Electrical safety intelligent detection management and control system suitable for constructional engineering

Abstract

The invention relates to the technical field of electrical safety, in particular to an electrical safety intelligent detection management and control system suitable for constructional engineering. Aligning data of different sampling frequencies in the dynamic parameter group by using a sliding window algorithm, and preprocessing and standardizing the aligned data to obtain initial electrical dynamic data; deleting noise in the initial electrical dynamic data based on an improved Kalman filtering algorithm; establishing a dual-channel deep network model according to the dual-channel architecture of the PCN parameterized convolutional network and the STAN space-time association network, and performing hyper-parameter optimization on the model by using a QPSO (Quantum Particle Swarm Optimization); and inputting the characteristic electrical dynamic data into the dual-channel deep network model for training to obtain an electrical safety score, and performing hierarchical management and control on the system based on the electrical safety score. The false alarm rate of old equipment is reduced, the service life of key equipment is prolonged, and the loss caused by electrical faults is reduced.

Description

Technical Field

[0001] The present invention relates to the field of electrical safety management, and in particular to an intelligent electrical safety detection and control system suitable for construction projects. Background Art

[0002] In the field of power system safety monitoring, traditional detection relies heavily on manual inspections. Manual inspections are affected by factors such as the inspector's professional level, work attitude, and fatigue level. Different inspectors may have different criteria for judging hidden dangers, which can easily lead to missed detections and false detections, and fail to timely detect potential electrical safety hazards. Existing technologies often use single-parameter threshold alarms. Alarms are triggered based solely on a single electrical parameter exceeding a set threshold. In reality, electrical systems operate in a complex manner, and complex hidden dangers occur frequently. Single-parameter alarms cannot make comprehensive judgments, resulting in a high false alarm rate and the possibility of missing real hidden dangers. Existing technologies also lack the ability to analyze the spatiotemporal correlations of electrical equipment operating data. The inability to effectively combine multi-dimensional information such as equipment operating time and spatial location makes it difficult to accurately determine the development trend of faults and the root causes of hidden dangers. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design an electrical safety intelligent detection and control system suitable for construction projects.

[0004] The technical solution of the present invention to achieve the above object is that, further, in the above-mentioned intelligent detection and control system for electrical safety of construction projects, the intelligent detection and control system for electrical safety of construction projects includes the following modules:

[0005] An electrical data acquisition module is used to utilize the electrical dynamic parameter group in the sensor array network acquisition system, align the data of different sampling frequencies in the dynamic parameter group using a sliding window algorithm, and preprocess and standardize the aligned data to obtain initial electrical dynamic data;

[0006] an electrical feature extraction module for removing noise from the initial electrical dynamic data based on an improved Kalman filter algorithm, extracting characteristic vectors of the current waveform harmonic distortion rate, the temperature gradient change rate, and the three-phase imbalance from the noise-removed data, and obtaining characteristic electrical dynamic data;

[0007] The network model building module is used to establish a PCN-STAN dual-channel deep network model based on the dual-channel architecture of the PCN parameterized convolutional network and the STAN spatiotemporal association network, and use the QPSO quantum particle swarm algorithm to optimize the model's hyperparameters to obtain the QPSO-PCN-STAN dual-channel deep network model;

[0008] The electrical safety judgment module is used to input the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and perform hierarchical control of the system based on the electrical safety score.

[0009] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical data acquisition module includes the following submodules:

[0010] Construction submodule, used to build an intelligent sensor array network, each node contains a high-frequency sampling module, a multi-source synchronization controller and an adaptive adjustment unit;

[0011] An acquisition submodule is configured to utilize a sensor array network to acquire a group of electrical dynamic parameters in the system, wherein the group of electrical dynamic parameters includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameter, electromagnetic interference intensity, and temperature gradient;

[0012] an alignment submodule, for defining a window width and a step size in a window function based on a sliding window algorithm, establishing a frequency-time mapping relationship, using cubic spline interpolation for high-frequency data, using Lagrange polynomial fitting for low-frequency data, establishing a unified timestamp sequence for the electrical dynamic parameter group, and obtaining aligned electrical dynamic data;

[0013] The cleaning submodule is used to clean the aligned data of outliers based on the improved Hampel filter, and standardize the cleaned data to obtain initial electrical dynamic data.

[0014] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical feature extraction module includes the following submodules:

[0015] The adjustment submodule is used to introduce a residual monitoring mechanism based on the improved Kalman filter algorithm, and adjust the process noise covariance matrix and the observation noise covariance matrix according to the real-time calculated state estimation residual;

[0016] State submodule, used to establish the state equation x of the initial electrical dynamic parameters k =Φx k-1 +ω k-1 , where Φ represents the state transfer matrix, ω represents the process noise, and x k-1 Represents the system state vector at time k-1, which is the state of the electrical dynamic parameters at the previous moment; ω k-1 represents the process noise vector, which is the noise introduced by the system internal uncertainties between time k-1 and time k;

[0017] Observation submodule, used to construct the observation equation z of the initial electrical dynamic parametersk =Hx k +v k , where H represents the observation matrix, v k represents the measurement noise, reflecting the noise introduced during the measurement at time k;

[0018] The calculation submodule is used to calculate the Kalman gain after assigning exponential decay weights to the initial electrical dynamic data in the prediction stage and the update stage, and perform state correction and covariance update on the gained data to obtain noise-reduced electrical dynamic data.

