Dynamic reactive power compensation method and system for low-voltage power distribution cabinet based on load characteristic analysis

Abstract

The present application relates to the technical field of reactive power compensation, and discloses a low-voltage power distribution cabinet dynamic reactive power compensation method and system based on load characteristic analysis, which comprises the following steps: determining the power factor based on the phase difference between the voltage signal and the current signal, determining the load type of each branch according to the load, performing adjacent difference on the second-level short-time sequence corresponding to each load type, performing continuous section fitting on the minute-level trend sequence, and performing period reconstruction on the hour-level period sequence, determining the harmonic influence sequence based on each harmonic component and the equivalent impedance of the system, performing step-by-step screening based on the difference between the reactive power compensation amount and the capacity of each capacitor bank, and removing the data that does not meet the minimum switching time interval constraint and the capacity change step constraint, and switching based on the switching control data set. The present application extracts three types of reactive power change characteristics, namely load instantaneous fluctuation, trend gradual change and periodical law, thereby improving the precision and adaptability of reactive power demand compensation.

Description

Technical Field

[0001] This invention relates to the field of reactive power compensation technology, and more specifically, to a dynamic reactive power compensation method and system for low-voltage distribution cabinets based on load characteristic analysis. Background Technology

[0002] Reactive power compensation in low-voltage switchgear is an important technology for achieving energy saving and loss reduction, stabilizing bus voltage, and ensuring power quality in low-voltage power distribution networks. It is widely used in low-voltage power distribution scenarios such as industrial power distribution, commercial buildings, and municipal utilities to meet the requirements of grid-connected operation safety.

[0003] Existing patent CN119813248A discloses a low-voltage reactive power compensation method and system for power distribution equipment. This scheme collects reactive load data of each phase on the low-voltage side of the power distribution equipment, constructs a reactive load time series model, and directly determines the capacitor grouping and switching strategy based on the prediction results. However, it only relies on the reactive load value for single-dimensional time series prediction and does not identify load characteristics such as load attributes and reactive power change rate, making it unable to adapt to the differentiated reactive power requirements of complex loads such as impulsive loads and periodic loads. Patent CN119362489A discloses a reactive power compensation control method for power distribution networks. This scheme obtains operating information such as frequency and harmonic components by performing analog-digital sampling and Fourier calculation on voltage and current signals. However, it does not incorporate the harmonic effect as a constraint into the reactive power compensation amount decision, which easily leads to the safety risk of resonance between capacitors and system harmonics, resulting in frequent switching, aggravating equipment losses, and reducing the stability of compensation. Existing low-voltage reactive power compensation either focuses only on the time-series prediction of reactive power values, ignoring load characteristic identification and harmonic constraints, or only achieves real-time harmonic detection and passive protection, failing to meet the needs of dynamic reactive power compensation under complex load scenarios.

[0004] Therefore, it is necessary to design a dynamic reactive power compensation method and system for low-voltage switchgear based on load characteristic analysis to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a dynamic reactive power compensation method and system for low-voltage distribution cabinets based on load characteristic analysis, aiming to solve the problem that it is difficult to adapt to the dynamic reactive power compensation requirements under complex load scenarios when facing complex and ever-changing load characteristics and power grid operating conditions.

[0006] In one aspect, the present invention proposes a dynamic reactive power compensation method for low-voltage switchgear based on load characteristic analysis, comprising:

[0007] The voltage signal, current signal and corresponding reactive power data of each branch are acquired. Based on the phase difference between the voltage signal and the current signal, the power factor is determined. The load is judged according to the power factor to determine the load type of each branch.

[0008] The reactive power data is segmented by a sliding time window. The reactive power change, rate of change, and fluctuation amplitude of branches with the same load type are classified to determine the reactive power characteristic data sequence. The reactive power characteristic data sequence is divided into second-level short-time sequence, minute-level trend sequence, and hour-level periodic sequence on a unified time axis. Adjacent difference is performed on the second-level short-time sequence corresponding to each load type, continuous segment fitting is performed on the minute-level trend sequence, and periodic reconstruction is performed on the hour-level periodic sequence. The results of adjacent difference, continuous segment fitting, and periodic reconstruction are fused based on the time order to determine the reactive power demand prediction sequence.

[0009] Discrete Fourier transform is performed on the voltage and current signals, and the harmonic influence sequence is determined based on each harmonic component and the system equivalent impedance. The harmonic influence sequence and the reactive power demand prediction sequence are compared under constraints to determine the reactive power compensation sequence. Based on the capacitor banks in the low-voltage distribution cabinet, the capacitor combination dataset is determined. The reactive power compensation is filtered step by step based on the difference between the capacity of each capacitor bank and the capacity of each capacitor bank. Data that does not meet the minimum switching time interval constraint and the capacity change step constraint are removed to determine the switching control dataset.

[0010] Switching is performed based on the aforementioned switching control dataset, and after switching, the updated voltage signal, current signal, and corresponding reactive power data of each branch are collected.

[0011] Furthermore, in determining the load type of each branch, the process includes: extracting the phase angles of the voltage signal and the current signal, using the voltage signal as a reference phase, comparing the phase of the current signal, determining the phase difference as positive when the phase of the current signal lags behind the phase of the voltage signal, and determining the phase difference as negative when the phase of the current signal leads the phase of the voltage signal. The power factor is determined based on the sign and range of the phase difference. The load type includes inductive loads and capacitive loads. Branches with a power factor greater than zero and a corresponding positive phase difference are identified as inductive loads, and branches with a power factor greater than zero and a corresponding negative phase difference are identified as capacitive loads.

[0012] Furthermore, in determining the reactive power characteristic data sequence, the process includes: determining a sliding time window based on a preset time length; moving the sliding time window; determining a reactive power data subsequence within each sliding time window; arranging the reactive power data subsequence in chronological order within each sliding time window; determining the reactive power change based on the difference between the first and last data points of the window; determining the instantaneous change rate based on the difference between adjacent sampling points and the sampling time interval; summing the instantaneous change rates in chronological order; calculating the mean of the summation results based on the number of sampling points to determine the change rate; determining the maximum and minimum values ​​of the reactive power data within each sliding time window; and determining the fluctuation amplitude based on the difference between the maximum and minimum values.

[0013] Furthermore, when dividing the reactive power characteristic data sequence into second-level short-time series, minute-level trend series, and hour-level periodic series on a unified time axis, the process includes: dividing data in the reactive power characteristic data sequence whose time interval is less than or equal to a second-level threshold into second-level short-time series; dividing data in the reactive power characteristic data sequence whose time interval is greater than a second-level threshold but less than a minute-level threshold into minute-level trend series; and dividing data in the reactive power characteristic data sequence whose time interval is greater than or equal to a minute-level threshold into hour-level periodic series, wherein the second-level threshold is less than the minute-level threshold.

[0014] Furthermore, in determining the reactive power demand forecast sequence, the process includes: arranging the second-level short-time sequences corresponding to each load type in time, performing differential calculations on adjacent sampling points to determine the differential value sequence, determining the direction of change based on the differential value sequence, dividing continuous differential values ​​with the same direction of change into segments to determine the short-time change segment sequence, dividing the minute-level trend sequence corresponding to each load type into several continuous segments based on time order, fitting each continuous segment to determine the fitting curve for the corresponding segment, determining the direction and degree of change for each segment based on the fitting curve, and merging adjacent segments with the same direction of change and a degree of change less than a threshold to determine the trend change sequence.

