A non-intrusive load event detection method, medium and terminal

Abstract

The application is suitable for the technical field of power system management, and relates to a non-intrusive load event detection method, medium and terminal. The application has simple process and convenient operation. Voltage and current signals are collected at a single measurement point to realize accurate detection of running events of various power equipment. S transform is used for time-frequency analysis to extract significant harmonic components. Then, an extended Kalman filter is constructed to dynamically estimate harmonic parameters. Finally, event detection is performed based on multi-feature fusion of filter residuals and power RMS values. The application significantly improves the event detection accuracy and efficiency in a complex load environment, and provides reliable technical support for load monitoring and energy consumption decomposition in an intelligent power system.

Description

Technical Field

[0001] This invention belongs to the field of power system management technology, and particularly relates to a non-intrusive load event detection method, medium and terminal. Background Technology

[0002] As an important component of the power system, the low-voltage distribution network directly serves end users. Its operational safety and stability are crucial to the reliability of power supply. With the development of smart grids, the monitoring and management of low-voltage distribution networks have become key to power system optimization. However, load event detection in low-voltage distribution networks faces many challenges, such as the inability to guarantee data quality, high real-time requirements for event detection, and complex and variable load environments.

[0003] Existing load event detection methods are mostly based on instantaneous power analysis, signal filtering, and time-frequency analysis. While these methods can identify load events to some extent, they still have the following shortcomings: Low accuracy and high false alarm and missed detection rates: Due to the influence of signal noise and nonlinear factors, traditional methods often struggle to achieve high-precision load event detection, resulting in high false alarm and missed detection rates; Difficulty in detecting multiple loads operating simultaneously: Existing methods struggle to handle event detection when multiple loads are operating simultaneously, especially in complex load environments, easily leading to erroneous detection and inefficient event judgment; Insufficient processing capacity, failing to meet the needs of large-scale systems: Traditional event detection methods have high computational complexity, unable to meet the real-time monitoring needs of large-scale power systems, limiting their application in smart grids; Insufficient feature extraction and fusion: Many existing methods fail to fully exploit the multi-dimensional features in signals, especially failing to effectively integrate time-frequency analysis and filtering techniques, resulting in insufficient detection capabilities. Patent application CN115219783A discloses a bilateral sliding window event detection method, which includes connecting an electrical data acquisition device in the circuit to acquire high-frequency current and voltage data at a sampling frequency of 1000 times per second, acquiring four sets of electrical data each time; after data preprocessing, the data is used as a window segment, and calculations and comparisons are performed on each set of data to identify the operating status of the equipment; based on extensive testing of the data, the effective power value is used for detection. This patent application also detects load events through power, and thus suffers from the same drawbacks as existing technologies.

[0004] Therefore, how to provide a load event detection method with high accuracy and efficiency is a problem that urgently needs to be solved by those in this technical field. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a non-invasive load event detection method to solve the problems of low accuracy and efficiency in load event detection in the prior art; in addition, this invention also provides a non-invasive load event detection medium and terminal.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a non-invasive method for detecting load events, comprising the following steps:

[0008] S10. Collect voltage and current signals at a single measurement point in the power system to obtain instantaneous power signals and preprocess the instantaneous power signals.

[0009] S20. The preprocessed instantaneous power signal is subjected to S-transform for time-frequency domain analysis to obtain its time-frequency representation. The harmonic frequency components with significant energy are selected from it. Each harmonic component is extracted in the frequency domain using a frequency-dependent Gaussian window function. The extracted harmonic components are then reconstructed by inverse S-transform to obtain a pure harmonic signal.

[0010] S30. Based on the reconstructed pure harmonic signal, a state-space model containing each harmonic parameter is constructed, and an extended Kalman filter is designed to achieve dynamic estimation of the system state. The filter continuously tracks the changes in harmonic parameters during prediction and updating, and extracts the filter residual sequence as important feature information for event detection.

