation manners

[0045] Please refer to Figure 1 , the embodiment of the present invention provides a technical solution: a multi - objective optimization method for a sensor array for magnetic field signal denoising. The method flow and the overall technical route are as Figure 2 and Figure 3 shown, and it includes the following steps: S1: Establish a magnetic field interference model under constrained modes and conduct a transient analysis to obtain the time - series data of the magnetic field signals collected by all sensors;

[0046] The magnetic field interference model under constrained modes in Step 1 is established by finite - element analysis software. The process is as follows: Use finite - element analysis software to build a magnetic field interference model. The built magnetic field interference model includes an interference source and a perturbed plane, and N + 1 sensors for detecting magnetic field signals are set on the surface of the perturbed plane, where 1 sensor is the main sensor located at the main position of the perturbed plane, and the remaining N sensors are candidate auxiliary sensors evenly distributed radially on the perturbed plane.

[0047] The process of obtaining the magnetic field signal data collected by all sensors in Step 1 is as follows: In the finite element analysis software, define the material properties of the established magnetic field interference model, and perform mesh division. Based on the selected element type and mesh density, apply fixed constraints and define the time-varying current load, and perform transient analysis to obtain the magnetic field signal data in three directions collected by all sensors.

[0048] Simulation: The prerequisite for realizing the optimal layout of auxiliary sensors is to obtain as much magnetic field distribution data as possible on the main position and its surrounding areas on the disturbed plane under electromagnetic interference. Use electromagnetic simulation analysis software, such as Ansys, COMSOL, etc., to obtain the magnetic field distribution data on the disturbed plane under different electromagnetic interferences. For different interference sources, perform the following steps:

[0049] (1) Import the physical structure model of electromagnetic interference, which mainly includes: interference source, disturbed plane, and the main sensors and candidate auxiliary sensors arranged on the plane;

[0050] (2) Define the material properties of the established model, and select the appropriate element type and mesh density;

[0051] (3) Define the boundary conditions according to the operating conditions parameters of the interference source;

[0052] (4) Select the solver and set the solution parameters;

[0053] (5) Perform simulation calculations;

[0054] (6) Obtain the magnetic field signal data collected by all sensors under the current interference source, and export the data to Excel;

[0055] First, a detailed survey and measurement of the interference source scenario are required. Collect various size information in the scenario, including the length, width, and height of the disturbed plane to be detected, as well as the interference source type and location information. Then, use professional 3D modeling software, such as SolidWorks, to convert the collected size data into a visual 3D model. Finally, import the model into the finite element analysis software and execute the above steps.

[0056] For the physical structure simulation diagram of electromagnetic interference, the main - auxiliary sensor layout simulation diagram, and the simulation result diagram under electromagnetic interference are as Figure 4 、 Figure 5 and Figure 6 shown. Among them, the interference source is a cylindrical energized solenoid. Apply fixed constraints and define the time-varying current load to it. A main sensor is located at the center of the disturbed plane. Select 42 candidate auxiliary sensors evenly distributed radially on the disturbed plane. Perform transient analysis on the model to obtain the time series of the X / Y / Z three-axis magnetic field signals of all sensors.

[0057] S2: Set optimization metrics based on the magnetic field signal data, use the optimization metrics as the multi-objective fitness function, optimize the multi-objective fitness function, and obtain the optimal auxiliary sensor layout scheme when the fitness function value reaches the maximum;

[0058] In step 2, the optimization metrics include optimization metric 1 and optimization metric 2. Among them, optimization metric 1 is the comprehensive dynamic time warping (DTW) distance between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor, and the smaller this optimization metric is, the better; optimization metric 2 is the comprehensive Chatterjee correlation coefficient between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor, and the larger it is, the better;

[0059] The specific process is as follows:

[0060] Define the magnetic field signal data collected by the sensor to be output as a discrete time series of length len through the analog-to-digital converter: S = {s 1 , s 2 ,..., s len}.