[0015] Furthermore, in determining the reactive power demand forecast sequence, the process includes: dividing the hourly cycle sequence corresponding to each load type into several cycle intervals based on time order; normalizing the data within each cycle interval; aligning the data at corresponding time positions within different cycle intervals; superimposing the aligned data to determine the cycle change characteristic curve; reconstructing the data within different cycles based on the cycle change characteristic curve to determine the cycle change sequence; and splicing the short-term change segment sequence, trend change sequence, and cycle change sequence to determine the reactive power demand forecast sequence.

[0016] Furthermore, in determining the reactive power compensation sequence, the process includes: performing a discrete Fourier transform on the voltage and current signals to extract the fundamental component and each harmonic component, arranging each harmonic component in frequency order, determining the corresponding harmonic voltage component based on the arrangement result and the system equivalent impedance, determining the harmonic influence range based on the amplitude range of the harmonic voltage component, determining the reactive power compensation amount corresponding to each time point based on the machine learning model and the harmonic influence range, and constructing the reactive power compensation sequence from all the reactive power compensation amounts.

[0017] Furthermore, in determining the switching control dataset, the process includes: determining a capacitor combination dataset based on the capacity of the capacitor banks in the low-voltage distribution cabinet; obtaining the compensation capacity corresponding to each capacitor bank; calculating the difference between the compensation capacity and each reactive power compensation quantity in the reactive power compensation quantity sequence; sorting each capacitor bank based on the difference; selecting the capacitor bank with the smallest difference as a candidate switching combination; recording adjacent switching times; judging the time interval between adjacent switching times; removing data with a time interval smaller than the minimum switching time interval; removing data with capacity changes exceeding the step range; arranging the remaining candidate switching combinations according to the removal results; and determining the switching control dataset.

[0018] Furthermore, when switching based on the switching control dataset, the process includes: switching the capacitor bank in the low-voltage distribution cabinet based on the switching control dataset.

[0019] Compared with existing technologies, the advantages of this invention are as follows: It acquires voltage signals, current signals, and corresponding reactive power data, providing a data foundation for subsequent load type determination and compensation decisions. This ensures that all analyses are based on actual operating conditions, avoiding judgment biases caused by missing data. By determining the load type and distinguishing between inductive and capacitive loads, it avoids problems such as incorrect compensation direction and overcompensation caused by misjudgment of load attributes, providing a classification basis for subsequent reactive power analysis. The use of a sliding time window to segment reactive power data enables accurate extraction of the dynamic change characteristics of the load, avoiding the limitations of a single data dimension. Furthermore, by dividing the data into sequences at different time scales and combining the processing methods at each scale, it comprehensively captures the instantaneous fluctuations, gradual trends, and periodic patterns of the load, solving the shortcomings of traditional single-dimensional analysis in adapting to complex loads. Furthermore, by extracting harmonic information through discrete Fourier transform and incorporating the harmonic effects into the compensation decision, the targetedness and stability of reactive power compensation are ensured, as well as the operational safety of low-voltage switchgear. By collecting and updating closed-loop data and adapting to load changes in real time, the reliability and adaptability of dynamic compensation are improved, thus meeting the safety requirements of grid-connected operation.

[0020] On the other hand, this application also provides a dynamic reactive power compensation system for low-voltage switchgear based on load characteristic analysis, used to apply the above-mentioned dynamic reactive power compensation method for low-voltage switchgear based on load characteristic analysis, including:

[0021] The acquisition and analysis unit is configured to acquire the voltage signal, current signal and corresponding reactive power data of each branch, determine the power factor based on the phase difference between the voltage signal and the current signal, determine the load based on the power factor, and determine the load type of each branch.

[0022] The first processing unit is configured to perform sliding time window segmentation processing on the reactive power data, classify the reactive power change, change rate and fluctuation amplitude of the same load type branch, determine the reactive power characteristic data sequence, and divide the reactive power characteristic data sequence into second-level short-time sequence, minute-level trend sequence and hour-level periodic sequence on a unified time axis. The second-level short-time sequence corresponding to each load type is subjected to adjacent difference, the minute-level trend sequence is subjected to continuous segment fitting and the hour-level periodic sequence is subjected to periodic reconstruction. The results of adjacent difference, continuous segment fitting and periodic reconstruction are fused based on time order to determine the reactive power demand prediction sequence.

[0023] The second processing unit is configured to perform discrete Fourier transform on the voltage and current signals, determine the harmonic influence sequence based on each harmonic component and the system equivalent impedance, compare the harmonic influence sequence with the reactive power demand prediction sequence to determine the reactive power compensation sequence, determine the capacitor combination dataset based on the capacitor banks in the low-voltage distribution cabinet, perform step-by-step filtering based on the reactive power compensation and the capacity difference of each capacitor bank, and remove data that does not meet the minimum switching time interval constraint and the capacity change step constraint to determine the switching control dataset.

[0024] The compensation processing unit is configured to perform switching based on the switching control dataset, and to collect updated voltage signals, current signals and corresponding reactive power data of each branch after switching.

[0025] It is understandable that the above-mentioned method and system for dynamic reactive power compensation of low-voltage distribution cabinets based on load characteristic analysis have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis, provided as an embodiment of the present invention;

[0028] Figure 2 This is a functional block diagram of a low-voltage switchgear dynamic reactive power compensation system based on load characteristic analysis, provided for an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] See Figure 1 As shown in some embodiments of this application, a dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis includes:

[0032] S100: Acquire the voltage signal, current signal and corresponding reactive power data of each branch, determine the power factor based on the phase difference between the voltage signal and the current signal, determine the load based on the power factor, and determine the load type of each branch.

[0033] S200: The reactive power data is segmented by a sliding time window. The reactive power change, rate of change, and fluctuation amplitude of the same load type branch are classified to determine the reactive power characteristic data sequence. The reactive power characteristic data sequence is divided into second-level short-time sequence, minute-level trend sequence, and hour-level periodic sequence on a unified time axis. Adjacent difference is performed on the second-level short-time sequence corresponding to each load type, continuous segment fitting is performed on the minute-level trend sequence, and periodic reconstruction is performed on the hour-level periodic sequence. The results of adjacent difference, continuous segment fitting, and periodic reconstruction are fused based on the time order to determine the reactive power demand forecast sequence.

[0034] S300: Perform Discrete Fourier Transform on voltage and current signals, and determine the harmonic influence sequence based on each harmonic component and the system equivalent impedance. Compare the harmonic influence sequence with the reactive power demand prediction sequence to determine the reactive power compensation sequence. Based on the capacitor banks in the low-voltage distribution cabinet, determine the capacitor combination dataset. Perform step-by-step screening based on the reactive power compensation and the capacity difference of each capacitor bank, and remove data that do not meet the minimum switching time interval constraint and the capacity change step constraint to determine the switching control dataset.

[0035] S400: Performs switching based on the switching control dataset, and collects updated voltage signals, current signals and corresponding reactive power data for each branch after switching.