[0011] S40. Using the filter residual and instantaneous power signal, calculate the RMS value within the sliding window to form multi-scale energy characteristics. Use the peak detection algorithm to analyze the RMS signal at different scales to identify possible event moments. For each detected peak, verify the detection results by comparing the power change amplitude before and after the event and setting a power change threshold, thereby distinguishing real load events from noise or occasional fluctuations.

[0012] S50. Statistically analyze the characteristic information of various load events recorded during the detection process, calculate the detection accuracy, recall rate and F1 score, and evaluate the algorithm's performance in terms of real-time performance and computational complexity.

[0013] Furthermore, in step S10, the instantaneous power signal expression is as follows:

[0014]

[0015] Among them, S n Let ω be the amplitude of the nth harmonic component, and ω0 be the fundamental angular frequency (2π·60Hz). Let be the phase angle of the nth harmonic, M be the maximum harmonic order considered, and ε(t) be the noise and nonlinear interference term.

[0016] Furthermore, the specific steps of step S20 are as follows:

[0017] S201. Harmonic Extraction: Perform an S-transform on the original signal p(t) to transform the signal from the time domain to the time-frequency domain, obtaining the time-frequency representation S(τ, f); the formula for the S-transform is as follows:

[0018] S202. Significant Harmonic Selection: Select the top M significant harmonic frequencies f with the largest power in the time-frequency representation S(τ, f). n The selection strategy is to choose the frequency that contributes the most based on the peak value of the spectrum;

[0019] S203. Signal Reconstruction: In the frequency domain, the Gaussian window is moved to the target frequency position through a spectrum shifting operation. The shifted Gaussian window is multiplied with the S-transform result to extract the time-domain signal of the corresponding frequency component. All extracted harmonic components are superimposed and reconstructed to obtain a pure signal containing the main harmonic components.

[0020] Furthermore, the specific steps in step S30 are as follows:

[0021] S301. State-space model construction: Establish a state vector containing the sine and cosine coefficients of all harmonics and the DC component. The state transition process of the system is described by the state equation, indicating how the current state evolves to the next moment. The relationship between the state and the observation is described by the observation equation, indicating that the variables in the state space are mapped to the measurement space.

[0022] S302, Extended Kalman Filter: In the prediction step, the state at the current time is first predicted based on the state estimate of the previous time step, and the error covariance matrix of the state estimate is predicted at the same time. Then, the optimal Kalman gain is calculated based on the difference between the new observation value and the predicted value.

[0023] S303, Adaptive Mechanism and Residual Output: An adaptive update mechanism for the measurement noise covariance is introduced, and the process noise covariance also adopts an adaptive update method.

[0024] Furthermore, the specific steps of step S40 are as follows:

[0025] S401, RMS Calculation: Perform sliding window RMS calculation on the power signal and RMS calculation on the residual signal of the Kalman filter.

[0026] S402, Peak Detection: Calculate the energy within the target window of the RMS signal, the energy in the neighborhood below the target window, and the energy in the neighborhood above the target window. When the energy of the target window is significantly higher than the energy of the neighborhood, it is determined to be a peak.

[0027] S403. Event Verification: For each detected peak location, calculate the average power change before and after the event, and set a relative change threshold to confirm the event.

[0028] Furthermore, step S50 specifically includes:

[0029] Detection performance metrics: Calculate the accuracy of event detection, which is the proportion of correctly detected events to the total number of detected events; calculate the recall of event detection, which is the proportion of correctly detected events to the total number of actual events; combine accuracy and recall to calculate the F1 score to evaluate detection performance.

[0030] Time performance evaluation: Calculate the average latency of event detection, quantify the real-time processing capability of the detection system, and calculate the average processing time for each sampling point.

[0031] In a second aspect, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0032] Thirdly, the present invention also provides an electronic terminal, comprising: a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to cause the terminal to perform the method described above.

