A Method and System for Identifying High-Resistance Grounding Faults in Coal Mines Based on PINN and MLE Modifications

Abstract

This invention proposes a method and system for identifying high-resistivity grounding faults in coal mines based on PINN and MLE corrections, relating to the field of power system relay protection technology. It addresses the problem of accurately solving the Holmes-Duffing oscillator bifurcation threshold. The method includes calculating the zero-sequence current before and after the fault to obtain the change in zero-sequence current of the faulted line; using the change in zero-sequence current as input to a physical information neural network (PIN), and the bifurcation threshold and state sequence as network outputs. The PSN uses the Holmes-Duffing oscillator system differential equation as physical constraints; employing the Benettin algorithm to calculate the maximum Lyapunov exponent of the oscillator system state for each line; and determining the chaotic state of the system based on the positive and negative characteristics of the maximum Lyapunov exponent. This invention utilizes the PSN to accurately solve the bifurcation threshold, and corrects the network through the maximum Lyapunov exponent to directly determine the line fault, thus improving the accuracy and practicality of high-resistivity grounding fault identification.

Description

Technical Field

[0001] This invention belongs to the field of power system relay protection technology, and particularly relates to a method and system for identifying high-resistance grounding faults in coal mines based on PINN and MLE correction. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] When a high-resistance grounding fault occurs in a coal mine power grid, the large transition resistance of the fault circuit results in low amplitude and weak characteristics of the zero-sequence current generated before and after the fault. This current is easily interfered with by background noise such as the system's own load current, electromagnetic interference, and line distributed capacitance. Although existing technologies have applied the high sensitivity of the Holmes-Duffing oscillator system to weak signals for identification, the following problems still exist in fault identification using the Holmes-Duffing oscillator system: First, the current technology for solving the bifurcation threshold usually uses a trial-and-error method. After obtaining the threshold through human experience, external excitation is added to verify its accuracy. This not only results in low accuracy in calculating the bifurcation threshold but is also a cumbersome process. Second, after determining the bifurcation threshold, fault line selection relies on manual judgment of the fault result using phase diagrams. When the phase diagram is close to the critical point of the bifurcation threshold, it may exhibit a weakly chaotic or quasi-periodic state, which can easily lead to misjudgment of the identification result. Therefore, when identifying high-resistance grounding faults based on the Holmes-Duffing oscillator system, there are problems such as low accuracy in solving the bifurcation threshold, cumbersome process, and misjudgment of the final result. Existing methods are difficult to apply in practical engineering. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method and system for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction. The Holmes-Duffing oscillator system state equation is embedded into a Physical Information Neural Network (PINN) as a physical constraint. The maximum Lyapunov exponent is calculated based on the network output. The faulty line is directly identified and the network is corrected based on the exponent calculation result. This method can accurately solve for the bifurcation threshold and reduce computational complexity and misjudgment of identification results.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction; A method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE corrections includes: Calculate the zero-sequence current before and after the fault to obtain the change in zero-sequence current of the faulted line. The zero-sequence current change is used as the input of the physical information neural network, and the bifurcation threshold and state sequence are used as the network output. The physical information neural network uses the Holmes-Duffing oscillator system differential equation as the physical constraint. The Benettin algorithm is used to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and the chaotic state of the system is determined based on the positive and negative characteristics of the maximum Lyapunov exponent. The bifurcation threshold of the physical information neural network output is verified and corrected by the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

[0006] Furthermore, the differential equation of the Holmes-Duffing oscillator system is specifically as follows:

