A switch cabinet opening and closing timing optimization method based on load fluctuation

Abstract

The application discloses a switch cabinet opening and closing timing optimization method based on load fluctuation, and belongs to the field of power system automatic control. In view of the problem that the existing technology lacks prospective load prediction, resulting in the mismatch between opening and closing operations and load changes, a load fluctuation prediction model is constructed, and a rolling prediction strategy is adopted to obtain future load values; a mapping relationship database of load trend and timing parameters is established, and the initial timing parameters are queried by taking the load change rate and the fluctuation amplitude as indexes; the running state information and the initial parameters are input into an optimization algorithm, and the optimal timing parameters are solved under the constraint condition, with the minimum electrical stress and the maximum equipment life as the targets; and the optimal parameters are used to drive the power waveform, the transmission mechanism and the delay trigger time for pre-adjustment control, so that the switch equipment can be operated at the most favorable moment of the predicted load. The application realizes the accurate matching between the operation timing and the load fluctuation, effectively reduces the electrical stress, reduces the contact ablation, and prolongs the service life of the equipment.

Description

Technical Field

[0001] This application belongs to the field of power system automation control technology, specifically relating to a method for optimizing the opening and closing timing of switchgear based on load fluctuations. Background Technology

[0002] As the core power distribution equipment in the power system, the timing parameters of the switching operation of the switchgear directly affect the reliability and service life of the equipment. In actual operation, the load current often exhibits frequent fluctuations. How to achieve precise optimization of the switching timing under load change conditions is a technical problem that urgently needs to be solved.

[0003] Common switchgear optimization control methods mainly focus on improving mechanical characteristics, adjusting drive current waveforms, and overall response to load changes. However, these methods are all based on current or historical load conditions and lack the ability to proactively analyze and predict future load fluctuation trends. They cannot optimize operation timing parameters in advance before drastic load fluctuations occur, leading to a mismatch between the timing of opening and closing operations and load changes. This results in electrical stress concentration, accelerated aging of equipment insulation, and in severe cases, operational failures or equipment damage. This technical bottleneck, which lacks a predictive timing adjustment mechanism, has become a key problem restricting the safe and reliable operation of switchgear. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for optimizing the opening and closing sequence of switchgear based on load fluctuations, which can effectively solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for optimizing the opening and closing sequence of switchgear based on load fluctuations, the method comprising the following steps: A load fluctuation prediction model is constructed based on standardized load data sequences. The predicted load values ​​within a preset time period are predicted using the load fluctuation prediction model, and the parameters of the load fluctuation prediction model are updated using a rolling prediction strategy. Establish a mapping relationship library between load trend and circuit breaker timing parameters. Calculate the load change rate and load fluctuation amplitude based on the predicted load value. Use the load change rate and load fluctuation amplitude as indexes to query the mapping relationship library to obtain the initial timing parameter combination. The data structure of the mapping relationship library adopts a multidimensional lookup table or a KD-tree index structure. The operating status information of the switchgear is obtained, and the operating status information and the initial timing parameters are combined as input parameters of the optimization algorithm. Under the premise of satisfying the constraints, the optimization is performed with the objectives of minimizing electrical stress and maximizing equipment life to generate the optimal timing parameter combination. Based on the optimal timing parameter combination, the drive parameters of the opening and closing operation to be performed are pre-adjusted to control the switching equipment so that the switching equipment completes the opening and closing operation at the most favorable time of the predicted load.

[0006] Furthermore, the rolling forecasting strategy also includes a forecasting error monitoring mechanism. The forecasting error monitoring mechanism calculates the root mean square error between the forecasted load value and the actual load value collected in each forecasting period. When the root mean square error of multiple consecutive forecasting periods exceeds a preset threshold, the model retraining process is triggered, and the load fluctuation forecasting model is retrained using the accumulated latest complete dataset.

[0007] Furthermore, the load fluctuation prediction model adopts a two-layer long short-term memory network structure, which includes two long short-term memory unit layers arranged in series. The first layer of long short-term memory unit receives the standardized load data sequence as input and extracts the local dependencies and short-term fluctuation characteristics of the load data in time series. The second layer of long short-term memory unit performs high-level feature abstraction based on the output of the first layer of long short-term memory unit to capture the long-term dependencies and periodic change patterns in the load sequence. Alternatively, the load fluctuation prediction model adopts a converter encoder-decoder structure, which employs a multi-head self-attention mechanism. The encoder part is composed of alternating stacks of multi-head self-attention layers and feedforward neural network layers. The decoder part generates a load prediction sequence for future time moments step by step through a masked self-attention mechanism based on the high-dimensional feature representation output by the encoder part and the generated prediction values.

[0008] Furthermore, the establishment of the mapping relationship library between load trends and circuit breaker timing parameters includes: Historical switchgear operation data were collected, and simulation data was generated by modeling the opening and closing process of the switchgear using electromagnetic transient simulation software and combined simulation experiments. Each record includes load status parameters, operation timing parameters and operation result evaluation indicators. Based on the historical switchgear operation data and the simulation data, the mapping relationship is modeled using a piecewise linear interpolation method or a neural network fitting method to obtain the mapping relationship library; When querying the mapping relationship library, the load change rate and the load fluctuation amplitude are used as query points. Nearest neighbor search is performed in the mapping relationship library. The nearest neighbor search uses the Euclidean distance between the query point and the candidate point as the criterion for determining the degree of matching. The time series parameter combination corresponding to the candidate point with the smallest Euclidean distance is returned as the initial time series parameter combination.

[0009] Further, the step of acquiring the operating status information of the switchgear and combining the operating status information with the initial timing parameters as input parameters for the optimization algorithm includes: The current temperature of the current-carrying element is used as the upper limit of the equipment temperature threshold in the constraint conditions. The current position of the switching mechanism and the cumulative number of circuit breaker operations are used as the lower limit of the safe operating interval in the constraint conditions. The energy storage state of the closing spring and the current and voltage values ​​of the drive motor are used as the reference state inputs for optimizing the action speed curve.

[0010] Furthermore, the optimization algorithm employs either particle swarm optimization or gradient descent optimization, and the objective of minimizing electrical stress and maximizing equipment lifespan includes: Maximizing equipment lifespan is converted into minimizing equipment lifespan loss. The normalized electrical stress value and the normalized lifespan loss value are multiplied by their respective weighting coefficients and then added together to obtain the comprehensive objective function value. The minimum of the comprehensive objective function value is taken as the optimization objective.

[0011] Furthermore, the constraints include the action time window, the equipment temperature threshold, and the safe operation interval; The constraint conditions are handled using the penalty function method, which adds a penalty term to the objective function: For inequality constraints, the penalty term is zero when the optimization parameter value satisfies the inequality constraint, and the penalty term is the distance of the parameter from the constraint boundary multiplied by a preset penalty coefficient when the inequality constraint is violated. For equality constraints, the penalty term is the absolute value of the difference between the two sides of the equation multiplied by a preset penalty coefficient; The summation of all penalty terms is added to the objective function to form an augmented objective function that satisfies the constraints.

[0012] Furthermore, the pre-adjustment of drive parameters for the opening and closing operation to be performed based on the optimal timing parameter combination includes: Adjust the output waveform parameters of the drive power supply, including amplitude and frequency. Adjusting the amplitude changes the peak current of the drive current, and adjusting the frequency changes the pulse width and pulse interval of the drive current pulse. In addition, the control parameters of the transmission mechanism are adjusted. The control parameters of the transmission mechanism include the optimization of the action speed curve and the preset of the closing angle. The optimization of the action speed curve is achieved by adjusting the speed control of the drive motor. The action speed curve includes three stages: acceleration, constant speed and deceleration. The preset of the closing angle is achieved by the feedback signal of the position sensor. The angle position that minimizes the contact resistance when the contacts are closed is selected as the target closing angle.