[0033] Compared with existing technologies, the non-invasive load event detection method, medium, and terminal provided by this invention have at least the following advantages:

[0034] Traditional event detection methods suffer from low accuracy and efficiency. This invention, with its simple process and convenient operation, achieves accurate detection of various electrical equipment operation events by acquiring voltage and current signals at a single measurement point. The core of the method is to utilize S-transform for time-frequency analysis to extract significant harmonic components, then construct an extended Kalman filter to dynamically estimate harmonic parameters, and finally perform multi-feature fusion for event detection based on filter residuals and power RMS values. This method significantly improves the accuracy and efficiency of event detection under complex load environments, providing reliable technical support for load monitoring and energy consumption decomposition in smart power systems. This invention combines S-transform and extended Kalman filtering to more accurately extract and estimate the harmonic components of load signals, thereby improving the accuracy of event detection. It can cope with noise, interference, and nonlinear changes in the load environment, enhancing the robustness of the algorithm. The use of real-time state estimation and feature fusion ensures the real-time performance of the algorithm, making it suitable for the load monitoring needs of practical power systems. Attached Figure Description

[0035] To more clearly illustrate the solution of the present invention, a brief introduction will be given to the drawings used in the description of the embodiments below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 This is a flowchart of a non-invasive load event detection method provided in an embodiment of the present invention. Detailed Implementation

[0037] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0039] This invention provides a non-invasive load event detection method, applied to the detection of non-invasive load events in complex load environments. The non-invasive load event detection method includes the following steps:

[0040] S10. Collect voltage and current signals at a single measurement point in the power system to obtain instantaneous power signals and preprocess them; S20. Perform time-frequency domain analysis on the preprocessed instantaneous power signals using S-transform to obtain their time-frequency representation, select significant harmonic frequency components, extract each harmonic component in the frequency domain using a frequency-dependent Gaussian window function, and then reconstruct the extracted harmonic components using inverse S-transform to obtain pure harmonic signals; S30. Based on the reconstructed pure harmonic signals, construct a state-space model containing each harmonic parameter, and design an extended Kalman filter to achieve dynamic estimation of the system state. The filter continuously tracks the changes in harmonic parameters during prediction and updating. Simultaneously, the filtered residual sequence is extracted as an important feature information for event detection; S40, using the filtered residual and instantaneous power signal, the RMS value within the sliding window is calculated to form multi-scale energy features. The peak detection algorithm is used to analyze the RMS signal at different scales to identify possible event moments. For each detected peak, the power change amplitude before and after the event is compared, and a power change threshold is set to verify the detection results, thereby distinguishing real load events from noise or occasional fluctuations; S50, the feature information of various load events recorded during the detection process is statistically analyzed, and the detection accuracy, recall, and F1 score are calculated. At the same time, the performance of the algorithm in terms of real-time performance and computational complexity is evaluated.

[0041] This invention features a simple process and convenient operation. By acquiring voltage and current signals at a single measurement point, it achieves accurate detection of various electrical equipment operation events. Time-frequency analysis using S-transform is employed to extract significant harmonic components. Then, an extended Kalman filter is constructed to dynamically estimate harmonic parameters. Finally, event detection is performed based on multi-feature fusion using the filter residual and power RMS value. This invention significantly improves the accuracy and efficiency of event detection under complex load environments, providing reliable technical support for load monitoring and energy consumption decomposition in intelligent power systems.

[0042] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0043] This invention provides a non-invasive load event detection method, applied to the detection of non-invasive load events in complex load environments, such as... Figure 1 As shown, in this embodiment, the non-invasive load event detection method includes the following steps:

[0044] S10. Data Preprocessing: At a single measurement point in the power system, voltage and current signals are synchronously acquired, and the instantaneous power signal is calculated using their product. This signal serves as the basis for subsequent analysis. By preprocessing the acquired signal (such as noise reduction and normalization), the data quality is ensured to meet the requirements of subsequent S-transform and state estimation, thereby providing accurate raw input for the system.

