A non-contact electric induction and infrared temperature measurement combined electric detection anti-misoperation system

Abstract

The application discloses a non-contact capacitive sensing and infrared temperature measurement fusion electricity testing anti-misoperation system and belongs to the technical field of electricity testing of a power system. The application firstly performs first-order and second-order derivation on a normalized original capacitive signal and takes an absolute value, and calculates a capacitive stable value sequence; then obtains first-order and second-order fluctuation value sequences based on a local window variance of a change rate and a change unstable value, and further obtains a capacitive time sequence internal coordination value sequence; simultaneously extracts a thermal input effective energy from a normalized infrared temperature signal, and generates a thermal input effective energy sequence; subsequently obtains a capacitive-infrared time sequence coordination value sequence in combination with the capacitive stable value and the thermal input effective energy; finally processes the capacitive stable value sequence, the thermal input effective energy sequence, the capacitive time sequence internal coordination value sequence and the capacitive-infrared time sequence coordination value sequence by using an electricity testing classification neural network, and outputs an electricity testing state result. The application effectively improves electricity testing precision and provides reliable guarantee for power operation safety.

Description

Technical Field

[0001] This invention relates to the field of power system voltage detection technology, specifically to a non-contact voltage detection and error prevention system that integrates capacitive sensing and infrared temperature measurement. Background Technology

[0002] In the operation and maintenance of power systems, voltage testing is a crucial preliminary step to ensure personnel safety and stable equipment operation. Its core objective is to accurately distinguish between the actual energized state and the induced energized state of equipment, avoiding safety accidents such as electric shock and short circuits caused by misjudgment. As power systems develop towards intelligence and unmanned operation, non-contact voltage testing technology is gradually replacing traditional contact voltage testing as the mainstream technology due to its advantages such as operational safety, no need to contact equipment conductors, and adaptability to complex operating conditions.

[0003] In existing technologies, non-contact voltage detection systems based on the principle of single-capacitance induction have emerged. These systems collect capacitance signals between the device surface and the detection probe, and use changes in the amplitude of the capacitance signal to determine whether the device is energized. The core logic is as follows: when the detection probe is close to a real energized device, a stable capacitive coupling is formed between the device and the probe, and the capacitance signal amplitude exhibits a smooth change characteristic; when close to an inductively energized device, the capacitive coupling relationship is unstable, and the capacitance signal amplitude fluctuates erratically. By setting an amplitude threshold, the energized state can be determined.

[0004] However, the aforementioned voltage detection system based on single capacitance induction has significant technical drawbacks: in complex power environments, electromagnetic interference around the equipment can cause irregular disturbances in the capacitance signal. Some stable stray electric field disturbances are difficult to distinguish from the stable characteristics of the capacitance signal under actual energized conditions, easily leading to false voltage detection. Therefore, existing technologies suffer from low voltage detection accuracy. Summary of the Invention

[0005] To address the aforementioned shortcomings in the existing technology, the present invention provides a non-contact capacitance sensing and infrared temperature measurement fusion voltage detection and error prevention system, which solves the problem of low voltage detection accuracy in the existing technology.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows: a non-contact capacitance sensing and infrared temperature measurement fusion voltage detection and error prevention system, comprising: a capacitance feature extraction subsystem for performing first-order derivative and taking the absolute value of the normalized original capacitance signal to obtain the rate of change, then performing second-order derivative and taking the absolute value to obtain the unstable value, and taking the ratio of the rate of change to the unstable value at the same moment as the stable value of the capacitance to obtain the sequence of stable values ​​of the capacitance.

[0007] The capacitance collaborative feature extraction subsystem is used to obtain first-order and second-order fluctuation value sequences based on the rate of change and the local window variance of the unstable value, and to obtain the capacitance time series internal collaborative value sequence based on the two sequences.

[0008] The infrared thermal feature quantization subsystem is used to extract the effective thermal input energy from the normalized infrared temperature signal to obtain the effective thermal input energy sequence.

[0009] The capacitor-infrared feature extraction subsystem is used to obtain the capacitor-infrared time-series co-value sequence based on the capacitor stability value and the effective energy of thermal input.

[0010] The classification subsystem is used to process the capacitance stable value sequence, thermal input effective energy sequence, capacitance time series internal coordinating value sequence, and capacitance-infrared time series coordinating value sequence using a voltage detection classification neural network to obtain the voltage detection state result.

[0011] Furthermore, the process of obtaining the first-order and second-order fluctuation value sequences in the capacitance-coordinated feature extraction subsystem includes:

[0012] Obtain the variance of each rate of change within the local window to get the first-order fluctuation value;

[0013] Obtain the variance of each unstable value in the local window to get the second-order fluctuation value;

[0014] Arrange the first-order fluctuation values ​​at each moment in chronological order to obtain the first-order fluctuation value sequence.

[0015] Arrange the second-order fluctuation values ​​at each time point in chronological order to obtain the second-order fluctuation value sequence.