[0013] Furthermore, the pre-adjustment drive parameter control also includes setting a delayed trigger time: Based on the predicted load sequence output by the load fluctuation prediction model, the target time when the load value is in a favorable range is determined. The criteria for determining the favorable range are at least one of the following: the load value is lower than the action threshold, the load change rate is in a stable range, or the load fluctuation amplitude is less than the allowable value. Subtracting the inherent time delay required for operation execution from the target time yields the delayed trigger time, where the inherent time delay includes the drive current setup time, the mechanism action response time, and the contact movement time. When the delayed trigger time arrives, the drive control system performs the opening and closing operation according to the pre-configured drive parameters.

[0014] Furthermore, updating the parameters of the load fluctuation prediction model using a rolling forecasting strategy includes: At the end of each prediction period, the newly collected standardized load data is appended to the end of the historical sequence, while the earliest historical data point at the beginning of the historical sequence is removed, keeping the input window length constant. The parameters of the load fluctuation prediction model are updated using an incremental learning approach. A small batch of training samples is formed using newly collected data and some historical data. The mean square error between the predicted output of the load fluctuation prediction model and the actual load value is calculated. The gradient of the loss function relative to each model parameter of the load fluctuation prediction model is calculated based on the mean square error, and the model parameters are adjusted according to a preset learning rate. Alternatively, a periodic retraining approach is adopted, in which the load fluctuation prediction model is fully trained using the latest complete dataset accumulated at fixed time intervals.

[0015] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention constructs a load fluctuation prediction model and adopts a rolling prediction strategy to achieve forward-looking prediction of future load trends, enabling the timing adjustment of opening and closing of circuit breakers to respond to load changes in advance, effectively avoiding mismatch between operation and load fluctuations, and reducing the risk of electrical stress concentration and equipment insulation aging caused by sudden load changes.

[0016] 2. This invention establishes a mapping relationship library between load trends and opening / closing timing parameters and performs multi-objective optimization by combining operating status information. Under the constraints of action time window, temperature threshold and safe operation interval, the invention generates optimal timing parameters with the goal of minimizing electrical stress and maximizing equipment life, which significantly reduces contact erosion and mechanical wear and extends the overall service life of the switchgear.

[0017] 3. Based on optimal timing parameters, this invention pre-adjusts and controls the drive power waveform, transmission mechanism speed, and delayed triggering time, enabling the switching equipment to accurately complete the opening and closing actions at the predicted most favorable load time, thereby improving the success rate of operation and the reliability of system operation, and reducing the risk of equipment damage and power outages caused by improper operation. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical architecture of a switchgear opening and closing sequence optimization method based on load fluctuations; Figure 2 A schematic diagram illustrating the core principle framework of a load fluctuation prediction model; Figure 3 A schematic diagram of the logical framework for mapping the load trend to the timing parameters of opening and closing circuit breakers; Figure 4 A schematic diagram of the process framework for dynamic timing parameter optimization and pre-adjustment of driving parameter control. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 4 The present invention will be further described in detail below with reference to specific embodiments.

[0020] This invention provides a method for optimizing the opening and closing timing of switchgear based on load fluctuations. The method includes the following core steps: First, deploying a high-precision load acquisition module to obtain historical load data and preprocessing it; second, constructing a load fluctuation prediction model based on standardized load data sequences to predict future load changes; then, establishing a mapping relationship library between load trends and opening / closing timing parameters to obtain an initial combination of timing parameters; next, optimizing and adjusting the initial timing parameters based on operating status information to generate optimal timing parameters; finally, pre-adjusting the drive parameters for opening and closing operations based on the optimal timing parameters to ensure that the switchgear completes its actions at the most favorable load moment. The specific implementation is as follows.

[0021] First, in step S1, a high-precision load acquisition module is deployed to acquire historical load data in real time. The acquired load data is then preprocessed to obtain a standardized load data sequence. This step, through the sequential execution of three stages, provides input data with a unified format and reliable quality for the subsequent load fluctuation prediction model. Specifically, this step includes the following operations: Step S101: Deploy a high-precision load acquisition module, employing a combination of high-precision current transformers and voltage sensors to acquire load data. The high-precision current transformer is installed in the incoming or outgoing circuit of the switchgear to sense the current signal flowing through the conductor in real time. The voltage sensor is connected across the busbar or load side to monitor voltage amplitude changes in real time.

[0022] The output signals from the current transformer and voltage sensor are processed by a signal conditioning circuit to adjust amplitude, match impedance, and provide electrical isolation before being transmitted to the data acquisition unit for digital processing. The data acquisition unit integrates an analog-to-digital converter to convert analog signals into digital signals with a conversion accuracy of at least 16 bits. The sampling frequency is set to at least 1000Hz, meaning at least 1000 data points are collected per second.

[0023] Step S102 involves real-time data cleaning of the acquired raw load data. This cleaning operation is performed in real-time on the embedded processor of the data acquisition unit or on a separately deployed data processing server. The cleaning process first checks the current and voltage values ​​at each sampling time for reasonableness, determining if any physically impossible values ​​exist, including negative current values, current values ​​exceeding the sensor's range, and voltage values ​​below a safety threshold. Abnormal data points deemed unreasonable are directly discarded and marked as invalid data.

[0024] Simultaneously, a sliding window statistical method is used to test the continuity of the data sequence. A fixed-length sliding window is set, covering several consecutive data points preceding the current sampling point. The arithmetic mean and standard deviation of all data points within the window are calculated, and the interval formed by adding and subtracting a preset multiple of the standard deviation from the window mean is taken as the normal fluctuation range. The preset multiple is set based on the statistical distribution characteristics of historical field data. When the value of the current data point at the end of the window exceeds the normal fluctuation range, the data point is determined to be an outlier and is removed.

[0025] The above anomaly detection method is derived from the concept of standard scores in statistics. Standard scores measure the degree of deviation of a data point from the mean of its dataset; they are defined as the difference between the data point and the mean divided by the standard deviation. This step applies this statistical principle to a sliding window framework, using the local mean and local standard deviation of the data within the window as reference benchmarks to measure the degree of deviation of the current data point. The deviation is calculated as follows:

[0026] in, This represents the sampled data value at the current moment; This represents the arithmetic mean of all data points within the current sliding window; This represents the standard deviation of all data points within the current sliding window. It is calculated as the square root of the average of the sum of the squares of the differences between each data point and the mean. This indicates the degree of dispersion of the current sampled value relative to the statistical center of the window; this value is dimensionless. If the value is greater than 3, the current data point is determined to be an outlier and is removed.

[0027] The data cleaning algorithm also includes anomaly detection logic based on historical statistical characteristics. This logic uses previously accumulated normal historical data to build a statistical model, which stores the statistical mean and standard deviation of historical load data at different time periods. For the currently collected data point, the corresponding time period is first determined based on its timestamp, and then the deviation between the current data value and the historical mean of that time period is calculated. If the absolute value of the deviation exceeds the product of a preset multiple and the historical standard deviation of that time period, the current data point is determined to be an anomaly and is removed.

[0028] After the above-mentioned anomaly detection mechanism is used for screening, outliers and noise interference in the original load data are removed, resulting in cleaned load data.

[0029] Step S103 involves filtering the cleaned load data to remove high-frequency interference components. The digital filtering algorithm is executed on the data processing server. The algorithm type is selected as either a finite impulse response filter or an infinite impulse response filter, and the specific parameters are configured according to the actual load characteristics.

[0030] Before filtering, a spectral analysis is performed on the cleaned data sequence to determine the effective frequency range of the load signal and the frequency distribution characteristics of interference noise. Based on the spectral analysis results, a bandpass filter is designed. The passband frequency range covers the main frequency components of load changes, while the stopband frequency range covers power frequency interference and its higher harmonic frequencies, interference frequencies caused by mechanical vibration, and electromagnetic noise frequencies.