[0045] Specifically, in this embodiment, in the voltage signal v(t) and current signal i(t) acquired at a single measurement point, changes in the load's on / off states will produce characteristic changes in the instantaneous power signal. These changes can be described by the following mathematical model:

[0046]

[0047] Among them, S n Let ω be the amplitude of the nth harmonic component, and ω0 be the fundamental angular frequency (2π·60Hz). Let be the phase angle of the nth harmonic, M be the maximum harmonic order considered, and ε(t) be the noise and nonlinear interference term.

[0048] S20. Harmonic Analysis and Reconstruction: The preprocessed instantaneous power signal is analyzed in the time-frequency domain using S-transform to obtain its time-frequency representation. The harmonic frequency components with significant energy are selected, and each harmonic component is extracted in the frequency domain using a frequency-dependent Gaussian window function. The extracted harmonic components are then reconstructed using inverse S-transform to obtain a clean harmonic signal. This step effectively removes noise interference from the original signal, highlights the main harmonic characteristics, and provides a high-quality signal input for subsequent state estimation.

[0049] Specifically, in this embodiment, step S20 consists of the following steps:

[0050] S201. Harmonic Extraction: First, perform an S-transform on the original signal p(t) to transform the signal from the time domain to the time-frequency domain, obtaining the time-frequency representation S(τ, f). The formula for the S-transform is as follows:

[0051]

[0052] Where g(tT, f) is a frequency-dependent Gaussian window:

[0053]

[0054] Where f is the frequency, τ is the time shift parameter representing the time window for time-frequency analysis, i is the imaginary unit, p(t) is the instantaneous power signal, and t is the actual signal time. S(τ, f) is a two-dimensional matrix describing the energy distribution of the signal at different time points and frequencies.

[0055] S202. Significant Harmonic Selection: Select the top M significant harmonic frequencies f with the largest power in the time-frequency representation S(τ, f). n These frequencies correspond to the dominant frequency components of the signal. The selection strategy is to choose the frequency that contributes the most based on the peak values ​​of the spectrum. The selection process can be expressed as:

[0056]

[0057] Among them, f n Let S(τ, f) be the nth significant harmonic frequency, M be the number of significant frequencies selected, and |S(τ, f)|| 2 Let f be the energy of the signal at frequency f and time τ.

[0058] S203, Signal Reconstruction: In the frequency domain, the Gaussian window is moved to the target frequency position through a spectrum shifting operation. The shifted spectrum can be represented as:

[0059]

[0060] The time-domain signal of the corresponding frequency component is extracted by multiplying the shifted Gaussian window with the S-transform result.

[0061] p harmonic (t,f n )=S(τ,f n )·g shifted (t,f n );

[0062] Finally, all extracted harmonic components are superimposed and reconstructed to obtain a pure signal containing the main harmonic components:

[0063]

[0064] The reconstructed signal p harmonic (t) preserves the main harmonic features of the original signal while suppressing noise interference, providing a high-quality input for subsequent state estimation.

[0065] S30. State Estimation: Based on the reconstructed pure harmonic signal, a state-space model containing various harmonic parameters (such as sine and cosine coefficients and DC components) is constructed, and an extended Kalman filter is designed to achieve dynamic estimation of the system state. The filter continuously tracks the changes in harmonic parameters during prediction and updating, and extracts the filter residual sequence as important feature information for event detection. This step effectively captures the changes in harmonic parameters caused by load events through real-time dynamic modeling, providing a reliable state estimation basis for event determination.

[0066] Specifically, in this embodiment, step S30 consists of the following steps:

[0067] S301. State-space model construction: First, establish the state vector, which contains the sine and cosine coefficients of all harmonics and the DC component:

[0068] x k =[A 1，k B 1，k A 2，k B 2，k A M，k B M，k C k ] T ;

[0069] Among them, A n，k B n，k These are the sine and cosine coefficients of the nth harmonic, respectively, C. k M represents the DC component, M represents the number of harmonics considered, and k represents the time index.

[0070] Furthermore, in this embodiment, the system's state transition process is described by a state equation, which indicates how the current state evolves to the next moment:

[0071] x k+1 =Φx k +w k ;

[0072] Where Φ is the state transition matrix, w k It is process noise that follows a zero-mean Gaussian distribution.

