A method for detecting electric shock current of living organisms based on long short-term memory neural network

Abstract

The present invention relates to a method for detecting electric shock current of a living body based on a long short-term memory neural network, comprising the following steps: a) building a living body electric shock test platform, and collecting residual current waveform data during the test; b) using a wavelet decomposition method to perform noise reduction processing on the original signal of the residual current waveform data collected in step a); c) randomly dividing the signal processed by step b) into a training set and a test set according to a ratio of 3:1, and using the training set data as the input of the long short-term memory neural network model to train the long short-term memory neural network model; and using the trained long short-term memory neural network model to test the data of the test set. Compared with CNN and BP networks, the long short-term memory neural network (LSTM) of the present invention has obvious advantages in the number of training times and average calculation time.

Description

Technical Field

[0001] The invention relates to a steel structure manufacturing line device, and more specifically, to a method for detecting electric shock current of a living body based on a long short-term memory neural network. Background Art

[0002] With the rapid development of society, the scale of power equipment and power lines continues to expand, and the resulting fire and electric shock accidents often occur. Therefore, the safety of power equipment has become one of the hot issues focused on by the power industry. At present, residual current protection equipment and leakage current detection technology have made a lot of breakthrough progress, but it has not fundamentally solved the problem of low correct commissioning rate of residual current protection equipment. At present, most residual current protectors mainly judge the occurrence of electric shock accidents by the size of the effective value of the total residual current, lack the ability to distinguish the type of electric shock, and may cause false operation when the electrical environment of the residual current protection device is affected. In addition, in some special occasions, the leakage current of the power grid may reduce the total residual current, causing the residual current protection device to refuse to operate. To solve these problems and prevent the occurrence of electric shock and leakage accidents, it is necessary to develop residual current protection devices from the two aspects of residual current and electric shock current of living organisms. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a method for detecting electric shock current of living organisms based on long short-term memory neural network. This method provides a reference for researching and developing a new generation of residual current protection devices for electric shock of living organisms, which has important practical significance for electricity safety.

[0004] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0005] The method for detecting electric shock current of a living being based on a long short-term memory neural network comprises the following steps:

[0006] a) Build a life form electric shock test platform, simulate life form electric shock under different conditions by controlling current, humidity and contact area, and collect residual current waveform data during the test;

[0007] b) Using the wavelet decomposition method to perform noise reduction on the original signal of the residual current waveform data collected in step a), the basic steps of the wavelet decomposition method mainly include the following parts:

[0008] 1) Perform wavelet decomposition on the original signal and calculate the wavelet decomposition coefficients;

[0009] 2) Threshold quantization of wavelet decomposition coefficients to discard the noise components concentrated in high-frequency coefficients;

[0010] 3) Reconstruct the low-frequency coefficients of wavelet decomposition and the high-frequency coefficients of each layer after threshold processing;

[0011] Assume that the collected electric shock current waveform signal of the living body is a one-dimensional discrete signal:

[0012] f(n)=s(n)+ε(n) (1)

[0013] Where s(n) is the original signal, ε(n) is the noise signal, and it obeys N(0,σ 2 )distributed;

[0014] The collected one-dimensional electrical signal f(n) is decomposed by wavelet using Mallat's pyramid algorithm, and the scale coefficient f k J＝1 and wavelet coefficient d k J+1 They are:

[0015]

[0016] Where J is the number of decomposition layers, h and g are the low-pass and high-pass filters corresponding to the scaling function and wavelet function, respectively, N is the length of discrete sampling data, n = 0, 1, ..., N-1, k = 0, 1, ..., N-1;

[0017] The wavelet coefficients d obtained by decomposition k J , soft threshold processing is adopted to obtain the estimated wavelet coefficient D k J for:

[0018]

[0019] The threshold of the soft clipping function is

[0020] The wavelet coefficient D after threshold processing k J Reconstruct and recover the original signal estimate:

[0021]

[0022] Where h′ and g′ are reconstruction coefficients, which are the conjugate transpose of h and g. [5] ;

[0023] Through the above wavelet decomposition and noise reduction process, the original signal of most residual current waveform data can be processed for noise reduction, thereby removing noise interference;

[0024] c) randomly divide the signal processed by step b) into a training set and a test set in a ratio of 3:1, and use the training set data as the input of the long short-term memory neural network model to train the long short-term memory neural network model; during the training process, the error limit is set to 1×10-7, the optimizer selects Adam, and the Dropout algorithm is introduced to regularize the long short-term memory neural network model, and the trained long short-term memory neural network model is used to test the data of the test set, and the long short-term memory neural network is based on the recurrent neural network and adds control gates, namely a forget gate, an input gate, and an output gate.