[0036] Specifically, the system acquires the voltage, current, and corresponding reactive power data for each branch. The voltage signal represents the analog voltage data collected from each branch of the low-voltage distribution cabinet. For example, in an industrial plant's low-voltage distribution cabinet, this includes the 220V power frequency voltage signal collected from the motor branch and the 220V voltage signal collected from the capacitor compensation branch. The current signal represents the analog current data collected from each branch of the low-voltage distribution cabinet. For example, this includes the 8A current signal collected during motor operation and the 3A current signal collected from the lighting branch. The reactive power data represents the numerical reactive power value corresponding to each branch of the low-voltage distribution cabinet. For example, the reactive power for the motor branch is 25kvar, and the reactive power for the equipment branch is 15kvar. The power factor is determined based on the phase difference between voltage and current signals. The phase difference represents the angular deviation between the voltage and current signals in the time domain. For example, with a voltage signal phase of 0° as a reference, if the motor branch current signal lags by 36°, the phase difference is +36°. The power factor represents the ratio of active power to apparent power in the circuit. Again, using a phase difference of 36° as an example, the corresponding power factor is cos36°≈0.81. Load type indicates whether the branch load is inductive or capacitive. Inductive loads include motors and fans, while capacitive loads include capacitors and frequency converters. By collecting electrical parameters and calculating phase and power factors, accurate classification of load attributes is achieved, thus distinguishing the reactive power demand differences between inductive and capacitive loads. This avoids overcompensation and undercompensation problems caused by incorrect load attribute judgment, providing a classification basis for subsequent reactive power characteristic analysis. The reactive power data is segmented by a sliding time window. The sliding time window segmentation is a method of continuously sliding and capturing reactive power data according to a preset fixed time window. For example, if the sliding window duration is set to 10 seconds, the latest 10 seconds of reactive power data is captured for analysis every 1 second. The reactive power changes, rates of change, and amplitudes of branches with the same load type are categorized. The reactive power change represents the difference between the first and last values ​​of reactive power within the same sliding time window. For example, if the reactive power changes from 20 kvar to 32 kvar within a 10-second window, the reactive power change is 12 kvar. The rate of change represents the average instantaneous rate of change of reactive power within the same sliding time window. Again, using a total change of 12 kvar within the window as an example, with a 10-second duration, the rate of change is 1.2 kvar / second. The amplitude of fluctuation represents the difference between the maximum and minimum reactive power values ​​within the same sliding time window. For example, if the maximum reactive power value is 35 kvar and the minimum is 18 kvar, the amplitude of fluctuation is 17 kvar. This is used to determine the reactive power characteristic data sequence. The reactive power characteristic data sequence integrates the time-series data of reactive power change characteristic parameters of branches with the same load type. By extracting the core characteristics of reactive power change through the sliding window, accurate quantification of the load's dynamic characteristics is achieved, avoiding the risk of relying solely on reactive power numerical analysis while ignoring dynamic change patterns.The reactive power characteristic data series are divided into second-level short-time series, minute-level trend series, and hour-level periodic series on a unified time axis. The second-level short-time series is a series of reactive power characteristic data with small time intervals, the minute-level trend series is a series of reactive power characteristic data with large time intervals, and the hour-level periodic series represents a series of reactive power characteristic data with the largest time intervals. The second-level short-time series corresponding to each load type are subjected to adjacent difference analysis, the minute-level trend series are subjected to continuous segment fitting, and the hour-level periodic series are subjected to periodic reconstruction. This is because the second-level short-time series mainly reflects the instantaneous impact and rapid fluctuation characteristics of the load. Adjacent difference analysis can accurately extract the reactive power difference between adjacent sampling points, quickly identify the direction and magnitude of instantaneous reactive power change, and adapt to the rapid response requirements of impact loads such as motor start-up and shutdown and sudden equipment commissioning. The minute-level trend series mainly reflects the slow and gradual change characteristics of the load and the continuous adjustment of the operating conditions. Continuous segment fitting can eliminate small fluctuation interference and clearly restore the overall trend of reactive power change, which is suitable for the gradual operating conditions of load commissioning in batches and power rising and falling steadily. The hour-level periodic series mainly represents the periodic characteristics of load changing by shift and day and night. Periodic reconstruction can align data from multiple periodic periods and restore stable periodic change patterns, which is suitable for periodic load scenarios such as factory production and building power consumption. By processing the three different time scale sequences accordingly, the reactive power variation characteristics of instantaneous load fluctuation, gradual trend change, and periodic regularity can be extracted comprehensively and accurately. This avoids the shortcomings of single-dimensional time series prediction, which cannot adapt to the differentiated reactive power demand of complex loads, and improves the accuracy and adaptability of reactive power demand compensation.

[0037] Understandably, by fusing data on instantaneous changes at the second level, trend changes at the minute level, and periodic changes at the hour level, a reactive power demand forecast sequence that can take into account periodic patterns, gradual trends, and instantaneous changes is determined. Discrete Fourier Transform (DFT) is performed on voltage and current signals. DFT converts voltage and current signals in the time domain into harmonic components in the frequency domain. For example, a 50Hz power frequency voltage and current signal is converted into amplitude and phase data for the 3rd, 5th, and 7th harmonics. The harmonic impact sequence is determined based on the harmonic components and the system's equivalent impedance. The system's equivalent impedance represents the combined equivalent resistance and reactance of the low-voltage distribution system. For example, the measured equivalent impedance of the distribution system is 0.4Ω. The harmonic impact sequence is the harmonic interference time sequence data generated by combining the harmonic components with the system's equivalent impedance. For example, a 5th harmonic current of 10A multiplied by the equivalent impedance of 0.4Ω yields a 5th harmonic voltage of 4V. The impact of each harmonic is calculated and integrated into the harmonic impact sequence. The harmonic impact sequence and reactive power demand prediction sequence are compared under constraints. This constraint comparison uses the harmonic impact sequence as a limiting condition to verify and adjust the values ​​of the reactive power demand prediction sequence, thereby determining the reactive power compensation sequence. The reactive power compensation sequence represents the set of reactive power compensation values ​​required at each time point after harmonic constraint verification. Using harmonic impact as a hard constraint condition for reactive power compensation solves the safety risk of resonance that can easily occur if harmonics are not incorporated into compensation decisions, and avoids the risk of resonance between capacitors and system harmonics, thus ensuring the operational safety of the power distribution system. Based on the capacitor banks within the low-voltage distribution cabinet, a capacitor combination dataset is determined. This dataset represents the set of capacity combinations formed by the switching of all capacitor banks within the cabinet. For example, if the distribution cabinet has three capacitor banks of 10kvar, 20kvar, and 30kvar, the combinations include 10kvar, 20kvar, 30kvar, 10+20=30kvar, 10+30=40kvar, 20+30=50kvar, and 10+20+30=60kvar. All combinations are then integrated to form the capacitor combination dataset. A step-by-step selection process is performed based on the reactive power compensation and the capacity difference between each capacitor bank. This step-by-step selection process chooses the optimal capacitor bank in ascending order of capacity difference. Data that does not meet the minimum switching time interval constraint and the capacity change step constraint is eliminated. Specifically, the minimum switching time interval constraint represents the minimum allowable time interval between two consecutive capacitor switching operations. For example, if the minimum switching time interval is 30 seconds, and the previous switching time was at the 10th second, then if the current switching time is within 40 seconds, the data is eliminated.The capacity change step constraint represents the maximum allowable amplitude of capacity change during capacitor switching. For example, each capacity change should not exceed 20 kvar. If switching directly from 10 kvar to 40 kvar results in a change of 30 kvar, exceeding the constraint, the data is discarded. This finalizes the switching control dataset, which is a set of capacitor switching data that meets both the minimum switching time interval constraint and the capacity change step constraint. By using dual constraints to filter switching strategies, the problems of frequent switching and compensation oscillations are avoided, while simultaneously reducing capacitor losses and improving the stability and reliability of the compensation process. Based on the switching control dataset, capacitor banks are switched on or off. After switching, updated voltage and current signals and corresponding reactive power data for each branch are collected, forming a complete closed-loop control system of acquisition, analysis, constraint, execution, and feedback. This adapts to the dynamic reactive power compensation needs under complex load scenarios and improves the accuracy of reactive power demand compensation.