[0073] The relationship between state and observation is given by the observation equation, which maps variables in the state space to the measurement space:

[0074] p k =Hx k +v k ;

[0075] Where H is the observation matrix, p k It is the power measurement value at time k, v k It measures noise.

[0076] S302, Extended Kalman Filter: In the prediction step, the current state is first predicted based on the state estimate from the previous time step.

[0077]

[0078] in, Let k be the prior state estimate at time k. This is the posterior state estimate at time k-1.

[0079] Simultaneously predict the error covariance matrix of the state estimate:

[0080] P k|k-1 =ΦP k-1|k-1 Φ T +Q k ;

[0081] Among them, P k|k-1 Let P be the prior error covariance matrix. k-1|k-1 Let Q be the posterior error covariance matrix of the previous time step. k Let be the process noise covariance matrix.

[0082] Furthermore, in this embodiment, the optimal Kalman gain is calculated based on the difference between the new observation and the predicted value:

[0083] K k =P k|k-1 H T HP k|k-1 H T +R k ) -1 ;

[0084] Among them, K k For Kalman gain, R k To measure the noise covariance matrix.

[0085] The optimal state estimate is obtained by correcting the state estimate using Kalman gain.

[0086]

[0087] in, Let k be the posterior state estimate at time k. To measure the residual.

[0088] Update the error covariance matrix of the state estimate:

[0089] P k|k =(IK k H)P k|k-1 ;

[0090] Among them, P k|k Let I be the posterior error covariance matrix at time k, and let I be the identity matrix.

[0091] S303, Adaptive Mechanism and Residual Output: Introducing an adaptive update mechanism for the measurement noise covariance:

[0092]

[0093] Where α is the smoothing factor (with a value of 0.15), R k-1 This is the measurement noise covariance matrix of the previous time step.

[0094] Similarly, the process noise covariance is also updated adaptively:

[0095]

[0096] Among them, Q k-1 Let be the process noise covariance matrix of the previous time step.

[0097] In the Kalman filtering process, special attention is paid to the measurement residual sequence:

[0098]

[0099] Where, ∈ k The residual sequence is used to measure the residual. This residual sequence is particularly sensitive to abrupt changes in the power signal and is an important feature for subsequent event detection.

[0100] S40. Event Detection and Confirmation: Using the filtered residual and instantaneous power signal, the root mean square (RMS) value within the sliding window is calculated to form multi-scale energy characteristics. A peak detection algorithm is used to analyze the RMS signals at different scales to identify possible event moments. For each detected peak, the detection results are further verified by comparing the power change amplitude before and after the event and setting a power change threshold. This distinguishes real load events from noise or occasional fluctuations, ensuring the accuracy and robustness of the detection results.

[0101] Specifically, in this embodiment, step S40 consists of the following steps:

[0102] S401, RMS Calculation: Event detection is based on two key signals: a power signal and a residual signal. First, a sliding window RMS calculation is performed on the power signal to evaluate the effective value changes of the signal.

[0103]

[0104] Where N is the window length (usually 1-2 power grid cycles), p i Let be the instantaneous power value at time i, and k be the current time.

[0105] Similarly, RMS calculations are performed on the residual signal of the Kalman filter to capture anomalous changes in state estimation:

[0106]

[0107] Where, ∈ i Let be the residual value at time i.

[0108] S402, Peak Detection: Calculation of energy within the target window of the RMS signal:

[0109]

[0110] Where w is the target window width (usually 3-5 power grid cycles), E target The energy within the target window.

[0111] Furthermore, in this embodiment, the energy in the region below the target window is calculated:

[0112]

[0113] Where l is the neighborhood range length (usually taken as 5 times the width of the target window).

[0114] Calculate the energy of the neighborhood above the target window:

[0115]

[0116] When the energy of the target window is significantly higher than that of its neighbors, it is considered a peak value.