[0007] In the formula, The damping coefficient is... Angular frequency, p For the bifurcation threshold, For input, It is a state sequence; Furthermore, the loss function of the physical information neural network is composed of a weighted sum of the differential equation residual loss term, the initial condition loss term, and the data loss term; Furthermore, the residual loss term of the differential equation is obtained by substituting the derivative obtained by automatic differentiation into the differential equation of the Holmes-Duffing oscillator system to calculate the residual; the initial condition loss term consists of the deviation between the network output at the initial moment and the given initial displacement and initial velocity; the data loss term consists of the deviation between the network output and the measured data. Furthermore, the Holmes-Duffing oscillator system is transformed into a first-order state space, the Jacobian matrix is ​​solved, and a four-dimensional extended system is constructed. The fourth-order Runge-Kutta method was used for numerical integration. The perturbation vector was normalized at a preset renormalization interval and the logarithmic amplification factor was recorded. The maximum Lyapunov exponent estimate is obtained by averaging the logarithmic magnification factor over the effective integration time. Furthermore, using the positive and negative characteristics of the maximum Lyapunov exponent as a condition, if the chaotic state of each line determined based on the current bifurcation threshold is different from that of the known faulty line, the parameters in the loss function of the physical information neural network are adjusted and retrained; if the chaotic state of each line determined based on the current bifurcation threshold is the same as that of the known faulty line, the bifurcation threshold is output, and the fault line selection is completed based on the threshold. Furthermore, if the maximum Lyapunov exponent is greater than zero, the system is determined to be in a chaotic state; if the maximum Lyapunov exponent is less than or equal to zero, the system is determined to be in a periodic or quasi-periodic state.

[0008] The second aspect of the present invention provides a coal mine high-resistivity grounding fault identification system based on PINN and MLE correction.

[0009] A coal mine high-resistivity grounding fault identification system based on PINN and MLE correction includes a data processing module, a model building module, a maximum Lyapunov exponent solving module, and a correction identification module. The data processing module is configured to: calculate the zero-sequence current before and after the fault, and obtain the change in zero-sequence current of the faulted line. The model building module is configured to take the zero-sequence current change as the input of the physical information neural network, and take the bifurcation threshold and state sequence as the network output. The physical information neural network takes the Holmes-Duffing oscillator system differential equation as the physical constraint. The maximum Lyapunov exponent calculation module is configured to: use the Benettin algorithm to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and determine the chaotic state of the system based on the positive and negative characteristics of the maximum Lyapunov exponent; The correction identification module is configured to: reverse-verify and correct the bifurcation threshold of the physical information neural network output by the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

[0010] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a coal mine high-resistivity grounding fault identification method based on PINN and MLE correction as described in the first aspect of the present invention.

[0011] A fourth aspect of the present invention provides a chip system including one or more processors, said one or more processors being invoked to perform steps in implementing a method for identifying high-resistance grounding faults in coal mines based on PINN and MLE correction as described in the first aspect of the present invention.

[0012] The above one or more technical solutions have the following beneficial effects: The proposed method for identifying high-resistivity grounding faults in coal mines, based on PINN and MLE corrections, introduces a physical information neural network for solution. By embedding the differential equations of the Holmes-Duffing oscillator system with external excitation as physical constraints into the physical information neural network, it overcomes the limitations of purely numerical iterative or purely data-driven models. This method achieves efficient approximation and solution of bifurcation boundaries while adhering to the inherent physical constraints of the system. It exhibits good generalization ability and accuracy in practical engineering, providing a novel approach and theoretical basis for the reliable identification of high-resistivity grounding faults in coal mine power grids.

[0013] Using the maximum Lyapunov exponent as the basis and verification standard for identifying chaotic and periodic states, the Benettin algorithm is used to calculate the maximum Lyapunov exponent. Compared with the traditional method of fault identification through phase diagrams, this method can solve the problem of misjudgment caused by manually selecting faulty lines and improve the accuracy of the final fault identification. On the other hand, by using the maximum Lyapunov exponent to inversely correct the output of the PINN network, the reliability of the solved bifurcation threshold is ensured, and the problem of the inability to interpret neural network models in application is solved.