[0016] Furthermore, the process of obtaining the intra-temporal cooperative value sequence of capacitance in the capacitance cooperative feature extraction subsystem includes:

[0017] In the first-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain a local segment of the first-order wave.

[0018] In the second-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain the local segment of the second-order wave.

[0019] Calculate the coefficient of variation of local segments of first-order and second-order waves belonging to the same moment;

[0020] The variational coordination coefficients are normalized to obtain the internal coordination values ​​of the capacitor timing sequence.

[0021] Arrange the capacitor timing internal coordination values ​​at each time point in chronological order to obtain the capacitor timing internal coordination value sequence.

[0022] Furthermore, the formula for calculating the internal coordination value of the capacitor timing is as follows: ,

[0023] in, For the first The internal coordination value of the capacitance timing at time 1. Let be the coefficient of variation at time t. It is a natural constant. This is the proportionality coefficient;

[0024] The formula for calculating the coefficient of variation is: ,

[0025] in, Let be the covariance of the local segment of the first-order wave at time t and the local segment of the second-order wave at time t. Let be the variance of a local segment of the first-order fluctuation at time t. Let be the variance of a local segment of the second-order fluctuation at time t. For a local segment of the first-order wave at time t, Let t be a local segment of the second-order wave at time t.

[0026] Furthermore, the process of obtaining the effective energy sequence of the thermal input in the infrared thermal feature quantization subsystem includes:

[0027] Calculate the first time derivative of the normalized infrared temperature signal and take the non-negative value to obtain the instantaneous heat input value;

[0028] Calculate the second time derivative of the normalized infrared temperature signal and take its absolute value to obtain the thermal input fluctuation value;

[0029] The mean of instantaneous heat input values ​​in a local window is obtained to determine the heat input duration.

[0030] The variance of the heat input fluctuation values ​​in a local window is obtained to determine the heat input stability.

[0031] Calculate the effective energy of the heat input based on the heat input duration and heat input stability;

[0032] Arrange the effective heat input energy at each moment in chronological order to obtain the effective heat input energy sequence.

[0033] Furthermore, the formula for calculating the effective energy of heat input is: ,

[0034] in, The effective energy of the heat input at time t. Let be the duration of heat input at time t. For heat input stability, Denominator parameter.

[0035] Furthermore, the process of obtaining the capacitance-infrared temporal co-value sequence in the capacitance-infrared feature extraction subsystem includes:

[0036] The capacitor stability value is nonlinearly compressed to obtain the capacitor stability compressed value;

[0037] The effective heat input energy is nonlinearly compressed to obtain the effective heat input compression energy;

[0038] Multiplying the stable compression value of the capacitor by the effective compression energy of the thermal input yields the capacitor-infrared timing synergy value;

[0039] The capacitance-infrared time-series coordinated values ​​at each time point are arranged in chronological order to obtain the capacitance-infrared time-series coordinated value sequence.

[0040] Furthermore, the electrophysiological classification neural network includes: a first sequence feature extraction and fusion unit, a second sequence feature extraction and fusion unit, a feature enhancement unit, a fully connected layer, and a Softmax classification layer;

[0041] The first sequence feature extraction and fusion unit is used to extract and fuse features from the capacitance stable value sequence and the thermal input effective energy sequence to obtain the capacitance-infrared multi-branch basic fusion features.

[0042] The second sequence feature extraction and fusion unit is used to extract and fuse features from the capacitor time series internal co-value sequence and the capacitor-infrared time series co-value sequence to obtain the capacitor-infrared multi-branch co-fusion feature.

[0043] The feature enhancement unit is used to enhance the basic fusion feature of the capacitor-infrared multi-branch fusion feature by employing the capacitor-infrared multi-branch collaborative fusion feature to obtain the capacitor-infrared multi-branch enhanced fusion feature.

[0044] The fully connected layer is used to map the capacitance-infrared multi-branch enhancement fusion features to obtain the voltage detection state features;

[0045] The Softmax classification layer is used to classify the voltage detection status features to obtain the voltage detection status results.

[0046] Furthermore, both the first sequence feature extraction and fusion unit and the second sequence feature extraction and fusion unit include: a first feature extraction module, a second feature extraction module, a multiplier, a reshape layer, and a multi-branch feature extraction module;

[0047] The first feature extraction module is used to extract features from the input sequence to obtain the first shallow features;

[0048] The second feature extraction module is used to extract features from the input sequence to obtain the second shallow features;

[0049] The multiplier is used to perform element-wise multiplication of the first shallow feature and the second shallow feature to obtain a one-dimensional shallow fused feature.

[0050] The Reshape layer is used to transform the dimension of one-dimensional shallow fusion features to obtain two-dimensional shallow fusion features.

[0051] The multi-branch feature extraction module is used to extract multi-branch features from two-dimensional shallow fusion features to obtain capacitance-infrared multi-branch basic fusion features or capacitance-infrared multi-branch collaborative fusion features.