[0038] In some embodiments of this application, determining the load type of each branch includes: extracting the phase angle of the voltage signal and the current signal, using the voltage signal as a reference phase, comparing the phase of the current signal, determining a positive phase difference when the phase of the current signal lags behind the phase of the voltage signal, and a negative phase difference when the phase of the current signal leads the phase of the voltage signal, determining the power factor based on the sign and range of the phase difference, and identifying the load type as inductive load and capacitive load. Branches with a power factor greater than zero and a corresponding positive phase difference are identified as inductive loads, and branches with a power factor greater than zero and a corresponding negative phase difference are identified as capacitive loads.

[0039] Specifically, the phase angles of the voltage and current signals are extracted. The phase angle represents the time-domain angular position of the voltage and current signals within the power frequency cycle. The voltage signal is used as the reference phase, representing a unified benchmark signal for phase comparison. Using the voltage signal as the sole benchmark for phase comparison eliminates the confusion caused by multiple benchmark comparisons and ensures consistency in load type determination for all branches. The phases of the current signals are compared. When the current signal phase lags behind the voltage signal phase, the phase difference is determined to be positive; when the current signal phase leads the voltage signal phase, the phase difference is determined to be negative. The phase difference represents the time-domain angular deviation between the voltage and current signals. Its sign is used to clarify the lag or lead relationship between the current and voltage. The sign of the phase difference is determined through numerical comparison, enabling real-time locking of the relative phase relationship between current and voltage, providing a basis for distinguishing load attributes. After determining the sign of the phase difference, the power factor is determined based on the sign and range of the phase difference. The power factor represents the ratio of active power to apparent power in the circuit, which can be directly calculated from the cosine of the phase difference. The calculation of the power factor based on the phase difference improves the efficiency of load determination. Load types include inductive and capacitive loads. Branches with a power factor greater than zero and a corresponding positive phase difference are classified as inductive loads, while branches with a power factor greater than zero and a corresponding negative phase difference are classified as capacitive loads. This is determined by the electrical phase characteristics and reactive power transmission patterns of inductive and capacitive loads in low-voltage power distribution systems, and also matches the control logic requirements of dynamic reactive power compensation. In other words, under normal operating conditions of a low-voltage power distribution system, the load power factor is always greater than zero, indicating that the load is in a normal power consumption state and there are no abnormal situations such as reverse power backflow, which can be directly used as a prerequisite for judgment. A positive phase difference indicates that the current signal phase lags behind the voltage signal phase, which is an electrical characteristic of inductive devices such as motors, fans, and reactors. Inductive loads consume inductive reactive power and are the main targets for reactive power compensation in low-voltage power distribution networks. Accurately identifying them is essential to provide a basis for subsequent calculation of inductive reactive power demand and the implementation of capacitor compensation. A negative phase difference indicates that the current signal phase leads the voltage signal phase. This is an electrical characteristic of capacitive devices such as parallel capacitors and power capacitor banks. Capacitive loads will send inductive reactive power to the system. If misjudgment occurs, it will lead to incorrect compensation direction, overcompensation and resonance problems. Accurate judgment can avoid repeated compensation of capacitive loads and ensure the accuracy of the compensation strategy.Using voltage signals as a unified reference phase to determine the sign of phase difference, and combining phase difference with power factor to distinguish load types, it can unify the phase determination benchmark for the entire branch. Based on the dual determination of phase difference and power factor, it can accurately distinguish between inductive and capacitive loads, avoiding problems such as reversed compensation direction, overcompensation, and undercompensation caused by incorrect load type judgment. At the same time, it provides a load classification basis for reactive power characteristic analysis, multi-timescale reactive power demand prediction, and harmonic constraint compensation decision-making, improving the adaptability and compensation accuracy of dynamic reactive power compensation to complex loads.

[0040] In some embodiments of this application, determining the reactive power characteristic data sequence includes: determining a sliding time window based on a preset time length; moving the sliding time window; determining a reactive power data subsequence within each sliding time window; arranging the reactive power data subsequences in time order within each sliding time window; determining the reactive power change based on the difference between the first and last data points of the window; determining the instantaneous change rate based on the difference between adjacent sampling points and the sampling time interval; summing the instantaneous change rates based on time order; calculating the average of the summed results based on the number of sampling points to determine the change rate; determining the maximum and minimum values ​​of the reactive power data within each sliding time window; and determining the fluctuation amplitude based on the difference between the maximum and minimum values.

[0041] Specifically, a sliding time window is an analysis interval for capturing reactive power data, constructed according to a pre-set fixed time length. The time length can be dynamically set according to the low-voltage distribution cabinet. By moving the sliding time window, reactive power data subsequences are determined within each sliding time window. A reactive power data subsequence represents a local reactive power data set captured by a single sliding time window, sorted by acquisition time. With a 10-second sliding time window and moving once every second, reactive power data subsequences for different time periods such as 0-10 seconds, 1-11 seconds, and 2-12 seconds can be captured sequentially. Continuously moving the window achieves full coverage analysis of reactive power data, ensuring the completeness of feature extraction. Within each sliding time window, reactive power data subsequences are arranged chronologically, and the reactive power change is determined based on the difference between the data at the beginning and end of the window. The reactive power change represents the difference between the reactive power data at the start and end of a single sliding time window. For example, in a 10-second sliding time window, if the reactive power data at the beginning is 20 kvar and at the end is 30 kvar, the difference is 10 kvar, meaning the reactive power change within that window is 10 kvar. This direct quantification of the overall increase or decrease in reactive power through the difference between the beginning and end of the window intuitively reflects the total reactive power change of the load during that period, providing a quantitative indicator for load dynamic characteristic analysis. The instantaneous rate of change is determined based on the difference between adjacent sampling points and the sampling time interval. The instantaneous rate of change represents the ratio of the difference between two adjacent reactive power sampling points to the corresponding sampling time interval within a single sliding time window. For example, if the reactive power data difference between adjacent sampling points is 2 kvar and the sampling time interval is 0.5 seconds, the instantaneous rate of change is 4 kvar / second. The instantaneous rates of change are summed based on time sequence, and the mean of the summation is calculated based on the number of sampling points to determine the rate of change. The rate of change represents the average of all instantaneous rates of change within a single sliding time window. For example, if there are 9 instantaneous rates of change within the window, the summation result is 18 kvar / s, and with 9 sampling points, the mean rate of change is 2 kvar / s. This mean calculation quantifies the average rate of change of reactive power of the load within that time period, reflecting the overall trend of reactive power change. The maximum and minimum values ​​of reactive power data are determined within each sliding time window, and the fluctuation amplitude is determined based on the difference between the maximum and minimum values. The fluctuation amplitude represents the maximum fluctuation range of reactive power data within a single sliding time window. For example, if the maximum reactive power value within the window is 35 kvar and the minimum is 18 kvar, the difference is 17 kvar, meaning the fluctuation amplitude of that window is 17 kvar. This intuitively quantifies the severity of reactive power fluctuations and identifies the differentiated characteristics of impulsive and stable loads.The reactive power changes, rates of change, and fluctuation amplitudes of branches with the same load type are categorized. By extracting and classifying multi-dimensional characteristic parameters, the reactive dynamic characteristics of the load are fully quantified. This avoids the shortcomings of relying solely on reactive power numerical analysis while ignoring dynamic change patterns, and improves the adaptability and compensation accuracy of dynamic reactive power compensation for complex loads.