[0014] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0015] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0016] Figure 1 This is a flowchart of a coal mine high-resistance grounding fault identification method based on PINN and MLE correction in an embodiment of the present invention. Figure 2 This is a logic diagram of the fault identification method in an embodiment of the present invention; Figure 3 This is an equivalent model diagram of the coal mine system in an embodiment of the present invention; Figure 4 This is an equivalent circuit diagram of the faulty circuit in an embodiment of the present invention; Figure 5(a) is a waveform diagram of the zero-sequence current difference when the grounding resistance is 90KΩ in an embodiment of the present invention; Figure 5(b) is a waveform diagram of the zero-sequence current difference when the grounding resistance is 50KΩ in an embodiment of the present invention; Figure 6 This is a graph of the training loss function of the PINN network in an embodiment of the present invention; Figure 7 This is a state sequence diagram predicted by the PINN network in an embodiment of the present invention; Figure 8(a) is a phase 1 diagram of the line when the grounding resistance is 90KΩ in an embodiment of the present invention; Figure 8(b) is a state diagram of the two phases of the line when the grounding resistance is 90KΩ in an embodiment of the present invention; Figure 8(c) is a three-phase diagram of the line when the grounding resistance is 90KΩ in an embodiment of the present invention; Figure 8(d) is a state diagram of the 4-phase line when the grounding resistance is 90KΩ in an embodiment of the present invention; Figure 9(a) is a phase 1 diagram of the line when the grounding resistance is 50KΩ in an embodiment of the present invention; Figure 9(b) is a state diagram of the two phases of the line when the grounding resistance is 50KΩ in an embodiment of the present invention; Figure 9(c) is a three-phase diagram of the line when the grounding resistance is 50KΩ in an embodiment of the present invention; Figure 9(d) is a state diagram of the 4-phase line when the grounding resistance is 50KΩ in an embodiment of the present invention; Figure 10(a) is a phase 1 diagram of the line when the grounding resistance is 25KΩ in an embodiment of the present invention; Figure 10(b) is a state diagram of the two phases of the line when the grounding resistance is 25KΩ in an embodiment of the present invention; Figure 10(c) is a three-phase diagram of the line when the grounding resistance is 25KΩ in an embodiment of the present invention; Figure 10(d) is a state diagram of the 4-phase line when the grounding resistance is 25KΩ in an embodiment of the present invention; Figure 11(a) is a phase 1 diagram of the line when the grounding resistance is 5KΩ in an embodiment of the present invention; Figure 11(b) is a state diagram of the two phases of the line when the grounding resistance is 5KΩ in an embodiment of the present invention; Figure 11(c) is a three-phase diagram of the line when the grounding resistance is 5KΩ in an embodiment of the present invention; Figure 11(d) is a 4-phase diagram of the line when the grounding resistance is 5KΩ in an embodiment of the present invention; Detailed Implementation The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0018] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0019] Example 1 This embodiment discloses a method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction; like Figure 1 , 2 As shown, the method described in this embodiment specifically includes: Step S1: Calculate the zero-sequence current before and after the fault to obtain the change in zero-sequence current of the faulted line.

[0020] Step S1 specifically includes the following: like Figure 3 As shown, an equivalent simulation model consisting of one busbar and four outgoing lines was built, based on a section of a 10kV power supply line in a Chinese mine. In single-phase grounding faults, the line admittance has a negligible impact on the system when studying steady-state characteristics; therefore, the line resistance and inductance can be ignored during steady-state analysis.

[0021] During normal system operation, the zero-sequence current within the system is negligible. However, when a ground fault occurs in a cable, a non-negligible zero-sequence current is generated. When an asymmetrical fault occurs, the voltage and current in each phase of the circuit will change differently. Therefore, when analyzing the zero-sequence component of the system, the symmetry of the three phases must be considered. Thus, the symmetrical component method is used to address system faults. Assuming a ground fault occurs in a phase of a cable, the fault point can be equivalently represented by connecting a grounding resistor between the faulty phase and the ground. R d The equivalent circuit of the faulty circuit is as follows: Figure 4 As shown.

[0022] Based on Kirchhoff's current quantification, the zero-sequence current of the faulted line is obtained as follows: (1) In the formula, , , , Zero-sequence voltage, For each phase voltage, It is the zero-sequence current. Based on symmetry, we can obtain... To simplify calculations, let And the sum of the three-phase voltages is always zero, that is Substituting the above conditions into equation (2) and simplifying, we get: (2) When no ground fault occurs, there is no ground resistance in the equivalent circuit. Similarly, according to Kirchhoff's current law, the zero-sequence current when no fault occurs is: (3) in This refers to the zero-sequence current of a non-faulty line when no fault has occurred. This represents the zero-sequence voltage of the non-faulty line before the fault occurred.