[0061] Optimization metric 1 is:

[0062]

[0063] Set two time series with time series lengths of i and j respectively: and

[0064] Then, calculate the dynamic time warping distance dtw(i, j) between them through the following recursive formula:

[0065]

[0066] Optimization metric 2 is:

[0067]

[0068] Among them, Index_1 represents optimization metric 1, and Index_2 represents optimization metric 2;

[0069] n sensor represents the number of selected auxiliary sensors, and m axis represents the magnetic field intensity direction (1 represents the X-axis, 2 represents the Y-axis, 3 represents the Z-axis);

[0070] represents the length of the m axis axis data collected by the main sensor, represents the m collected by the nth selected auxiliary sensoraxis The length of the shaft data;

[0071] Expressed as time series data S a The i-th element in Similarly;

[0072] Define w as the Sakoe-Chiba band constraint window parameter, and the symmetric band region ∣i - j∣ ≤ w;

[0073] l g Denote the index array S after sorting the time series S order In, the rank corresponding to the g-th element of the time series S, that is, the number of elements in the sequence S greater than this element, is expressed by the formula:

[0074]

[0075] Among them, |() represents the indicator function, S order [h] represents the size corresponding to the h-th element of the time series S in the index array S order In, S order [g] is the same.

[0076] r g Is a copy of l g The difference is that there will be different handling methods when dealing with ties (there are equal elements during the sorting process). If the g-th element of the time series S is equal to other elements, then: randomly sort and select values among these equal elements.

[0077] The optimal auxiliary sensor layout scheme refers to the optimal auxiliary sensor layout scheme selected from N candidate auxiliary sensor layouts; the genetic algorithm with adaptive double structure coding is used to optimize the multi-objective fitness function.

[0078] The optimal sensor layout scheme in this example refers to the 4 optimal auxiliary sensor layouts selected from 42 candidate auxiliary sensor positions, and the optimization method used is a genetic algorithm with a new type of adaptive double structure coding;

[0079] The process of optimizing the multi-objective fitness function using the genetic algorithm with adaptive double structure coding is as follows: S21: Initialize the population size, the maximum number of iterations, the number of candidate auxiliary sensors, the number of selected auxiliary sensors, the mutation selection probability, the crossover selection probability, the optimization index weight ratio, the number of elite individuals, and the stagnation threshold number of iterations;

[0080] The following are some parameters of the genetic algorithm with a new type of adaptive double structure coding:

[0081]

[0082]

[0083] S22: Construct the individuals of the dual - structure - encoded population. The dual - structure of an individual chromosome consists of two rows: the position encoding and the confirmation code. The position encoding indicates the selectable layout positions of the auxiliary sensors, and the confirmation code indicates whether the position is selected or not.

[0084] The position encodings of 42 candidate auxiliary sensors are as shown in Figure 7 Then a dual - structure - encoded structure is:

[0085]

[0086] S23: Set the multi - objective fitness function.

[0087] S231: Take Optimization Index 1 and Optimization Index 2 as the multi - objective fitness function, where Optimization Index 1 is taken as negative and Optimization Index 2 is taken as positive.

[0088]

[0089] S232: Normalize the calculated Optimization Index 1 and Optimization Index 2, and weight them according to the optimization - index weight ratio to obtain the comprehensive fitness value of a single chromosome. The formula expression is:

[0090] Integrated_Fitness = [weight index_1 , weight index_2 × [Norm Index_1 , Norm Index_2 T ;

[0091] Among them, weight index_1 and weight index_2 respectively represent the proportion of the optimization - index weights, and Norm Index_1 and Norm Index_2 respectively represent the normalization results of Optimization Index 1 and Optimization Index 2 within the population.

[0092] S24: Optimize the multi - objective fitness function to obtain the optimal layout scheme of the auxiliary sensors when the fitness - function value is maximized.

[0093] S241: Calculate the comprehensive fitness values of all chromosomes in the initial population. Through roulette - wheel selection, partially - mapped crossover, inverse - bit mutation, and using the elitist - preservation strategy, retain the number of elite individuals of the chromosomes with the highest fitness values to generate a new population.