[0042] In some embodiments of this application, when dividing reactive power characteristic data sequences into second-level short-time sequences, minute-level trend sequences, and hour-level periodic sequences on a unified time axis, the following steps are included: dividing data in the reactive power characteristic data sequence whose time interval is less than or equal to a second-level threshold into second-level short-time sequences; dividing data in the reactive power characteristic data sequence whose time interval is greater than a second-level threshold but less than a minute-level threshold into minute-level trend sequences; and dividing data in the reactive power characteristic data sequence whose time interval is greater than or equal to a minute-level threshold into hour-level periodic sequences, wherein the second-level threshold is less than the minute-level threshold.

[0043] Specifically, all reactive power characteristic data sequences are first aligned to a unified time axis. The unified time axis represents a shared, continuous, and unique time reference axis for all reactive power characteristic data. All data points are sorted according to the same time scale, eliminating the division deviation caused by inconsistent time references for data from different time periods. Data with time intervals less than or equal to a second-level threshold in the reactive power characteristic data sequence are divided into second-level short-time sequences. The time interval represents the time difference between two adjacent data points in the reactive power characteristic data sequence, and the second-level threshold represents the preset time threshold for dividing the second-level short-time sequences. The second-level threshold is less than the minute-level threshold. The second-level short-time sequence represents a time series set composed of reactive power characteristic data with time intervals less than or equal to the second-level threshold. For example, if the second-level threshold is set to 1 second, and the time intervals between adjacent reactive power characteristic data points are 0.2 seconds, 0.5 seconds, and 1 second, and all meet the condition of being less than or equal to the second-level threshold, then all of this data is classified into the second-level short-time sequence. By accurately filtering out high-frequency, short-interval instantaneous characteristic data, the changing patterns of instantaneous load impacts and rapid fluctuations are matched. Furthermore, by accurately selecting mid-frequency and medium-interval trend characteristic data, the system matches the changing patterns of slow, gradual load changes and continuous adjustments in operating conditions. Secondly, by selecting low-frequency and long-interval periodic characteristic data, it matches the periodic changes of loads operating on a shift-by-shift, day-night cycle. By categorizing data hierarchically by time interval length, the system can accurately separate the instantaneous fluctuation characteristics, gradual trend characteristics, and periodic patterns of the load, avoiding the shortcomings of single-dimensional time-series analysis that cannot account for multiple types of load changes. This improves the adaptability and accuracy of dynamic reactive power compensation for complex loads.

[0044] In some embodiments of this application, determining the reactive power demand forecast sequence includes: arranging the second-level short-time sequences corresponding to each load type in time, performing differential calculation on adjacent sampling points to determine the differential value sequence, determining the direction of change based on the differential value sequence, dividing continuous differential values ​​with the same direction of change into segments to determine the short-time change segment sequence, dividing the minute-level trend sequence corresponding to each load type into several continuous segments based on time order, fitting each continuous segment to determine the fitting curve of the corresponding segment, determining the direction and degree of change of each segment based on the fitting curve, and merging adjacent segments with the same direction of change and a degree of change less than a threshold to determine the trend change sequence.

[0045] In some embodiments of this application, determining the reactive power demand forecast sequence includes: dividing the hourly cycle sequence corresponding to each load type into several cycle intervals based on time order; normalizing the data in each cycle interval; aligning the data at corresponding time positions in different cycle intervals; superimposing the aligned data to determine the cycle change characteristic curve; reconstructing the data in different cycles based on the cycle change characteristic curve to determine the cycle change sequence; and splicing the short-time change segment sequence, trend change sequence, and cycle change sequence to determine the reactive power demand forecast sequence.