[0031] The filtering algorithm employs zero-phase filtering to eliminate phase distortion introduced by the filter, ensuring that the filtered data sequence remains aligned with the original signal on the time axis. After this processing, high-frequency interference components are removed, resulting in smoothed load data.

[0032] Step S104 involves standardizing the smoothed load data. Standardization eliminates the effects of differences in sensor range, data acquisition channel gain, and physical dimensions, ensuring comparability of load data from different sources. The standardization algorithm employs either min-max normalization or Z-score standardization.

[0033] The min-maximum normalization method linearly maps the original data to the interval between 0 and 1, based on the global minimum and global maximum values ​​in the original data sequence. Let any data point in the smoothed load data sequence be... The global minimum value in this data sequence is The global maximum value is The normalized value after mapping By and Difference divided by and The difference is achievable.

[0034] Z-score standardization is derived from the standard normal transformation in statistics, which converts a set of data into a form with a mean of 0 and a standard deviation of 1. The transformation method is as follows:

[0035] in, Indicates the first element in the smoothed afterload data sequence The original values ​​of each data point; This represents the arithmetic mean of the data sequence; The standard deviation of the data series is calculated as the square root of the sum of the squares of the differences between each data point and the mean. This represents the standardized value of the data point after Z-score standardization; this value is dimensionless.

[0036] After standardization, the smoothed load data is transformed into a standardized load data sequence. The parameters required for standardization, including minimum, maximum, mean, and standard deviation, need to be persistently stored in the data processing system so that the same standardization transformation can be performed on the original data in subsequent real-time forecasting.

[0037] In summary, step S1 transforms the raw load data into a standardized load data sequence by deploying a high-precision acquisition module and executing a preprocessing procedure encompassing data cleaning, filtering, and standardization. This sequence accurately reflects the actual load variation patterns and supports the construction of load fluctuation prediction models and the optimization of circuit breaker timing in subsequent steps.

[0038] For step S2 in the method of this embodiment, a load fluctuation prediction model is constructed based on a standardized load data sequence, and the load change trend within a preset time period is predicted using this model to obtain the predicted load value. The following steps are followed.

[0039] Step S201: Construct the core algorithm engine for the load fluctuation prediction model. The core algorithm engine adopts a two-layer long short-term memory network structure or a converter encoder-decoder structure.

[0040] The two-layer Long Short-Term Memory (LSTM) network structure comprises two cascaded LSM unit layers. Each unit layer contains three gating mechanisms: an input gate, a forget gate, and an output gate. The first LSM unit receives a standardized workload data sequence as input, extracts the temporal local dependencies and short-term fluctuation features of the workload data, and outputs a feature vector that is passed to the second LSM unit. The second unit performs high-level feature abstraction based on the output of the first layer, capturing long-term dependencies and periodic change patterns in the workload sequence, and outputs the final feature representation for prediction computation.

[0041] The transducer encoder-decoder architecture employs a self-attention mechanism. The encoder consists of alternating stacks of multi-head self-attention layers and feedforward neural network layers. The multi-head self-attention layers multiply the input sequence by the query weight matrix, key weight matrix, and value weight matrix, respectively, to obtain the query matrix, key matrix, and value matrix.

[0042] The attention weights are calculated as follows: The query matrix and the transpose of the key matrix are multiplied to obtain the attention score matrix; each element in the attention score matrix is ​​scaled by dividing it by the square root of the key vector dimension; then, a flexible maximum function is applied to each row of the scaled attention score matrix to convert it into a probability distribution form of the attention weight matrix; finally, the attention weight matrix is ​​multiplied by the value matrix to obtain the weighted aggregated output feature matrix. The attention calculation uses the scaled dot product attention formula:

[0043] in, Represents the query matrix; Represents the key matrix; Represents a value matrix; The dimension of the key vector is a scalar integer; This represents a matrix multiplication operation between the query matrix and the transpose of the key matrix. Represents the square root of the dimension of the key vector; This represents a flexible maximum function that normalizes the input vector into a probability distribution; This represents the output feature matrix of the attention mechanism.

[0044] The multi-head attention mechanism uses multiple parallel attention heads. Each attention head performs the attention calculations described above using independently initialized query weight matrices, key weight matrices, and value weight matrices. The outputs of each attention head are concatenated along the feature dimension to form the multi-head attention output. The decoder part generates the load prediction sequence for future time periods step by step using a masked self-attention mechanism based on the high-dimensional feature representation of the encoder output and the generated prediction values.

[0045] Step S202: Set the input window length for the historical load sequence, dividing the standardized load data sequence into multiple input window samples in chronological order. The input window length is set to 5 to 10 minutes. The sliding step size of the input window is set to 1 / 2 or 1 / 3 of the window length, with temporal overlap between adjacent input windows. Each input window sample contains all load sampling point data within that time window.

[0046] Step S203 involves using the model's feature extraction layer to extract features from the input window samples. Each input window sample is then fed into the load fluctuation prediction model, where the model's feature extraction layer extracts time-series features of the load data. For a two-layer long short-term memory network structure, the first layer extracts low-level time-series features from the original load data, including local fluctuation features, gradient change features, and periodic fluctuation features. The second layer extracts high-level semantic features based on the low-level features, including trend change features, periodic cycle features, and abnormal fluctuation features.

[0047] The hidden state vector output by each layer encodes the load sequence information up to the current time step. For the transformer encoder-decoder structure, feature extraction is performed through a multi-layer self-attention mechanism and a feedforward neural network layer as described in step S201.

[0048] Step S204: The predicted load value for a preset time period is generated by the prediction output layer. The prediction output layer calculates the predicted load value for future time periods based on the high-level feature representation obtained from the feature extraction layer. For the two-layer long short-term memory network structure, the output layer uses a linear mapping to convert the hidden state vector output by the second layer into a predicted value. This linear mapping is implemented through a fully connected layer, which multiplies each dimension of the hidden state vector by its corresponding weight coefficient, sums the results, and adds a bias term to obtain the predicted load value.

[0049] The calculation method for the output of scalar prediction values ​​is as follows:

[0050] in, The first hidden state vector represents the first hidden state. Each dimension value; This represents the total dimension of the hidden state vector; Indicates the relationship with the first The weight coefficients corresponding to each hidden state dimension; This represents the bias term of the fully connected layer; This represents the scalar value of the predicted load output by the model. For the predicted sequence output in vector form, the above linear mapping is performed for each prediction time step, and the scalar predicted values ​​output at each time step are combined to form the predicted sequence vector.

[0051] The prediction time range is set to 1 to 5 minutes into the future, and the prediction time resolution is consistent with the input sampling resolution. If the prediction time range exceeds the capability of the model's single output, a rolling prediction strategy is used for multi-step prediction.

[0052] Step S205: The rolling prediction strategy is used to continuously update the model. After each prediction cycle, the newly collected standardized load data is appended to the end of the historical sequence, while the earliest historical data point at the beginning of the sequence is removed. The input window length is kept constant, and then feature extraction and prediction calculation are re-executed.

[0053] Model parameters are updated using either incremental learning or periodic retraining. Incremental learning involves using newly collected data and a subset of historical data to create mini-batch training samples based on existing model parameters. The mean squared error between the model's predicted output and the actual load is calculated, and the gradient of the loss function relative to each model parameter is calculated based on the error value. The parameters are then adjusted along the gradient direction according to a preset learning rate. Periodic retraining uses the latest complete dataset accumulated at fixed time intervals for full model training.

[0054] The rolling forecasting strategy also includes a forecasting error monitoring mechanism. This mechanism calculates the root mean square error between the predicted value output by the model and the actual load value collected subsequently in each forecasting period, outputting a scalar error index. When the error index exceeds a preset threshold for multiple consecutive forecasting periods, a model retraining process is triggered to retrain the model using the latest accumulated complete dataset; or an anomaly detection process is triggered.