[0094] ​S242: Propose a non - linear adaptive regulation mechanism by quantifying the population distribution characteristics in real time;

[0095] By statistically analyzing the standard deviation and range distribution of the population fitness, construct a composite diversity evaluation index, and use the Sigmoid function to achieve the smooth dynamic regulation of the crossover probability P c and the mutation probability P m . The formula is as follows:

[0096]

[0097] Among them, set the comprehensive fitness value of the current population as . There is:

[0098]

[0099] P c is the crossover probability, P m is the mutation probability, D is the composite diversity index, e is the natural constant, P size represents the population size, f p represents the comprehensive fitness value of the p - th chromosome in the population, f max represents the maximum comprehensive fitness in the population, f min represents the minimum comprehensive fitness in the population, μ is the average fitness, σ is the fitness standard deviation, Δ is the range normalization factor;

[0100] S243: In S241, add the non - linear adaptive regulation mechanism and repeat this step until the maximum number of iterations is reached or the maximum - fitness chromosome in the population does not change after the stagnation threshold number of iterations, obtain the large - fitness chromosome, and deconstruct the double - structure encoding to obtain the optimal auxiliary sensor layout.

[0101] The curve graphs of each index in the optimization processes of the standard GA algorithm and the new GA algorithm are as Figure 8 shown, where a represents the result of the optimization process of the standard GA algorithm, b represents the result of the optimization process of the new GA algorithm, Figure 9 represents the comprehensive comparison graph of the optimization processes of the standard GA algorithm and the new GA algorithm. From the trend of the graph, it can be seen that the new GA algorithm reaches a higher performance value faster under the same number of iterations, while the standard GA algorithm improves slowly. This shows that the adaptive mechanism can accelerate the search efficiency and reduce resource consumption. The final obtained optimal sensor layout is as Figure 10 .

[0102] S3: Based on the optimal auxiliary sensor layout scheme, establish the main sensor and the optimal auxiliary sensor array;

[0103] S4: In the environment affected by the interference source, process the noise signals of the optimal auxiliary sensor array based on the trained deep learning network model to obtain the denoised magnetic field signal of the main sensor.

[0104] In the optimal auxiliary sensor layout scheme, the layout of the main sensor is close to the position to be detected. The optimal auxiliary sensor layout is obtained by comprehensively optimizing using a multi-objective fitness function, and this optimal auxiliary sensor layout changes with the position and type of the interference source. The environment affected by the interference source means that the current-carrying conductor is not energized, and only the interference source generates a magnetic field.

[0105] During the training process of the deep learning network model, use the noise signal corresponding to the optimal auxiliary sensor in the noise signals of the optimal auxiliary sensor array as the input, and the noise signal corresponding to the main sensor as the output. Use a large amount of data to perform supervised training on the network. Among them, both the main sensor noise and the optimal auxiliary sensor noise are time series data. The input of the deep learning network is a one-dimensional time series signal of the number of auxiliary sensors, and the output is a one-dimensional time series data of a single-channel main sensor.

[0106] Process the noise signals of the optimal auxiliary sensor array based on the trained deep learning network model to obtain the denoised magnetic field signal of the main sensor. The process is as follows: Use the trained deep learning network model to reconstruct the noise signals of the optimal auxiliary sensor array and output the noise signal at the main position; subtract the disturbed signal obtained by the main sensor from the noise signal at the main position to obtain the denoised magnetic field signal.

[0107] In the state where the environment is affected by the interference source and the current-carrying conductor is energized to generate a magnetic field, use the trained deep learning network model to reconstruct the noise at the main position from the noise of the auxiliary sensor array. According to Ampere's law, when the current-carrying conductor is in the energized state, a magnetic field proportional to the current intensity will be formed around it. The main sensor is arranged above the current-carrying conductor. In the environment affected by the interference source, the signal collected by the main sensor is the superposition of the magnetic fields generated by the current-carrying conductor and the interference source, as Figure 10 shown by the red curve part in the figure, and the signal collected by the auxiliary sensor is the magnetic field of the interference source at its position. Use the trained deep learning network model to reconstruct the noise at the main position from the noise of the auxiliary sensor, as Figure 11 shown by the green curve part in the figure.