[0019] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical feature extraction module further includes the following submodules:

[0020] The analysis submodule is used to analyze the current waveform in the noise-reduced electrical dynamic data through FFT and calculate the total harmonic distortion rate THD;

[0021]

[0022] Where n represents the highest order of harmonics, I h It represents the effective value of the hth harmonic current, where h=,3,...,n represents the magnitude of each harmonic current except the fundamental current; I1 represents the effective value of the fundamental current;

[0023] The acquisition submodule is used to obtain the temperature gradient change rate in the data using the sliding window difference method and calculate ΔT / Δt=(T t -T t-n ) / (nΔt), where the window length n is adaptively adjusted according to the sampling frequency, where ΔT / Δt represents the temperature gradient change rate, T t Indicates the temperature value measured at the current time t, T t-n represents the temperature value n time intervals ago, n represents the sliding window length, and Δt represents the sampling time interval;

[0024] The extraction submodule is used to extract the ratio of the negative sequence component to the positive sequence component in the data based on the symmetrical component method to obtain the three-phase imbalance;

[0025] The iterative submodule is used to truncate the residuals exceeding 3σ in the total harmonic distortion rate, temperature gradient change rate and three-phase imbalance, and stop the iteration when the residual variance exceeds the threshold after 10 consecutive iterations to obtain characteristic electrical dynamic data.

[0026] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical feature extraction module includes the following units:

[0027] Convolution kernel unit, the convolution kernel used in the PCN-STAN dual-channel deep network model is initialized using the He normal distribution and the activation function uses PReLU;

[0028] The attention unit is used to introduce the DCA dynamic channel attention module into the PCN parameterized convolutional network of the model and learn channel weights through the SE-Net concept;

[0029] The spatiotemporal association unit is used to segment the electrical feature sequence into time windows using the STAN spatiotemporal association network in the PCN-STAN dual-channel deep network model, extract local temporal features using 1D convolution, and obtain the physical connection relationship between devices using the GCN graph convolutional network;

[0030] The channel classification unit is used to process the current and voltage waveform data using the PCN channel and the temperature and harmonic timing data using the STAN channel.

[0031] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical feature extraction module further includes the following units:

[0032] A hyperparameter optimization unit is used to optimize the hyperparameters of the model using the QPSO quantum particle swarm algorithm and adjust the shrinkage factor using a nonlinear decreasing strategy. The hyperparameters include at least the convolution kernel size, network depth, learning rate, and number of attention heads.

[0033] The hyperparameter update unit is used to apply Lévy flight perturbations to the global optimal particle during the hyperparameter optimization process, encode the hyperparameters into quantum bit probability amplitudes, and update the hyperparameters of the model through quantum rotating gates to obtain the QPSO-PCN-STAN dual-channel deep network model.

[0034] Furthermore, in the above-mentioned intelligent electrical safety detection and control system applicable to construction projects, the electrical feature extraction module includes the following units:

[0035] A data input unit, configured to input the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training;

[0036] The data training unit is used to timestamp-align the local features of the current waveform extracted by the PCN channel with the global features of the temperature time series generated by the STAN channel, using interpolation to compensate for sampling frequency differences. The self-attention mechanism is used to map the harmonic distortion rate, three-phase imbalance, and temperature gradient into a unified high-dimensional space.

[0037] An indicator establishment unit is used to establish a scoring indicator system including at least harmonic mutation index, heat accumulation risk value, three-phase imbalance, insulation degradation trend and load fluctuation coefficient;

[0038] A hierarchical management and control unit is used to obtain an electrical safety score based on model training, and to divide the electrical safety score into a three-level management and control system, including at least a normal state with a score ≥85, a warning state with a score of 70≤<85, and an alarm state with a score of <70; the system is hierarchically managed based on the three states.

[0039] Furthermore, in implementing the above-mentioned method of an intelligent electrical safety detection and control system applicable to construction projects, the method includes the following steps:

[0040] Using the electrical dynamic parameter group in the sensor array network acquisition system, aligning data of different sampling frequencies in the dynamic parameter group using a sliding window algorithm, preprocessing and standardizing the aligned data to obtain initial electrical dynamic data;

[0041] Deleting noise from the initial electrical dynamic data based on an improved Kalman filter algorithm, extracting characteristic vectors of current waveform harmonic distortion rate, temperature gradient change rate, and three-phase imbalance from the noise-deleted data, and obtaining characteristic electrical dynamic data;

[0042] Based on the dual-channel architecture of PCN parameterized convolutional network and STAN spatiotemporal association network, a PCN-STAN dual-channel deep network model is established. The QPSO quantum particle swarm algorithm is used to optimize the model's hyperparameters, resulting in a QPSO-PCN-STAN dual-channel deep network model.

[0043] The characteristic electrical dynamic data is input into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and the system is hierarchically controlled based on the electrical safety score.