[0023] The change in zero-sequence current of the faulted line is obtained by subtracting the zero-sequence current before and after the fault: (4) Similarly, the change in zero-sequence current of a non-faulty circuit is: (5) The derived zero-sequence current changes between faulty and non-faulty lines show differences in both amplitude and phase. In practical engineering, when a high-resistance grounding fault occurs in a coal mine power grid, the zero-sequence current change is at the microampere level. Especially in the coal mine environment, the zero-sequence current change, as a tiny signal, is highly susceptible to interference, making signal identification difficult. The Holmes-Duffing oscillator system, however, is highly sensitive to tiny signals and is a useful tool for identifying them and faults. It can distinguish the subtle differences in zero-sequence current changes between faulty and non-faulty lines by outputting phase diagram states, making the differences visible. Therefore, this application uses the zero-sequence current changes before and after each line fault as input, outputs four phase diagrams through the Holmes-Duffing oscillator system, and inputs a suitable bifurcation threshold for the system. p This allows the phase diagrams of faulty and non-faulty lines to be in different chaotic states.

[0024] In existing fault identification processes based on Holmes-Duffing oscillator systems, the selection of bifurcation thresholds largely relies on human experience. Furthermore, after selecting the threshold, external excitation needs to be added to the differential equations to determine the final result. Therefore, existing methods for solving bifurcation thresholds suffer from the following problems: First, the bifurcation threshold needs to be selected through trial and error based on human experience, which is not only inefficient but also makes it difficult to guarantee the accuracy of the threshold. Second, existing fault identification processes require the addition of external excitations after determining the bifurcation threshold, necessitating repeated iterative adjustments, making the process extremely cumbersome.

[0025] Therefore, the bifurcation threshold is solved using the method in step S2: Step S2: The change in zero-sequence current is used as the input to the physical information neural network, and the bifurcation threshold and state sequence are used as the network output. The physical information neural network uses the Holmes-Duffing oscillator system differential equation as the physical constraint.

[0026] Step S2 specifically includes the following: The dynamic equation of the Holmes-Duffing oscillator system is the differential equation shown in equation (6). In this application, the state equation of the Holmes-Duffing oscillator system is embedded into PINN. The PINN algorithm is used, and a fully connected neural network is employed to train the bifurcation threshold of the Holmes-Duffing oscillator system through the neural network.

[0027] (6) Using the zero-sequence current difference between each line as the external excitation of the Holmes-Duffing oscillator system, we obtain equation (7). (7) in, The damping coefficient is... Angular frequency, p For the bifurcation threshold, For input, It is a state sequence.

[0028] During neural network training, the accuracy of the output is ensured by continuously reducing the loss function. In the PINN network of this application, the loss function includes differential equation residual loss. Initial condition loss and data loss During training, the loss function is minimized to ensure more accurate training output values.

[0029] PINN's automatic differentiation is used to calculate... and At the same time, the initial displacement is given. initial velocity Therefore, the residual loss of the differential equation is obtained. for: (8) Initial condition loss for: (9) Data loss for: (10) Therefore, the total loss is: (11) in, This represents the average of the squared residuals over all training samples. The state sequence at time zero. Let be the state sequence at any time t. These are actual measured data. These are the weighting coefficients. The preferred value is 0.6. The preferred value is 0.2, which is used to balance physical constraints and data fitting.

[0030] During training, the Adam optimizer is used to train to minimize Training stops when the loss converges or the maximum number of iterations is reached. After training, the final bifurcation threshold is output. p Simultaneously, through network forward propagation, any ( t, Δ I The state sequence corresponding to 0) .

[0031] Once the neural network is trained, it possesses the ability to accurately map the dynamics of complex systems, directly and quickly outputting the bifurcation threshold. This method effectively solves the problem that the bifurcation threshold in complex nonlinear systems is difficult or even impossible to obtain analytically. Compared with traditional fixed-step-size parameter scanning simulation methods, which rely on massive amounts of computationally expensive iterative simulations, resulting in low efficiency and accuracy limited by the step size selection, the neural network, once trained, provides near-instantaneous evaluation, improving efficiency and, thanks to its powerful nonlinear fitting capabilities, offering more accurate bifurcation threshold predictions.