[0108] Subtract the noise signal at the main position reconstructed by the auxiliary sensor array from the disturbed signal obtained by the main sensor to obtain the denoised magnetic field signal. As Figure 12 shown in the figure, the red line in the figure is the denoised signal, and the black line is the ideal signal. Figure 13To achieve the error comparison diagram of the denoised signal obtained by the method of the present invention compared with the random layout of the auxiliary sensors. The smaller the error indicators RMSE (root mean square error) and MAE (mean absolute error) in the figure, the better the denoising effect. It can be seen that through the scientific sensor layout strategy, the method of the present invention effectively improves the signal acquisition efficiency and reduces the environmental noise interference, so that excellent performance can still be maintained in complex scenarios.

[0109] An electronic device, comprising: a processor; and a memory in which computer program instructions are stored, and when the computer program instructions are run by the processor, the processor is caused to execute the multi-objective optimization method of the sensor array for magnetic field signal denoising as described above.

[0110] A computer-readable storage medium for storing a program, which when executed by a processor implements the multi-objective optimization method of the sensor array for magnetic field signal denoising as described above.

[0111] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to the flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0113] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, causing a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 or steps for implementing the functions specified in a block or a plurality of blocks.

[0115] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic creative concept. Therefore, the appended claims are intended to be construed to cover the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0116] It is obvious that those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A multi-objective optimization method for a sensor array for magnetic field signal denoising, characterized in that It includes the following steps: S1: Establish a magnetic field interference model in the constrained mode and conduct a transient analysis to obtain the time series data of the magnetic field signals collected by all sensors; S2: Set optimization indicators based on the magnetic field signal data, use the optimization indicators as the multi-objective fitness function, optimize the multi-objective fitness function, and obtain the optimal layout plan of the auxiliary sensors when the fitness function value reaches the maximum; S3: Based on the optimal layout plan of the auxiliary sensors, establish the main sensor and the optimal auxiliary sensor array; S4: Under the influence of the interference source environment, process the noise signals of the optimal auxiliary sensor array based on the trained deep learning network model to obtain the denoised magnetic field signals of the main sensor.

2. The multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 1, characterized in that, The magnetic field interference model in the constrained mode in step 1 is established by finite element analysis software, and the process is as follows: Use finite element analysis software to build a magnetic field interference model. The magnetic field interference model includes the interference source and the perturbed plane, and N+1 sensors for detecting magnetic field signals are set on the surface of the perturbed plane. Among them, 1 sensor is the main sensor located at the main position of the perturbed plane, and the remaining N sensors are candidate auxiliary sensors evenly distributed radially on the perturbed plane.

3. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 2, characterized in that, The process of obtaining the magnetic field signal data collected by all sensors in step 1 is as follows: In the finite element analysis software, define the material properties of the established magnetic field interference model, perform mesh division, based on the selected element type and mesh density, apply fixed constraints and define the current load that changes with time, and conduct a transient analysis to obtain the magnetic field signal data in three directions collected by all sensors.

4. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 2, characterized in that, The optimization indicators in step 2 include optimization indicator 1 and optimization indicator 2. Among them, optimization indicator 1 is the comprehensive dynamic time warping distance between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor, and optimization indicator 2 is the comprehensive Chatterjee correlation coefficient between the sensor magnetic field signal data of the selected candidate auxiliary sensor layout and the main sensor; The optimal layout plan of the auxiliary sensors refers to the optimal layout plan of the auxiliary sensors selected from the N candidate auxiliary sensor layouts; An adaptive dual-structure encoding genetic algorithm is used to optimize the multi-objective fitness function.

5. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 4, characterized in that, The process of using the adaptive dual-structure encoding genetic algorithm to optimize the multi-objective fitness function is as follows: S21: Initialize the population size, maximum number of iterations, number of candidate auxiliary sensors, number of selected auxiliary sensors, mutation selection probability, crossover selection probability, optimization indicator weight ratio, number of elite individuals, and stagnation threshold iteration number; S22: Construct the individual of the dual-structure encoding population. The dual-structure of the individual chromosome consists of two rows of position encoding and confirmation code. Among them, the position encoding indicates the selectable layout positions of the auxiliary sensors, and the confirmation code is whether to select this position; S23: Set the multi-objective fitness function; S24: Optimize the multi-objective fitness function to obtain the optimal layout plan of the auxiliary sensors when the fitness function value reaches the maximum.

6. The multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 5, wherein The specific process of setting the multi-objective fitness function in S23 is as follows: S231: Take Optimization Index 1 and Optimization Index 2 as the multi-objective fitness function, where Optimization Index 1 is negative and Optimization Index 2 is positive; S232: Normalize the calculated Optimization Index 1 and Optimization Index 2, and weight them according to the optimization index weight ratio to obtain the comprehensive fitness value of a single chromosome. The formula expression is: Integrated_Fitness=[weight index_1 ,weight index_2 ×[Norm Index_1 ,Norm Index_2 T ;​ Among them, weight index_1 and weight index_2 respectively represent the proportion of the optimization index weights, Norm Index_1 and Norm Index_2 respectively represent the normalization results of optimization index 1 and optimization index 2 within the population.

7. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 5, characterized in that, S24: Optimize the multi-objective fitness function to obtain the optimal auxiliary sensor layout scheme when the fitness function value reaches the maximum. The specific process is as follows: S241: Calculate the comprehensive fitness values of all chromosomes in the initial population, perform roulette wheel selection, partially matched crossover, and inverse mutation, and use the elite preservation strategy to retain the number of elite individuals of the chromosomes with the highest fitness values to generate a new population; S242: Propose a non-linear adaptive adjustment mechanism by real-time quantifying the population distribution characteristics; Construct a composite diversity evaluation index by statistically analyzing the standard deviation and range distribution of population fitness, and use the Sigmoid function to achieve the smooth dynamic adjustment of the crossover probability P c and the mutation probability P m , and the formula is expressed as: Among them, the comprehensive fitness value of the contemporary population is set to There is: P c is the crossover probability, P m is the mutation probability, D is the composite diversity index, e is the natural constant, P size represents the population size, f p represents the comprehensive fitness value of the p-th chromosome in the population, f max represents the maximum comprehensive fitness in the population, f min represents the minimum comprehensive fitness in the population, μ is the average fitness, σ is the fitness standard deviation, and Δ is the range normalization factor; S243: In S241, add the non-linear adaptive adjustment mechanism and repeat this step until the maximum iteration number is reached or the maximum fitness chromosome in the population does not change after the stagnation threshold iteration number. Obtain the large fitness chromosome and deconstruct the double structure encoding to obtain the optimal auxiliary sensor layout.

8. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 1, characterized in that: In the optimal auxiliary sensor layout scheme, the main sensor layout is close to the detected position. The optimal auxiliary sensor layout is obtained by comprehensively optimizing the multi-objective fitness function, and this optimal auxiliary sensor layout changes with the position and type of the interference source; The interference source affecting the environment means that the current-carrying conductor is not energized and only the interference source generates a magnetic field.

9. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 1, characterized in that, During the training process of the deep learning network model, the noise signal corresponding to the optimal auxiliary sensor in the noise signal of the optimal auxiliary sensor array is used as the input, and the noise signal corresponding to the main sensor is used as the output.

10. A multi-objective optimization method for a sensor array for magnetic field signal denoising according to claim 9, characterized in that, Based on the trained deep learning network model, process the noise signal of the optimal auxiliary sensor array to obtain the denoised magnetic field signal of the main sensor. The process is as follows: Use the trained deep learning network model to reconstruct the noise signal of the optimal auxiliary sensor array and output the noise signal at the main position; Subtract the noise signal at the main position from the disturbed signal obtained by the main sensor to obtain the denoised magnetic field signal.

Images