[0117] E target >γ(E inferior +E superior );

[0118] Wherein, γ is the decision threshold (usually taken as 0.5-0.7).

[0119] S403. Event Verification: For each detected peak location, calculate the average power change before and after the event.

[0120] ΔP=|P post -P pre |;

[0121] Among them, P pre Let P be the average power over the N periods preceding the event. post The average power over N periods after the event.

[0122] Set a relative change threshold for event confirmation:

[0123] ΔP>θP pre ;

[0124] Here, θ is the relative change threshold (usually taken as 0.1-0.2), and only peak values ​​where the power change exceeds this threshold are considered valid events.

[0125] S50. Performance Evaluation: Statistical analysis of the characteristic information of various load events recorded during the detection process is performed to calculate performance indicators such as detection accuracy, recall, and F1 score. At the same time, the performance of the algorithm in terms of real-time performance and computational complexity is evaluated.

[0126] Specifically, in this embodiment, step S50 includes:

[0127] Detection performance metrics: First, calculate the event detection accuracy, which is the proportion of correctly detected events out of the total number of detected events.

[0128]

[0129] Where TP (True Positive) represents the number of correctly detected events, and FP (False Positive) represents the number of false alarms.

[0130] Calculate the recall rate of event detection, which is the proportion of correctly detected events out of the total number of actual events:

[0131]

[0132] Here, FN (False Negative) represents the number of missed events.

[0133] The F1 score is calculated by combining accuracy and recall to evaluate detection performance.

[0134]

[0135] Furthermore, in this embodiment, the time performance is evaluated by calculating the average latency of event detection.

[0136]

[0137] Where, N e t represents the total number of events. detect，i Let t be the detection time of the i-th event. true，i Let be the actual time when the i-th event occurs.

[0138] To quantify the real-time processing capability of the detection system, the average processing time for each sampling point is calculated.

[0139]

[0140] Where N is the total number of sampling points, t k The time required to process the kth sampling point.

[0141] The following example, using the electricity usage of a resident in a smart community, illustrates the specific implementation process of the non-intrusive load event detection method provided in this embodiment of the invention. The resident's household is equipped with a single-phase smart meter with a sampling frequency of 6.4kHz, a voltage range of 0-380V, a current range of 0-60A, and a resolution of 0.1%. The test duration is 1 hour, including the switching operations of typical household appliances such as air conditioners and electric water heaters.

[0142] Step 1: Data Preprocessing

[0143] The data acquisition parameter settings are shown in Table 1:

[0144] Table 1

[0145] sampling frequency 6.4kHz Standard smart meter sampling frequency Voltage range 0-380V Single-phase smart meter voltage range Current range 0-60A Single-phase smart meter current range resolution 0.1％ Measurement accuracy Grid frequency 50Hz my country's power grid standard frequency Points per cycle 128 6400Hz / 50Hz

[0146] Raw data records: An example of raw data (sampling interval 156.25μs, partial data) is shown in Table 2:

[0147] Table 2

[0148] 0.000 220.5 0.52 114.66 0.156 218.3 0.51 111.33 0.312 215.8 0.50 107.90 0.468 212.4 0.49 104.08

[0149] Step 2: Harmonic Analysis and Reconstruction

[0150] The S-transform parameter settings are shown in Table 3:

[0151] Table 3

[0152] power signal frequency 2×Grid Frequency 100Hz Gaussian window width 2× power signal frequency 200Hz Harmonic range 1 / 2 × power signal frequency 50Hz Significant harmonic quantity Selection based on frequency domain amplitude 10

[0153] The harmonic analysis results are shown in Table 4. The characteristics of the main harmonic components are as follows:

[0154] Table 4

[0155] 1 50 215.6 0.52 3 150 42.3 1.57 5 250 11.8 2.36 7 350 5.2 3.14

[0156] State estimation: The Kalman filter parameters are shown in Table 5:

[0157] Table 5

[0158] State vector dimension 2 × harmonic number + 1 21 Measurement noise standard deviation maximum signal value × 1e-2 5W Process noise standard deviation identity matrix 1 Initial state covariance maximum signal value × 1e-2 × identity matrix diag(100) Adaptive factor α The plan stipulates 0.15