[0052] Furthermore, the expression for the feature enhancement unit is: ,

[0053] in, For capacitive-infrared multi-branch enhancement fusion features, This is a fusion feature based on capacitive-infrared multi-branch architecture. It features multi-branch synergistic fusion of capacitive and infrared technologies. For element-wise multiplication, It is a 1×1 convolutional layer. It is a Sigmoid layer.

[0054] The beneficial effects of this invention are as follows:

[0055] 1. This invention, through the cooperation of a capacitance feature extraction subsystem and a capacitance collaborative feature extraction subsystem, not only extracts the basic time-domain features of rate of change and unstable values ​​from the original capacitance signal, but also obtains first-order and second-order fluctuation value sequences through local window variance analysis, and further constructs the internal collaborative value sequence of the capacitance time series. Compared with the existing technology that relies solely on the single feature of capacitance signal amplitude for judgment, the multi-dimensional capacitance time series features extracted by this invention can characterize the dynamic change law and intrinsic correlation of the capacitance signal, increasing the distinguishability between "true charged stationary time series mode" and "pseudo-stationary features superimposed by interference", thus weakening the influence of interference on capacitance detection judgment at the feature level.

[0056] 2. This invention employs an infrared thermal feature quantification subsystem to extract the effective energy sequence of thermal input. Real electrical equipment generates stable and continuous thermal input due to the thermal effect of current, resulting in a regular pattern in the effective energy of the thermal input. In contrast, inductively coupled equipment has no actual current flowing through it and therefore does not generate stable thermal input; its thermal signal is only affected by ambient temperature and exhibits no obvious pattern. Electromagnetic interference only affects the capacitance signal and cannot change the fact that the equipment has a real thermal input. By constructing a cross-modal time-series collaborative value sequence through the capacitance-infrared feature extraction subsystem, dual verification of "capacitance features + infrared features" can be achieved, effectively reducing misjudgments caused by interference with a single capacitance feature and improving the accuracy of voltage detection.

[0057] 3. This invention employs a classification subsystem and a voltage detection classification neural network to fuse the capacitor stable value sequence, the effective energy sequence of thermal input, the capacitor time-series internal cooperative value sequence, and the capacitor-infrared time-series cooperative value sequence. This fully exploits the correlation information between features of different modes and dimensions, enabling a comprehensive assessment of the voltage detection status. Compared to the threshold determination method of existing technologies, the multi-feature fusion capability of the neural network can adapt to diverse interference scenarios in complex power environments. Through the collaborative verification of multiple features, it outputs accurate voltage detection results, significantly improving the system's voltage detection accuracy and environmental adaptability. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a non-contact capacitance sensing and infrared temperature measurement integrated voltage detection and error prevention system.

[0059] Figure 2 This is a schematic diagram of the structure of a neural network for classifying electrical signals.

[0060] Figure 3 This is a schematic diagram of the structure of the first sequence feature extraction and fusion unit;

[0061] Figure 4 This is a schematic diagram of the structure of the second sequence feature extraction and fusion unit;

[0062] Figure 5 This is a schematic diagram of the structure of the first feature extraction module and the second feature extraction module;

[0063] Figure 6 This is a schematic diagram of the multi-branch feature extraction module. Detailed Implementation

[0064] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0065] like Figure 1 As shown, a non-contact capacitance sensing and infrared temperature measurement fusion voltage detection and error prevention system includes: a capacitance feature extraction subsystem, a capacitance collaborative feature extraction subsystem, an infrared thermal feature quantification subsystem, a capacitance-infrared feature extraction subsystem, and a classification subsystem;

[0066] The capacitance feature extraction subsystem is used to perform first-order differentiation and take the absolute value of the normalized original capacitance signal to obtain the rate of change, and then perform second-order differentiation and take the absolute value to obtain the unstable value. The ratio of the rate of change to the unstable value at the same moment is used as the stable value of the capacitance to obtain the sequence of stable values ​​of capacitance.

[0067] The capacitance collaborative feature extraction subsystem is used to obtain first-order and second-order fluctuation value sequences based on the rate of change and the local window variance of the unstable value, and to obtain the capacitance time series internal collaborative value sequence based on the two sequences.

[0068] The infrared thermal feature quantization subsystem is used to extract the effective thermal input energy from the normalized infrared temperature signal to obtain the effective thermal input energy sequence.

[0069] The capacitor-infrared feature extraction subsystem is used to obtain the capacitor-infrared time-series co-value sequence based on the capacitor stability value and the effective energy of thermal input.

[0070] The classification subsystem is used to process the capacitance stable value sequence, thermal input effective energy sequence, capacitance time series internal coordinating value sequence, and capacitance-infrared time series coordinating value sequence using a voltage detection classification neural network to obtain the voltage detection state result.

