A power distribution cabinet energy consumption monitoring method, system and storage medium

Abstract

The application relates to the technical field of power monitoring and Internet of Things, and discloses a power distribution cabinet energy consumption monitoring method, system and storage medium. The method comprises the following steps: synchronously sampling currents of all circuits through an Internet of Things device to obtain original time domain sequences; performing time-frequency analysis to obtain time-frequency spectrum matrices of the circuits; determining a circuit dominant frequency range according to time-frequency spectrum energy distribution and selecting energy consumption characteristic sequences; aligning the characteristic sequences of all circuits according to time to construct a unified characteristic data set; performing power spectrum density analysis based on the data set to obtain overall energy consumption distribution information; and judging abnormal fluctuations according to the energy consumption distribution information and generating an early warning report when an abnormality is detected. The application effectively improves the accuracy and timeliness of power distribution cabinet energy consumption monitoring.

Description

Technical Field

[0001] This application relates to the fields of power monitoring and Internet of Things technology, and in particular to a method, system and storage medium for monitoring the energy consumption of a power distribution cabinet. Background Technology

[0002] In modern power systems, distribution cabinets, as core equipment for power distribution and protection, directly affect the stability and security of power supply. With the popularization of Internet of Things (IoT) technology, distribution cabinet management based on intelligent monitoring has become a key direction for industry development, especially in the field of multi-circuit energy consumption monitoring and anomaly early warning. Accurately grasping the power consumption and potential risks of each circuit is particularly important. Research in this area is not only related to the operational efficiency of power systems, but also closely related to the safety of industrial production and daily life.

[0003] However, current methods for multi-circuit monitoring of distribution cabinets often struggle to meet the data acquisition and analysis needs in complex environments, especially when multiple circuits are operating simultaneously, making it difficult to guarantee data synchronization and integrity. Many technologies, when processing information from multiple circuits, are prone to errors in judging energy consumption distribution due to time discrepancies or signal interference, thus affecting the timely detection of anomalies. This limitation often leaves distribution cabinet management in a reactive state when facing unexpected problems.

[0004] A deeper technical challenge lies in extracting crucial information that truly reflects energy consumption characteristics from data from multiple circuits. Each circuit's current signal contains different frequency components and fluctuation patterns, and these signals influence each other, increasing the complexity of the analysis. Without accurately separating and identifying the frequency ranges representing the main energy consumption characteristics, it's impossible to fully grasp the overall energy consumption distribution and determine which circuit is experiencing anomalies. For example, in a distribution cabinet, the current signal of a certain circuit may exhibit irregular fluctuations due to equipment aging, but these fluctuations may be masked by normal signals from other circuits, causing the problem to be overlooked and potentially leading to greater fault risks.

[0005] Therefore, how to accurately extract and analyze the energy consumption characteristics of each circuit in a multi-circuit environment, while identifying hidden abnormal fluctuations and accurately locating the problem circuit, has become a key issue in the intelligent monitoring and management of power distribution cabinets. Summary of the Invention

[0006] This application provides a method, system, and storage medium for monitoring the energy consumption of power distribution cabinets, which can improve the accuracy and timeliness of energy consumption monitoring of power distribution cabinets.

[0007] In a first aspect, this application provides a method for monitoring the energy consumption of a power distribution cabinet, the method comprising: S1. High-speed synchronous sampling of current signals of each circuit in the power distribution cabinet is performed through IoT acquisition devices to obtain the original time-domain sampling sequence data of each circuit; S2. Perform time-frequency analysis on the original time-domain sampling sequence data to obtain the time-frequency spectrum matrix corresponding to each loop; S3. Based on the energy distribution characteristics of the time-frequency matrix, determine the dominant frequency range of each loop, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range; S4. Align the energy consumption characteristic sequences of all circuits according to the time axis to form a unified characteristic dataset of multiple circuits in the distribution cabinet. S5. Perform power spectral density analysis based on the unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet; S6. Determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and generate an early warning monitoring report when an abnormality is detected.

[0008] Secondly, this application provides a system for monitoring the energy consumption of a power distribution cabinet, the system comprising: The data acquisition module is used to perform high-speed synchronous sampling of the current signals of each circuit in the power distribution cabinet through IoT acquisition devices to obtain the original time-domain sampling sequence data of each circuit. The time-frequency analysis module is used to perform time-frequency analysis on the original time-domain sampling sequence data to obtain the time-frequency spectrum matrix corresponding to each loop; The frequency extraction module is used to determine the dominant frequency range of each loop based on the energy distribution characteristics of the time-spectrum matrix, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range. The feature alignment module is used to align the energy consumption feature sequences of all circuits according to the time axis to form a unified feature dataset of multiple circuits in the distribution cabinet. The energy consumption analysis module is used to perform power spectral density analysis based on the unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet; The anomaly warning module is used to determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and to generate an early warning monitoring report when an anomaly is detected.

[0009] Thirdly, this application provides a computer device comprising: a memory and at least one processor, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via a bus, and when the machine-readable instructions are executed by the processor, the steps of the above-described power distribution cabinet energy consumption monitoring method are performed.

[0010] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned method for monitoring the energy consumption of a power distribution cabinet.

[0011] Compared with the prior art, the beneficial effects of the present invention are at least as follows: 1. By using high-speed synchronous sampling through IoT acquisition devices, the timing accuracy and data integrity of multi-loop current signals are ensured, solving the data synchronization problem caused by time deviation or signal interference in traditional methods.

[0012] 2. Through time-frequency analysis and frequency extraction technology, this invention can accurately identify the dominant frequency range of each circuit, and then select a representative energy consumption characteristic sequence, thus avoiding the problem of not being able to accurately distinguish the energy consumption fluctuations of different circuits in traditional methods.

[0013] 3. This method enables real-time monitoring of overall energy consumption distribution and automatically generates early warning reports when abnormal fluctuations are detected, ensuring that potential fault risks can be detected in a timely manner during the operation of the distribution cabinet and improving the safety of the power system.

[0014] 4. By aligning the energy consumption characteristic sequences of each circuit according to the time axis to form a unified dataset, the multi-circuit energy consumption analysis of the distribution cabinet becomes more efficient, providing reliable basic data support for subsequent energy consumption analysis and optimization. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for monitoring the energy consumption of a power distribution cabinet according to this application; Figure 2 The following are power spectral density analysis diagrams for each loop in this application; Figure 3 This is an analysis diagram of the overall energy consumption distribution of this application versus that of a normal template; Figure 4 This is a schematic diagram of the structure of a power distribution cabinet energy consumption monitoring system according to this application; Figure 5 This is a schematic block diagram of the structure of a power distribution cabinet energy consumption monitoring device according to this application. Detailed Implementation

[0017] This application provides a method, system, and storage medium for monitoring the energy consumption of a power distribution cabinet. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of a method for monitoring the energy consumption of a power distribution cabinet in this application includes: Step S1: Use an IoT acquisition device to perform high-speed synchronous sampling of the current signals of each circuit in the power distribution cabinet to obtain the original time-domain sampling sequence data of each circuit.

[0019] In one specific embodiment, the process of performing step S1 may specifically include the following steps: The main control unit of the IoT acquisition device generates a unified sampling clock signal and synchronously distributes the sampling clock signal to the corresponding acquisition channel of each loop. Control each acquisition channel to start analog-to-digital conversion at the same time, and acquire the instantaneous current value of each circuit in parallel; The continuously acquired instantaneous current values ​​are arranged in chronological order to generate the original time-domain sampling sequence data of each loop that is strictly aligned on the time axis.