[0044] Furthermore, in the above-mentioned electrical safety intelligent detection and control method applicable to construction projects, the method includes the following steps:

[0045] Build an intelligent sensor array network, where each node contains a high-frequency sampling module, a multi-source synchronization controller, and an adaptive adjustment unit;

[0046] Using a sensor array network to collect an electrical dynamic parameter group in the system, the electrical dynamic parameter group includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameters, electromagnetic interference intensity and temperature gradient;

[0047] Based on the sliding window algorithm, the window width and step size in the window function are defined, a frequency-time mapping relationship is established, cubic spline interpolation is used for high-frequency data, Lagrange polynomial fitting is used for low-frequency data, and a unified timestamp sequence is established for the electrical dynamic parameter group to obtain aligned electrical dynamic data;

[0048] The aligned data are cleaned of outliers based on the improved Hampel filter, and the cleaned data are standardized to obtain the initial electrical dynamic data.

[0049] Furthermore, in the above-mentioned intelligent detection and control method for electrical safety in construction projects, the step of acquiring voice data in a real-time environment through a microphone array includes:

[0050] The model is optimized for hyperparameters using the QPSO quantum particle swarm algorithm, and the shrinkage factor is adjusted using a nonlinear decreasing strategy. The hyperparameters include at least the convolution kernel size, network depth, learning rate, and number of attention heads.

[0051] During the hyperparameter optimization process, Lévy flight perturbations are applied to the global optimal particle, the hyperparameters are encoded as quantum bit probability amplitudes, and the hyperparameters of the model are updated through quantum rotating gates to obtain the QPSO-PCN-STAN dual-channel deep network model.

[0052] Its beneficial effects are as follows: by utilizing the electrical dynamic parameter group in the sensor array network acquisition system, using a sliding window algorithm to align the data of different sampling frequencies in the dynamic parameter group, preprocessing and standardizing the aligned data to obtain initial electrical dynamic data; using an improved Kalman filter algorithm to remove noise from the initial electrical dynamic data, extracting the characteristic vectors of the current waveform harmonic distortion rate, temperature gradient change rate, and three-phase imbalance in the noise-removed data, and obtaining characteristic electrical dynamic data; establishing a PCN-STAN dual-channel deep network model based on the dual-channel architecture of the PCN parameterized convolutional network and the STAN spatiotemporal association network, and using the QPSO quantum particle swarm algorithm to optimize the model's hyperparameters to obtain a QPSO-PCN-STAN dual-channel deep network model; inputting the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and performing hierarchical control of the system based on the electrical safety score. 1. Improved noise suppression efficiency: The noise suppression effect is improved through the dynamically adjusted Q / R matrix, effectively eliminating the impact of transient interference such as arc discharge on feature extraction. 2. Hidden Fault Identification: Based on a dual-channel network, a spatiotemporal attention mechanism is used to improve the accuracy of identifying the correlation between harmonic distortion rate and temperature gradient, enabling the capture of early fault characteristics such as poor contact. 3. Dynamic Optimization of Safety Assessment: A scoring model based on the equipment lifecycle attenuation factor is introduced to reduce the false alarm rate of older equipment while extending the service life of critical equipment. 4. Doubled System Interaction Efficiency: A three-level control strategy is used to achieve a 200ms closed-loop response from risk perception to action, reducing losses caused by electrical failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0054] Figure 1 This is a schematic diagram of a first embodiment of an intelligent electrical safety detection and control system applicable to construction projects in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of a second embodiment of an intelligent electrical safety detection and control system applicable to construction projects in accordance with an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of a third embodiment of an intelligent electrical safety detection and control system suitable for construction projects in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", and "" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0059] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, an electrical safety intelligent detection and control system suitable for construction projects includes the following modules:

[0060] 101. An electrical data acquisition module, configured to utilize the electrical dynamic parameter group in the sensor array network acquisition system, align data of different sampling frequencies in the dynamic parameter group using a sliding window algorithm, and preprocess and standardize the aligned data to obtain initial electrical dynamic data;

[0061] Specifically, this embodiment also includes a construction submodule for constructing an intelligent sensor array network, each node including a high-frequency sampling module, a multi-source synchronous controller and an adaptive adjustment unit;

[0062] An acquisition submodule is used to acquire an electrical dynamic parameter group in the system using a sensor array network, wherein the electrical dynamic parameter group includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameter, electromagnetic interference intensity, and temperature gradient;

[0063] The alignment submodule is used to define the window width and step size in the window function based on the sliding window algorithm, establish a frequency-time mapping relationship, use cubic spline interpolation for high-frequency data, use Lagrange polynomial fitting for low-frequency data, establish a unified timestamp sequence for the electrical dynamic parameter group, and obtain aligned electrical dynamic data;

[0064] The cleaning submodule is used to clean the aligned data of outliers based on the improved Hampel filter, and standardize the cleaned data to obtain initial electrical dynamic data.