[0071] In this embodiment, the device for acquiring the original capacitance signal is a capacitive proximity sensor, and the device for acquiring the infrared temperature signal is an infrared temperature sensor. The maximum capacitance signal value is found in the original capacitance signal, and all values ​​in the original capacitance signal are divided by the maximum capacitance signal value to obtain the normalized original capacitance signal. Similarly, the maximum temperature signal value is found in the infrared temperature signal, and all values ​​in the infrared temperature signal are divided by the maximum temperature signal value to obtain the normalized infrared temperature signal.

[0072] In this embodiment, the rate of change is: ,in, For the first rate of change at time, To take the absolute value, The normalized original capacitance signal, is the differential symbol, and t is time.

[0073] The value of instability is: ,in, Let be the unstable value at time t.

[0074] The stable value of the capacitor is: ,in, Let be the stable capacitance value at time t. Let be the rate of change at time t. Let t be the unstable value at time t. Arrange the stable values ​​of the capacitor at each time in chronological order to obtain the sequence of stable values ​​of the capacitor.

[0075] Under actual energized conditions, due to the continuous flow of current, the capacitance signal exhibits continuous and consistent change characteristics. Its first-order rate of change remains at a high level, while the second-order unstable value is relatively small, resulting in an overall larger capacitance stability value that changes gradually over time.

[0076] Under induced charging conditions, the capacitance signal is mainly affected by the coupling of the ambient electric field and transient disturbances. The rate of change is intermittent and has a low amplitude. The proportion of unstable second-order changes is relatively high, which leads to the overall small capacitance stability value and significant fluctuations.

[0077] In this embodiment, the process of obtaining the first-order and second-order fluctuation value sequences in the capacitance-coordinated feature extraction subsystem includes:

[0078] The variance of each rate of change in the local window is obtained to obtain the first-order fluctuation value. Specifically, a local window of length T is set with the current time as the center, and the variance of each rate of change in the local window is obtained to obtain the first-order fluctuation value corresponding to the current time. T is a positive integer. In this embodiment, the length of the local window is taken as the data of 5 times.

[0079] To obtain the second-order fluctuation value, the variance of each unstable value in the local window is obtained. Specifically, a local window of length T is set with the current time as the center, and the variance of each unstable value in the local window is obtained to obtain the second-order fluctuation value corresponding to the current time.

[0080] Arrange the first-order fluctuation values ​​at each moment in chronological order to obtain the first-order fluctuation value sequence.

[0081] Arrange the second-order fluctuation values ​​at each time point in chronological order to obtain the second-order fluctuation value sequence.

[0082] The rate of change calculated by this invention With unstable values It can quantify the instantaneous change intensity and the fluctuation of the rate of change of the capacitance signal respectively; the ratio of the two constitutes the capacitance stability value, which can directly reflect the stability of the capacitance coupling relationship.

[0083] When the device is actually energized, a stable capacitive coupling is formed between the device and the sensor. The instantaneous changes in the capacitive signal are smooth and the fluctuation is low. The corresponding stable value of the capacitance is always at a high level and the fluctuation range is narrow. When the device is induced to be energized, the capacitive coupling relationship is not supported by a stable current. The instantaneous changes in the signal are chaotic and the fluctuation is violent. The corresponding stable value of the capacitance is at a low level.

[0084] This invention obtains first-order and second-order fluctuation value sequences by calculating the rate of change and the variance of unstable values ​​within a local window. These sequences can reflect the overall fluctuation pattern of the capacitance signal over a continuous time period, rather than isolated single-moment characteristics.

[0085] Under actual charged conditions, the temporal changes of the capacitance signal exhibit strong regularity, with the variance value within the local window remaining at a low level, and the corresponding fluctuation value sequence remaining stable without significant abrupt changes. Under induced charged conditions, the capacitance signal exhibits irregular fluctuations due to the influence of the ambient electric field, with the variance value within the local window remaining at a high level, and the corresponding fluctuation value sequence exhibiting frequent and chaotic abrupt changes.

[0086] In this embodiment, the process of obtaining the internal cooperative value sequence of capacitance time series in the capacitance cooperative feature extraction subsystem includes:

[0087] In the first-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain a local segment of the first-order wave.

[0088] In the second-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain the local segment of the second-order wave.

[0089] Calculate the coefficient of variation of local segments of first-order and second-order waves belonging to the same moment;

[0090] The variational coordination coefficients are normalized to obtain the internal coordination values ​​of the capacitor timing sequence.

[0091] Arrange the capacitor timing internal coordination values ​​at each time point in chronological order to obtain the capacitor timing internal coordination value sequence.

[0092] In this embodiment, the formula for calculating the internal coordination value of the capacitor timing is: ,

[0093] in, For the first The internal coordination value of the capacitance timing at time 1. Let be the coefficient of variation at time t. It is a natural constant. This is the proportionality coefficient;

[0094] The formula for calculating the coefficient of variation is: ,

[0095] in, Let be the covariance of the local segment of the first-order wave at time t and the local segment of the second-order wave at time t. Let be the variance of a local segment of the first-order fluctuation at time t. Let be the variance of a local segment of the second-order fluctuation at time t. For a local segment of the first-order wave at time t, Let t be a local segment of the second-order wave at time t.