[0025] As a preferred solution: In the long short-term memory neural network, define x t is the input at time t, h t is the hidden layer output at time t, f t is the forget gate at time t, i t is the input gate at time t, o t is the output gate at time t, s t is the cell state at time t, c t is the input state of the memory unit, σ is the Sigmoid function, and tanh is the hyperbolic tangent function; the long short-term memory neural network uses the forget gate to determine and delete the information that the neuron does not need. The forget gate f t The output value is calculated as:

[0026] f t =σ(W xf x t +W hf h t-1 +b f ) (5)

[0027] Where W xf , W hf is the weight coefficient of the forget gate, b f is the forget gate offset;

[0028] Based on the selection of neurons in the long short-term memory neural network structure through the forget gate, ti and h t-1 The product of the weight controls the neuron state i in the input gate t , and its calculation formula is:

[0029] i t =σ(W xi x ti +W xi h t-1 +b i ) (6)

[0030] Where W xi and Whi is the input gate weight coefficient, b i is the input gate offset;

[0031] Then use the tanh function to further update the neuron information, and then get the neuron state C of the memory unit at time t t , and its calculation formula is as follows:

[0032]

[0033]

[0034] Where W xc and W hc is the cell state weight coefficient, b c is the cell state offset;

[0035] The final output result is obtained by using the Sigmoid function to determine whether the neuron state is output. t , and its calculation formula is as follows:

[0036] o t =σ(W xo x t +W ho h t-1 +b0) (9)

[0037] Where W xo and W ho is the weight coefficient of the output gate, b o is the offset of the output gate;

[0038] The calculation formula of the hidden layer output at time t is:

[0039] h t =o t tanh(C t ) (10)

[0040] As a preferred solution: the living body electric shock test platform includes a power supply, a test object, a grounding plate, a power-on tool, an oscilloscope A, a current-limiting resistor, an auxiliary switch and a load which are connected in series to form a circuit; an oscilloscope B is also connected in parallel to the test object and the grounding plate.

[0041] As a preferred solution: the living organism electric shock test platform adopts a three-phase power supply and adds a load to simulate the load in the low-voltage power distribution system. The C-phase circuit is used to simulate an electric shock accident. The current limiting resistor is used to control the current size, and the oscilloscope is used to record the voltage and current passing through the living organism to be tested during the test.

[0042] As a preferred solution: the living body electric shock test platform obtains 400 groups of residual current waveform data, 100 groups of residual current waveform data with current amplitudes of 30mA, 100mA, 300mA, and 500mA respectively.

[0043] As a preferred solution: in the step b), the wavelet mother function of the filter selected for wavelet decomposition and denoising is db3 wavelet, and the parameters are set to J=3, σ 2 =0.67, t=4.772.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention proposes a method for detecting electric shock current of living organisms based on long short-term memory neural network. The residual current signal of electric shock is obtained through the electric shock test platform of living organisms, and then the residual current signal of electric shock is decomposed and denoised by wavelet, and then the special waveform current signal shown by the total residual current is used, and the long short-term memory neural network (LSTM) is trained in combination with its time domain characteristics and waveform characteristics to construct a model for electric shock recognition of living organisms and achieve good detection results. In addition, compared with CNN and BP networks, the long short-term memory neural network (LSTM) has obvious advantages in the number of training times and average calculation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings in the specification, which constitute a part of the present application, are used to provide a further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute a limitation on the present application.