[0065] 102. An electrical feature extraction module, configured to remove noise from the initial electrical dynamic data based on an improved Kalman filter algorithm, extract characteristic vectors of the current waveform harmonic distortion rate, temperature gradient change rate, and three-phase imbalance from the noise-removed data, and obtain characteristic electrical dynamic data;

[0066] Specifically, this embodiment also includes an adjustment submodule for introducing a residual monitoring mechanism based on an improved Kalman filter algorithm, and adjusting the process noise covariance matrix and the observation noise covariance matrix according to the state estimation residual calculated in real time;

[0067] State submodule, used to establish the state equation x of the initial electrical dynamic parameters k =Φx k-1 +ω k-1 , where Φ represents the state transfer matrix, ω represents the process noise, and x k-1 Represents the system state vector at time k-1, which is the state of the electrical dynamic parameters at the previous moment; ω k-1 represents the process noise vector, which is the noise introduced by the system internal uncertainties between time k-1 and time k;

[0068] Observation submodule, used to construct the observation equation z of the initial electrical dynamic parameters k =Hx k +v k , where H represents the observation matrix, v k represents the measurement noise, reflecting the noise introduced during the measurement at time k;

[0069] The calculation submodule is used to calculate the Kalman gain after assigning exponential decay weights to the initial electrical dynamic data in the prediction stage and the update stage, and perform state correction and covariance update on the gained data to obtain noise-reduced electrical dynamic data.

[0070] The analysis submodule is used to analyze the current waveform in the noise-reduced electrical dynamic data through FFT and calculate the total harmonic distortion rate THD;

[0071]

[0072] Where n represents the highest order of harmonics, I h It represents the effective value of the hth harmonic current, where h=,3,...,n represents the magnitude of each harmonic current except the fundamental current; I1 represents the effective value of the fundamental current;

[0073] The acquisition submodule is used to obtain the temperature gradient change rate in the data using the sliding window difference method and calculate ΔT / Δt=(T t -T t-n ) / (nΔt), where the window length n is adaptively adjusted according to the sampling frequency, where ΔT / Δt represents the temperature gradient change rate, T t Indicates the temperature value measured at the current time t, T t-n represents the temperature value n time intervals ago, n represents the sliding window length, and Δt represents the sampling time interval;

[0074] The extraction submodule is used to extract the ratio of the negative sequence component to the positive sequence component in the data based on the symmetrical component method to obtain the three-phase imbalance;

[0075] The iterative submodule is used to truncate the residuals exceeding 3σ in the total harmonic distortion rate, temperature gradient change rate and three-phase imbalance. When the residual variance exceeds the threshold after 10 consecutive iterations, the iteration is stopped to obtain the characteristic electrical dynamic data.

[0076] 103. A network model building module is used to build a PCN-STAN dual-channel deep network model based on the dual-channel architecture of the PCN parameterized convolutional network and the STAN spatiotemporal association network, and to optimize the model's hyperparameters using the QPSO quantum particle swarm algorithm to obtain a QPSO-PCN-STAN dual-channel deep network model;

[0077] Specifically, this embodiment also includes a convolution kernel unit, in which the convolution kernel of the PCN-STAN dual-channel deep network model is initialized using the He normal distribution and the activation function uses PReLU;

[0078] The attention unit is used to introduce the DCA dynamic channel attention module into the PCN parameterized convolutional network of the model and learn channel weights through the SE-Net concept;

[0079] The spatiotemporal association unit is used to segment the electrical feature sequence into time windows using the STAN spatiotemporal association network in the PCN-STAN dual-channel deep network model, extract local temporal features using 1D convolution, and obtain the physical connection relationship between devices using the GCN graph convolutional network;

[0080] The channel classification unit is used to process the current and voltage waveform data using the PCN channel and the temperature and harmonic timing data using the STAN channel.

[0081] The hyperparameter optimization unit is used to optimize the model's hyperparameters using the QPSO quantum particle swarm algorithm and adjust the shrinkage factor using a nonlinear decreasing strategy. The hyperparameters include at least the convolution kernel size, network depth, learning rate, and number of attention heads.

[0082] The hyperparameter update unit is used to apply Lévy flight perturbations to the global optimal particle during the hyperparameter optimization process, encode the hyperparameters into quantum bit probability amplitudes, and update the hyperparameters of the model through quantum rotating gates to obtain the QPSO-PCN-STAN dual-channel deep network model.

[0083] Specifically, the model is introduced as follows:

[0084] 1. Model architecture design principles:

[0085] 1. PCN parameterized convolutional network:

[0086] Core idea: Break through the fixed size limitation of traditional convolution kernels, build a dynamically adjustable parameterized convolution structure, and adjust the size, shape and receptive field of the convolution kernel in real time according to the adaptive characteristics of the input features.

[0087] Dynamic convolution kernel generator: Automatically generates convolution kernel parameters based on the spectral analysis results of the feature graph (such as the proportion of high-frequency components). For example, when detecting current harmonic distortion, the kernel width is automatically expanded to capture broadband features.

[0088] Multi-scale feature fusion: Large, medium, and small convolution kernels are deployed in parallel within the same layer, and features of different granularities are dynamically fused through attention weights.

[0089] Applicable scenarios: Focus on local refined feature extraction, such as capturing detailed features such as current waveform mutation points and temperature gradient extremes.