[0032] Although step S2 provides a more detailed method for predicting the bifurcation threshold, in fault identification based on the Holmes-Duffing oscillator system, after determining the accurate bifurcation threshold, the final phase diagram is still needed to identify the fault point. However, in practical applications, subsequent fault line selection still relies on manual observation of the phase diagram to determine whether the system is in a chaotic or periodic state. When the system is operating near the bifurcation threshold, the phase trajectory often exhibits a weakly chaotic or quasi-periodic pattern, and the boundary between chaos and periodicity is extremely blurred. Relying solely on manual judgment makes it difficult to make accurate judgments, and it is very easy to confuse the state of faulty lines with that of non-faulty lines, leading to misjudgments in line selection.

[0033] The Lyapunov exponents are average exponents about the evolution of linearized perturbations, so changes in sign are often related to dynamical bifurcation. When system parameters reach critical values, they can trigger changes in stability. There is a set of Lyapunov exponents for high-dimensional systems. The maximum Lyapunov index (MLE) is . This indicates that two very close initial conditions separate exponentially over time, and the system is highly sensitive to initial conditions (sensitive to initial values), which is one of the key characteristics of chaotic dynamics. This indicates that the error neither grows exponentially nor decays exponentially (which usually corresponds to the phase shift in the neutral direction or periodic orbit of a conservative system). This indicates that the error is exponentially contracting, and the orbit is attracted to a stable equilibrium point or a stable periodic orbit (attractor), making it insensitive to initial conditions.

[0034] The criterion for determining chaotic states using the Lyapunov exponent property is: the existence of a positive Lyapunov exponent indicates that the system is chaotic; in other words, a system is chaotic if its maximum Lyapunov exponent is greater than zero. Therefore, by solving for the maximum Lyapunov exponent, we can identify the output phase diagram state of the Holmes-Duffing oscillator system under different inputs. This also serves as a constraint on the PINN algorithm, further ensuring the accuracy of the PINN network output.

[0035] Furthermore, when calculating the maximum Lyapunov exponent, if the bifurcation threshold and state sequence obtained by PINN are used directly to calculate the maximum Lyapunov exponent, in chaotic states, the calculation of the perturbation vector will grow exponentially over time, resulting in numerical overflow. This causes the calculation results to deviate from the true dynamic characteristics, ultimately making normal calculation impossible.

[0036] Therefore, the method in step S3 is used to calculate the maximum Lyapunov exponent: Step S3: Use the Benettin algorithm to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and determine the chaotic state of the system based on the positive and negative characteristics of the maximum Lyapunov exponent.

[0037] Step S3 specifically includes the following: Step S3.1: Transform the Holmes-Duffing oscillator system into a first-order state space: Let y1 = x in step S2 Then the system is transformed into a first-order state-space form: (12) Abbreviated as ,in .

[0038] Step S3.2, construct the linearized variational equation: Regarding the status Solving the Jacobian matrix ; (13) Let the disturbance vector be... The evolution satisfies the linear differential equation: (14) Right now: (15) Step S3.3, Build the extended system: Combining the original system with the variational equations yields a four-dimensional extended system: (16) Set initial conditions ,For example The perturbation vector is initialized to a unit vector, such as... .

[0039] Step S3.4, calculate the maximum Lyapunov exponent based on the Benettin method: Step S3.4.1, set parameters: Total time is T total Normalization time interval τ: For periodically driven systems, it is usually taken as one driving cycle τ = 2π / ω; Transient discard time T transient It is used to eliminate the effects of initial transients.

[0040] Numerical integration employs the fourth-order Runge-Kutta method, with a relative tolerance set to... The absolute tolerance is set to The maximum step size is set as the signal sampling interval.

[0041] Step S3.4.2, Initialization: Current time State vector ; Perturbation vector Normalization; instantaneous logarithmic amplification factor =0; Effective cumulative step counter k =0.