[0020] Specifically, the IoT data acquisition device is deployed inside or near the power distribution cabinet. Its main control unit incorporates a high-precision temperature-compensated crystal oscillator to generate a unified sampling clock signal with a frequency stability of ±0.5ppm. This sampling clock signal is simultaneously distributed to the corresponding acquisition channels of each loop via a low-voltage differential signal bus. The clock transmission lines use an equal-length wiring method to ensure that the phase difference of the clock signal received by each channel is less than 1 nanosecond, thereby ensuring that all acquisition channels strictly share the same clock source and that the clock edges are aligned. Each acquisition channel is independently configured with a Hall effect-based closed-loop current sensor and a 16-bit successive approximation analog-to-digital converter. The Hall effect current sensor linearly converts the AC current in the loop from 0 to 100 amps into a voltage signal from 0 to 5 volts, with a conversion accuracy of ±0.5% of full scale. This voltage signal is then filtered by a second-order Butterworth anti-aliasing filter to remove high-frequency noise before being input to the analog-to-digital converter. The cutoff frequency of the anti-aliasing filter is set to 0.4 times the sampling frequency to prevent high-frequency signals from aliasing into the low-frequency band and affecting measurement accuracy. The main control unit controls each acquisition channel to initiate analog-to-digital conversion simultaneously via a synchronous trigger pulse. This trigger pulse is generated by the main control unit's timer module, with a rise time difference of less than 10 nanoseconds, and is sent in parallel to all acquisition channels via a dedicated trigger bus. Upon receiving the rising edge of the trigger pulse, the analog-to-digital converter of each acquisition channel immediately samples the current input voltage and initiates a successive approximation conversion process. The conversion time is 1 microsecond. After conversion, an interrupt signal is generated to notify the main control unit to read the data. Through this mechanism, all channels achieve true parallel acquisition of the instantaneous current value of their respective circuits, completely eliminating the inherent time misalignment problem of traditional polling acquisition methods. The analog-to-digital converter discretizes the continuous current signal at a sampling frequency of 12800Hz. This sampling frequency is chosen based on the Nyquist sampling theorem, which can completely retain frequency components up to 6400Hz in the current signal, sufficient to cover the common power frequency fundamental wave and its higher harmonic components in power systems. Each sampling moment yields a 16-bit precise digital current amplitude, with a quantization resolution of 1 / 65536 of full scale, capable of accurately capturing minute current fluctuations. Taking the actual deployment of a power distribution cabinet in an industrial park as an example, this cabinet contains eight outgoing circuits, supplying power to the injection molding workshop, stamping workshop, assembly workshop, air compressor station, lighting system, air conditioning system, office area, and backup line, respectively. An IoT data acquisition device continuously collects current signals from each circuit in parallel at a sampling rate of 12800Hz, generating 12800 current amplitude data points per second for each circuit. Taking the injection molding workshop circuit as an example, its effective current value during normal operation is approximately 85 amps, corresponding to a standard sinusoidal instantaneous current waveform with a peak current of approximately 120 amps. Within a continuous sampling time of one second, this circuit obtains 12800 sampling points, which are arranged chronologically to form the original time-domain sampling sequence data of this circuit. Similarly, the other seven circuits also each obtain 12800 sampling points within the same second.Because the sampling times of each channel are strictly synchronized, the 5000th sampling point of loop 1 corresponds to the same physical time as the 5000th sampling point of loop 2. Therefore, the original time-domain sampling sequence data of all loops are strictly aligned on the time axis. During an actual monitoring process, at 10:23:15 AM, the stamping workshop loop experienced an inrush current due to the startup of a large stamping machine. The amplitude of the sampling point at the corresponding moment in its sequence data jumped instantaneously from 45 amperes during normal operation to 210 amperes, and then decayed back to the normal range within 0.2 seconds. Because the data of each loop is strictly aligned, by comparing the sampling data of other loops at the same time, it was found that no abnormal fluctuations were observed in the injection molding workshop loop, assembly workshop loop, etc., thus accurately determining that the inrush current only occurred in the stamping workshop loop and other loops were unaffected. If a traditional asynchronous acquisition method is used, there will be millisecond-level deviations in the sampling times of each loop, which may lead to misjudging the impact of the stamping workshop loop as multiple loops being abnormal simultaneously, or failing to accurately locate the source of the anomaly. Through the aforementioned high-speed synchronous sampling process, high-precision and highly consistent multi-loop current data were obtained. The original time-domain sampling sequence data of each loop completely preserved the time-domain waveform information of the current signal, including characteristics such as amplitude, phase, and distortion. Moreover, the data of each loop naturally possesses time alignment characteristics. This data foundation effectively solves the technical problem of difficulty in joint analysis of multi-loop data caused by inconsistent sampling times in traditional acquisition methods. It provides reliable input data for accurately identifying the timing of current abrupt changes and harmonic distortion frequency bands in subsequent time-frequency analysis, ensuring the accuracy of feature extraction and anomaly localization.

[0021] Step S2: Perform time-frequency analysis on the original time-domain sampled sequence data to obtain the time-frequency spectrum matrix corresponding to each loop.

[0022] In one specific embodiment, the process of performing step S2 may specifically include the following steps: For the original time-domain sampled sequence data, discrete time sequences are extracted separately for each loop; Set a time window length that determines the length of the signal segment analyzed in each Fourier transform and an overlap rate that determines the step size of the time window sliding; The discrete time series is divided into a series of time window data segments with overlapping ratios based on the length of the time window. Apply a window function to each time window data segment, and perform a fast Fourier transform on each windowed data segment to calculate the energy intensity of the signal at different frequency points within the corresponding window. Summarize the calculation results for all time windows. The calculation results are arranged in chronological order to construct a two-dimensional matrix as the time-frequency matrix of the corresponding loop.

[0023] Specifically, discrete time series are extracted from the original time-domain sampling sequence data of each circuit, which is strictly aligned on the time axis. Taking the eight circuits of a distribution cabinet in an industrial park as an example, the original time-domain sampling sequence data of each circuit contains 12,800 current amplitude points collected within one second. These amplitude points are arranged in chronological order to form the discrete time series of the circuit. Each element in the series corresponds to the instantaneous current value at a sampling moment, and the time interval between adjacent sampling points is the reciprocal of the sampling frequency, i.e., 78.125 microseconds. This discrete time series completely preserves the time-domain waveform information of the current signal and serves as the input basis for subsequent time-frequency analysis.

[0024] For the discrete-time series of each loop, two key parameters need to be set: time window length and overlap rate. The time window length determines the length of the signal segment analyzed in each Fourier transform. Its selection requires a balance between time resolution and frequency resolution; a longer time window results in higher frequency resolution but lower time resolution, and vice versa. In this embodiment, considering that power signal analysis needs to simultaneously focus on the fundamental frequency wave and its harmonic components, the time window length is set to 256 sampling points, corresponding to a time length of 256 × 78.125 microseconds = 20 milliseconds, which is exactly one power frequency cycle. This window length provides a frequency resolution of 50 Hz, sufficient to distinguish the fundamental frequency wave and its harmonics. The overlap rate determines the sliding step size between adjacent time windows. A higher overlap rate allows for a smaller sliding step size, resulting in smoother time-frequency change tracking. In this embodiment, the overlap rate is set to 75%, that is, 192 sampling points overlap between adjacent windows, and the sliding step size is 64 sampling points corresponding to 5 milliseconds, so that there is a new analysis window every 5 milliseconds, which can effectively capture the transient change characteristics of the current signal.

[0025] Based on a set time window length of 256 points and an overlap rate of 75%, the discrete time series is divided into a series of time window data segments. For a discrete time series of 12800 points, the first time window contains sampling points 1 to 256, the second time window starts from point 65 to point 320 according to a sliding step of 64 points, and so on, generating a total of 197 time window data segments, each with a length of 256 sampling points. These data segments cover the entire 1-second sampling duration, and adjacent windows maintain a 75% overlap, ensuring the continuity and integrity of the time-frequency analysis.

[0026] A window function is applied to each time window data segment to reduce the spectral leakage effect caused by signal truncation. This embodiment uses the Hanning window, a cosine window function characterized by a smooth decay to zero at both ends of the data segment, effectively suppressing sidelobe leakage. Each time window data segment is multiplied by the Hanning window function to obtain the windowed data segment. This windowed segment has a smooth amplitude transition at both ends, reducing the high-frequency components introduced by truncation.

[0027] A Fast Fourier Transform (FFT) is performed on each windowed data segment. FFT is an efficient algorithm for implementing Discrete Fourier Transform (DFT), capable of converting time-domain signals to the frequency domain. A 256-point FFT is performed on a windowed data segment of length 256 points, resulting in 256 complex spectral values. These spectral values ​​correspond to 256 discrete frequency points within the range of 0Hz to a sampling frequency of 12800Hz. The squared modulus of each complex spectral value represents the energy intensity of that frequency component within the current time window. Taking the injection molding workshop circuit as an example, under normal operating conditions, the energy intensity corresponding to the fundamental frequency of 50Hz is mainly concentrated in the low-frequency region, while the energy intensity of higher harmonic components is relatively low. Within the time window of the stamping workshop circuit due to the inrush current generated during equipment startup, its spectral distribution exhibits broadband characteristics, with a significant increase in energy intensity at multiple frequency points.

[0028] The calculation results for all 197 time windows are summarized, with each window generating a set of energy intensity values ​​for 256 frequency points. These calculation results are arranged in chronological order to construct a two-dimensional matrix as the time-spectrum matrix of the circuit. The row indices of this matrix correspond to the start times of the 197 time windows, the column indices correspond to the 256 discrete frequency points, and the matrix element values ​​are the signal energy intensities at the corresponding time and frequency positions. For the injection molding workshop circuit, its time-spectrum matrix shows that the energy is mainly concentrated in a few columns at low frequencies and changes steadily over time. For the stamping workshop circuit, its time-spectrum matrix shows that the energy diffuses to higher frequencies in multiple columns at the row corresponding to the equipment startup time, intuitively reflecting the broadband transient characteristics of the impact current.