[0055] In summary, step S2, by constructing a load fluctuation prediction model and employing a rolling prediction strategy, achieves continuous prediction output of load change trends for the next 1 to 5 minutes. The prediction results provide data support for subsequent steps, including querying the mapping relation library and determining the initial combination of time-series parameters.

[0056] The next step, S3, is to establish a mapping relationship library between load trends and circuit breaker timing parameters, and to query the mapping relationship library based on the predicted load value to obtain the initial timing parameter combination, including the following implementation steps.

[0057] Step S301: Collect historical switchgear operation data and analyze the correlation between load parameters and time sequence parameters. Collect a large amount of historical switchgear operation data. Data sources include factory test data provided by switchgear manufacturers, operation record data accumulated by power companies, and load response characteristic experimental data.

[0058] Each historical operation record includes load status parameters at the time of operation, timing parameters of operation execution, and evaluation indicators of operation results. Load status parameters include load current, load voltage, load active power, load reactive power before and after the operation, as well as the load change rate and load fluctuation amplitude calculated from the above values.

[0059] The operation timing parameters include the trigger time of the opening and closing operation, the rise time of the driving current, the peak current of the driving current, the pulse width of the driving current, the interval between adjacent operations, the timing sequence of multi-phase operations, and the critical node time of the action speed curve. The operation result evaluation indicators include whether the operation was successfully completed, the record of contact welding events, the duration of the arc, the degree of contact erosion, and the post-operation electrical stress assessment value.

[0060] By statistically analyzing historical data, a correlation model is established between load state parameters, operation timing parameters, and operation results.

[0061] Step S302: Based on the simulation verification results, a multi-dimensional mapping relationship library is established, and electromagnetic transient simulation software is used to model the opening and closing process of the switchgear. The simulation model includes an electrical network model, a switch arc model, a mechanism motion model, and a thermodynamic model. The electrical network model describes the electrical characteristics of the transmission line and distribution network. The switch arc model uses dynamic arc resistance to describe the nonlinear characteristics of the arc ignition and extinction processes. The mechanism motion model describes the motion process of the contacts and transmission mechanism. The thermodynamic model calculates the temperature rise and thermal stress distribution of the contacts during operation.

[0062] Combined simulation experiments were conducted by changing the load and timing parameters in the simulation model. The load parameters covered the normal load range, overload range, and range with severe load fluctuations. The timing parameters covered different combinations of action sequences, interval times, and drive current waveforms. The operational performance evaluation results corresponding to different timing parameters under different load conditions were collected.

[0063] Based on simulation experimental data, a multi-dimensional mapping relationship library is established between load change rate, load level, load fluctuation amplitude and action sequence, interval time, and drive current waveform.

[0064] Step S303: Model the mapping relationship to form a queryable mapping relationship library structure. Use piecewise linear interpolation or neural network fitting method to model the mapping relationship.

[0065] The piecewise linear interpolation method divides the load parameter space into multiple sub-regions, and uses a linear function to approximate the mapping relationship within each sub-region. The division of sub-regions is determined based on the sensitivity and nonlinearity of the load parameters, with finer-grained divisions used in regions where the mapping relationship changes drastically.

[0066] Neural network fitting methods use multilayer perceptrons or radial basis function networks to approximate the mapping relationship. The number of nodes in the input layer equals the load parameter dimension, including the load change rate and load fluctuation amplitude. The number of nodes in the output layer equals the timing parameter dimension, including the action sequence encoding, interval time value, and drive current waveform parameters.

[0067] The training data comes from the historical operation records collected in step S301 and the simulation experimental data generated in step S302. The input of the training samples is the load change rate and the load fluctuation amplitude, and the output is the corresponding combination of time series parameters. The neural network training uses the error backpropagation algorithm or the genetic algorithm for parameter optimization.

[0068] The mapping database uses either a multidimensional lookup table or a KD-tree index structure. The multidimensional lookup table discretizes the load parameter space into a grid, with each grid cell storing the corresponding combination of time-series parameters. The KD-tree index structure organizes the mapping relationships into a balanced binary tree, implementing query operations through recursive space partitioning. When new load parameters are input, the above data structure returns the corresponding combination of time-series parameters.

[0069] Step S304: Calculate the load change rate and load fluctuation amplitude based on the predicted load value. Calculate the estimated load change rate and load fluctuation amplitude at the current moment based on the predicted load sequence output in step S2.

[0070] The load change rate is calculated by dividing the difference between the predicted load value and the current actual load value by the prediction time interval.

[0071] Load fluctuation amplitude is calculated using one of three methods: standard deviation, the difference between peak and trough values, or the coefficient of variation. The standard deviation is calculated by taking the predicted load values ​​at each moment within the prediction time window, calculating their arithmetic mean, then averaging the sum of squares of the differences between each predicted load value and the average value, and finally taking the square root. The difference between peak and trough values ​​is measured by taking the maximum and minimum predicted load values ​​within the prediction time window and calculating the difference between them. The coefficient of variation is calculated by dividing the standard deviation by the arithmetic mean of the predicted load sequence. The calculated load change rate and load fluctuation amplitude are used as index keys for querying the mapping database.

[0072] Step S305: Based on the calculated load change rate and load fluctuation amplitude, query the mapping relationship library to retrieve the corresponding initial time series parameter combination.

[0073] The query operation performs nearest neighbor search or interval interpolation in the mapping relationship database based on the load change rate and load fluctuation range.

[0074] If a multidimensional lookup table structure is used, the corresponding table entry is accessed directly based on the discretized index value. If a KD-tree index structure is used, the leaf node containing the query point is searched in the KD-tree, and an exact match or linear interpolation is performed in the mapping relationship within the leaf node.

[0075] For nearest neighbor search, the Euclidean distance between the query point and the candidate point is used as the criterion for determining the degree of matching. The Euclidean distance is calculated as follows:

[0076] in, This indicates the load change rate at the query point. This represents the load change rate of candidate points in the mapping relation library. This indicates the load fluctuation range of the query point. This indicates the load fluctuation range of candidate points in the mapping relation library. This represents the Euclidean distance between the query point and the candidate points. During the nearest neighbor search, the Euclidean distance between the query point and each candidate point is calculated sequentially, and the candidate point with the smallest distance is selected as the combination that best matches the current load state.

[0077] The query results return the initial timing parameter combination that has the smallest Euclidean distance from the current load state, including the recommended action sequence, recommended interval time, and recommended drive current waveform parameters. This initial combination serves as the input for optimization and adjustment in step S4.

[0078] In summary, step S3 establishes a mapping relationship library between load trends and time-series parameters through historical data analysis and simulation verification. The load change rate and load fluctuation amplitude calculated from the predicted load values ​​are used as indexes for querying to obtain an initial combination of time-series parameters matching the current load state. This initial combination serves as the starting point for optimization and adjustment in subsequent steps.

[0079] The next step, S4, involves obtaining the current operating status information of the switchgear and optimizing the initial timing parameter combination based on this information to generate the optimal timing parameters. Specifically, this step includes the following sub-operations: Step S401: The current temperature information of the current-carrying element is acquired through a temperature sensor. The temperature sensor, which uses a thermocouple or infrared temperature measuring element, is installed on or near the surface of the current-carrying element to monitor the temperature of current-carrying components such as contacts, busbars, and disconnect switches in real time. The temperature data is transmitted to the processing unit through the data acquisition channel. The processing unit converts the raw signal into a temperature value according to the calibration parameters of the temperature sensor.