[0090] 2. STAN spatiotemporal association network

[0091] Core idea: Integrate spatial topological relationships (connection structure of distribution equipment) and temporal evolution laws (trend of temperature change with load) to establish spatiotemporal joint analysis capabilities for electrical systems.

[0092] Graph structure modeling: Circuit breakers, transformers and other equipment in the power distribution system are abstracted as graph nodes, and cable connection relationships are used as edges to construct an electrical topology diagram.

[0093] Temporal Causal Convolution: A convolutional layer with temporal constraints is used to ensure that feature extraction relies only on historical data, avoiding future information leakage.

[0094] 2. Dual-channel connection mechanism

[0095] 1. Data diversion strategy:

[0096] PCN channel input: high-dimensional data such as original waveforms and temperature field distribution maps, retaining local details.

[0097] STAN channel input: structured data such as equipment topology diagrams and time series parameters (hourly load changes).

[0098] 2. Feature interaction fusion:

[0099] Intermediate layer interaction: Cross-channel attention gates are set at the third and sixth layers of the network to dynamically calculate the correlation weights of the two-channel features. For example, when an abnormal temperature is detected at a node, the feature weight of that node in the topology graph is automatically increased.

[0100] Output layer fusion: An adaptive weighted voting mechanism is used to dynamically allocate the final decision weight according to the confidence scores (such as the predicted probability variance) output by the two channels, with the high-confidence channel taking the dominant position.

[0101] 3. Joint training mechanism

[0102] Loss function coupling: We design a triple supervisory signal consisting of local feature loss (PCN output), spatiotemporal association loss (STAN output), and joint consistency loss, forcing the two channels to reach consensus while maintaining expertise.

[0103] Gradient coordination: Through gradient masking technology, the gradients of the two channels are dynamically scaled during backpropagation to prevent one channel from overly dominating the training process.

[0104] 3. QPSO Quantum Particle Swarm Optimization Strategy

[0105] 1. Core hyperparameter set:

[0106] The convolution kernel growth rate of the PCN channel, the number of graph attention heads of the STAN channel, the weight decay coefficient of the cross-channel fusion layer, the convolution kernel size, the network depth, the learning rate, and the number of attention heads;

[0107] 2. Principle of quantization improvement:

[0108] Quantum state encoding: Representing each hyperparameter combination as a probability amplitude of quantum bits, enabling simultaneous exploration of multiple potential optimal regions.

[0109] Potential well constraint mechanism: The quantum potential well model is introduced to enable particles to gradually converge to the optimal solution neighborhood during the search process, avoiding invalid random walks.

[0110] Three-stage optimization process:

[0111] 1. Global exploration stage:

[0112] The particle swarm conducts wide-area quantum tunneling search within the preset hyperparameter space to quickly locate potential areas.

[0113] 2. Local development stage:

[0114] A refined grid scan is performed on the potential area, and the Metropolis criterion is used to accept suboptimal solutions to avoid premature convergence.

[0115] 3. Dynamic balance stage:

[0116] Monitor the validation set loss curve in real time. When a performance plateau is detected, automatically expand the search dimension (add optimization objectives or release constraints).

[0117] 104. An electrical safety judgment module is used to input characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and to perform hierarchical control of the system based on the electrical safety score.

[0118] Specifically, this embodiment also includes a data input unit for inputting characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training;

[0119] The data training unit is used to timestamp-align the local features of the current waveform extracted by the PCN channel with the global features of the temperature time series generated by the STAN channel, using interpolation to compensate for sampling frequency differences. The self-attention mechanism is used to map the harmonic distortion rate, three-phase imbalance, and temperature gradient into a unified high-dimensional space.

[0120] An indicator establishment unit is used to establish a scoring indicator system including at least harmonic mutation index, heat accumulation risk value, three-phase imbalance, insulation degradation trend and load fluctuation coefficient;

[0121] The hierarchical management and control unit is used to obtain electrical safety scores based on model training. The system is divided into a three-level management and control system according to the electrical safety scores, including at least a normal state with a score ≥85, a warning state with a score of 70≤<85, and an alarm state with a score of <70; the system is hierarchically managed based on the three states.

[0122] Specifically, this embodiment also includes:

[0123] 1. Feature data fusion strategy:

[0124] Spatiotemporal feature alignment: The local features of the current waveform extracted by the PCN channel are timestamp-aligned with the global features of the temperature time series generated by the STAN channel, and the interpolation method is used to compensate for the sampling frequency difference.

[0125] Multimodal embedding: The harmonic distortion rate (frequency domain feature), three-phase imbalance (spatial feature) and temperature gradient (time domain feature) are mapped to a unified high-dimensional space through the self-attention mechanism.

[0126] 2. Training mechanism design

[0127] Two-stage training strategy:

[0128] Pre-training stage: Use historical fault datasets (including typical scenarios such as arcs and short circuits) to perform transfer learning on PCN-STAN and freeze the shallow network parameters.

[0129] Fine-tuning stage: Sliding window incremental training is used, and the window length matches the thermal time constant of the device (for example, the thermal inertia of copper conductors corresponds to a 20-minute window).