[0029] Through the aforementioned time-frequency analysis process, the one-dimensional time-domain current signal is converted into a two-dimensional time-spectrum matrix, clearly revealing the frequency composition of the signal over time. This solves the technical problems of traditional time-domain analysis's inability to capture signal frequency-domain characteristics and traditional frequency-domain analysis's inability to pinpoint the exact moment of anomaly occurrence. The time-spectrum matrix not only fully preserves the signal's amplitude information but also provides joint distribution characteristics in both time and frequency dimensions. This provides intuitive and reliable data support for subsequent identification of energy concentration areas in each loop and determination of the dominant frequency range, effectively improving the detection sensitivity and positioning accuracy of anomalies such as current mutations and harmonic distortion.

[0030] Step S3: Based on the energy distribution characteristics of the time-frequency matrix, determine the dominant frequency range of each loop, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range.

[0031] In one specific embodiment, the process of performing step S3 may specifically include the following steps: Analyze the energy distribution characteristics of the time spectrum matrix, identify the regions where energy is concentrated on the frequency axis, and determine the corresponding dominant frequency range for each loop; Determine whether the width of the dominant frequency range exceeds a preset span threshold and whether the dominant frequency range contains high-frequency fluctuation features above a preset high-frequency boundary. If so, the original time-domain sampling sequence data of the loop is processed by multi-resolution decomposition technology to obtain a wideband fluctuation feature sequence that can simultaneously characterize low-frequency trends and high-frequency transients, and the wideband fluctuation feature sequence is used as the energy consumption feature sequence of the loop. If not, the original time-domain sampled sequence data of the loop is subjected to narrowband filtering and downsampling to obtain a low-frequency characteristic component sequence that reflects the energy consumption trend, and the low-frequency characteristic component sequence is used as the energy consumption characteristic sequence of the loop.

[0032] Specifically, based on the time-spectrum matrix of each loop obtained in step S2, the energy distribution characteristics of the time-spectrum matrix are analyzed to identify the regions where energy is concentrated on the frequency axis, and the corresponding dominant frequency range is determined for each loop. The rows of the time-spectrum matrix correspond to time windows, and the columns correspond to discrete frequency points. The matrix element value is the signal energy intensity at that time and frequency position. By accumulating energy along the time axis of the time-spectrum matrix, the cumulative energy value at each frequency point is calculated, which is the sum of the energy intensities of that frequency point across all time windows, yielding the total energy contribution at that frequency point. Taking the injection molding workshop loop as an example, after time-axis accumulation, the energy in its time-spectrum matrix is ​​mainly concentrated at a few frequency points such as 50Hz, 150Hz, and 250Hz. The cumulative energy value at 50Hz is the highest, reaching over 85% of the total energy. The cumulative energy values ​​at 150Hz and 250Hz account for 8% and 4% of the total energy, respectively, while the energy contributions at other frequency points are relatively small. A continuous frequency range where the cumulative energy value exceeds a preset energy threshold of 0.8 is identified as an energy concentration area. The preset energy threshold is set based on 80% of the total energy of each circuit, i.e., finding the smallest continuous frequency range that can cover more than 80% of the total energy. For the injection molding workshop circuit, the cumulative energy in the range from 45Hz to 55Hz reaches 85% of the total energy, so this range is determined as the dominant frequency range, with a lower limit of 45Hz, an upper limit of 55Hz, and a dominant frequency range width of 10Hz.

[0033] After determining the dominant frequency range of each loop, the signal characteristics of that loop are further assessed to select an appropriate feature extraction method. The assessment criteria include whether the width of the dominant frequency range exceeds a preset span threshold, and whether the dominant frequency range contains high-frequency fluctuations above a preset high-frequency boundary. The preset span threshold is set based on the multiples of the fundamental frequency. In this embodiment, 10 times the fundamental frequency of 50Hz, i.e., 500Hz, is used as the span threshold to measure the dispersion of the signal frequency distribution. If the width of the dominant frequency range exceeds 500Hz, it indicates that the signal frequency distribution is relatively dispersed. The preset high-frequency boundary is set based on the multiples of the fundamental frequency. In this embodiment, 20 times the fundamental frequency of 50Hz, i.e., 1000Hz, is used as the high-frequency boundary to define the starting frequency of high-frequency fluctuations. If there are frequency components above 1000Hz within the dominant frequency range, it indicates that the signal contains high-frequency fluctuation characteristics. Taking the stamping workshop circuit as an example, during equipment startup, the time-spectrum matrix shows that energy diffuses from low frequency to high frequency within multiple time windows. The lower limit of the dominant frequency range is 45Hz, the upper limit reaches 1800Hz, and the width is 1755Hz, significantly exceeding the preset span threshold of 500Hz. Furthermore, the upper limit of 1800Hz is higher than the preset high-frequency boundary of 1000Hz. Therefore, both judgment conditions are met simultaneously, indicating that there are high-frequency transient characteristics in the circuit signal, requiring a feature extraction method that can simultaneously retain low-frequency and high-frequency information. In addition, the preset span threshold and preset high-frequency boundary are not fixed but are pre-configured according to the application scenario of the distribution cabinet and the characteristics of the monitored load. For example, for industrial distribution cabinets mainly containing power frequency motor loads, the signal energy is usually concentrated in the low-frequency band. The span threshold can be set to 10 times the fundamental frequency (50Hz), i.e., 500Hz, and the high-frequency boundary can be set to 20 times the fundamental frequency, i.e., 1000Hz, to distinguish between conventional harmonics and high-frequency transient interference. For circuits containing equipment such as frequency converters, their operating frequency range may be wider. In such cases, the span threshold and high-frequency boundary can be appropriately increased based on the highest output frequency of the frequency converter. The threshold can be configured by maintenance personnel during system initialization based on the actual monitoring results.

[0034] For loops that meet the judgment criteria, multi-resolution decomposition technology is used for layered processing. In this embodiment, multi-resolution decomposition technology is implemented using discrete wavelet transform, a time-frequency analysis method that can decompose a signal into components of different frequencies. Through multi-scale decomposition of wavelet basis functions, the original signal is separated layer by layer into low-frequency approximate components and high-frequency detail components. The input data for discrete wavelet transform is the original time-domain sampling sequence data of the loop, which has a length of 12800 points and a sampling frequency of 12800Hz. A suitable wavelet basis function for power signal analysis is selected. In this embodiment, the db4 wavelet from the Daubechies wavelet system is selected. This wavelet basis has orthogonality, tight support, and good regularity, and can effectively capture transient features and local abrupt changes in power signals. The number of decomposition layers is determined to be 6 layers based on the sampling frequency and the frequency range of interest. Each layer of decomposition further decomposes the low-frequency approximate components of the previous layer into new low-frequency approximate components and high-frequency detail components. The first level of decomposition processes the original sequence, generating a high-frequency detail component D1 and a low-frequency approximation component A1, where D1 corresponds to the high-frequency band of 3200Hz to 6400Hz. The second level of decomposition processes A1, generating D2, corresponding to 1600Hz to 3200Hz and A2. The third level of decomposition generates D3, corresponding to 800Hz to 1600Hz and A3. The fourth level of decomposition generates D4, corresponding to 400Hz to 800Hz and A4. The fifth level of decomposition generates D5, corresponding to 200Hz to 400Hz and A5. The sixth level of decomposition generates D6, corresponding to 100Hz to 200Hz and A6, where A6 corresponds to the lowest frequency band of 0Hz to 100Hz. Through these six levels of decomposition, the original signal is decomposed into six high-frequency detail components in six different frequency bands and a low-frequency approximation component in the lowest frequency band. To obtain a broadband fluctuation feature sequence that can simultaneously characterize low-frequency trends and high-frequency transients, all high-frequency detail components D1 to D6, except for the last-level low-frequency approximation component A6, are reconstructed and superimposed with the last-level low-frequency approximation component A6. The reconstruction process employs inverse discrete wavelet transform, converting each level of detail component back to the time domain from the wavelet domain, resulting in time-domain signal components of the same length as the original sequence. Specifically, inverse transforms are performed on the high-frequency detail components D1 to D6 to obtain the reconstructed high-frequency time-domain components, denoted as HD1, HD2, HD3, HD4, HD5, and HD6, respectively; an inverse transform is performed on the low-frequency approximation component A6 to obtain the reconstructed low-frequency time-domain component, denoted as HA6. Thus, we obtain seven time-domain components corresponding to different frequency bands of the original signal. Since the amplitude dimensions of each level of detail component are the same but their energy contributions differ, normalization processing is required before superposition to reasonably reflect the contribution of each frequency band to the overall characteristics. The energy proportion weight of each level of detail component is dynamically determined based on its contribution to the total energy of the original signal. Specifically, calculate the total energy E_total of the original signal, and the energies E_HD1 to E_HD6 and E_HA6 of each reconstructed component. Then, calculate the weighting coefficients for each component. =E_HDi / E_total(i=1...6), =E_HA6 / E_total. Then, the amplitude of each reconstructed component is divided by its maximum absolute value to normalize the amplitude to the range of -1 to 1. Then, each normalized component is multiplied by its corresponding energy proportion weight. Finally, all weighted components are summed point by point to generate a broadband fluctuation feature sequence. This sequence is still 12800 points long, but each data point comprehensively reflects the complete energy consumption information from low frequency to high frequency. HA6 contributes to the low-frequency energy consumption trend, while HD1 to HD6 contribute to the high-frequency transient characteristics of different frequency bands. Each frequency band is assigned a corresponding weight according to its actual energy contribution, so that the broadband fluctuation feature sequence retains the energy distribution characteristics of the original signal while highlighting the relative importance of each frequency band in the overall characteristics. For ease of understanding, let's take a typical signal segment from the stamping workshop circuit during equipment startup as an example. The dynamic calculation of the energy proportion of each layer component is as follows: D12%, D23%, D35%, D48%, D512%, D615%, A655%. The corresponding weighting coefficients are 0.02, 0.03, 0.05, 0.08, 0.12, 0.15, and 0.55.