[0046] Specifically, the second-level short-time series corresponding to each load type are time-sorted. Time sorting means that all data within the second-level short-time series are sorted sequentially according to the collection time, thus ensuring the continuity and regularity of the data time sequence. Differential calculation is performed on adjacent sampling points, and the differential value sequence represents the numerical sequence composed of the differential calculation results of all adjacent sampling points. For example, if the sampling point values ​​of the second-level short-time series are 20kvar, 24kvar, and 26kvar respectively, the differential calculation yields a differential value sequence of +4kvar and +2kvar. By accurately extracting the instantaneous change amplitude of reactive power at the second scale, the instantaneous fluctuation characteristics of impact loads such as motor start-up and shutdown and sudden equipment commissioning are quickly captured. Based on the differential value sequence, the direction of change is determined. The direction of change indicates that a positive differential value represents an increase in reactive power, and a negative differential value represents a decrease in reactive power. This clarifies the instantaneous increase and decrease trend of reactive power. Continuous difference values ​​with consistent changing directions are segmented, meaning that continuous difference values ​​with the same changing direction are grouped into an independent segment, thus determining the short-term change segment sequence. The short-term change segment sequence is a second-level feature sequence composed of multiple continuous segments with consistent instantaneous changing directions. By identifying instantaneous change characteristics at the second scale, the reliability of short-term feature identification is improved. The minute-level trend sequence corresponding to each load type is divided into several continuous segments based on time order. Continuous segments represent uninterrupted data segments in the minute-level trend sequence divided by time. Decomposing long-term trend data into short segments reduces the overall fitting complexity and improves the accuracy of trend fitting. Fitting is performed on each continuous segment using methods such as least squares fitting and polynomial fitting. The fitted curve represents the reactive power change pattern of a single continuous segment. Fitting can eliminate small random fluctuations in minute-level data. Based on the fitted curve, the changing direction and degree of change of each segment are determined. The degree of change is represented by the absolute value of the slope of the fitted curve, used to quantify the rate of reactive power change. Adjacent segments with consistent direction of change and a change level less than a threshold are merged. The trend change sequence is a minute-level feature sequence composed of merged, regularized, gradually changing trend segments. For the hourly periodic sequences corresponding to each load type, the data is divided into several periodic intervals based on time sequence. Long-cycle time-series data is decomposed into periodic units, and the data within each periodic interval is normalized. Normalization scales the data from different periodic intervals to a uniform numerical range, eliminating numerical deviations caused by differences in load intensity across different periods. Data at corresponding time positions within different periodic intervals are aligned, meaning data at the same relative time point within different periodic intervals are mapped one-to-one.The aligned data are superimposed, which involves summing the values ​​of multiple data points at the same time position after alignment to determine the periodic variation characteristic curve. This curve represents the reactive power variation pattern within a standard load cycle, incorporating common patterns from multiple cycles. For example, in a factory power distribution network with an 8-hour cycle, the hourly cycle sequence is sequentially divided into several independent cycle intervals, such as 8:00-16:00, 16:00-24:00, and 8:00-16:00 the next day. The reactive power data for the first cycle interval is 10-30 kvar, and for the second cycle interval it is 15-35 kvar. Normalization is then applied to both sets of data. The data is scaled to a standard range of 0-1. Then, the data at corresponding time points within different cycle intervals are aligned. This means that the reactive power data for the 1st, 2nd, ... 8th hour of each cycle interval are matched one-to-one, ensuring data consistency at the same time point across different cycles. The normalized data for the 2nd hour of the three cycle intervals after alignment are 0.2, 0.22, and 0.21. The average of these values ​​after superposition is 0.21. This process is repeated for all time points, and the superposition results are arranged chronologically to determine the cycle variation characteristic curve. By integrating the common reactive power variation patterns across multiple cycles, random fluctuations and abnormal interference within individual cycles are eliminated. Reconstruction uses the cycle variation characteristic curve as a benchmark to restore the reactive power data for each independent cycle interval. Using the period duration corresponding to the periodic variation characteristic curve as the matching basis, the time span of each period interval to be reconstructed is aligned to ensure that the time length of each period interval to be reconstructed is completely consistent with the periodic variation characteristic curve. For example, if the periodic variation characteristic curve is a standard period template generated based on 8 hours, each period interval to be reconstructed is also a complete 8-hour running segment. This ensures that the reconstructed data conforms to the inherent periodic variation law of the load, eliminating random fluctuations and abnormal interference within a single cycle. After completing the data reconstruction of all period intervals, the reconstructed data of each cycle are sequentially and continuously spliced ​​according to the time order of a unified time axis to finally determine the periodic variation sequence. The periodic variation sequence is a time-series data set that is composed of the reconstructed data of each cycle and completely represents the reactive power variation law of the load cycle. By fusing features of multiple time scales, it simultaneously covers all types of load variation characteristics, including instantaneous impacts, slow gradual changes, and periodic repetitions, ensuring that the reactive power demand forecast sequence can accurately reflect the long-term periodic variation characteristics of the load and improving the adaptability of reactive power compensation.

[0047] In some embodiments of this application, determining the reactive power compensation sequence includes: performing discrete Fourier transform on the voltage and current signals to extract the fundamental component and each harmonic component, arranging each harmonic component in frequency order, determining the corresponding harmonic voltage component based on the arrangement result and the system equivalent impedance, determining the harmonic influence range according to the amplitude range of the harmonic voltage component, determining the reactive power compensation amount corresponding to each time point based on the machine learning model and the harmonic influence range, and constructing all reactive power compensation amounts into a reactive power compensation sequence.

[0048] Specifically, a Discrete Fourier Transform (DFT) is performed on the voltage and current signals to extract the fundamental component and harmonic components. The fundamental component represents the basic voltage and current electrical components at the 50Hz power frequency in the power distribution system, while each harmonic component represents the harmonic voltage and current electrical components whose frequencies are integer multiples of the fundamental frequency. This provides frequency domain data support for harmonic impact analysis. The harmonic components are arranged in frequency order, which sorts them from smallest to largest harmonic order to avoid data confusion and subsequent calculation errors. Based on the arrangement and the system's equivalent impedance, the corresponding harmonic voltage components are determined. These harmonic voltage components are the harmonic voltages generated when each harmonic current flows through the system's equivalent impedance. The value of a harmonic voltage component is equal to the product of the corresponding harmonic current component and the system's equivalent impedance. For example, if the 5th harmonic current component is 10A and the system's equivalent impedance is 0.4Ω, the 5th harmonic voltage component is calculated as follows: The value is 4V, which accurately quantifies the actual voltage interference generated by each harmonic in the system, intuitively reflecting the degree of harmonic impact on the power distribution system. The harmonic influence range is determined according to the amplitude range of the harmonic voltage component, that is, the harmonic interference strength range is divided according to the magnitude of the harmonic voltage component amplitude. For example, the harmonic voltage component amplitude is set as low harmonic influence range, 2V to 5V as medium harmonic influence range, and greater than 5V as high harmonic influence range. The degree of harmonic interference is quantified into graded ranges, providing a standardized basis for the harmonic constraint judgment of reactive power compensation. When determining the reactive power compensation sequence, the reactive power compensation amount corresponding to each time point is determined based on a machine learning model and the harmonic influence interval. The machine learning model is a computational model pre-trained using historical harmonic influence interval data, reactive power demand data, and safe reactive power compensation amount data as samples. This model uses the harmonic influence interval as a constraint and combines it with reactive power demand prediction data to output the safe reactive power compensation amount at each time point. The harmonic influence interval is incorporated as a constraint into the reactive power compensation amount calculation. The machine learning model achieves accurate matching between harmonic constraints and reactive power demand, avoiding the safety hazards of capacitor resonance and harmonic amplification with system harmonics. All reactive power compensation amounts are constructed into a reactive power compensation amount sequence. The reactive power compensation amount sequence is a set of reactive power compensation amount values ​​at each time point after harmonic constraints, arranged in a unified time axis order. The reactive power compensation amount sequence ensures that subsequent switching strategies meet both reactive power compensation requirements and system harmonic safe operation requirements, further improving the reliability of reactive power compensation.

[0049] It should be noted that the machine learning model in this application can preferably be implemented using existing technologies in the field, including decision tree models and convolutional neural networks, etc., and is not the focus of the improvement claimed in this application. The focus of the improvement in this application lies in its constraint processing logic using the harmonic influence range. In some embodiments, the machine learning model takes the harmonic influence range and reactive power demand as input and outputs the constrained reactive power compensation amount to avoid harmonic amplification and resonance risks. Those skilled in the art can complete the adaptation, replacement, or equivalent implementation based on the input-output relationship, parameter configuration rules, and calling sequence disclosed in this application, combined with existing disclosed technologies or conventional engineering methods, without affecting the implementation of the technical solution of this application.

[0050] In some embodiments of this application, determining the switching control dataset includes: determining a capacitor combination dataset based on the capacity of the capacitor banks in the low-voltage distribution cabinet; obtaining the compensation capacity corresponding to each capacitor bank; calculating the difference between the compensation capacity and each reactive power compensation quantity in the reactive power compensation quantity sequence; sorting each capacitor bank based on the difference; sequentially selecting the capacitor bank with the smallest difference as a candidate switching combination; recording adjacent switching times; judging the time interval between adjacent switching times; removing data with a time interval less than the minimum switching time interval; removing data with a capacity change exceeding the step range; arranging the remaining candidate switching combinations according to the removal results; and determining the switching control dataset.

[0051] In some embodiments of this application, when switching is performed based on the switching control dataset, the following is included: switching the capacitor bank in the low-voltage distribution cabinet based on the switching control dataset.