[0130] 3. Three-level control system

[0131] Normal state (score ≥85)

[0132] Start a periodic automatic review mechanism to update the assessment results every 30 minutes;

[0133] The visual interface displays a green safety mark and is simultaneously pushed to the SCADA system.

[0134] Warning status (70≤score<85):

[0135] Trigger in-depth diagnosis of device health and call knowledge graphs to analyze associated failure modes;

[0136] Automatically generate maintenance recommendations ("The temperature of the phase B connector is 4.2°C higher than that of phase A. Infrared retesting is recommended").

[0137] Alarm state (score <70):

[0138] Initiate multi-system linkage: cut off non-critical loads, activate fire extinguishing equipment, and push emergency work orders;

[0139] Generate a fault tracing report and mark the characteristic contribution (heat accumulation risk contribution is 62%);

[0140] 4. Dynamic adjustment mechanism

[0141] Threshold adaptation: The alarm threshold is adjusted based on the exponential decay function of the equipment's service life. The threshold decrease rate for old equipment is λ = 0.58.

[0142] Feedback optimization: Add false positive / missing cases to the reinforcement learning buffer and update the model parameters once a month.

[0143] Its beneficial effects are as follows: by utilizing the electrical dynamic parameter group in the sensor array network acquisition system, using a sliding window algorithm to align the data of different sampling frequencies in the dynamic parameter group, preprocessing and standardizing the aligned data to obtain initial electrical dynamic data; using an improved Kalman filter algorithm to remove noise from the initial electrical dynamic data, extracting the characteristic vectors of the current waveform harmonic distortion rate, temperature gradient change rate, and three-phase imbalance in the noise-removed data, and obtaining characteristic electrical dynamic data; establishing a PCN-STAN dual-channel deep network model based on the dual-channel architecture of the PCN parameterized convolutional network and the STAN spatiotemporal association network, and using the QPSO quantum particle swarm algorithm to optimize the model's hyperparameters to obtain a QPSO-PCN-STAN dual-channel deep network model; inputting the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training, obtaining an electrical safety score, and implementing hierarchical control of the system based on the electrical safety score. 1. Improved noise suppression efficiency: The noise suppression effect is improved through the dynamically adjusted Q / R matrix, effectively eliminating the impact of transient interference such as arc discharge on feature extraction. 2. Hidden Fault Identification: Based on a dual-channel network, a spatiotemporal attention mechanism is used to improve the accuracy of identifying the correlation between harmonic distortion rate and temperature gradient, enabling the capture of early fault characteristics such as poor contact. 3. Dynamic Optimization of Safety Assessment: A scoring model based on the equipment lifecycle attenuation factor is introduced to reduce the false alarm rate of older equipment while extending the service life of critical equipment. 4. Doubled System Interaction Efficiency: A three-level control strategy is used to achieve a 200ms closed-loop response from risk perception to action, reducing losses caused by electrical failures.

[0144] In this embodiment, please refer to Figure 2 In a second embodiment of an intelligent electrical safety detection and control system for construction projects according to an embodiment of the present invention, the electrical data acquisition module includes the following submodules:

[0145] Construction submodule, used to build an intelligent sensor array network, each node contains a high-frequency sampling module, a multi-source synchronization controller and an adaptive adjustment unit;

[0146] An acquisition submodule is used to acquire an electrical dynamic parameter group in the system using a sensor array network, wherein the electrical dynamic parameter group includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameter, electromagnetic interference intensity, and temperature gradient;

[0147] The alignment submodule is used to define the window width and step size in the window function based on the sliding window algorithm, establish a frequency-time mapping relationship, use cubic spline interpolation for high-frequency data, use Lagrange polynomial fitting for low-frequency data, establish a unified timestamp sequence for the electrical dynamic parameter group, and obtain aligned electrical dynamic data;

[0148] The cleaning submodule is used to clean the aligned data of outliers based on the improved Hampel filter, and standardize the cleaned data to obtain initial electrical dynamic data.

[0149] Its beneficial effect is that the noise suppression effect is improved through the dynamically adjusted Q / R matrix, effectively eliminating the influence of transient interference such as arc discharge on feature extraction.

[0150] In this embodiment, please refer to Figure 3 In a third embodiment of an intelligent electrical safety detection and control system for construction projects according to an embodiment of the present invention, the electrical feature extraction module includes the following subunits:

[0151] A data input unit is used to input characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training;

[0152] The data training unit is used to timestamp-align the local features of the current waveform extracted by the PCN channel with the global features of the temperature time series generated by the STAN channel, using interpolation to compensate for sampling frequency differences. The self-attention mechanism is used to map the harmonic distortion rate, three-phase imbalance, and temperature gradient into a unified high-dimensional space.

[0153] An indicator establishment unit is used to establish a scoring indicator system including at least harmonic mutation index, heat accumulation risk value, three-phase imbalance, insulation degradation trend and load fluctuation coefficient;

[0154] The hierarchical management and control unit is used to obtain electrical safety scores based on model training. The system is divided into a three-level management and control system according to the electrical safety scores, including at least a normal state with a score ≥85, a warning state with a score of 70≤<85, and an alarm state with a score of <70; the system is hierarchically managed based on the three states.