[0035] After normalizing each component, multiplying it by its corresponding weight and then superimposing them, a wideband fluctuation feature sequence is generated. This sequence still has 12800 points, but each data point comprehensively reflects complete energy consumption information from low to high frequencies. It retains the low-frequency energy consumption trend represented by A6 and also includes the high-frequency transient features captured by D1 to D6. For loops that do not meet the judgment criteria, such as the injection molding workshop loop, whose dominant frequency range is only 10Hz and does not contain high-frequency fluctuation features, it indicates that the loop signal is dominated by a low-frequency energy consumption trend, and multi-resolution decomposition is unnecessary. For such loops, narrowband filtering and downsampling are used to extract the low-frequency feature component sequence. The narrowband filter bank adopts a finite impulse response digital filter design, with the passband range set to 40Hz to 60Hz according to the dominant frequency range, to fully retain the signal components within the dominant frequency range of 45Hz to 55Hz, while filtering out high-frequency noise and non-interested frequency components above 60Hz. The filter order was set to 128, the transition band width to 20Hz, and the stopband attenuation to greater than 60dB, ensuring distortion-free signal passage within the passband and effective suppression of frequency components within the stopband. Applying this narrowband filter to the original time-domain sampled sequence data yielded a filtered signal, primarily containing low-frequency components in the 40Hz to 60Hz range. The filtered signal was then downsampled by a factor of 8, reducing the sampling frequency from 12800Hz to 1600Hz, with one point extracted every eight points to form a new sequence. The downsampled sequence length was one-eighth of the original sequence, i.e., 1600 points, significantly reducing the data volume. Furthermore, according to the Nyquist sampling theorem, a sampling frequency of 1600Hz is sufficient to completely preserve the highest 60Hz frequency components in the filtered signal, without information loss. This downsampled sequence is the low-frequency characteristic component sequence, with each data point reflecting the energy consumption trend of the loop within the dominant frequency range. The reduced data volume facilitates subsequent multi-loop data fusion and overall energy consumption analysis.

[0036] In one specific embodiment, the process of performing hierarchical processing using multi-resolution decomposition technology to obtain a broadband fluctuation feature sequence that can simultaneously characterize low-frequency trends and high-frequency transients can specifically include the following steps: The original time-domain sampled sequence data is subjected to multi-level wavelet decomposition, and each level of decomposition produces a high-frequency detail component and a low-frequency approximation component. The high-frequency detail components of all levels except the last-level low-frequency approximation component are reconstructed and superimposed with the last-level low-frequency approximation component to generate a broadband fluctuation feature sequence.

[0037] Specifically, for loops that satisfy the condition that the dominant frequency range exceeds a preset span threshold and contains high-frequency fluctuation characteristics, discrete wavelet transform (DWT) in multi-resolution decomposition technology is used for processing. The input data for DWT is the original time-domain sampling sequence data of the loop, which has a length of 12800 points and a sampling frequency of 12800Hz, completely recording the instantaneous current value changes within one second. Before performing wavelet decomposition, it is necessary to select a suitable wavelet basis function for power signal analysis. In this embodiment, the db4 wavelet from the Daubechies wavelet family is selected as the wavelet basis. The db4 wavelet is one of the orthogonal wavelet families constructed by IngridDaubechies. It has compact support, orthogonality, and good regularity. Its support length is 8, and its vanishing moment order is 4. It can effectively capture transient features and local abrupt changes in power signals, and is particularly suitable for analyzing non-stationary components such as impulses and oscillations in current signals. The determination of the number of decomposition layers needs to comprehensively consider the sampling frequency and the frequency range of interest. In this embodiment, the sampling frequency is 12800Hz. According to the Nyquist sampling theorem, the effective analysis frequency range of the signal is 0 to 6400Hz. To completely decompose the signal to the low-frequency band where the power frequency fundamental wave is located, the number of decomposition layers is set to 6. Each decomposition layer further decomposes the low-frequency approximation component of the previous layer into new low-frequency approximation components and high-frequency detail components.

[0038] The original time-domain sampled sequence data undergoes a first-level wavelet decomposition. The original sequence is passed through a low-pass filter and a high-pass filter constructed using the db4 wavelet to obtain the first-level low-frequency approximation component A1 and the first-level high-frequency detail component D1, respectively. The low-pass filter outputs the low-frequency components of the corresponding signal, with a frequency range of 0 to 3200 Hz; the high-pass filter outputs the high-frequency components of the corresponding signal, with a frequency range of 3200 Hz to 6400 Hz. The lengths of A1 and D1 are both half the length of the original sequence, i.e., 6400 points, because the downsampling operation in wavelet decomposition halves the data length. The second-level wavelet decomposition uses A1 as input, again passing it through low-pass and high-pass filters to obtain the second-level low-frequency approximation component A2 and the high-frequency detail component D2. The frequency range of A2 is 0 to 1600 Hz, and the frequency range of D2 is 1600 Hz to 3200 Hz, both with a length of 3200 points. The third-level decomposition takes A2 as input and yields A3 and D3. A3 has a frequency range of 0 to 800 Hz, and D3 has a frequency range of 800 Hz to 1600 Hz. Both have a length of 1600 points. The fourth-level decomposition takes A3 as input and yields A4 and D4. A4 has a frequency range of 0 to 400 Hz, and D4 has a frequency range of 400 Hz to 800 Hz. Both have a length of 800 points. The fifth-level decomposition takes A4 as input and yields A5 and D5. A5 has a frequency range of 0 to 200 Hz, and D5 has a frequency range of 200 Hz to 400 Hz. Both have a length of 400 points. The sixth-level decomposition takes A5 as input and yields A6 and D6. A6 has a frequency range of 0 to 100 Hz, and D6 has a frequency range of 100 Hz to 200 Hz. Both have a length of 200 points. Thus, through six-level wavelet decomposition, the original signal is decomposed into six high-frequency detail components D1 to D6 in different frequency bands and a low-frequency approximation component A6 in the lowest frequency band. D1 corresponds to the highest frequency band, reflecting the fast transient components in the signal. D2 to D6 correspond to the gradually decreasing frequency bands, reflecting the mid-to-high frequency fluctuation components. A6 corresponds to the lowest frequency band, reflecting the overall trend of the signal.