[0080] Step S402: Obtain the current position information of the switching mechanism through a position sensor. The position sensor may be a limit switch, a rotary transformer, or a photoelectric encoder, used to detect the current position state of the switching mechanism. The limit switch outputs discrete closed or open signals, while the rotary transformer and photoelectric encoder output continuous rotation angle or linear displacement signals, accurately representing the spatial coordinates of the contacts in the open, closed, or intermediate positions.

[0081] The operating status information also includes the cumulative number of circuit breaker operations, the energy storage status of the closing spring, and the current and voltage values ​​of the drive motor.

[0082] Step S403: The operating status information and initial timing parameters are combined as input parameters for the optimization algorithm. The input of the optimization algorithm includes two types of parameters. One type is non-adjustable constraint parameters, including the current current-carrying element temperature, the upper limit of the equipment temperature threshold, the lower limit of the safe operation interval, and the action time window boundary. These parameters reflect the safe operation limitations of the equipment and have fixed values ​​during the optimization process.

[0083] Another type is adjustable optimization parameters, including various timing parameters in the initial timing parameter combination, namely, action sequence, interval time, drive current waveform parameters, and action speed curve parameters. The task of the optimization algorithm is to find the optimal parameter values ​​that make the objective function optimal within the range of constraint parameters.

[0084] Step S404: Use particle swarm optimization algorithm or gradient descent optimization algorithm to find the optimal solution with the objective functions of minimizing electrical stress and maximizing equipment life.

[0085] Both electrical stress and equipment lifespan are functions of optimization parameters, but they have different dimensions. A weighted summation method is used to unify the two sub-objectives into a single scalar objective function. The objective of maximizing equipment lifespan is transformed into minimizing equipment lifespan loss. Then, the normalized electrical stress value and the normalized lifespan loss value are multiplied by their respective weighting coefficients and summed to obtain the comprehensive objective function value. The smaller this objective function value, the better the overall performance of the corresponding combination of time-series parameters in terms of both electrical stress and equipment lifespan.

[0086] To enable the optimization algorithm to quantitatively evaluate each combination of candidate timing parameters, mathematical calculation models for electrical stress and equipment life loss need to be established separately. The electrical stress is calculated using the arc energy during the contact opening and closing process as a metric. The arc energy is calculated using the following formula:

[0087] in, This represents the total energy generated by the electric arc during a single operation, measured in joules. Indicates the moment of arc initiation. Indicates the moment the electric arc extinguishes; This represents the instantaneous value of the arc voltage, the magnitude of which is related to the contact material, contact gap, and plasma state in the arc region, and its range is [value missing]. to ; This represents the instantaneous value of the load current flowing through the contacts. Arc voltage. Distance from contact The relationship between them is determined by the following empirical formula:

[0088] in, The cathode voltage drop constant is The potential gradient constant of the arc column is... and The value is determined based on the contact material; The contact opening distance is a time function, determined jointly by the transmission mechanism's operating speed curve and the drive current waveform parameters. Normalized electrical stress value. By calculating the electric arc energy Divided by reference arc energy get, Take the arc energy value corresponding to the standard operating sequence under rated operating conditions. It is a dimensionless value.

[0089] Equipment life loss is measured by the electrical wear of the contacts. The cumulative wear model based on arc energy and breaking current is used for calculation, and its mathematical expression is as follows:

[0090] in, This represents the cumulative electrical wear of the contacts, in units of any scale. This indicates the total number of opening and closing operations within the current statistical period; Indicates the first Arc energy of the next operation; Indicates the first The effective value of the load current when the operation is disconnected, in amperes; The arc energy wear coefficient, The wear coefficient is the square of the current. and The value is obtained by fitting standard electrical wear test data of the contact material. The current cumulative wear amount is... Divide by the manufacturer's specified maximum permissible wear of the contacts. The normalized equipment life loss value is obtained. , It is a dimensionless value.

[0091] Comprehensive objective function The expression is:

[0092] in, and These are the electrical stress weighting coefficient and the life loss weighting coefficient, respectively, satisfying... The specific value is determined based on the equipment's operating strategy; for electrical stress-sensitive scenarios, the value is... Lifetime-sensitive scenarios .

[0093] In particle swarm optimization, the position vector of each particle represents a candidate solution for a set of optimization parameters, and the dimension of the position vector is equal to the number of optimization parameters. The velocity vector has the same dimension as the position vector and determines the direction and step size of the particle's movement in each dimension.

[0094] During the iteration process, each dimension is updated independently, with no cross-coupling between dimensions. The fitness of each particle is calculated based on the objective function. Substituting the particle's position vector into the objective function expression yields the corresponding comprehensive objective function value; particles with lower fitness are closer to the optimal solution. The particle swarm optimization algorithm guides each particle to gradually approach the optimal solution during the iteration process by tracking the individual historical best position of each particle and the global best position of the entire swarm. When the number of iterations reaches a preset limit or the fitness of the global best position no longer improves in consecutive iterations, the iteration stops, and the current global best position is the optimal combination of time-series parameters.

[0095] In the gradient descent optimization algorithm, the objective function is treated as a multivariate function of the optimization parameters. The partial derivatives of the objective function with respect to each optimization parameter are calculated, and these partial derivatives together form the gradient vector. Each component of the gradient vector corresponds to the partial derivative value of each optimization parameter, and each component is calculated independently. The optimization parameters are iteratively updated along the negative direction of the gradient vector, with the step size of each update controlled by a preset learning rate. Iteration stops when the number of iterations reaches a preset upper limit or when the decrease in the objective function value over multiple iterations is less than a preset threshold. The current value of the optimization parameters is then considered the optimal combination of time-series parameters.

[0096] Step S405: Solve for the optimal combination of time series parameters under the constraints.

[0097] Constraints include operating time windows, equipment temperature thresholds, and safe operating intervals, and can take the form of equality constraints and inequality constraints. Equality constraints require certain parameters to satisfy specific relationships; for example, the operating sequence of multiphase switches must meet specific phase-to-phase interval requirements. Inequality constraints require certain parameters to be within specified ranges; for example, the temperature of current-carrying elements must not exceed a safe threshold, and the interval time must not be less than the minimum safe interval.

[0098] The penalty function method is used to handle constraints. A penalty term is added to the objective function, and the value of the penalty term is positively correlated with the degree of constraint violation. For each inequality constraint, when the optimization parameter value satisfies the constraint, the penalty term is 0; when the optimization parameter value violates the constraint, the penalty term is the distance of the parameter from the constraint boundary multiplied by a preset penalty coefficient.

[0099] For each equality constraint, the penalty term is the absolute value of the difference between the two sides of the equation multiplied by a preset penalty coefficient. All penalty terms are summed and added to the objective function to form the augmented objective function. The optimization algorithm searches for the best solution under the guidance of the augmented objective function. Violating constraints increases the value of the augmented objective function, allowing the algorithm to avoid infeasible regions. The final output solution is a feasible solution that satisfies the constraints.

[0100] Step S406: Output the optimal timing parameters that satisfy the constraints. The optimal timing parameters include the optimized action sequence, minimum interval time, optimal values ​​of the drive current waveform parameters, key node times of the action speed curve, and offset of the delayed trigger time. The action sequence is a discrete parameter, taking the encoded value that minimizes the objective function; the other parameters are continuous parameters, taking the values ​​corresponding to the iteration termination.

[0101] The processing unit encapsulates the optimal timing parameters into a control command data packet and sends it to the drive control system through the communication interface. The drive control system then performs precise control of the actuator based on the received parameters.

[0102] In this step, the criterion for determining the optimal solution is: the solution that minimizes the value of the augmented objective function while satisfying all constraints is the optimal combination of time series parameters.

[0103] Steps S403 to S406 sequentially complete the four stages of input organization, target construction and optimization, constraint processing, and result output, which together realize the transformation from the initial scheme to the optimal scheme that satisfies the constraints.