[0042] Step S3.4.3, cyclic integration until... T total : Use a numerical integrator from the current time t arrive t+τ The integral extension yields: (17) (18) in It is the system's state vector over a time period τ. It is the perturbation vector that has not been normalized after integration; Calculate the instantaneous logarithmic amplification factor: (19) if (That is, the transient period has passed), then: S = + L i (20) In the formula, S is the cumulative logarithmic amplification factor. This is the cumulative logarithmic scaling factor before this accumulation; Scale the integrated, unnormalized perturbation vector back to unit length:

[0043] Update status: (twenty two) Preferably, the instantaneous MLE estimate at the current time is stored: (twenty three) Step S3.4.4, calculate the maximum Lyapunov exponent: Total effective integration time after completing all loops The maximum Lyapunov exponent is estimated as follows: (twenty four) Step S3.4.5, Result Determination: like The system is in a chaotic state (adjacent trajectories are exponentially separated).

[0044] like The system is in a periodic or quasi-periodic state (the trajectory is bounded and insensitive to initial conditions).

[0045] In the process of solving the bifurcation threshold in step S2, since the loss function of the PINN network is a weighted sum of the differential equation residual loss, initial condition loss and data loss, it will have multiple local minima. During the model training process, it will easily get stuck in local optima and fail to converge to the true bifurcation threshold.

[0046] Therefore, the output of the PINN network is reverse-verified and corrected using the method in step S4: Step S4: Verify and correct the bifurcation threshold of the physical information neural network output by reverse verification of the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

[0047] Step S4 specifically includes the following: The Lyapunov exponential property is used as a criterion for judging chaos. The chaotic state of a system can be determined by the relationship between the exponent value and zero (i.e., its positive or negative characteristic). By calculating the exponent of the response input, the system state under that input can be determined. The bifurcation threshold under that input can be obtained using PINN training. p The calculated system state is compared with the system state determined by the exponent. If they are inconsistent, it indicates that the P-value output by PINN training is not accurate enough, and PINN is retrained and iterated continuously. p This continues until the PINN output state matches the exponential judgment state, ensuring that the fault selection result matches the actual simulation, thus playing a role in reverse verification and correction.

[0048] Case Analysis Simulation model building in MATLAB Figure 3 The corresponding simulation model uses one busbar and four outgoing lines (four cables) to collect training data through fault point simulation modeling. The Three-Phase Fault module is used in the simulation model to set the grounding resistance, the faulty line, and the fault occurrence time. In the simulation, the grounding resistance is set to 90 ohms. KΩ 50 KΩ 25 KΩ 5 KΩ The simulation was uniformly set to cause a ground fault on the first line, and the fault time was set to 0.1s.

[0049] Zero-sequence currents were collected for the four lines of the model after a fault occurred under four different grounding resistances. To reduce the impact of the fault inrush current on the data, the zero-sequence current data was processed after collection. The zero-sequence current of the two periods before the fault was subtracted from the zero-sequence current of the two periods after the fault to obtain the final data, i.e., the zero-sequence current difference of each line, and a dataset was established. The simulation frequency was 50Hz. The data before the fault was taken from 0.07s to 0.09s, and the data after the fault was taken from 0.11s to 0.13s. Figure 5 shows the waveform of the zero-sequence current difference. Figure 5(a) shows the data when the grounding resistance is 90 ohms. KΩ The data waveform diagram, Figure 5(b) shows a grounding resistance of 50 Ω. KΩ The data waveform diagram.

[0050] The dataset is used as input to PINN. The parameter settings in PINN are shown in Table 1. Table 1 PINN Training Parameters

[0051] Equation (7) is used as a physical constraint and embedded in the PINN network. Simultaneously, the loss function of the neural network is constructed based on the state equation of the Holmes-Duffing oscillator system. A fully connected neural network is used, and the Adam optimizer is employed for algorithm optimization. First, Adam (with an initial learning rate of 1e) is used... -3 Training 5e 4 -1e 5 To achieve initial convergence, L-BFGS is then used to refine the total loss, and PINN is trained until the loss converges, yielding the output result: the chaos threshold of the Holmes-Duffing oscillator system. p =0.68258191532, Figure 6 This is a convergence plot of the loss function. To clearly illustrate the convergence process, the plot only shows the results from the first 50 rounds. Figure 7 Predict x(t) for PINN.