[0155] Its beneficial effect is to evaluate the electrical safety status through an intelligent predictive model, and use a three-level management and control strategy to achieve a 200ms closed-loop response from risk perception to disposal, thereby reducing the losses caused by electrical failures.

[0156] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention, and such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent electrical safety detection and control system suitable for construction projects, characterized in that: The electrical safety intelligent detection and control system includes the following modules: An electrical data acquisition module is used to utilize the electrical dynamic parameter group in the sensor array network acquisition system, align the data of different sampling frequencies in the dynamic parameter group using a sliding window algorithm, and preprocess and standardize the aligned data to obtain initial electrical dynamic data; an electrical feature extraction module for removing noise from the initial electrical dynamic data based on an improved Kalman filter algorithm, extracting characteristic vectors of the current waveform harmonic distortion rate, the temperature gradient change rate, and the three-phase imbalance from the noise-removed data, and obtaining characteristic electrical dynamic data; The network model building module is used to establish a PCN-STAN dual-channel deep network model based on the dual-channel architecture of the PCN parameterized convolutional network and the STAN spatiotemporal association network, and use the QPSO quantum particle swarm algorithm to optimize the model's hyperparameters to obtain the QPSO-PCN-STAN dual-channel deep network model; The electrical safety judgment module is used to input the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and perform hierarchical control of the system based on the electrical safety score.

2. The electrical safety intelligent detection and control system for construction engineering according to claim 1, characterized in that: The electrical data acquisition module includes the following submodules: Construction submodule, used to build an intelligent sensor array network, each node contains a high-frequency sampling module, a multi-source synchronization controller and an adaptive adjustment unit; An acquisition submodule is configured to utilize a sensor array network to acquire a group of electrical dynamic parameters in the system, wherein the group of electrical dynamic parameters includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameter, electromagnetic interference intensity, and temperature gradient; an alignment submodule, for defining a window width and a step size in a window function based on a sliding window algorithm, establishing a frequency-time mapping relationship, using cubic spline interpolation for high-frequency data, using Lagrange polynomial fitting for low-frequency data, establishing a unified timestamp sequence for the electrical dynamic parameter group, and obtaining aligned electrical dynamic data; The cleaning submodule is used to clean the aligned data of outliers based on the improved Hampel filter, and standardize the cleaned data to obtain initial electrical dynamic data.

3. The electrical safety intelligent detection and control system suitable for construction engineering according to claim 1, characterized in that: The electrical feature extraction module includes the following submodules: The adjustment submodule is used to introduce a residual monitoring mechanism based on the improved Kalman filter algorithm, and adjust the process noise covariance matrix and the observation noise covariance matrix according to the real-time calculated state estimation residual; State submodule, used to establish the state equation x of the initial electrical dynamic parameters k =Φx k-1 +ω k-1 , where Φ represents the state transfer matrix, ω represents the process noise, and x k-1 Represents the system state vector at time k-1, which is the state of the electrical dynamic parameters at the previous moment; ω k-1 represents the process noise vector, which is the noise introduced by the system internal uncertainties between time k-1 and time k; Observation submodule, used to construct the observation equation z of the initial electrical dynamic parameters k =Hx k +v k , where H represents the observation matrix, v k represents the measurement noise, reflecting the noise introduced during the measurement at time k; The calculation submodule is used to calculate the Kalman gain after assigning exponential decay weights to the initial electrical dynamic data in the prediction stage and the update stage, and perform state correction and covariance update on the gained data to obtain noise-reduced electrical dynamic data.

4. The electrical safety intelligent detection and control system for construction engineering according to claim 1, characterized in that: The electrical feature extraction module also includes the following submodules: The analysis submodule is used to analyze the current waveform in the noise-reduced electrical dynamic data through FFT and calculate the total harmonic distortion rate THD; Where n represents the highest order of harmonics, I h It represents the effective value of the hth harmonic current, where h=,3,...,n represents the magnitude of each harmonic current except the fundamental current; I1 represents the effective value of the fundamental current; The acquisition submodule is used to obtain the temperature gradient change rate in the data using the sliding window difference method and calculate ΔT / Δt=(T t -T t-n ) / (nΔt), where the window length n is adaptively adjusted according to the sampling frequency, where ΔT / Δt represents the temperature gradient change rate, T t Indicates the temperature value measured at the current time t, T t-n represents the temperature value n time intervals ago, n represents the sliding window length, and Δt represents the sampling time interval; The extraction submodule is used to extract the ratio of the negative sequence component to the positive sequence component in the data based on the symmetrical component method to obtain the three-phase imbalance; The iterative submodule is used to truncate the residuals exceeding 3σ in the total harmonic distortion rate, temperature gradient change rate and three-phase imbalance, stop the iteration when the residual variance exceeds the threshold after 10 consecutive iterations, and obtain characteristic electrical dynamic data.