[0104] Finally, in step S5, the drive parameters are pre-adjusted based on the optimal timing parameters for the closing or opening operation to be performed, so that the switchgear completes the operation at the most favorable time predicted by the load. Specifically, this step includes the following sub-operations: Step S501: Adjust the output waveform parameters of the drive power supply in advance according to the predicted load trend. The drive power supply is a controllable DC power supply or pulse power supply composed of power semiconductor devices. The adjustment of the output waveform parameters is achieved by changing the drive signal of the power semiconductor devices, including amplitude adjustment and frequency adjustment. Amplitude adjustment changes the peak current magnitude of the drive current. Frequency adjustment changes the time distribution characteristics of the drive current pulses, including pulse width, pulse interval, and number of pulses.

[0105] The pre-adjustment strategy selects matching waveform parameter configurations based on the predicted load trend. When the predicted load is on an upward trend, the drive current amplitude is increased in advance to prepare for a larger electromagnetic thrust. When the predicted load is on a downward trend, the drive current amplitude is reduced to match the reduced contact pressure requirements.

[0106] To achieve effective driving and control of the switchgear's opening and closing operations using the aforementioned drive power supply output waveform, this embodiment employs a servo motor drive scheme to modify the original spring operating mechanism of the switchgear. Specifically, a permanent magnet synchronous servo motor is installed at the end of the energy storage shaft of the switchgear operating mechanism. The output shaft of this servo motor is rigidly connected coaxially to the energy storage shaft of the operating mechanism via a reducer. The reduction ratio of the reducer is determined based on the required closing operation power of the operating mechanism and the rated torque of the servo motor. An absolute encoder is installed on the servo motor to detect the angular displacement and angular velocity of the motor rotor in real time. The encoder resolution is no less than 65,536 pulses per revolution. The servo motor is fixed to the switchgear operating mechanism frame via a motor bracket. The motor bracket uses an adjustable mounting base, allowing the coaxiality between the motor axis and the energy storage shaft axis to be adjusted within a range of ±0.1mm.

[0107] The servo motor's drive circuit receives control signals from the processing unit, including the target speed, target angular displacement, and acceleration time constant. The drive circuit generates a pulse-width modulation (PWM) signal based on the control signal, which, after power amplification, drives the three-phase windings of the servo motor, achieving precise control of the motor's speed, torque, and angular displacement. The servo motor's output torque is amplified by a reducer and transmitted to the energy storage shaft. The energy storage shaft drives the cam assembly in the operating mechanism to rotate. The cam assembly converts the rotational motion into linear motion of the contact assembly via a linkage mechanism, thereby driving the moving contact to perform opening and closing actions. The power transmission path is: servo motor → reducer → energy storage shaft → cam assembly → linkage mechanism → moving contact.

[0108] Step S502: Adjust the control parameters of the transmission mechanism, which includes a drive motor, a reducer, a linkage mechanism, and a contact assembly. Adjusting the control parameters includes optimizing the operating speed curve and presetting the closing angle.

[0109] The optimization of the motion speed curve is achieved through servo motor speed-time curve control. The processing unit generates a target speed curve based on optimal timing parameters, which includes three stages: acceleration, constant speed, and deceleration. In the acceleration stage, the processing unit outputs an increasing speed setpoint to the servo motor drive circuit, accelerating the motor from a standstill to the preset target speed. The acceleration in this stage is determined by the acceleration time constant. In the constant speed stage, the processing unit maintains a constant speed setpoint, allowing the motor to run at the target speed at a constant speed. In the deceleration stage, the processing unit outputs a decreasing speed setpoint to the servo motor drive circuit, decelerating the motor from the target speed to a preset termination speed. The deceleration in this stage is determined by the deceleration time constant. The actual speed of the servo motor is fed back to the processing unit in real time via an absolute encoder. The processing unit uses a proportional-integral-derivative closed-loop control algorithm to correct speed deviations in real time, ensuring that the deviation between the actual speed curve and the target speed curve does not exceed a preset allowable range. The specific values ​​of the target speed, acceleration time constant, and deceleration time constant are determined by the critical node times of the motion speed curve in the optimal timing parameters.

[0110] The preset closing angle is achieved through precise control of the angular displacement of the servo motor. The processing unit determines the target angular position for contact closure based on optimal timing parameters. The target angular displacement of the servo motor rotor corresponding to this position is calculated using the kinematic model of the transmission mechanism. The kinematic model describes the correspondence between the angular displacement of the servo motor rotor and the linear displacement of the moving contact, which is determined by the reduction ratio of the reducer, the cam profile curve, and the geometric parameters of the linkage mechanism. The processing unit sends the calculated target angular displacement of the servo motor rotor as a position setpoint to the servo motor drive circuit. The drive circuit uses a position closed-loop control algorithm to drive the motor to rotate to the target angular displacement. The absolute encoder provides real-time feedback on the actual angular displacement of the motor rotor. The drive circuit compares the actual angular displacement with the position setpoint. When the deviation exceeds a preset threshold, the drive circuit automatically adjusts the motor winding current to eliminate the deviation. Through the above position closed-loop control, the moving contact is precisely stopped at the angular position where the contact resistance is minimized when the contact is closed. The contact pressure corresponding to this angular position is not lower than the minimum pressure value required to ensure current carrying capacity. The adjustment range of the preset closing angle is determined according to the allowable fine adjustment range of the mechanical structure of the switchgear operating mechanism. Fine adjustments of no more than ±2° electrical angle are made based on the closed position.

[0111] In the aforementioned servo motor drive scheme, the processing unit completes the pre-adjustment settings of all drive parameters before operation execution, including the parameters of each stage of the target speed curve, the target angular displacement value, and the issuance of closed-loop control parameters. Parameters are transmitted to the servo motor drive circuit via a high-speed communication interface, and the drive circuit receives and stores and verifies the parameters locally. After the pre-adjustment settings are completed, the processing unit issues a start command when the delayed trigger time arrives, and the servo motor performs the opening and closing drive operation according to the pre-configured speed curve and angular displacement target.

[0112] Step S503: Set the delayed trigger time according to the adjusted drive parameters. The processing unit determines the target time when the load value is in a favorable range based on the predicted load sequence output by the load fluctuation prediction model. The criteria for determining a favorable range include three conditions: the load value is lower than the action threshold, the load change rate is in a stable range, and the load fluctuation amplitude is less than the allowable value. Meeting any one of these conditions determines that the load is in a favorable range.

[0113] To achieve the engineering feasibility and adaptive adjustment capability of the aforementioned action threshold, stability range, and allowable value, this embodiment adopts a dynamic tuning method based on the statistical characteristics of historical load data. The action threshold is tuned as follows: the processing unit maintains a fixed-length window of recent historical load data, with a window duration of 24 hours, covering at least one complete production cycle. At the beginning of each forecast cycle, the processing unit reads all load data within the window, calculates the arithmetic mean and standard deviation of these data, and then calculates the action threshold according to the following formula:

[0114] in, This indicates the action threshold for the current prediction period. This represents the arithmetic mean of the load data within the window. This represents the standard deviation of the load data within the window. The threshold coefficient ranges from 0.5 to 1.5, determined based on the severity of load fluctuations. A larger value is used when load fluctuations are severe, and a smaller value is used when load fluctuations are mild. Once new actual load data is collected, this data is appended to the end of the historical window, while the oldest historical data point at the front of the window is removed, maintaining a constant window length. This enables adaptive updates of the action threshold in each prediction cycle.

[0115] The method for setting the stable interval is as follows: The processing unit calculates the standard deviation of the load change rate in the recent historical load data window, and constructs a stable interval with the mean load change rate as the center and a preset multiple of the standard deviation as the radius. The lower bound of the stable interval is the mean load change rate minus twice the standard deviation, and the upper bound is the mean load change rate plus twice the standard deviation. The mean load change rate and standard deviation are recalculated based on the latest historical data window in each forecast period to achieve adaptive updating of the stable interval.