[0096] This invention extracts first- and second-order fluctuation segments within a local window, centered on each moment. This allows for focusing on continuous time segments at the current moment. For real energized equipment, the fluctuations in its capacitance signal exhibit strong temporal continuity, and the first- and second-order fluctuation segments within the same local window will display synchronized low-fluctuation characteristics. However, the fluctuations of inductively energized equipment are unpredictable, and the two fluctuation segments within the local window show no stable correlation. Therefore, the variation coordination coefficient is large for real energized equipment and small for inductively energized equipment.

[0097] In this embodiment, the process of obtaining the effective energy sequence of thermal input in the infrared thermal feature quantization subsystem includes:

[0098] Calculate the first-order time derivative of the normalized infrared temperature signal and take the non-negative value to obtain the instantaneous thermal input value: ,in, Let be the instantaneous heat input value at time t. For normalized infrared temperature signals, The differential symbol, For time, To obtain the maximum value;

[0099] Calculate the second time derivative of the normalized infrared temperature signal and take its absolute value to obtain the thermal input fluctuation value: ,in, Let be the heat input fluctuation value at time t. To take the absolute value;

[0100] To obtain the heat input duration, the mean of instantaneous heat input values ​​in a local window is obtained. Specifically, a local window of length T is set with the current time as the center, and the mean of each instantaneous heat input value in the local window is obtained to obtain the heat input duration corresponding to the current time.

[0101] To obtain the thermal input stability, the variance of the thermal input fluctuation values ​​in a local window is obtained. Specifically, a local window of length T is set with the current time as the center, and the variance of each thermal input fluctuation value in the local window is obtained to obtain the thermal input stability corresponding to the current time.

[0102] Calculate the effective energy of the heat input based on the heat input duration and heat input stability;

[0103] Arrange the effective heat input energy at each moment in chronological order to obtain the effective heat input energy sequence.

[0104] In this embodiment, the formula for calculating the effective energy of heat input is: ,

[0105] in, The effective energy of the heat input at time t. Let be the duration of heat input at time t. For heat input stability, This is a denominator parameter used to avoid the denominator being zero. .

[0106] Real electrical equipment, due to the Joule heating effect of current, continuously generates a stable heat input, corresponding to... If the temperature remains consistently in the positive range, and no actual current flows through the inductively charged equipment, it cannot generate active heat input; its temperature change is only affected by the environment. Mostly 0.

[0107] This invention characterizes the "duration intensity" and "fluctuation pattern" of heat input based on heat input duration (mean instantaneous heat input rate) and heat input stability (variance of heat input fluctuation value) calculated using a local window. Under actual charged conditions, heat input is continuous and stable, with the mean instantaneous heat input value (heat input duration) within the local window at a high level, and the variance of heat input fluctuation value (heat input stability) at a low level, reflecting the characteristic of "continuous and stable heat input." Under induced charged conditions, there is no active heat input, heat input duration approaches 0, and heat input stability exhibits irregular fluctuations due to environmental interference, but remains at a relatively high overall level. This time-series statistical feature further amplifies the difference in thermal properties between actual and induced charged conditions, enhancing the feature's anti-interference capability.

[0108] In this embodiment, the process of obtaining the capacitance-infrared temporal co-value sequence in the capacitance-infrared feature extraction subsystem includes:

[0109] The capacitor stability value is nonlinearly compressed to obtain the capacitor stability compressed value;

[0110] The effective heat input energy is nonlinearly compressed to obtain the effective heat input compression energy;

[0111] Multiplying the stable compression value of the capacitor by the effective compression energy of the thermal input yields the capacitor-infrared timing synergy value: ,in, For the first The capacitance-infrared timing coordination value at time t, The effective energy of the heat input at time t. For the first The capacitance stability value at each time point is obtained by arranging the capacitance-infrared time-series coordinated values ​​in chronological order to obtain the capacitance-infrared time-series coordinated value sequence.

[0112] This invention is achieved through and The nonlinear compression maps the two types of features to the (0,1) interval and uses the calculation method of "capacitive stable compression value × effective compression energy of thermal input". The synergistic value will approach 1 only when both the capacitor and infrared features are at the "true charging level" at the same time; if either feature is at the "induced charging level", the synergistic value will be significantly weakened.

[0113] like Figure 2 As shown, the electrophysiological classification neural network includes: a first sequence feature extraction and fusion unit, a second sequence feature extraction and fusion unit, a feature enhancement unit, a fully connected layer, and a Softmax classification layer;

[0114] The first sequence feature extraction and fusion unit is used to extract and fuse features from the capacitance stable value sequence and the thermal input effective energy sequence to obtain the capacitance-infrared multi-branch basic fusion features.