5. The electrical safety intelligent detection and control system suitable for construction engineering according to claim 1, characterized in that: The electrical feature extraction module includes the following units: Convolution kernel unit, the convolution kernel used in the PCN-STAN dual-channel deep network model is initialized using the He normal distribution and the activation function uses PReLU; The attention unit is used to introduce the DCA dynamic channel attention module into the PCN parameterized convolutional network of the model and learn channel weights through the SE-Net concept; The spatiotemporal association unit is used to segment the electrical feature sequence into time windows using the STAN spatiotemporal association network in the PCN-STAN dual-channel deep network model, extract local temporal features using 1D convolution, and obtain the physical connection relationship between devices using the GCN graph convolutional network; The channel classification unit is used to process the current and voltage waveform data using the PCN channel and the temperature and harmonic timing data using the STAN channel.

6. The electrical safety intelligent detection and control system for construction engineering according to claim 1, characterized in that: The electrical feature extraction module also The following units are included: A hyperparameter optimization unit is used to optimize the hyperparameters of the model using the QPSO quantum particle swarm algorithm and adjust the shrinkage factor using a nonlinear decreasing strategy. The hyperparameters include at least the convolution kernel size, network depth, learning rate, and number of attention heads. The hyperparameter update unit is used to apply Lévy flight perturbations to the global optimal particle during the hyperparameter optimization process, encode the hyperparameters into quantum bit probability amplitudes, and update the hyperparameters of the model through quantum rotating gates to obtain the QPSO-PCN-STAN dual-channel deep network model.

7. The electrical safety intelligent detection and control system for construction engineering according to claim 1, characterized in that: The electrical feature extraction module includes the following units: A data input unit, configured to input the characteristic electrical dynamic data into the QPSO-PCN-STAN dual-channel deep network model for training; The data training unit is used to timestamp-align the local features of the current waveform extracted by the PCN channel with the global features of the temperature time series generated by the STAN channel, using interpolation to compensate for sampling frequency differences. The self-attention mechanism is used to map the harmonic distortion rate, three-phase imbalance, and temperature gradient into a unified high-dimensional space. An indicator establishment unit is used to establish a scoring indicator system including at least harmonic mutation index, heat accumulation risk value, three-phase imbalance, insulation degradation trend and load fluctuation coefficient; A hierarchical management and control unit is used to obtain an electrical safety score based on model training, and to divide the electrical safety score into a three-level management and control system, including at least a normal state with a score ≥85, a warning state with a score of 70≤<85, and an alarm state with a score of <70; the system is hierarchically managed based on the three states.

8. A method for implementing an intelligent electrical safety detection and control system for construction projects as claimed in claim 1, characterized in that: The method comprises the following steps: Using the electrical dynamic parameter group in the sensor array network acquisition system, aligning data of different sampling frequencies in the dynamic parameter group using a sliding window algorithm, preprocessing and standardizing the aligned data to obtain initial electrical dynamic data; Deleting noise from the initial electrical dynamic data based on an improved Kalman filter algorithm, extracting characteristic vectors of current waveform harmonic distortion rate, temperature gradient change rate, and three-phase imbalance from the noise-deleted data, and obtaining characteristic electrical dynamic data; Based on the dual-channel architecture of PCN parameterized convolutional network and STAN spatiotemporal association network, a PCN-STAN dual-channel deep network model is established. The QPSO quantum particle swarm algorithm is used to optimize the model's hyperparameters, resulting in a QPSO-PCN-STAN dual-channel deep network model. The characteristic electrical dynamic data is input into the QPSO-PCN-STAN dual-channel deep network model for training to obtain an electrical safety score, and the system is hierarchically controlled based on the electrical safety score.

9. A method for implementing an intelligent electrical safety detection and control system applicable to construction projects as claimed in claim 1, characterized in that: The method comprises the following steps: Build an intelligent sensor array network, where each node contains a high-frequency sampling module, a multi-source synchronization controller, and an adaptive adjustment unit; Using a sensor array network to collect an electrical dynamic parameter group in the system, the electrical dynamic parameter group includes at least current harmonic distortion rate, insulation degradation index, three-phase imbalance, environmental coupling parameters, electromagnetic interference intensity and temperature gradient; Based on the sliding window algorithm, the window width and step size in the window function are defined, a frequency-time mapping relationship is established, cubic spline interpolation is used for high-frequency data, Lagrange polynomial fitting is used for low-frequency data, and a unified timestamp sequence is established for the electrical dynamic parameter group to obtain aligned electrical dynamic data; The aligned data are cleaned of outliers based on the improved Hampel filter, and the cleaned data are standardized to obtain the initial electrical dynamic data.

10. A method for implementing an intelligent electrical safety detection and control system applicable to construction projects as claimed in claim 1, characterized in that: The method comprises the following steps: The model is optimized for hyperparameters using the QPSO quantum particle swarm algorithm, and the shrinkage factor is adjusted using a nonlinear decreasing strategy. The hyperparameters include at least the convolution kernel size, network depth, learning rate, and number of attention heads. During the hyperparameter optimization process, Lévy flight perturbations are applied to the global optimal particle, the hyperparameters are encoded as quantum bit probability amplitudes, and the hyperparameters of the model are updated through quantum rotating gates to obtain the QPSO-PCN-STAN dual-channel deep network model.

Images