[0116] The tolerance value is set as follows: the processing unit calculates the 90th percentile of the load fluctuation amplitude in the recent historical load data window, and uses this percentile as the tolerance value for the current forecast period. The 90th percentile of the load fluctuation amplitude is recalculated in each forecast period based on the latest historical data window, achieving adaptive updating of the tolerance value.

[0117] Subtracting the inherent time delay required for operation execution from the target time yields the set value for the delayed trigger time. The inherent time delay includes three components: drive current setup time, mechanism action response time, and contact movement time. The duration of each component is pre-calibrated through actual measurement or simulation methods.

[0118] To overcome the mechanical response drift caused by factors such as contact wear, lubrication deterioration, and ambient temperature changes during long-term equipment operation, this embodiment uses a position sensor installed on the switchgear transmission mechanism to collect the timestamp of the contact reaching the closed or open position in real time after each opening and closing operation. This timestamp is then compared with the time the processing unit issues the operation execution command to calculate the actual total delay time of the operation. The actual total delay time is calculated as follows:

[0119] in, Indicates the first The actual total delay time for this operation. Indicates the first The timestamp of the contact reaching the correct position during this operation. Indicates the first The moment when the processing unit issues the operation execution instruction in this operation.

[0120] The processing unit stores the actual total delay time of each operation in non-volatile memory, forming a historical delay time series. The processing unit maintains a fixed-length historical delay time window, with the window size set to the most recent 50 operations. Before each delay trigger calculation, the processing unit reads the actual total delay time of all records in the historical delay time window and calculates its arithmetic mean. and standard deviation Then, the inherent time delay is dynamically compensated according to the following formula:

[0121] in, This represents the initial inherent time delay pre-calibrated through actual measurement or simulation. Indicates the amount of time drift compensation. Determined as follows: the arithmetic mean of the actual total delay times of the most recent 50 operations. With the initial inherent time delay The absolute value of the deviation between them is less than three standard deviations. hour, Values When the absolute value of the deviation is greater than or equal to three standard deviations When this occurs, an abnormal alarm is triggered, indicating that the operating mechanism may have a malfunction or severe wear and tear, requiring inspection and maintenance.

[0122] In addition to the statistical compensation based on historical operating data mentioned above, this embodiment also includes an ambient temperature compensation mechanism. The processing unit monitors the ambient temperature in real time using a temperature sensor installed inside the switch cabinet and corrects the inherent time delay according to the following linear compensation model:

[0123] in, This indicates the inherent time delay after temperature compensation. This indicates the current ambient temperature, in degrees Celsius. This indicates the reference ambient temperature during initial calibration, in degrees Celsius. This is the temperature compensation coefficient, representing the relative rate of change of the inherent time delay for every 1 degree Celsius change in temperature. The value of is determined experimentally based on the linear expansion coefficient of the transmission mechanism material and the viscosity-temperature characteristics of the lubricating medium, ranging from 0.001 / °C to 0.005 / °C. The inherent time delay ultimately used for calculating the delayed trigger moment is the temperature-compensated value. .

[0124] Through the dual mechanisms of historical delay statistical compensation and ambient temperature compensation, the processing unit can dynamically correct the inherent time delay by using the latest actual action feedback data each time it calculates the delay trigger time, forming a closed-loop control of "prediction-trigger-execution-monitoring-correction", ensuring that the accuracy of the delay trigger time does not decrease due to mechanical time drift during long-term operation of the equipment.

[0125] The final calculation method for the delay trigger time is as follows:

[0126] in, Indicates the final delay trigger time. This indicates the target time when the load value is within a favorable range. This indicates the inherent time delay after compensation for both temperature and wear. To provide a safety margin, the value ranges from 50 to 200 milliseconds, used to compensate for random fluctuations not covered by statistical compensation mechanisms.

[0127] After the pre-adjustment settings are completed, the processing unit enters a waiting state to monitor load changes in real time. When the delayed trigger time arrives, the processing unit issues an operation execution command, driving the control system to perform opening and closing operations according to the pre-configured parameters.

[0128] Step S504: Before the operation is executed, all drive parameters are pre-adjusted and set, and the operation is executed when the delayed trigger time arrives.

[0129] The pre-adjustment settings include the distribution of drive power parameters, transmission mechanism control parameters, and the setting of delay trigger times. Parameter distribution is transmitted to the corresponding control unit via a high-speed communication interface. After receiving the parameters, the control unit stores and verifies them locally. Parameter verification ensures that the received parameter values ​​are within the valid range; if an anomaly is detected, a retransmission is requested or an alarm is triggered.

[0130] After the pre-adjustment settings are completed, the processing unit enters a waiting state to monitor load changes in real time. When the delayed trigger time arrives, the processing unit issues an operation execution command, driving the control system to perform opening and closing operations according to the pre-configured parameters.

[0131] In summary, step S5, through the sequential implementation of four stages—drive power supply waveform pre-adjustment, transmission mechanism parameter pre-setting, delay trigger time calculation, and pre-adjustment parameter issuance and execution—transforms the optimal timing parameters into specific control parameters such as the drive power supply output waveform, transmission mechanism action characteristics, and delay trigger time. This enables the switchgear to accurately execute closing or opening operations at the predicted favorable load time, achieving precise matching between the operation timing and load fluctuations. The switchgear's closing and opening operations are executed according to the optimized timing parameters at the most favorable load time, reducing electrical stress, minimizing contact erosion, and extending equipment lifespan.

[0132] To facilitate understanding, the technical solution of the present invention will be further explained below in conjunction with specific application scenarios.

[0133] In the power distribution system of an industrial user, a batch of medium-voltage switchgear is installed to control the power supply to various load equipment on the production line. The load of this power distribution system has obvious fluctuation characteristics, mainly manifested as sudden load changes caused by the starting and stopping of high-power motors, load pulses from the periodic operation of welding equipment, and slow load changes in the workshop lighting and air conditioning systems. Under the traditional opening and closing operation control mode, the switchgear performs opening and closing actions according to a fixed operating sequence, failing to fully consider the impact of load fluctuations. This can easily lead to operational failures or equipment damage when the load changes drastically.

[0134] After adopting the switchgear opening and closing timing optimization method based on load fluctuations provided by this invention, the data acquisition unit deployed on the switchgear side continuously monitors the current and voltage signals of the incoming load, with the sampling frequency set to no less than 1000Hz to ensure accurate capture of load fluctuations. The acquired raw load data undergoes real-time cleaning and filtering to remove outliers and noise interference, followed by standardization to generate a standardized load data sequence.

[0135] The load fluctuation prediction model in this embodiment adopts a two-layer long short-term memory network structure, with an input window length set to 10 minutes. This means that the model is input with the most recent 10 minutes of standardized load data, and the output is the load forecast value for the next 3 minutes. The load fluctuation prediction model uses a rolling prediction strategy for continuous updates, incrementally updating the model every minute using the latest collected data to maintain prediction accuracy.

[0136] Before a planned load switching operation, the processing unit establishes a mapping relationship library between load trends and closing / opening timing parameters based on historical operation data and simulation verification results. It then queries the library to obtain the initial timing parameter combination based on the currently predicted load change trend. Simultaneously, the processing unit acquires the switchgear's operating status information, including contact temperature and mechanism position status. This operating status information and the initial timing parameter combination are used as input to the optimization algorithm. The optimization algorithm employs particle swarm optimization, aiming to minimize electrical stress and maximize equipment lifespan, and solves for the optimal timing parameter combination under constraints such as the action time window and equipment temperature threshold. The obtained optimal timing parameters indicate that the closing operation will be delayed for a certain period, ensuring that the operation is initiated when the load drops to a low point.