[0039] After obtaining the decomposed components at each level, it is necessary to reconstruct the high-frequency detail components of all levels except for the last-level low-frequency approximation component A6, and then superimpose them with A6 to generate a broadband wavelet feature sequence. The reconstruction process uses inverse discrete wavelet transform to convert each level of detail components back from the wavelet domain to the time domain, obtaining time-domain signal components of the same length as the original sequence. Since the lengths of the decomposed components at each level are different, the inverse discrete wavelet transform restores each level of components to the original sampling rate through upsampling and filtering operations. Specifically, when performing the inverse transform on D1, zeros are inserted between two adjacent points to achieve a 2x upsampling, restoring its length to 12800 points. Then, it is filtered by a reconstruction high-pass filter constructed from the db4 wavelet to obtain the reconstructed high-frequency component HD1. HD1 retains the information in the 3200Hz to 6400Hz frequency band of the original signal and has the same length as the original sequence. Similarly, the inverse transform of D2 yields HD2 corresponding to the 1600Hz to 3200Hz frequency band, the inverse transform of D3 yields HD3 corresponding to the 800Hz to 1600Hz frequency band, the inverse transform of D4 yields HD4 corresponding to the 400Hz to 800Hz frequency band, the inverse transform of D5 yields HD5 corresponding to the 200Hz to 400Hz frequency band, the inverse transform of D6 yields HD6 corresponding to the 100Hz to 200Hz frequency band, and the inverse transform of A6 yields HA6 corresponding to the 0Hz to 100Hz frequency band.

[0040] Since the amplitude dimensions of each reconstructed component are the same but their energy contributions differ, normalization processing is required before superposition to reasonably reflect the contribution of each frequency band to the overall characteristics. The total energy of the original signal is calculated, i.e., the sum of squares of the amplitudes at each point in the original sequence. Then, the energy of each reconstructed component is calculated separately, i.e., the sum of squares of the amplitudes of HD1 to HD6 and HA6. The energy percentage weight of each component is the component's energy divided by the total energy of the original signal, resulting in energy percentages of D1 to D6 of 0.02, 0.03, 0.05, 0.08, 0.12, and 0.15, respectively, and A6's energy percentage is 0.55. The amplitude of each reconstructed component is divided by its maximum absolute value to normalize the amplitude to the range of -1 to 1. Then, each normalized component is multiplied by its corresponding energy percentage weight. Finally, all weighted components are summed point by point to generate a broadband fluctuation characteristic sequence. The sequence has the same length as the original sequence, 12,800 points. Each data point comprehensively reflects the complete energy consumption information from low frequency to high frequency. HA6 contributes to the low-frequency energy consumption trend, while HD1 to HD6 contribute to the high-frequency transient characteristics of different frequency bands. Each frequency band is assigned a corresponding weight according to its actual energy contribution, so that the broadband fluctuation feature sequence not only retains the energy distribution characteristics of the original signal, but also highlights the relative importance of each frequency band in the overall characteristics. Taking the stamping workshop circuit as an example, during the equipment startup process, the high-frequency energy in the original signal increases significantly. The corresponding energy proportions of D1 to D3 rise from 0.01, 0.02, and 0.03 during normal operation to 0.05, 0.08, and 0.12. Through the above weighted superposition method, these high-frequency characteristics are fully reflected in the broadband fluctuation feature sequence, enabling subsequent analysis to accurately capture the impact characteristics caused by equipment startup.

[0041] Through the aforementioned multi-resolution decomposition and reconstruction superposition process, a broadband fluctuation feature sequence capable of simultaneously characterizing low-frequency trends and high-frequency transients is generated, overcoming the limitation of traditional single-scale analysis in failing to simultaneously consider low-frequency energy consumption trends and high-frequency transient characteristics. This feature sequence not only fully preserves the overall energy consumption variation trend of the signal but also highlights the local fluctuation characteristics of each frequency band, providing informative and representative input data for the subsequent construction of a unified feature dataset for multiple loops. This significantly improves the sensitivity of abnormal fluctuation detection and the accuracy of abnormal loop localization.

[0042] Step S4: Align the energy consumption feature sequences of all circuits according to the time axis to form a unified feature dataset of multiple circuits in the distribution cabinet.

[0043] In one specific embodiment, the process of performing step S4 may specifically include the following steps: Identify the time axis labels corresponding to the determined energy consumption characteristic sequences of each loop, and align the energy consumption characteristic sequences of all loops in the time dimension according to the time axis labels; The aligned energy consumption feature sequences are combined into a multi-dimensional data structure, which serves as a unified feature dataset for multiple circuits in the distribution cabinet.

[0044] Specifically, after extracting the energy consumption feature sequences of each loop, these feature sequences need to be integrated into a unified dataset for overall energy consumption analysis. It is necessary to identify the time axis labels corresponding to the determined energy consumption feature sequences of each loop. The time axis label refers to the actual sampling time or relative time position information corresponding to each feature sequence data point. Since a unified sampling clock signal is generated by the main control unit of the IoT acquisition device in step S1, and each acquisition channel starts analog-to-digital conversion at the same time to acquire the instantaneous current values ​​of each loop in parallel, the original time-domain sampling sequence data of all loops are naturally and strictly aligned on the time axis. The subsequent time-frequency analysis in step S2 and the feature extraction process in step S3 are both performed while maintaining the time order. Whether it is the low-frequency feature component sequence obtained through narrowband filtering and downsampling, or the wideband fluctuation feature sequence obtained through multi-resolution decomposition technology, both inherit the time reference of the original data. Taking the low-frequency feature component sequence as an example, it is obtained by downsampling the original sequence by 8 times. The nth data point in the downsampled sequence corresponds to the time of the 8nth sampling point in the original sequence. Taking the broadband wavelet feature sequence as an example, it is obtained by reconstructing and superimposing wavelet components at each level. The reconstruction process maintains the same length and sampling rate as the original sequence. Therefore, the nth data point in the sequence has the same time index as the nth sampling point in the original sequence.

[0045] After identifying the time axis labels of the energy consumption characteristic sequences of each loop, the energy consumption characteristic sequences of all loops are aligned in the time dimension according to these time axis labels. Since the time base of the characteristic sequences of each loop is consistent, the alignment process is essentially a continuation of verifying and maintaining the original synchronization, without the need for complex interpolation or resampling operations. Specifically, the characteristic sequences of each loop are arranged according to the same time index to ensure that the characteristic data of different loops at the same time have the same time label. For the injection molding workshop loop, its low-frequency characteristic component sequence length is 1600 points, the time index is from t=0 to t=1 seconds, and each point corresponds to a time interval of 0.625 milliseconds; for the stamping workshop loop, its wideband fluctuation characteristic sequence length is 12800 points, the time index is also from t=0 to t=1 seconds, and each point corresponds to a time interval of 78.125 microseconds. Since the time base of both is the same, alignment can be achieved through the time index, that is, the nth characteristic point of the injection molding workshop loop corresponds to the same physical time as the 16nth characteristic point of the stamping workshop loop. To ensure the accuracy of the alignment, it can be further verified whether the start time and sampling duration of the feature sequences of each loop are consistent. In this embodiment, the start time of the feature sequences of each loop is the same absolute moment, and the sampling duration is 1 second. Therefore, the time axis is perfectly matched.

[0046] The aligned energy consumption feature sequences are combined into a multi-dimensional data structure, serving as a unified feature dataset for multiple circuits in the distribution cabinet. However, the energy consumption feature sequences of different circuits may have different data lengths due to different feature extraction methods (for example, the low-frequency feature sequence obtained after narrowband filtering downsampling has a length of 1600 points, while the wideband feature sequence obtained after wavelet reconstruction has a length of 12800 points). To facilitate subsequent joint analysis of multiple circuits, it is necessary to unify all feature sequences to the same time resolution. In this embodiment, the longest feature sequence length of 12800 points among all circuits is used as the benchmark. Shorter feature sequences are upsampled and interpolated, for example, using a linear interpolation method, inserting equally spaced points between adjacent points to extend their length to 12800 points while maintaining the basic shape of the signal waveform. After interpolation, the feature sequences of all circuits have the same time axis (i.e., 12800 time points).

[0047] The characteristic sequences of each circuit, after being standardized in length, are arranged by circuit number to form a two-dimensional matrix of 12,800 rows and 8 columns. Each row corresponds to a unified time point, and each column corresponds to a circuit. The matrix elements are the energy consumption characteristic values ​​of the circuit at that time point. This two-dimensional matrix is ​​the unified characteristic dataset of the multiple circuits in the distribution cabinet, which fully records the energy consumption characteristic information of the eight circuits at each moment within one second.