[0115] The second sequence feature extraction and fusion unit is used to extract and fuse features from the capacitor time series internal co-value sequence and the capacitor-infrared time series co-value sequence to obtain the capacitor-infrared multi-branch co-fusion feature.

[0116] The feature enhancement unit is used to enhance the basic fusion feature of the capacitor-infrared multi-branch fusion feature by employing the capacitor-infrared multi-branch collaborative fusion feature to obtain the capacitor-infrared multi-branch enhanced fusion feature.

[0117] The fully connected layer is used to map the capacitance-infrared multi-branch enhancement fusion features to obtain the voltage detection state features;

[0118] The Softmax classification layer is used to classify the voltage detection status features to obtain the voltage detection status results.

[0119] The results of voltage testing include: actual charged state, induced charged state, and no charged state.

[0120] This invention separates basic features and collaborative features through first and second sequence feature extraction and fusion units. The feature enhancement unit guides the enhancement of basic fusion features with collaborative fusion features, which can focus on the key features of real electrical charge and amplify their difference from induced electrical charge. Subsequently, through the processing of fully connected layers and Softmax classification layers, the accurate mapping and probabilistic classification of features to electrical detection states are achieved.

[0121] like Figure 3 and 4 As shown, both the first sequence feature extraction and fusion unit and the second sequence feature extraction and fusion unit include: a first feature extraction module, a second feature extraction module, a multiplier, a reshape layer, and a multi-branch feature extraction module;

[0122] The first feature extraction module is used to extract features from the input sequence to obtain the first shallow features;

[0123] The second feature extraction module is used to extract features from the input sequence to obtain the second shallow features;

[0124] The multiplier is used to perform element-wise multiplication of the first shallow feature and the second shallow feature to obtain a one-dimensional shallow fused feature.

[0125] The Reshape layer is used to transform the dimension of one-dimensional shallow fusion features to obtain two-dimensional shallow fusion features.

[0126] The multi-branch feature extraction module is used to extract multi-branch features from two-dimensional shallow fusion features to obtain capacitance-infrared multi-branch basic fusion features or capacitance-infrared multi-branch collaborative fusion features.

[0127] The Reshape layer is a tensor shape transformation layer.

[0128] This invention extracts shallow features from the input sequence through the first and second feature extraction modules respectively. The element-wise multiplication operation of the multiplier can strengthen the correlation information between the two types of shallow features and initially realize feature fusion. The dimensional transformation of the Reshape layer converts the one-dimensional fused features into a two-dimensional form that is adapted to multi-branch extraction. The multi-branch feature extraction module further extracts multi-scale and multi-dimensional depth features from the two-dimensional features, and finally obtains the capacitance-infrared multi-branch basic / cooperative fusion features.

[0129] like Figure 3 As shown, in the first sequence feature extraction and fusion unit: the input end of the first feature extraction module is used to input the capacitance stable value sequence, the input end of the second feature extraction module is used to input the thermal input effective energy sequence, and the output end of the multi-branch feature extraction module is used to output the capacitance-infrared multi-branch basic fusion feature.

[0130] like Figure 4 As shown, in the second sequence feature extraction and fusion unit: the input end of the first feature extraction module is used to input the capacitance time-series internal collaborative value sequence, the input end of the second feature extraction module is used to input the capacitance-infrared time-series collaborative value sequence, and the output end of the multi-branch feature extraction module is used to output the capacitance-infrared multi-branch collaborative fusion feature.

[0131] like Figure 5 As shown, the first feature extraction module and the second feature extraction module have the same structure, both including a first convolutional layer, a second convolutional layer and a third convolutional layer connected in sequence.

[0132] The kernel size of the first convolutional layer is 1×1, the kernel size of the second convolutional layer is 1×3, and the kernel size of the third convolutional layer is 1×3.

[0133] like Figure 6As shown, the multi-branch feature extraction module includes: a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, and a Concat layer. The input of the fourth convolutional layer is connected to the inputs of the fifth and seventh convolutional layers, respectively, and serves as the input of the multi-branch feature extraction module. The output of the fifth convolutional layer is connected to the input of the sixth convolutional layer. The input of the eighth convolutional layer is connected to the output of the seventh convolutional layer, and its output is connected to the input of the ninth convolutional layer. The input of the Concat layer is connected to the outputs of the fourth, sixth, and ninth convolutional layers, respectively, and its output serves as the output of the multi-branch feature extraction module.

[0134] The kernel size of the fourth convolutional layer is 3×3, the kernel size of the fifth convolutional layer is 1×1, the kernel size of the sixth convolutional layer is 3×3, the kernel size of the seventh convolutional layer is 1×1, the kernel size of the eighth convolutional layer is 3×3, and the kernel size of the ninth convolutional layer is 3×3.