[0047] Figure 1 It is a structural schematic diagram of the electric shock test platform of the present invention;

[0048] Figure 2 It is a structural diagram of the long short-term memory neural network of the present invention;

[0049] Figure 3 This is the residual current waveform diagram under normal conditions of the present invention;

[0050] Figure 4 This is a residual current waveform diagram when electric shock occurs in the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0052] The method for detecting electric shock current of a living being based on a long short-term memory neural network comprises the following steps:

[0053] Step a), building a living body electric shock test platform, the living body electric shock test platform uses the animal body as the electric shock test object, simulates the electric shock of the living body under different conditions by controlling the current, humidity, and contact area, and collects the residual current waveform data during the test; Figure 1 As shown, the living body electric shock test platform includes a power supply, a test object, a grounding plate, a power-on tool, an oscilloscope A, a current-limiting resistor, an auxiliary switch and a load which are connected in series to form a loop; an oscilloscope B is also connected in parallel to the test object and the grounding plate; the living body electric shock test platform adopts a three-phase power supply, and a load is added to simulate the load in the low-voltage power distribution system, the C-phase circuit is taken to simulate the electric shock accident, the current-limiting resistor is used to control the current size, and the oscilloscope is used to record the voltage and current passing through the tested life in the test; the living body electric shock test platform obtains 400 groups of residual current waveform data, 100 groups of residual current waveform data with current amplitudes of 30mA, 100mA, 300mA, and 500mA respectively, and the residual current waveform diagrams under normal conditions and the residual current waveform diagrams when electric shock occurs are shown as follows respectively. Figure 3 and Figure 4 shown.

[0054] Step b), using a wavelet decomposition method to perform noise reduction processing on the original signal of the residual current waveform data collected in step a), the basic steps of the wavelet decomposition method mainly include the following parts:

[0055] 1) Perform wavelet decomposition on the original signal and calculate the wavelet decomposition coefficients;

[0056] 2) Threshold quantization of wavelet decomposition coefficients to discard the noise components concentrated in high-frequency coefficients;

[0057] 3) Reconstruct the low-frequency coefficients of wavelet decomposition and the high-frequency coefficients of each layer after threshold processing;

[0058] Assume that the collected electric shock current waveform signal of the living body is a one-dimensional discrete signal:

[0059] f(n)=s(n)+ε(n) (1)

[0060] Where s(n) is the original signal, ε(n) is the noise signal, and it obeys N(0,σ 2 )distributed;

[0061] The collected one-dimensional electrical signal f(n) is decomposed by wavelet using Mallat's pyramid algorithm, and the scale coefficient f k J＝1 and wavelet coefficient d k J+1 They are:

[0062]

[0063] Where J is the number of decomposition layers, h and g are the low-pass and high-pass filters corresponding to the scaling function and wavelet function, respectively, N is the length of discrete sampling data, n = 0, 1, ..., N-1, k = 0, 1, ..., N-1;

[0064] The wavelet coefficients d obtained by decomposition k J , soft threshold processing is adopted to obtain the estimated wavelet coefficient D k J for:

[0065]

[0066] The threshold of the soft clipping function is

[0067] The wavelet coefficient D after threshold processing k J Reconstruct and recover the original signal estimate:

[0068]

[0069] Where h′, g ′ is the reconstruction coefficient, which is the conjugate transpose of h and g [5] ;

[0070] The above wavelet decomposition and denoising process can reduce the noise of most of the original signals of the residual current waveform data, thereby removing the noise interference. Among them, the wavelet mother function of the filter selected for wavelet decomposition and denoising is db3 wavelet, and the parameters are set to J=3, σ 2 =0.67, t=4.772.

[0071] The residual current signal recorded by the actual residual current protection device inevitably has noise interference, so it is necessary to reduce the noise of the original signal. Wavelet decomposition has the multi-resolution property that can gradually observe the signal from coarse to fine, which has a good effect on characterizing the non-stationary characteristics of the signal and can retain the real electrical signal characteristics of the special waveform of electric shock to the greatest extent.