[0048] In another implementation, considering that the characteristic sequences of different loops may have different physical meanings and dimensions—for example, the low-frequency characteristic component sequence reflects the current amplitude after narrowband filtering, while the broadband fluctuation characteristic sequence reflects the comprehensive characteristics of multi-band weighted superposition, and their numerical ranges may differ—the characteristic sequences of each loop are normalized in amplitude before combination to facilitate subsequent power spectral density analysis, while retaining absolute energy information. Specifically, the energy reference value of each loop's characteristic sequence is first recorded, for example, its maximum value or root mean square value is used as the reference amplitude of the loop; then, the characteristic sequence of each loop is divided by its own reference value to obtain a dimensionless relative change sequence, with the value range uniformly set to 0~1 or fluctuating around 1. During power spectral density analysis, the power spectral density calculation result of each loop is multiplied back to its corresponding energy reference value, thereby restoring the absolute energy contribution of each loop. Taking the injection molding workshop loop as an example, its low-frequency characteristic component sequence has a maximum value of 85 amperes, and the reference value is recorded as 85; the broadband fluctuation characteristic sequence of the stamping workshop loop has a maximum value of 210, and the reference value is recorded as 210. After dividing each by its respective benchmark value, the sequence values ​​are all between 0 and 1, and the uniformity of dimensions facilitates joint analysis. In the subsequent step S5, the power spectral density calculation is performed by multiplying the power spectral density results of each loop back by 85 and 210, respectively, so as to accurately reflect the difference in the actual energy consumption of the two loops in the overall energy consumption distribution.

[0049] The unified feature dataset constructed through the above process brings together energy consumption characteristic information that was originally scattered across different loops and time points under a single data framework, achieving temporal alignment and dimensional uniformity of multi-loop feature data. The formation of this dataset enables horizontal comparative analysis between different loops, such as comparing the magnitude of feature values ​​of each loop at the same time to identify abnormal loops. It also provides a standardized input data format for subsequent power spectral density analysis, achieving an organic transition from single-loop monitoring to multi-loop joint analysis and laying a data foundation for obtaining overall energy consumption distribution information.

[0050] Step S5: Perform power spectral density analysis based on the unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet.

[0051] In one specific embodiment, the process of performing step S5 may specifically include the following steps: Extract feature sequences from each loop from a unified feature dataset; The power spectral density estimation algorithm is used to process the characteristic sequences of each loop and calculate the distribution of the signal power of the corresponding loop in the frequency domain. By summarizing the distribution of signal power in each circuit, we can obtain overall energy consumption distribution information that reflects the energy distribution of the entire distribution cabinet in different frequency ranges.

[0052] Specifically, based on the unified feature dataset constructed in step S4, feature sequences for each loop are extracted from this dataset. The unified feature dataset is a two-dimensional matrix of 12800 rows and 8 columns, where each column corresponds to a feature sequence for one loop, and each row corresponds to a time point. Taking the injection molding workshop of loop 1 as an example, its corresponding column vector is extracted as the feature sequence for that loop. This sequence has a length of 12800 points, a sampling frequency of 12800Hz, and a time span of 1 second. Each data point reflects the energy consumption characteristic value of that loop at the current moment. Similarly, feature sequences for loops 2 to 8 are extracted sequentially, with each sequence maintaining the same time base and sampling rate as the original data. These feature sequences may be low-frequency feature component sequences or broadband fluctuation feature sequences, but all have undergone feature extraction in step S3 and length unification and normalization processing in step S4, possessing the same time resolution and a unified dimensional range.

[0053] For each loop's characteristic sequence, a power spectral density estimation algorithm is used to calculate the distribution of the corresponding loop's signal power in the frequency domain. Power spectral density describes the relationship between signal power and frequency, reflecting the energy contribution of each frequency component in the signal. In this embodiment, the periodogram method based on Fast Fourier Transform (FFT) is used for power spectral density estimation. This method obtains the power spectral density by performing a FFT on the signal sequence, calculating the square of the amplitude, and dividing by the frequency resolution. Specifically, a FFT is performed on the 12800-point characteristic sequence. Before the FFT, a Hanning window is applied to the sequence to reduce spectral leakage. The Hanning window is a cosine window function that can smoothly decay to zero at both ends of the data, effectively suppressing spectral energy diffusion caused by finite length truncation. The windowed sequence yields 12800 complex spectral values ​​through FFT, corresponding to a frequency range of 0Hz to 12800Hz. However, since the sampling frequency is 12800Hz, according to the Nyquist sampling theorem, the effective analysis frequency range is 0Hz to 6400Hz. Therefore, the first 6400 spectral values ​​are actually used for subsequent analysis. The squared modulus of each complex spectral value represents the power of that frequency component. Dividing the squared modulus by the frequency resolution yields the power spectral density value. The frequency resolution is the sampling frequency divided by the sequence length, i.e., 12800Hz / 12800 = 1Hz. Therefore, the unit of power spectral density is power per hertz. Taking the injection molding workshop in loop 1 as an example, its power spectral density calculation shows a significant peak at 50Hz, corresponding to the fundamental frequency component, smaller peaks at 150Hz and 250Hz, corresponding to the third and fifth harmonic components, and power spectral density values ​​close to zero at other frequencies. The characteristic sequence corresponding to the equipment startup period in the stamping workshop in loop 2 shows a broadband power spectral density distribution, with high power spectral density values ​​ranging from 50Hz to 2000Hz, reflecting the broadband characteristics of the inrush current.

[0054] After calculating the power spectral density of each circuit, the results need to be summarized to obtain the overall energy consumption distribution information of the distribution cabinet. The summarization method involves directly superimposing the power spectral densities of each circuit to obtain the total power spectral density of the entire distribution cabinet under the combined effect of all circuits. Specifically, for each frequency point, the power spectral density values ​​of the eight circuits at that frequency point are added together to obtain the total power spectral density value for that frequency point. Taking the 50Hz frequency point as an example, the power spectral density values ​​of circuits 1 to 8 at that point are 85, 45, 32, 28, 56, 42, 38, and 30, respectively. The sum is 356, indicating that the total power density of the entire distribution cabinet at the 50Hz frequency point is 356 units. The same superposition operation is performed on each frequency point within the range of 0Hz to 6400Hz to obtain a total power spectral density sequence of length 6400 points. This sequence represents the overall energy consumption distribution information of the distribution cabinet. This information is presented in the frequency domain, providing a complete description of the distribution characteristics of the power distribution cabinet's energy across different frequency ranges, including the concentration near the fundamental frequency, the distribution pattern of harmonic frequency bands, and fluctuations in the high-frequency band. For example, in the overall energy consumption distribution information, if the peak value at 50Hz is much higher than other frequency points, it indicates that the power distribution cabinet is mainly subjected to power frequency loads; if there are abnormal increases at odd harmonic frequencies such as 150Hz and 250Hz, it indicates a harmonic distortion problem; if there is a broad-spectrum increase in the high-frequency band, it may correspond to an impulsive load or equipment failure.

[0055] During the superposition process, considering the potential differences in capacity among circuits, a weighted superposition method can be adopted, assigning different weighting coefficients based on the rated capacity or actual load level of each circuit. The weighting coefficients are determined based on the historical operating data of each circuit, using the ratio of the circuit's rated capacity to the total capacity as the weight. The rated capacities of the eight circuits in the distribution cabinet are 100kVA, 80kVA, 60kVA, 50kVA, 70kVA, 55kVA, 45kVA, and 40kVA, with a total capacity of 500kVA. The weighting coefficients for each circuit are 0.2, 0.16, 0.12, 0.1, 0.1, 0.14, 0.11, 0.09, and 0.08, respectively. Before superposition, the power spectral density of each circuit is multiplied by its corresponding weighting coefficient, and then summed to obtain the weighted overall energy consumption distribution information. The weighting method allows circuits with larger capacities to contribute more significantly to the overall energy consumption distribution, more accurately reflecting the actual operating status of the distribution cabinet. Regardless of whether direct superposition or weighted superposition is used, the final overall energy consumption distribution information is presented as a curve with frequency as the horizontal axis and power spectral density as the vertical axis, or stored as a data sequence corresponding to frequency points and power spectral density values ​​for subsequent anomaly detection.

[0056] Through the aforementioned power spectral density analysis process, the time-domain characteristics of the unified feature dataset from multiple circuits are transformed into frequency-domain distribution information, achieving an organic integration from single-circuit time-domain characteristics to multi-circuit frequency-domain distribution. The overall energy consumption distribution information not only reveals the distribution pattern of energy in the distribution cabinet across various frequency ranges but also provides a global quantitative basis for subsequent abnormal fluctuation detection by comparing the power spectral density contributions of each circuit. This makes the identification of abnormal phenomena such as phase loss characteristics, harmonic distortion, and deviations in power consumption patterns more accurate and reliable. (Reference) Figure 2 The figure shows the power spectral density analysis of each loop, and it can be seen that the spectrum of loop 2 is significantly different.

[0057] Step S6: Determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and generate an early warning monitoring report when an anomaly is detected.