[0159] Step 4: Event Detection Settings and Results

[0160] The RMS calculation parameters are shown in Table 6:

[0161] Table 6

[0162] Calculation window One power grid cycle (50Hz) 128 points Movement step size 1 / 4 cycle 32 points Resolution mode full_cycle -

[0163] The peak detection parameters are shown in Table 7:

[0164] Table 7

[0165] Factor Recommended Solution 5 Width Recommended Solution 3 Power change threshold 10% of the reference power 100W relative change threshold Recommended Solution 0.1

[0166] The event detection log is shown in Table 8:

[0167] Table 8

[0168] 300.15 Air conditioner on 2180 125 150 600.12 Water heater turned on 1985 115 120 900.18 Air conditioner off -2175 128 180 1200.14 Water heater off -1980 112 140

[0169] Step 5: Performance Evaluation Results

[0170] Detection performance metrics: The algorithm performance evaluation is shown in Table 9:

[0171] Table 9

[0172] Accuracy TP / (TP+FP) 100％ Recall TP / (TP+FN) 100％ F1 score 2×(P×R) / (P+R) 1.0 Average detection delay Σ(Detection time - Actual time) / Number of events 147.5ms Single point processing time Total processing time / Number of sampling points 0.85ms Real-time processing margin (1 - processing time / sampling period) × 100% 86.9％

[0173] Experimental results show that the non-intrusive load event detection method provided in this embodiment of the invention can accurately detect the switching events of household appliances under the conditions of a sampling frequency of 6.4kHz and a power grid frequency of 50Hz. The detection delay is less than 200ms, and the computational complexity is moderate, making it suitable for implementation in smart meters.

[0174] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods in this embodiment.

[0175] This invention also provides an electronic terminal, including: a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal performs any of the methods in this embodiment.

[0176] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0177] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0178] Compared with existing technologies, traditional event detection methods have lower accuracy and efficiency in the non-intrusive load event detection method, medium, and terminal described in the above embodiments. This invention features a simple process and convenient operation. By acquiring voltage and current signals at a single measurement point, it achieves accurate detection of various electrical equipment operation events. The core of the method is to use S-transform for time-frequency analysis to extract significant harmonic components, then construct an extended Kalman filter to dynamically estimate harmonic parameters, and finally perform multi-feature fusion event detection based on filter residuals and power RMS values. This method significantly improves the accuracy and efficiency of event detection in complex load environments, providing reliable technical support for load monitoring and energy consumption decomposition in smart power systems. This invention combines S-transform and extended Kalman filtering to more accurately extract and estimate the harmonic components of load signals, thereby improving the accuracy of event detection. It can cope with noise, interference, and nonlinear changes in the load environment, enhancing the robustness of the algorithm. The use of real-time state estimation and feature fusion ensures the real-time performance of the algorithm, making it suitable for the load monitoring needs of actual power systems.

[0179] Obviously, the embodiments described above are merely preferred embodiments of the present invention, and not all embodiments. The accompanying drawings illustrate preferred embodiments of the present invention, but do not limit the scope of the patent. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this invention.

Claims

1. A non-invasive method for detecting load events, characterized in that, Includes the following steps: S10. Collect voltage and current signals at a single measurement point in the power system to obtain instantaneous power signals and preprocess the instantaneous power signals. S20. The preprocessed instantaneous power signal is subjected to S-transform for time-frequency domain analysis to obtain its time-frequency representation. The harmonic frequency components with significant energy are selected from it. Each harmonic component is extracted in the frequency domain using a frequency-dependent Gaussian window function. The extracted harmonic components are then reconstructed by inverse S-transform to obtain a pure harmonic signal. S30. Based on the reconstructed pure harmonic signal, a state-space model containing each harmonic parameter is constructed, and an extended Kalman filter is designed to achieve dynamic estimation of the system state. The filter continuously tracks the changes in harmonic parameters during prediction and updating, and extracts the filter residual sequence as important feature information for event detection. S40. Using the filter residual and instantaneous power signal, calculate the RMS value within the sliding window to form multi-scale energy characteristics. Use the peak detection algorithm to analyze the RMS signal at different scales to identify possible event moments. For each detected peak, verify the detection results by comparing the power change amplitude before and after the event and setting a power change threshold, thereby distinguishing real load events from noise or occasional fluctuations. S50. Statistically analyze the characteristic information of various load events recorded during the detection process, calculate the detection accuracy, recall rate and F1 score, and evaluate the algorithm's performance in terms of real-time performance and computational complexity.