[0135] In this embodiment, the expression for the feature enhancement unit is: ,

[0136] in, For capacitive-infrared multi-branch enhancement fusion features, This is a fusion feature based on capacitive-infrared multi-branch architecture. It features multi-branch synergistic fusion of capacitive and infrared technologies. For element-wise multiplication, It is a 1×1 convolutional layer. It is a Sigmoid layer.

[0137] This invention uses the Sigmoid activation function to... Generate attention weights, and then... Element-wise multiplication enhances the key features of actual charging and precisely amplifies the differences in features between actual and induced charging.

[0138] In this embodiment, the lengths of the capacitor stable value sequence, the thermal input effective energy sequence, the capacitor time-series internal cooperative value sequence, and the capacitor-infrared time-series cooperative value sequence are consistent, which facilitates the processing of the electrodetection classification neural network.

[0139] This invention, through the combined use of a capacitance feature extraction subsystem and a capacitance collaborative feature extraction subsystem, not only extracts the basic time-domain features of rate of change and unstable values ​​from the original capacitance signal, but also obtains first-order and second-order fluctuation value sequences through local window variance analysis, and further constructs the internal collaborative value sequence of the capacitance time series. Compared with existing technologies that rely solely on the single feature of capacitance signal amplitude for determination, the multi-dimensional capacitance time series features extracted by this invention can characterize the dynamic change law and intrinsic correlation of the capacitance signal, increasing the distinguishability between "true charged stationary time series patterns" and "pseudo-stationary features superimposed by interference," thus weakening the impact of interference on capacitance detection at the feature level.

[0140] This invention employs an infrared thermal feature quantification subsystem to extract the effective energy sequence of thermal input. Real-world electrical equipment generates stable and continuous thermal input due to the thermal effect of current, resulting in a regular pattern in the effective energy of the thermal input. In contrast, inductively coupled equipment has no actual current flowing through it and therefore does not generate stable thermal input; its thermal signal is only affected by ambient temperature and exhibits no clear pattern. Electromagnetic interference only affects the capacitance signal and cannot alter the fact that the equipment has a real thermal input. By constructing a cross-modal time-series collaborative value sequence through a capacitance-infrared feature extraction subsystem, dual verification using "capacitance features + infrared features" can be achieved, effectively reducing misjudgments caused by interference with a single capacitance feature and improving the accuracy of voltage detection.

[0141] This invention employs a classification subsystem and a voltage detection classification neural network to fuse the capacitor stable value sequence, the effective energy sequence of thermal input, the capacitor time-series internal cooperative value sequence, and the capacitor-infrared time-series cooperative value sequence. This fully exploits the correlation information between features of different modes and dimensions, enabling a comprehensive assessment of the voltage detection status. Compared to the threshold determination method of existing technologies, the multi-feature fusion capability of the neural network can adapt to diverse interference scenarios in complex power environments. Through the collaborative verification of multiple features, it outputs accurate voltage detection results, significantly improving the system's voltage detection accuracy and environmental adaptability.

[0142] The above are merely preferred embodiments of the present invention and are not intended to limit the present 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 non-contact capacitance sensing and infrared temperature measurement integrated voltage detection and error prevention system, characterized in that, include: The capacitance feature extraction subsystem is used to perform first-order differentiation and take the absolute value of the normalized original capacitance signal to obtain the rate of change, and then perform second-order differentiation and take the absolute value to obtain the unstable value. The ratio of the rate of change to the unstable value at the same moment is used as the stable value of the capacitance to obtain the sequence of stable values ​​of capacitance. The capacitance collaborative feature extraction subsystem is used to obtain first-order and second-order fluctuation value sequences based on the rate of change and the local window variance of the unstable value, and to obtain the capacitance time series internal collaborative value sequence based on the two sequences. The infrared thermal feature quantization subsystem is used to extract the effective thermal input energy from the normalized infrared temperature signal to obtain the effective thermal input energy sequence. The process of obtaining the effective energy sequence of thermal input in the infrared thermal feature quantization subsystem includes: Calculate the first time derivative of the normalized infrared temperature signal and take the non-negative value to obtain the instantaneous heat input value; Calculate the second time derivative of the normalized infrared temperature signal and take its absolute value to obtain the thermal input fluctuation value; The mean of instantaneous heat input values ​​in a local window is obtained to determine the heat input duration. The variance of the heat input fluctuation values ​​in a local window is obtained to determine the heat input stability. Calculate the effective energy of the heat input based on the heat input duration and heat input stability; Arrange the effective heat input energy at each moment in chronological order to obtain the effective heat input energy sequence. The capacitor-infrared feature extraction subsystem is used to obtain the capacitor-infrared time-series co-value sequence based on the capacitor stability value and the effective energy of thermal input. The classification subsystem is used to process the capacitor stable value sequence, the thermal input effective energy sequence, the capacitor time series internal coordinating value sequence, and the capacitor-infrared time series coordinating value sequence using a voltage detection classification neural network to obtain the voltage detection status result. The electrodetection classification neural network includes: a first sequence feature extraction and fusion unit, a second sequence feature extraction and fusion unit, a feature enhancement unit, a fully connected layer, and a Softmax classification layer; The first sequence feature extraction and fusion unit is used to extract and fuse features from the capacitance stable value sequence and the thermal input effective energy sequence to obtain the capacitance-infrared multi-branch basic fusion features. The second sequence feature extraction and fusion unit is used to extract and fuse features from the capacitor time series internal co-value sequence and the capacitor-infrared time series co-value sequence to obtain the capacitor-infrared multi-branch co-fusion feature. The feature enhancement unit is used to enhance the basic fusion feature of the capacitor-infrared multi-branch fusion feature by employing the capacitor-infrared multi-branch collaborative fusion feature to obtain the capacitor-infrared multi-branch enhanced fusion feature. The fully connected layer is used to map the capacitance-infrared multi-branch enhancement fusion features to obtain the voltage detection state features; The Softmax classification layer is used to classify the voltage detection status features to obtain the voltage detection status results.

2. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, The process of obtaining first-order and second-order fluctuation value sequences in the capacitance-coordinated feature extraction subsystem includes: Obtain the variance of each rate of change within the local window to get the first-order fluctuation value; Obtain the variance of each unstable value in the local window to get the second-order fluctuation value; Arrange the first-order fluctuation values ​​at each moment in chronological order to obtain the first-order fluctuation value sequence. Arrange the second-order fluctuation values ​​at each time point in chronological order to obtain the second-order fluctuation value sequence.

3. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, The process of obtaining the intra-temporal cooperative value sequence of capacitance in the capacitance cooperative feature extraction subsystem includes: In the first-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain a local segment of the first-order wave. In the second-order wave value sequence, data of a local window length is extracted with each time point as the center to obtain the local segment of the second-order wave. Calculate the coefficient of variation of local segments of first-order and second-order waves belonging to the same moment; The variational coordination coefficients are normalized to obtain the internal coordination values ​​of the capacitor timing sequence. Arrange the capacitor timing internal coordination values ​​at each time point in chronological order to obtain the capacitor timing internal coordination value sequence.

4. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 3, characterized in that, The formula for calculating the internal coordination value of the capacitor timing is: , in, For the first The internal coordination value of the capacitance timing at time 1. For the first The coefficient of variation at time 10:00 It is a natural constant. This is the proportionality coefficient; The formula for calculating the coefficient of variation is: , in, For the first The local segment of the first-order wave at time t and the t The covariance of a local segment of the second-order fluctuation at time t. For the first The variance of a local segment of the first-order fluctuation at time t. For the first The variance of a local segment of the second-order fluctuation at time t. For the first The local segment of the first-order fluctuation at time t. For the first The local segment of the second-order oscillation at time t.

5. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, The formula for calculating the effective energy of heat input is: , in, For the first The effective energy input at any time For the first Duration of heat input at any given moment For heat input stability, is the denominator parameter.

6. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, The process of obtaining the capacitance-infrared temporal co-value sequence in the capacitance-infrared feature extraction subsystem includes: The capacitor stability value is nonlinearly compressed to obtain the capacitor stability compressed value; The effective heat input energy is nonlinearly compressed to obtain the effective heat input compression energy; Multiplying the stable compression value of the capacitor by the effective compression energy of the thermal input yields the capacitor-infrared timing synergy value; The capacitance-infrared time-series coordinated values ​​at each time point are arranged in chronological order to obtain the capacitance-infrared time-series coordinated value sequence.

7. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, Both the first sequence feature extraction and fusion unit and the second sequence feature extraction and fusion unit include: a first feature extraction module, a second feature extraction module, a multiplier, a reshape layer, and a multi-branch feature extraction module; The first feature extraction module is used to extract features from the input sequence to obtain the first shallow features; The second feature extraction module is used to extract features from the input sequence to obtain the second shallow features; The multiplier is used to perform element-wise multiplication of the first shallow feature and the second shallow feature to obtain a one-dimensional shallow fused feature. The Reshape layer is used to transform the dimension of one-dimensional shallow fusion features to obtain two-dimensional shallow fusion features. The multi-branch feature extraction module is used to extract multi-branch features from two-dimensional shallow fusion features to obtain capacitance-infrared multi-branch basic fusion features or capacitance-infrared multi-branch collaborative fusion features.

8. The non-contact capacitive sensing and infrared temperature measurement fusion voltage detection and error prevention system according to claim 1, characterized in that, The expression for the feature enhancement unit is: , in, For capacitive-infrared multi-branch enhancement fusion features, This is a fusion feature based on capacitive-infrared multi-branch architecture. It features multi-branch synergistic fusion of capacitive and infrared technologies. For element-wise multiplication, It is a 1×1 convolutional layer. It is a Sigmoid layer.

Images