[0058] In one specific embodiment, the process of performing step S6 may specifically include the following steps: The overall energy consumption distribution information is compared with the preset normal energy consumption distribution template to detect whether there are phase loss characteristics or deviations in power consumption patterns. If a phase loss feature is detected, a phase loss warning signal is generated, and the location of the abnormal circuit is determined based on the circuit with a sudden drop in power in the overall energy consumption distribution information. If a deviation from the power consumption pattern is detected, an abnormal power consumption warning signal is generated, and the type of abnormality is determined based on the distribution characteristics of the deviation in the frequency domain. The output includes an early warning monitoring report containing the location and type of abnormal loop.

[0059] Specifically, based on the overall energy consumption distribution information of the distribution cabinet obtained in step S5, a preset normal energy consumption distribution template is established. This template is obtained by statistically analyzing data from the distribution cabinet under historical normal operating conditions, reflecting the typical energy consumption distribution characteristics of the distribution cabinet under normal conditions. The specific construction method involves collecting overall energy consumption distribution information of the distribution cabinet during 30 consecutive days of normal operation, selecting data from stable power consumption periods each day, averaging the power spectral density values ​​at each frequency point for each day, and then averaging the 30-day average to obtain the baseline power spectral density value for each frequency point. Simultaneously, the standard deviation of each frequency point is calculated as the fluctuation range. Taking the 50Hz frequency point as an example, the 30-day average baseline value is 350, and the standard deviation is 25, so the normal fluctuation range is 325 to 375. The same baseline values ​​and fluctuation ranges are established for other frequency points to form a complete normal energy consumption distribution template, which includes the baseline power spectral density values ​​for each frequency point and their allowable upper and lower limits for fluctuation.

[0060] After obtaining the current overall energy consumption distribution information, it is compared with a preset normal energy consumption distribution template at each frequency point to detect any significant deviations. The detection process first calculates the difference between the power spectral density value of each current frequency point and the template reference value. If the difference exceeds three times the standard deviation, it is initially determined that the frequency point may be abnormal. To eliminate transient interference or misjudgment, it is necessary to further verify the duration of the abnormal state: if the power spectral density value of the frequency point continuously exceeds three times the standard deviation and the duration exceeds a preset time threshold (e.g., 0.5 seconds), it is finally determined that the frequency point is abnormal. At the same time, the overall pattern is analyzed to see if there are any phase loss characteristics or power consumption pattern deviation characteristics. The phase loss characteristic is manifested as the power of a certain circuit suddenly dropping to zero or close to zero, which is reflected in the overall energy consumption distribution information as an overall decrease in the power spectral density value near the fundamental frequency. Since the power spectral density of each circuit has been calculated separately before superposition, the circuit with a sudden drop in power can be identified by tracing back the power spectral density contribution of each circuit. Taking a certain test as an example, the overall energy consumption distribution information showed that the total power spectral density at the 50Hz frequency point dropped sharply from the normal value of 350 to 180. By checking the contribution value of each circuit at this frequency point, it was found that the contribution value of circuit 3 dropped from 56 to 3, while the contribution values ​​of other circuits remained basically unchanged. Based on this, it was determined that circuit 3 had a phase loss characteristic. The determination of the phase loss characteristic also needs to meet the duration condition, that is, the power drop state lasts for more than 0.5 seconds, in order to eliminate misjudgments caused by instantaneous disturbances.

[0061] Deviance in power consumption patterns manifests as changes in the shape of energy consumption distribution, including abnormal increases in harmonic components and shifts in frequency band energy distribution. When detecting deviations, the similarity between the current overall energy consumption distribution and a normal template is calculated. The similarity calculation method uses cosine similarity, treating the current overall energy consumption distribution as a 6400-dimensional vector, and the normal template as a 6400-dimensional vector as well. The cosine of the angle between the two vectors is used as the similarity index. If the cosine similarity is lower than a preset threshold of 0.85, a deviation in power consumption pattern is identified. After identifying a deviation, the anomaly type is further determined based on the frequency domain distribution characteristics of the deviation. If the power spectral density values ​​at odd harmonic frequencies such as 150Hz and 250Hz exceed the template upper limit, and the fundamental frequency component is normal, it is identified as an abnormal harmonic distortion, possibly caused by nonlinear loads such as frequency converters and rectifiers. If abnormal energy accumulation occurs in the 0.5Hz to 5Hz ultra-low frequency band, it is identified as an abnormal load fluctuation, possibly caused by the start-up and shutdown of large equipment or drastic load changes. If a wide-spectrum rise occurs above 1000Hz in the high-frequency band, it is determined to be an abnormal high-frequency interference, which may be caused by equipment aging, poor contact, or a precursor to a short circuit fault. Taking a certain test as an example, the overall energy consumption distribution information shows that the power spectral density value at the 150Hz frequency point is 120, while the template reference value is 45 and the upper limit is 60. At the same time, the value at the 250Hz frequency point is 85, the template upper limit is 50, and the value at the fundamental frequency of 50Hz is 355, which is within the normal range. Based on this, it is determined to be an abnormal harmonic distortion.

[0062] Upon detecting a phase loss characteristic, a phase loss warning signal is immediately generated. This signal includes an anomaly type identifier of "phase loss," a timestamp of the anomaly occurrence, the affected circuit number, and a comparison of the current power value with the normal value. Simultaneously, the location of the abnormal circuit is determined based on the circuit corresponding to the sudden power drop in the overall energy consumption distribution information. By tracing back the power spectral density contribution value of each circuit near the fundamental frequency, the circuit with the abnormally low contribution value is located. Upon detecting a deviation from the power consumption pattern, an power consumption anomaly warning signal is generated. This signal includes an anomaly type identifier, such as "harmonic distortion anomaly," "load fluctuation anomaly," or "high-frequency interference anomaly," as well as information such as the abnormal frequency range, a quantitative indicator of the deviation degree, and suggested inspection directions. The anomaly type is determined based on the aforementioned frequency domain distribution characteristic analysis results. The quantitative indicator of the deviation degree can be expressed as the ratio of the power spectral density value at the abnormal frequency point to the upper limit of the template. For example, a deviation of 120 / 60 = 2.0 at 150Hz indicates a deviation of one time from the normal range. (Reference) Figure 3 This figure shows an analysis of the overall energy consumption distribution versus the normal template.

[0063] The final output includes an early warning monitoring report containing the location and type of the abnormal circuit. This report is stored in structured data format and also generates a visual interface for maintenance personnel to view. The report content includes monitoring time, distribution cabinet number, abnormal type, abnormal circuit number, abnormal frequency range, deviation degree, confidence score, and recommended remedial measures. The confidence score is calculated based on the significance and duration of the abnormal characteristics. The confidence score for phase loss characteristics is positively correlated with the magnitude and duration of the power drop, while the confidence score for power consumption pattern deviation characteristics is positively correlated with the degree of similarity deviation and the number of abnormal frequency points. Recommended remedial measures are automatically matched according to the abnormal type: for phase loss anomalies, it is recommended to check the status of the circuit breaker and line connections; for harmonic distortion anomalies, it is recommended to check the operating status of nonlinear load equipment or consider installing harmonic filters; for load fluctuation anomalies, it is recommended to check the start-up and shutdown records of large equipment; and for high-frequency interference anomalies, it is recommended to check the insulation condition and contact reliability of the equipment.

[0064] Through the above-mentioned anomaly judgment and early warning mechanism, abnormal conditions such as phase loss faults and deviations in power consumption mode during the operation of the distribution cabinet can be identified in a timely manner, and abnormal circuits can be accurately located and abnormal types can be identified. This provides clear handling guidance for operation and maintenance personnel, effectively shortens the fault investigation time, reduces the risk of equipment damage and safety accidents caused by the failure to detect abnormalities in a timely manner, and ensures the safe and stable operation of the power distribution system.

[0065] It is understood that the executing entity of this application can be a system for monitoring the energy consumption of a power distribution cabinet, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0066] The above describes a method for monitoring the energy consumption of a distribution cabinet according to an embodiment of this application. The following describes a system for monitoring the energy consumption of a distribution cabinet according to an embodiment of this application. Please refer to [link / reference]. Figure 4 One embodiment of a power distribution cabinet energy consumption monitoring system according to this application includes: The data acquisition module is used to perform high-speed synchronous sampling of the current signals of each circuit in the power distribution cabinet through IoT acquisition devices to obtain the original time-domain sampling sequence data of each circuit. The time-frequency analysis module is used to perform time-frequency analysis on the original time-domain sampled sequence data to obtain the time-frequency spectrum matrix corresponding to each loop. The frequency extraction module is used to determine the dominant frequency range of each loop based on the energy distribution characteristics of the time-spectrum matrix, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range. The feature alignment module is used to align the energy consumption feature sequences of all circuits according to the time axis to form a unified feature dataset of multiple circuits in the distribution cabinet. The energy consumption analysis module is used to perform power spectral density analysis based on a unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet; The anomaly warning module is used to determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and to generate an early warning monitoring report when an anomaly is detected.