[0072] A recurrent neural network (RNN) is a recursive neural network that takes a sequence as input. It has obvious advantages over traditional neural networks in dealing with conventional sequence problems. However, its disadvantages are also obvious. When the number of recursive layers is large, it is easy to cause gradient explosion and vanishing problems, which will seriously affect the accuracy of the trained model. Therefore, a long and short-term memory neural network (LSTM) is proposed based on the traditional recurrent neural network. As an improved recurrent neural network, the LSTM neural network can adaptively capture time series data with long-distance dependencies and nonlinear dynamic changes. The overall structure of the LSTM neural network is similar to that of the RNN, but the calculation method in the hidden layer of the LSTM neural network is different. Control gates are added to the LSTM neural network, namely the forget gate, input gate, and output gate. The main function of the forget gate is to enable the LSTM network to forget and delete some irrelevant information, thus solving the long-term dependency problem of the traditional recurrent neural network. The structure of the LSTM neural network is as follows: Figure 2 shown.

[0073] x in the figure t is the input at time t, h t is the hidden layer output at time t, f t is the forget gate at time t, i t is the input gate at time t, o t is the output gate at time t, s t is the cell state at time t, c t is the input state of the memory unit, σ is the Sigmoid function, and tanh is the hyperbolic tangent function; the long short-term memory neural network uses the forget gate to determine and delete the information that the neuron does not need. The forget gate f t The output value is calculated as:

[0074] f t =σ(W xf x t +W hf h t-1 +b f ) (5)

[0075] Where W xf , W hf is the weight coefficient of the forget gate, b f is the forget gate offset;

[0076] Based on the selection of neurons in the long short-term memory neural network structure through the forget gate, ti and h t-1The product of the weight controls the neuron state i in the input gate t , and its calculation formula is:

[0077] i t =σ(W xi x ti +W xi h t-1 +b i ) (6)

[0078] Where W xi and W hi is the input gate weight coefficient, b i is the input gate offset;

[0079] Then use the tanh function to further update the neuron information, and then get the neuron state C of the memory unit at time t t , and its calculation formula is as follows:

[0080]

[0081]

[0082] Where W xc and W hc is the cell state weight coefficient, b c is the cell state offset;

[0083] The final output result is obtained by using the Sigmoid function to determine whether the neuron state is output. t , and its calculation formula is as follows:

[0084] o t =σ(W xo x t +W ho h t-1 +b0) (9)

[0085] Where W xo and W ho is the weight coefficient of the output gate, b o is the offset of the output gate;

[0086] The calculation formula of the hidden layer output at time t is:

[0087] h t =o t tanh(C t ) (10)

[0088] In order to improve the performance of the model to a large extent, the network parameters of the number of layers and nodes of the LSTM hidden layer are tuned. Therefore, two sets of experimental tests with restrictive conditions are set. In the first set of experiments, in order to determine the optimal number of hidden layers, the number of nodes in each layer is set to a fixed value of 32, and other conditions remain unchanged. By setting the number of hidden layers from 1 to 4, the MAE value is observed, and the final number of hidden layers is determined by MAE. The experimental results are shown in Table 1. The same method is used to determine the number of hidden layer nodes. The hidden layer is set to a fixed value, and other conditions remain unchanged. The number of hidden layer nodes is set to 8, 16, 32, and 64, and the MAE value is observed. The number of hidden layer nodes is determined by MAE. The experimental results are shown in Table 2.

[0089] Table 1. MAE table of hidden layer number

[0090] Current number of hidden layers Number of hidden layer nodes MAE 1 64 0.1873 2 64 0.1965 3 64 0.2133 4 64 0.2371

[0091] Table 2 MAE table of hidden layer node number

[0092] Number of hidden layers Current number of hidden layer nodes MAE 1 8 0.2384 1 16 0.2254 1 32 0.2176 1 64 0.2589

[0093] Generally, the larger the number of hidden layers in a model, the more information can be obtained, and the higher the accuracy of the trained model. However, as the number of layers increases, the probability of overfitting and other problems also increases. From Table 1, it can be seen that when the number of hidden layers is 1, the MAE error evaluation is the smallest, so the number of hidden layers is determined to be 1. Similarly, the number of nodes will also affect the learning effect during training. From Table 2, it can be seen that as the number of nodes increases, the MAE continues to decrease, but when it exceeds a certain value, the training effect of the model will be reduced. When the number of hidden layer nodes is 32, the MAE is the smallest, so the number of hidden layer nodes is determined to be 32.