2. The non-invasive load event detection method according to claim 1, characterized in that, In step S10, the instantaneous power signal expression is as follows: Among them, S n Let ω be the amplitude of the nth harmonic component, and ω0 be the fundamental angular frequency (2π·60Hz). Let be the phase angle of the nth harmonic, M be the maximum harmonic order considered, and ε(t) be the noise and nonlinear interference term.

3. The non-invasive load event detection method according to claim 2, characterized in that, The specific steps of step S20 are as follows: S201. Harmonic Extraction: Perform an S-transform on the original signal p(t) to transform the signal from the time domain to the time-frequency domain, obtaining the time-frequency representation S(τ,f); the formula for the S-transform is as follows: S202. Significant Harmonic Selection: Select the top M significant harmonic frequencies f with the largest power in the time-frequency representation S(τ, f). n The selection strategy is to choose the frequency that contributes the most based on the peak value of the spectrum; S203. Signal Reconstruction: In the frequency domain, the Gaussian window is moved to the target frequency position through a spectrum shifting operation. The shifted Gaussian window is multiplied with the S-transform result to extract the time-domain signal of the corresponding frequency component. All extracted harmonic components are superimposed and reconstructed to obtain a pure signal containing the main harmonic components.

4. The non-invasive load event detection method according to claim 1, characterized in that, The specific steps in step S30 are as follows: S301. State-space model construction: Establish a state vector containing the sine and cosine coefficients of all harmonics and the DC component. The state transition process of the system is described by the state equation, indicating how the current state evolves to the next moment. The relationship between the state and the observation is described by the observation equation, indicating that the variables in the state space are mapped to the measurement space. S302, Extended Kalman Filter: In the prediction step, the state at the current time is first predicted based on the state estimate of the previous time step, and the error covariance matrix of the state estimate is predicted at the same time. Then, the optimal Kalman gain is calculated based on the difference between the new observation value and the predicted value. S303, Adaptive Mechanism and Residual Output: An adaptive update mechanism for the measurement noise covariance is introduced, and the process noise covariance also adopts an adaptive update method.

5. The non-invasive load event detection method according to claim 1, characterized in that, The specific steps of step S40 are as follows: S401, RMS Calculation: Perform sliding window RMS calculation on the power signal and RMS calculation on the residual signal of the Kalman filter. S402, Peak Detection: Calculate the energy within the target window of the RMS signal, the energy in the neighborhood below the target window, and the energy in the neighborhood above the target window. When the energy of the target window is significantly higher than the energy of the neighborhood, it is determined to be a peak. S403. Event Verification: For each detected peak location, calculate the average power change before and after the event, and set a relative change threshold to confirm the event.

6. The non-invasive load event detection method according to claim 1, characterized in that, Step S50 specifically includes: Detection performance metrics: Calculate the accuracy of event detection, which is the proportion of correctly detected events to the total number of detected events; calculate the recall of event detection, which is the proportion of correctly detected events to the total number of actual events; combine accuracy and recall to calculate the F1 score to evaluate detection performance. Time performance evaluation: Calculate the average latency of event detection, quantify the real-time processing capability of the detection system, and calculate the average processing time for each sampling point.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

8. An electronic terminal, characterized in that, include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to cause the terminal to perform the method as described in any one of claims 1 to 6.

Images