[0067] Through the collaborative work of various modules, efficient and accurate energy consumption monitoring of the distribution cabinet is achieved. The data acquisition module uses IoT devices to perform high-speed synchronous sampling of current signals from each circuit within the distribution cabinet, ensuring data synchronization and integrity and avoiding sampling errors caused by time deviations or interference. The time-frequency analysis module performs time-frequency analysis on the raw data, generating a time-spectrum matrix for each circuit, accurately displaying the frequency components of each circuit and providing a reliable basis for subsequent energy consumption feature extraction. The frequency extraction module further analyzes the energy distribution of the time-spectrum matrix, determines the dominant frequency range of each circuit, and selects energy consumption feature sequences based on this to accurately reflect the energy consumption status of the circuit. The feature alignment module aligns the energy consumption feature sequences of all circuits along the time axis, generating a unified dataset that provides stable data support for subsequent energy consumption analysis. The energy consumption analysis module uses power spectral density analysis to obtain overall energy consumption distribution information for the distribution cabinet, helping to understand the energy consumption trends of each circuit. Finally, the anomaly warning module monitors the overall energy consumption distribution in real time; if abnormal fluctuations are detected, it promptly generates an early warning report and locates the abnormal circuit. Through the coordinated operation of these modules, this invention not only improves the accuracy of energy consumption monitoring in distribution cabinets, but also effectively provides early warning of potential power failures, ensuring the safe operation of the power system.

[0068] above Figure 4 The present invention provides a detailed description of a power distribution cabinet energy consumption monitoring system from the perspective of modular functional entities. The following describes a power distribution cabinet energy consumption monitoring device from the perspective of hardware processing.

[0069] Reference Figure 5 This invention also provides a device for monitoring the energy consumption of a power distribution cabinet. This device can be a server, and its internal structure can be as follows: Figure 5 As shown, the power distribution cabinet energy consumption monitoring device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the power distribution cabinet energy consumption monitoring device includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the power distribution cabinet energy consumption monitoring device stores the data corresponding to this embodiment. The network interface of the power distribution cabinet energy consumption monitoring device is used for communication with external terminals via network connection. When the computer program is executed by the processor, it can implement the above-described method.

[0070] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the device for monitoring the energy consumption of a power distribution cabinet to which the present invention is applied.

[0071] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for monitoring the energy consumption of a power distribution cabinet.

[0072] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring the energy consumption of a power distribution cabinet, characterized in that, The method includes: S1. High-speed synchronous sampling of current signals of each circuit in the power distribution cabinet is performed through IoT acquisition devices to obtain the original time-domain sampling sequence data of each circuit; S2. Perform time-frequency analysis on the original time-domain sampling sequence data to obtain the time-frequency spectrum matrix corresponding to each loop; S3. Based on the energy distribution characteristics of the time-frequency matrix, determine the dominant frequency range of each loop, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range; S4. Align the energy consumption characteristic sequences of all circuits according to the time axis to form a unified characteristic dataset of multiple circuits in the distribution cabinet. S5. Perform power spectral density analysis based on the unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet; S6. Determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and generate an early warning monitoring report when an abnormality is detected.

2. The method according to claim 1, characterized in that, S1 includes: The main control unit of the IoT acquisition device generates a unified sampling clock signal and synchronously distributes the sampling clock signal to the corresponding acquisition channel of each loop. Control each acquisition channel to start analog-to-digital conversion at the same time, and acquire the instantaneous current value of each circuit in parallel; The continuously acquired instantaneous current values ​​are arranged in chronological order to generate the original time-domain sampling sequence data of each loop that is strictly aligned on the time axis.

3. The method according to claim 1, characterized in that, S2 includes: For the original time-domain sampled sequence data, discrete time sequences are extracted for each loop. Set a time window length that determines the length of the signal segment analyzed in each Fourier transform and an overlap rate that determines the step size of the time window sliding; The discrete time series is divided into a series of time window data segments with the specified overlap rate according to the specified time window length. Apply a window function to each time window data segment, and perform a fast Fourier transform on each windowed data segment to calculate the energy intensity of the signal at different frequency points within the corresponding window. Summarize the calculation results for all time windows. The calculation results are arranged in chronological order to construct a two-dimensional matrix as the time-frequency matrix of the corresponding loop.

4. The method according to claim 3, characterized in that, S3 includes: Analyze the energy distribution characteristics of the time-frequency matrix, identify the regions where energy is concentrated on the frequency axis, and determine the corresponding dominant frequency range for each loop; Determine whether the width of the dominant frequency range exceeds a preset span threshold and whether the dominant frequency range contains high-frequency fluctuation features above a preset high-frequency boundary; If so, the original time-domain sampling sequence data of the loop is processed by multi-resolution decomposition technology to obtain a wideband fluctuation feature sequence that can simultaneously characterize low-frequency trends and high-frequency transients, and the wideband fluctuation feature sequence is used as the energy consumption feature sequence of the loop. If not, the original time-domain sampled sequence data of the loop is subjected to narrowband filtering and downsampling to obtain a low-frequency characteristic component sequence that reflects the energy consumption trend, and the low-frequency characteristic component sequence is used as the energy consumption characteristic sequence of the loop.

5. The method according to claim 4, characterized in that, The method employs multi-resolution decomposition technology for hierarchical processing to obtain a broadband fluctuation feature sequence capable of simultaneously characterizing low-frequency trends and high-frequency transients, including: The original time-domain sampled sequence data is subjected to multi-level wavelet decomposition, and each level of decomposition produces a high-frequency detail component and a low-frequency approximation component. The high-frequency detail components of all levels except the last-level low-frequency approximation component are reconstructed and superimposed with the last-level low-frequency approximation component to generate the broadband fluctuation feature sequence.

6. The method according to claim 1, characterized in that, S4 includes: Identify the time axis label corresponding to the determined energy consumption feature sequence of each loop, and align the energy consumption feature sequences of all loops in the time dimension according to the time axis label; The aligned energy consumption feature sequences are combined into a multi-dimensional data structure, which serves as a unified feature dataset for multiple circuits in the distribution cabinet.

7. The method according to claim 1, characterized in that, S5 includes: Extract the feature sequences of each loop from the unified feature dataset; The power spectral density estimation algorithm is used to process the characteristic sequences of each loop and calculate the distribution of the signal power of the corresponding loop in the frequency domain. By summarizing the distribution of signal power in each circuit, we can obtain overall energy consumption distribution information that reflects the energy distribution of the entire distribution cabinet in different frequency ranges.

8. The method according to claim 7, characterized in that, S6 includes: The overall energy consumption distribution information is compared with a preset normal energy consumption distribution template to detect whether there are phase loss characteristics or deviations in power consumption patterns. If a phase loss feature is detected, a phase loss warning signal is generated, and the location of the abnormal circuit is determined based on the circuit with a sudden drop in power in the overall energy consumption distribution information. If a deviation from the power consumption pattern is detected, an abnormal power consumption warning signal is generated, and the type of abnormality is determined based on the distribution characteristics of the deviation in the frequency domain. The output includes an early warning monitoring report containing the location and type of abnormal loop.

9. A system for monitoring the energy consumption of a power distribution cabinet, used to implement the method as described in any one of claims 1-8, characterized in that, The system for monitoring the energy consumption of a power distribution cabinet includes: The data acquisition module is used to perform high-speed synchronous sampling of the current signals of each circuit in the power distribution cabinet through IoT acquisition devices to obtain the original time-domain sampling sequence data of each circuit. The time-frequency analysis module is used to perform time-frequency analysis on the original time-domain sampling sequence data to obtain the time-frequency spectrum matrix corresponding to each loop; The frequency extraction module is used to determine the dominant frequency range of each loop based on the energy distribution characteristics of the time-spectrum matrix, and select the energy consumption characteristic sequence of the loop based on the dominant frequency range. The feature alignment module is used to align the energy consumption feature sequences of all circuits according to the time axis to form a unified feature dataset of multiple circuits in the distribution cabinet. The energy consumption analysis module is used to perform power spectral density analysis based on the unified feature dataset to obtain the overall energy consumption distribution information of the distribution cabinet; The anomaly warning module is used to determine whether there are abnormal fluctuations based on the overall energy consumption distribution information, and to generate an early warning monitoring report when an anomaly is detected.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instruction is executed by the processor, it implements a method for monitoring the energy consumption of a power distribution cabinet as described in any one of claims 1-8.

Images