[0094] Step c), randomly divide the signal processed by step b) into a training set and a test set in a ratio of 3:1, and use the training set data as the input of the long short-term memory neural network model to train the long short-term memory neural network model; during the training process, the error limit is set to 1×10-7, and the optimizer selects Adam. The Adam optimizer can adaptively adjust the learning rate, greatly improving the speed of training and the stability of the training process. In addition, the Dropout algorithm is introduced. The main function of Dropout is to regularize the model, which can effectively prevent overfitting. During the model training process, Dropout will randomly block some neurons, so new neuron results will be formed and the neurons are directly independent of each other, which effectively solves the problem of local feature dependence and further improves the stability of the neural network.

[0095] The trained LSTM network was used to test 100 sets of test data. The results are shown in Table 3. The total mean square error is 3.3238×10-6, the average simulation time is 0.000034s, and the accuracy is 100%, which meets the actual engineering needs.

[0096] Table 3 Electric shock identification results

[0097] Serial number Expected Output Actual Output Mean square error 1 1 0.9154 <![CDATA[2.3401×10 -6 ]]> 2 1 0.9614 <![CDATA[5.3477×10 -7 ]]> 3 1 0.9912 <![CDATA[3.1564×10 -6 ]]> 4 1 0.9725 <![CDATA[1.2512×10 -6 ]]> 5 1 0.9851 <![CDATA[4.2511×10 -6 ]]> 6 1 0.9521 <![CDATA[2.7845×10 -7 ]]> 7 1 0.9754 <![CDATA[9.3574×10 -7 ]]> 8 1 0.9565 <![CDATA[3.9514×10 -6 ]]> 9 1 0.9921 <![CDATA[5.5484×10 -6 ]]> 10 1 0.9789 <![CDATA[1.3894×10 -6 ]]> … … … … 100 1 0.9920 <![CDATA[4.8924×10 -6 ]]>

[0098] In order to verify the effectiveness of the LSTM-based method for detecting electric shock of living organisms, CNN and BP neural networks were used to compare with the method proposed in this paper.

[0099] Table 4 Comparison of detection effects of different methods

[0100]

[0101]

[0102] As can be seen from Table 4, the proposed method has a higher accuracy and fewer iterations, indicating that this method has a better characterization ability for electric shock of living organisms. In addition, it is also better than CNN and BP neural networks in terms of average calculation time, with an average recognition time of only 0.000034.