[0052] Training PINN to obtain the threshold p Using the trained network, forward inference is performed on the required time series to obtain x(t) and its derivative. The results of the trained neural network and the zero-sequence current difference are used as inputs to the Benettin algorithm. The maximum Lyapunov exponent is calculated by numerically tracking the exponential growth of the initial disturbance. The estimated maximum Lyapunov exponent (MLE) is obtained through Equation 22. The maximum Lyapunov exponent (MLE) is compared with 0 to determine the phase diagram state of the Holmes-Duffing oscillator system under the corresponding zero-sequence current difference output. The phase diagram state is compared with the actual phase diagram state to correct the chaotic threshold P output by the trained PINN neural network.

[0053] To investigate the accuracy and universality of the parameter thresholds identified by the PINN neural network, the grounding resistance was set to 90Ω in the simulation. KΩ 50 KΩ 25 KΩ 5 KΩ The corresponding zero-sequence current difference data were collected, and the differential equations of the Holmes-Duffing oscillator system with external drive were solved using the variable step-size Runge-Kutta(4,5) method. The phase diagram state was then plotted to visualize the state of the oscillator system. Figures 8(a)-(d), 9(a)-(d), 10(a)-(d), and 11(a)-(d) show the phase diagram states of the four lines under four different grounding resistances.

[0054] The simulation sets line one as the faulty line. Analysis of the phase diagrams generated under four different grounding resistances shows that line one exhibits a large-cycle state under all four grounding resistances, while the other three lines show a chaotic state. At a grounding resistance of [missing information], [missing information]... , , At that time, the chaos level of the phase diagram state of line 2 is very small, but compared with the large-cycle state of line 1, it is still a chaotic state. The grounding resistance is... At that time, line one was in a large-cycle state, while the other three lines exhibited less chaos, but were still in a chaotic state. Overall, under the four grounding resistances, the phase diagram of the faulty line was the opposite of that of the non-faulty line, verifying the bifurcation threshold. p The accuracy of the method was verified, and the phase diagram states under these four grounding resistance conditions were analyzed, thus confirming the universality of the method.

[0055] Table 2 shows the positive and negative characteristics of the maximum Lyapunov exponent (MLE) under various grounding impedances calculated using the trained PINN. Based on the Lyapunov characteristic exponent standard for determining chaos, it is only necessary to determine the relationship between the maximum Lyapunov exponent (MLE) and zero. When the MLE exhibits a positive characteristic, it corresponds to a chaotic state; when the MLE exhibits a negative characteristic, it corresponds to a periodic state. After multiple iterations using the Benettin algorithm, the MLE characteristic results in Table 2 were obtained. Analysis of the MLE characteristics shows that the MLE of line one is negative. The MLEs of lines two, three, and four are all positive, indicating that line one is in a periodic state, while the other three lines are in a chaotic state. This is consistent with the phase diagram state of the actual Holmes-Duffing oscillator system, indicating that line one is a faulty line. The others are non-faulty lines, further verifying the bifurcation threshold identified by PINN. p The accuracy.

[0056] Table 2. MLE characteristics under various grounding impedances.

[0057] Example 2 This embodiment discloses a coal mine high-resistivity grounding fault identification system based on PINN and MLE correction, including a data processing module, a model building module, a maximum Lyapunov exponent solving module, and a correction identification module; The data processing module is configured to: calculate the zero-sequence current before and after the fault, and obtain the change in zero-sequence current of the faulted line. The model building module is configured to take the zero-sequence current change as the input of the physical information neural network, and take the bifurcation threshold and state sequence as the network output. The physical information neural network takes the Holmes-Duffing oscillator system differential equation as the physical constraint. The maximum Lyapunov exponent calculation module is configured to: use the Benettin algorithm to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and determine the chaotic state of the system based on the positive and negative characteristics of the maximum Lyapunov exponent; The correction identification module is configured to: reverse-verify and correct the bifurcation threshold of the physical information neural network output by the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

[0058] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a coal mine high-resistivity grounding fault identification method based on PINN and MLE correction as described in Embodiment 1 of this disclosure.