[0052] Specifically, the capacitor group dataset consists of all individual capacitors and multiple capacitor groups within a low-voltage distribution cabinet. The capacity of each capacitor group represents the rated compensation capacity of each capacitor in the low-voltage distribution cabinet. For example, if a low-voltage distribution cabinet is equipped with three capacitor groups of 10kvar, 20kvar, and 30kvar, its capacitor group dataset includes 10kvar (single 10kvar), 20kvar (single 20kvar), 30kvar (single 30kvar), 30kvar (10kvar + 20kvar), 40kvar (10kvar + 30kvar), and 50kvar (20kvar + 30kvar). The system includes 60kvar (10kvar + 20kvar + 30kvar) capacitor banks, comprehensively covering all possible combinations of compensation capacity to ensure that the best combination for reactive power compensation is found. The compensation capacity for each capacitor bank is obtained; this capacity represents the actual reactive power compensation provided by a single capacitor or multiple capacitor banks, i.e., the total capacity corresponding to each combination. The difference between the compensation capacity and the reactive power compensation in the reactive power compensation sequence is calculated. For example, if the reactive power compensation at a certain time point is 25kvar, the difference for the 10kvar combination is 15kvar, the difference for the 20kvar combination is 5kvar, and the difference for the 30kvar combination is 5kvar. By calculating the difference, the gap between each capacitor bank and the actual compensation requirement is quantified, providing a quantitative standard for selecting the optimal combination. Based on the magnitude of the difference, each capacitor bank is ranked, and the capacitor bank with the smallest difference is selected as the candidate switching combination. Furthermore, when determining the candidate switching combination, adjacent switching times are recorded. Adjacent switching times represent the switching time points corresponding to two consecutive candidate switching combinations. The minimum switching time interval represents the minimum allowable time limit for two consecutive switching operations of capacitors in the low-voltage distribution cabinet. For example, if the minimum switching time interval is set to 30 seconds, the previous switching time is at the 10th second, and the current candidate switching time is at the 35th second, the time interval of 25 seconds is less than 30 seconds, then the candidate... Selected switching combinations are eliminated to avoid repeated switching of capacitors in a short period of time, thereby reducing the loss of capacitor equipment and preventing compensation oscillations during reactive power compensation. Data with capacity changes exceeding the stepped range are also eliminated. The stepped range of capacity changes represents the maximum allowable range of compensation capacity change between two adjacent switching operations. For example, if the stepped range of capacity change is set to no more than 20 kvar, the compensation capacity of the previous switching combination is 10 kvar, and the compensation capacity of the current candidate switching combination is 40 kvar, and the capacity change is 30 kvar, which exceeds the stepped range, then the combination is eliminated. This is to prevent grid voltage fluctuations and reactive power compensation imbalances caused by sudden changes in compensation capacity, thereby improving the stability of the compensation process.Based on the elimination results, the remaining candidate switching combinations are arranged to determine the switching control dataset. The switching control dataset is a set of candidate switching combinations that meet all switching requirements after double constraint screening, arranged in chronological order. Based on the switching control dataset, the capacitor banks in the low-voltage distribution cabinet are switched, and the capacitor banks are either connected (connected to the circuit for compensation) or disconnected (removed from the circuit to stop compensation) to ensure the reliability of dynamic reactive power compensation, and finally achieve precise control of dynamic reactive power compensation of the low-voltage distribution cabinet.

[0053] In summary, the beneficial effects of this invention are as follows: It acquires voltage signals, current signals, and corresponding reactive power data, providing a data foundation for subsequent load type determination and compensation decisions, ensuring that all analyses are based on actual operating conditions and avoiding judgment biases caused by missing data. By determining the load type and distinguishing between inductive and capacitive loads, it avoids problems such as incorrect compensation direction and overcompensation caused by misjudgment of load attributes, providing a classification basis for subsequent reactive power analysis. The use of a sliding time window to segment reactive power data enables accurate extraction of the dynamic change characteristics of the load, avoiding the limitations of a single data dimension. Furthermore, by dividing the data into sequences at different time scales and combining the processing methods at each scale, it comprehensively captures the instantaneous fluctuations, gradual trends, and periodic patterns of the load, solving the shortcomings of traditional single-dimensional analysis in adapting to complex loads. Furthermore, by extracting harmonic information through discrete Fourier transform and incorporating the harmonic effects into the compensation decision, the targetedness and stability of reactive power compensation are ensured, as well as the operational safety of low-voltage switchgear. By collecting and updating closed-loop data and adapting to load changes in real time, the reliability and adaptability of dynamic compensation are improved, thus meeting the safety requirements of grid-connected operation.

[0054] In another preferred embodiment based on the above embodiments, see [reference] Figure 2 As shown, this embodiment provides a dynamic reactive power compensation system for low-voltage distribution cabinets based on load characteristic analysis, used to apply a dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis, including:

[0055] The data acquisition and analysis unit is configured to acquire the voltage signal, current signal and corresponding reactive power data of each branch, determine the power factor based on the phase difference between the voltage signal and the current signal, determine the load based on the power factor, and determine the load type of each branch.

[0056] The first processing unit is configured to perform sliding time window segmentation processing on reactive power data, classify the reactive power change, change rate and fluctuation amplitude of branches of the same load type, determine the reactive power characteristic data sequence, and divide the reactive power characteristic data sequence into second-level short-time sequence, minute-level trend sequence and hour-level periodic sequence on a unified time axis. It performs adjacent difference on the second-level short-time sequence corresponding to each load type, continuous segment fitting on the minute-level trend sequence and periodic reconstruction on the hour-level periodic sequence. Based on the time order, it merges the results of adjacent difference, continuous segment fitting and periodic reconstruction to determine the reactive power demand forecast sequence.

[0057] The second processing unit is configured to perform discrete Fourier transform on voltage and current signals, determine the harmonic influence sequence based on each harmonic component and the system equivalent impedance, compare the harmonic influence sequence with the reactive power demand prediction sequence to determine the reactive power compensation sequence, determine the capacitor combination dataset based on the capacitor banks in the low-voltage distribution cabinet, perform step-by-step screening based on the reactive power compensation and the capacity difference of each capacitor bank, and remove data that does not meet the minimum switching time interval constraint and the capacity change step constraint to determine the switching control dataset.

[0058] The compensation processing unit is configured to perform switching based on the switching control dataset, and to collect updated voltage signals, current signals and corresponding reactive power data of each branch after switching.