[0103] The present invention proposes a method for detecting electric shock current of living organisms based on long short-term memory neural network. The residual current signal of electric shock is obtained through the electric shock test platform of living organisms, and then the residual current signal of electric shock is decomposed and denoised by wavelet. Then, the special waveform current signal shown by the total residual current is used, and the LSTM is trained in combination with its time domain characteristics and waveform characteristics to construct a model for electric shock recognition of living organisms and achieve good detection results. In addition, compared with CNN and BP networks, LSTM has obvious advantages in the number of training times and average calculation time.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0105] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for detecting electric shock current of living organisms based on long short-term memory neural network, characterized in that: The steps include: a) Build a human-like electric shock test platform powered by 220V±20% power supply, simulate electric shock of human life under different conditions by controlling current, humidity and contact area, and collect residual current waveform data during the test; b) Using the wavelet decomposition method to perform noise reduction on the original signal of the residual current waveform data collected in step a), the basic steps of the wavelet decomposition method mainly include the following parts: 1) Perform wavelet decomposition on the original signal and calculate the wavelet decomposition coefficients; 2) Threshold quantization of wavelet decomposition coefficients to discard the noise components concentrated in high-frequency coefficients; 3) Reconstruct the low-frequency coefficients of wavelet decomposition and the high-frequency coefficients of each layer after threshold processing; Assume that the collected electric shock current waveform signal of the living body is a one-dimensional discrete signal: f(n)=s(n)+ε(n) (1) Where s(n) is the original signal, ε(n) is the noise signal, and it obeys N(0, σ 2 )distributed; The collected one-dimensional electrical signal f(n) is decomposed by wavelet using Mallat's pyramid algorithm, and the scale coefficient f k J＝1 and wavelet coefficient d k J+1 They are: Where J is the number of decomposition layers, h and g are the low-pass and high-pass filters corresponding to the scaling function and wavelet function, respectively, N is the length of discrete sampling data, n = 0, 1, ..., N-1, k = 0, 1, ..., N-1; The wavelet coefficients d obtained by decomposition k J , soft threshold processing is adopted to obtain the estimated wavelet coefficient D k J for: The threshold of the soft clipping function is The wavelet coefficient D after threshold processing k J Reconstruct and recover the original signal estimate: Where h′ and g′ are reconstruction coefficients, which are the conjugate transpose of h and g; Through the above wavelet decomposition and noise reduction process, the original signal of most residual current waveform data can be processed for noise reduction, thereby removing noise interference; c) The signal processed in step b) is randomly divided into a training set and a test set in a ratio of 3:1, and the training set data is used as the input of the long short-term memory neural network model to train the long short-term memory neural network model; during the training process, the error limit is set to 1×10 -7 , the optimizer selects Adam, and introduces the Dropout algorithm to regularize the long short-term memory neural network model. The trained long short-term memory neural network model is used to test the data of the test set. The long short-term memory neural network is based on the recurrent neural network and adds control gates, namely the forget gate, input gate and output gate; In the long short-term memory neural network, define x t is the input at time t, h t is the hidden layer output at time t, f t is the forget gate at time t, i t is the input gate at time t, o t is the output gate at time t, s t is the cell state at time t, c t is the input state of the memory unit, σ is the Sigmoid function, and tanh is the hyperbolic tangent function; the long short-term memory neural network uses the forget gate to determine and delete the information that the neuron does not need. The forget gate f t The output value is calculated as: f t =σ(W xf x t +W hf h t-1 +b f ) (5) Where W xf , W hf is the weight coefficient of the forget gate, b f is the forget gate offset; Based on the selection of neurons in the long short-term memory neural network structure through the forget gate, ti and h t-1 The product of the weight controls the neuron state i in the input gate t , and its calculation formula is: i t =σ(W xi x ti +W xi h t-1 +b i ) (6) Where W xi and W hi is the input gate weight coefficient, bi is the input gate offset; Then use the tanh function to further update the neuron information, and then get the neuron state C of the memory unit at time t t , and its calculation formula is as follows: Where W xc and W hc is the cell state weight coefficient, b c is the cell state offset; The final output result is obtained by using the Sigmoid function to determine whether the neuron state is output. t , and its calculation formula is as follows: o t =σ(W xo x t +W ho h t-1 +b0) (9) Where W xo and W ho is the weight coefficient of the output gate, b is the offset of the output gate; The calculation formula of the hidden layer output at time t is: h t =o t fishy(C) t ) (10); In the step b), the wavelet mother function of the filter selected for wavelet decomposition and denoising is db3 wavelet, and the parameters are set to J=3, σ 2 =0.67, t=4.

772.

2. The method for detecting electric shock current of a living being based on a long short-term memory neural network according to claim 1, characterized in that: The living body electric shock test platform comprises a power supply, a test object, a grounding plate, a power-on tool, an oscilloscope A, a current-limiting resistor, an auxiliary switch and a load which are connected in series to form a circuit; an oscilloscope B is also connected in parallel to the test object and the grounding plate.

3. The method for detecting electric shock current of a living being based on a long short-term memory neural network according to claim 1, characterized in that: The living organism electric shock test platform adopts a three-phase power supply and adds a load to simulate the load in the low-voltage power distribution system. The C-phase circuit is used to simulate an electric shock accident. The current limiting resistor is used to control the current size, and the oscilloscope is used to record the voltage and current passing through the living organism to be tested during the test.

4. The method for detecting electric shock current of a living being based on a long short-term memory neural network according to claim 1, characterized in that: The living body electric shock test platform obtains 400 sets of residual current waveform data, 100 sets of residual current waveform data with current amplitudes of 30mA, 100mA, 300mA, and 500mA respectively.

Images