[0059] Example 4 The chip system includes a processor for steps in a coal mine high-resistance grounding fault identification method based on PINN and MLE correction as described in Embodiment 1 of this disclosure. In one possible design, the chip system may further include a memory for storing program instructions and data necessary for inter-process communication. The chip system may be composed of chips or may include chips and other discrete devices.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE corrections, characterized in that, Calculate the zero-sequence current before and after the fault to obtain the change in zero-sequence current of the faulted line. The zero-sequence current change is used as the input of the physical information neural network, and the bifurcation threshold and state sequence are used as the network output. The physical information neural network uses the Holmes-Duffing oscillator system differential equation as the physical constraint. The Benettin algorithm is used to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and the chaotic state of the system is determined based on the positive and negative characteristics of the maximum Lyapunov exponent. The bifurcation threshold of the physical information neural network output is verified and corrected by the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

2. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 1, characterized in that, The differential equations of the Holmes-Duffing oscillator system are as follows: In the formula, The damping coefficient is... Angular frequency, p For the bifurcation threshold, For input, It is a state sequence.

3. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 1, characterized in that, The loss function of the physical information neural network is composed of a weighted sum of the differential equation residual loss term, the initial condition loss term, and the data loss term.

4. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 3, characterized in that, The residual loss term of the differential equation is obtained by substituting the derivative obtained by automatic differentiation into the differential equation of the Holmes-Duffing oscillator system to calculate the residual; the initial condition loss term consists of the deviation between the network output at the initial moment and the given initial displacement and initial velocity; the data loss term consists of the deviation between the network output and the measured data.

5. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 1, characterized in that, The Holmes-Duffing oscillator system is transformed into a first-order state space, the Jacobian matrix is ​​solved, and a four-dimensional extended system is constructed. The fourth-order Runge-Kutta method was used for numerical integration. The perturbation vector was normalized at a preset renormalization interval and the logarithmic amplification factor was recorded. The maximum Lyapunov exponent estimate is obtained by averaging the logarithmic magnification factor over the effective integration time.

6. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 1, characterized in that, Using the positive and negative characteristics of the maximum Lyapunov exponent as a condition, if the chaotic state of each line determined based on the current bifurcation threshold is different from that of the known faulty line, the parameters in the loss function of the physical information neural network are adjusted and retrained. If the chaotic state of each line determined based on the current bifurcation threshold is the same as that of a known faulty line, then the bifurcation threshold is output, and the faulty line selection is completed based on this threshold.

7. The method for identifying high-resistivity grounding faults in coal mines based on PINN and MLE correction as described in claim 1, characterized in that, If the maximum Lyapunov exponent is greater than zero, the system is determined to be in a chaotic state; if the maximum Lyapunov exponent is less than or equal to zero, the system is determined to be in a periodic or quasi-periodic state.

8. A coal mine high-resistance grounding fault identification system based on PINN and MLE correction, characterized in that, It includes a data processing module, a model building module, a maximum Lyapunov exponent solving module, and a correction and identification module; The data processing module is configured to: calculate the zero-sequence current before and after the fault, and obtain the change in zero-sequence current of the faulted line. The model building module is configured to take the zero-sequence current change as the input of the physical information neural network, and take the bifurcation threshold and state sequence as the network output. The physical information neural network takes the Holmes-Duffing oscillator system differential equation as the physical constraint. The maximum Lyapunov exponent calculation module is configured to: use the Benettin algorithm to calculate the maximum Lyapunov exponent of the oscillator system state corresponding to each line, and determine the chaotic state of the system based on the positive and negative characteristics of the maximum Lyapunov exponent; The correction identification module is configured to: reverse-verify and correct the bifurcation threshold of the physical information neural network output by the maximum Lyapunov exponent until the chaotic state determination result is consistent with the actual fault condition of the line.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the coal mine high-resistance grounding fault identification method based on PINN and MLE correction as described in any one of claims 1-7.

10. A chip system, characterized in that, It includes one or more processors, which are invoked to perform the steps in a coal mine high-resistivity grounding fault identification method based on PINN and MLE correction as described in any one of claims 1-7.

Images