[0059] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A dynamic reactive power compensation method for low-voltage switchgear based on load characteristic analysis, characterized in that, include: The voltage signal, current signal and corresponding reactive power data of each branch are acquired. Based on the phase difference between the voltage signal and the current signal, the power factor is determined. The load is judged according to the power factor to determine the load type of each branch. The reactive power data is segmented by a sliding time window. The reactive power change, rate of change, and fluctuation amplitude of branches with the same load type are classified to determine the reactive power characteristic data sequence. The reactive power characteristic data sequence is divided into second-level short-time sequence, minute-level trend sequence, and hour-level periodic sequence on a unified time axis. Adjacent difference is performed on the second-level short-time sequence corresponding to each load type, continuous segment fitting is performed on the minute-level trend sequence, and periodic reconstruction is performed on the hour-level periodic sequence. The results of adjacent difference, continuous segment fitting, and periodic reconstruction are fused based on the time order to determine the reactive power demand prediction sequence. Discrete Fourier transform is performed on the voltage and current signals, and the harmonic influence sequence is determined based on each harmonic component and the system equivalent impedance. The harmonic influence sequence and the reactive power demand prediction sequence are compared under constraints to determine the reactive power compensation sequence. Based on the capacitor banks in the low-voltage distribution cabinet, the capacitor combination dataset is determined. The reactive power compensation is filtered step by step based on the difference between the capacity of each capacitor bank and the capacity of each capacitor bank. Data that does not meet the minimum switching time interval constraint and the capacity change step constraint are removed to determine the switching control dataset. Switching is performed based on the aforementioned switching control dataset, and after switching, the updated voltage signal, current signal, and corresponding reactive power data of each branch are collected.

2. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 1, characterized in that, When determining the load type of each branch, the process includes: extracting the phase angles of the voltage and current signals, using the voltage signal as a reference phase, comparing the phase of the current signal, determining the phase difference as positive when the phase of the current signal lags behind the phase of the voltage signal, and determining the phase difference as negative when the phase of the current signal leads the phase of the voltage signal. The power factor is determined based on the sign and range of the phase difference. The load types include inductive loads and capacitive loads. Branches with a power factor greater than zero and a corresponding positive phase difference are identified as inductive loads, and branches with a power factor greater than zero and a corresponding negative phase difference are identified as capacitive loads.

3. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 2, characterized in that, When determining the reactive power characteristic data sequence, the process includes: determining a sliding time window based on a preset time length; moving the sliding time window to determine a reactive power data subsequence within each sliding time window; arranging the reactive power data subsequence in chronological order within each sliding time window; determining the reactive power change based on the difference between the first and last data points of the window; determining the instantaneous change rate based on the difference between adjacent sampling points and the sampling time interval; summing the instantaneous change rates in chronological order; calculating the mean of the summation results based on the number of sampling points; determining the maximum and minimum values ​​of the reactive power data within each sliding time window; and determining the fluctuation amplitude based on the difference between the maximum and minimum values.

4. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 3, characterized in that, When dividing the reactive power characteristic data sequence into second-level short-time series, minute-level trend series, and hour-level periodic series on a unified time axis, the process includes: dividing data in the reactive power characteristic data sequence whose time interval is less than or equal to a second-level threshold into second-level short-time series; dividing data in the reactive power characteristic data sequence whose time interval is greater than a second-level threshold but less than a minute-level threshold into minute-level trend series; and dividing data in the reactive power characteristic data sequence whose time interval is greater than or equal to a minute-level threshold into hour-level periodic series, wherein the second-level threshold is less than the minute-level threshold.

5. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 4, characterized in that, The process of determining the reactive power demand forecast sequence includes: arranging the second-level short-time sequences corresponding to each load type in time, performing differential calculations on adjacent sampling points to determine the differential value sequence, determining the direction of change based on the differential value sequence, dividing continuous differential values ​​with the same direction of change into segments to determine the short-time change segment sequence, dividing the minute-level trend sequence corresponding to each load type into several continuous segments based on time order, fitting each continuous segment to determine the fitting curve for the corresponding segment, determining the direction and degree of change for each segment based on the fitting curve, and merging adjacent segments with the same direction of change and a degree of change less than a threshold to determine the trend change sequence.

6. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 5, characterized in that, The process of determining the reactive power demand forecast sequence includes: dividing the hourly cycle sequence corresponding to each load type into several cycle intervals based on time order; normalizing the data within each cycle interval; aligning the data at corresponding time positions within different cycle intervals; superimposing the aligned data to determine the cycle change characteristic curve; reconstructing the data within different cycles based on the cycle change characteristic curve to determine the cycle change sequence; and splicing the short-term change segment sequence, trend change sequence, and cycle change sequence to determine the reactive power demand forecast sequence.

7. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 6, characterized in that, The process of determining the reactive power compensation sequence includes: performing a discrete Fourier transform on the voltage and current signals to extract the fundamental component and each harmonic component, arranging each harmonic component in frequency order, determining the corresponding harmonic voltage component based on the arrangement result and the system equivalent impedance, determining the harmonic influence range based on the amplitude range of the harmonic voltage component, determining the reactive power compensation amount corresponding to each time point based on the machine learning model and the harmonic influence range, and constructing the reactive power compensation sequence from all the reactive power compensation amounts.

8. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 7, characterized in that, The process of determining the switching control dataset includes: determining a capacitor combination dataset based on the capacity of the capacitor banks in the low-voltage distribution cabinet; obtaining the compensation capacity corresponding to each capacitor bank; calculating the difference between the compensation capacity and each reactive power compensation quantity in the reactive power compensation quantity sequence; sorting each capacitor bank based on the difference; selecting the capacitor bank with the smallest difference as a candidate switching combination; recording adjacent switching times; judging the time interval between adjacent switching times; removing data with a time interval smaller than the minimum switching time interval; removing data with capacity changes exceeding the step range; arranging the remaining candidate switching combinations according to the removal results; and determining the switching control dataset.

9. The dynamic reactive power compensation method for low-voltage distribution cabinets based on load characteristic analysis according to claim 8, characterized in that, When switching based on the switching control dataset, the process includes: switching the capacitor bank in the low-voltage distribution cabinet based on the switching control dataset.

10. A dynamic reactive power compensation system for low-voltage switchgear based on load characteristic analysis, used to apply the dynamic reactive power compensation method for low-voltage switchgear based on load characteristic analysis as described in any one of claims 1-9, characterized in that, include: The acquisition and analysis unit is configured to acquire the voltage signal, current signal and corresponding reactive power data of each branch, determine the power factor based on the phase difference between the voltage signal and the current signal, determine the load based on the power factor, and determine the load type of each branch. The first processing unit is configured to perform sliding time window segmentation processing on the reactive power data, classify the reactive power change, change rate and fluctuation amplitude of the same load type branch, determine the reactive power characteristic data sequence, and divide the reactive power characteristic data sequence into second-level short-time sequence, minute-level trend sequence and hour-level periodic sequence on a unified time axis. The second-level short-time sequence corresponding to each load type is subjected to adjacent difference, the minute-level trend sequence is subjected to continuous segment fitting and the hour-level periodic sequence is subjected to periodic reconstruction. The results of adjacent difference, continuous segment fitting and periodic reconstruction are fused based on time order to determine the reactive power demand prediction sequence. The second processing unit is configured to perform discrete Fourier transform on the voltage and current signals, determine the harmonic influence sequence based on each harmonic component and the system equivalent impedance, compare the harmonic influence sequence with the reactive power demand prediction sequence to determine the reactive power compensation sequence, determine the capacitor combination dataset based on the capacitor banks in the low-voltage distribution cabinet, perform step-by-step filtering based on the reactive power compensation and the capacity difference of each capacitor bank, and remove data that does not meet the minimum switching time interval constraint and the capacity change step constraint to determine the switching control dataset. The compensation processing unit is configured to perform switching based on the switching control dataset, and to collect updated voltage signals, current signals and corresponding reactive power data of each branch after switching.

Images