[0137] Based on optimal timing parameters, the processing unit pre-adjusts the output waveform of the drive power supply before operation, appropriately reducing the peak drive current to decrease electrical stress; simultaneously, it adjusts the speed curve of the transmission mechanism to increase the buffering effect of the deceleration section. The processing unit sets a delayed trigger time so that the switch completes the closing action when the predicted load is at its lowest point. In actual operation, the switch successfully completes closing in the low load range, reducing arc energy and contact erosion during the closing process, improving the success rate of operation, and extending the service life of the equipment.

[0138] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0139] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for optimizing the opening and closing timing of switchgear based on load fluctuations, characterized in that, include: A load fluctuation prediction model is constructed based on standardized load data sequences. The predicted load values ​​within a preset time period are predicted using the load fluctuation prediction model, and the parameters of the load fluctuation prediction model are updated using a rolling prediction strategy. Establish a mapping relationship library between load trend and circuit breaker timing parameters. Calculate the load change rate and load fluctuation amplitude based on the predicted load value. Use the load change rate and load fluctuation amplitude as indexes to query the mapping relationship library to obtain the initial timing parameter combination. The data structure of the mapping relationship library adopts a multidimensional lookup table or a KD-tree index structure. The operating status information of the switchgear is obtained, and the operating status information and the initial timing parameters are combined as input parameters of the optimization algorithm. Under the premise of satisfying the constraints, the optimization is performed with the objectives of minimizing electrical stress and maximizing equipment life to generate the optimal timing parameter combination. Based on the optimal timing parameter combination, the drive parameters of the opening and closing operation to be performed are pre-adjusted to control the switching equipment so that the switching equipment completes the opening and closing operation at the most favorable time of the predicted load.

2. The method for optimizing the switching timing of switchgear based on load fluctuations according to claim 1, characterized in that, The rolling forecasting strategy also includes a forecasting error monitoring mechanism. The forecasting error monitoring mechanism calculates the root mean square error between the forecasted load value and the actual load value collected in each forecasting period. When the root mean square error exceeds a preset threshold for multiple consecutive forecasting periods, the model retraining process is triggered, and the load fluctuation forecasting model is retrained using the latest accumulated complete dataset.

3. The method for optimizing the switching timing of switchgear based on load fluctuations according to claim 1, characterized in that, The load fluctuation prediction model adopts a two-layer long short-term memory network structure, which includes two long short-term memory unit layers arranged in series. The first layer of long short-term memory unit receives the standardized load data sequence as input and extracts the local dependencies and short-term fluctuation characteristics of the load data in time series. The second layer of long short-term memory unit performs high-level feature abstraction based on the output of the first layer of long short-term memory unit to capture the long-term dependencies and periodic change patterns in the load sequence. Alternatively, the load fluctuation prediction model adopts a converter encoder-decoder structure, which employs a multi-head self-attention mechanism. The encoder part is composed of alternating stacks of multi-head self-attention layers and feedforward neural network layers. The decoder part generates a load prediction sequence for future time moments step by step through a masked self-attention mechanism based on the high-dimensional feature representation output by the encoder part and the generated prediction values.

4. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 1, characterized in that, The establishment of the mapping relationship library between load trends and circuit breaker timing parameters includes: Historical switchgear operation data were collected, and simulation data was generated by modeling the opening and closing process of the switchgear using electromagnetic transient simulation software and combined simulation experiments. Each record includes load status parameters, operation timing parameters and operation result evaluation indicators. Based on the historical switchgear operation data and the simulation data, the mapping relationship is modeled using a piecewise linear interpolation method or a neural network fitting method to obtain the mapping relationship library; When querying the mapping relationship library, the load change rate and the load fluctuation amplitude are used as query points. Nearest neighbor search is performed in the mapping relationship library. The nearest neighbor search uses the Euclidean distance between the query point and the candidate point as the criterion for determining the degree of matching. The time series parameter combination corresponding to the candidate point with the smallest Euclidean distance is returned as the initial time series parameter combination.

5. The method for optimizing the switching timing of switchgear based on load fluctuations according to claim 1, characterized in that, The step of acquiring the operating status information of the switchgear and combining the operating status information with the initial timing parameters as input parameters for the optimization algorithm includes: The current temperature of the current-carrying element is used as the upper limit of the equipment temperature threshold in the constraint conditions. The current position of the switching mechanism and the cumulative number of circuit breaker operations are used as the lower limit of the safe operating interval in the constraint conditions. The energy storage state of the closing spring and the current and voltage values ​​of the drive motor are used as the reference state inputs for optimizing the action speed curve.

6. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 1, characterized in that, The optimization algorithm employs either particle swarm optimization or gradient descent optimization, and the objective of minimizing electrical stress and maximizing equipment lifespan includes: Maximizing equipment lifespan is converted into minimizing equipment lifespan loss. The normalized electrical stress value and the normalized lifespan loss value are multiplied by their respective weighting coefficients and then added together to obtain the comprehensive objective function value. The minimum of the comprehensive objective function value is taken as the optimization objective.

7. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 6, characterized in that, The constraints include the action time window, equipment temperature threshold, and safe operation interval; The constraint conditions are handled using the penalty function method, which adds a penalty term to the objective function: For inequality constraints, the penalty term is zero when the optimization parameter value satisfies the inequality constraint, and the penalty term is the distance of the parameter from the constraint boundary multiplied by a preset penalty coefficient when the inequality constraint is violated. For equality constraints, the penalty term is the absolute value of the difference between the two sides of the equation multiplied by a preset penalty coefficient; The summation of all penalty terms is added to the objective function to form an augmented objective function that satisfies the constraints.

8. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 1, characterized in that, The pre-adjustment of drive parameters for the opening and closing operations to be performed based on the optimal timing parameter combination includes: Adjust the output waveform parameters of the drive power supply, including amplitude and frequency. Adjusting the amplitude changes the peak current of the drive current, and adjusting the frequency changes the pulse width and pulse interval of the drive current pulse. In addition, the control parameters of the transmission mechanism are adjusted. The control parameters of the transmission mechanism include the optimization of the action speed curve and the preset of the closing angle. The optimization of the action speed curve is achieved by adjusting the speed control of the drive motor. The action speed curve includes three stages: acceleration, constant speed and deceleration. The preset of the closing angle is achieved by the feedback signal of the position sensor. The angle position that minimizes the contact resistance when the contacts are closed is selected as the target closing angle.

9. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 8, characterized in that, The pre-adjusted drive parameter control also includes setting a delayed trigger time: Based on the predicted load sequence output by the load fluctuation prediction model, the target time when the load value is in a favorable range is determined. The criteria for determining the favorable range are at least one of the following: the load value is lower than the action threshold, the load change rate is in a stable range, or the load fluctuation amplitude is less than the allowable value. Subtracting the inherent time delay required for operation execution from the target time yields the delayed trigger time, where the inherent time delay includes the drive current setup time, the mechanism action response time, and the contact movement time. When the delayed trigger time arrives, the drive control system performs the opening and closing operation according to the pre-configured drive parameters.

10. The method for optimizing the opening and closing sequence of switchgear based on load fluctuations according to claim 1, characterized in that, The step of updating the parameters of the load fluctuation prediction model using a rolling forecasting strategy includes: At the end of each prediction period, the newly collected standardized load data is appended to the end of the historical sequence, while the earliest historical data point at the beginning of the historical sequence is removed, keeping the input window length constant. The parameters of the load fluctuation prediction model are updated using an incremental learning approach. A small batch of training samples is formed using newly collected data and some historical data. The mean square error between the predicted output of the load fluctuation prediction model and the actual load value is calculated. The gradient of the loss function relative to each model parameter of the load fluctuation prediction model is calculated based on the mean square error, and the model parameters are adjusted according to a preset learning rate. Alternatively, a periodic retraining approach is adopted, in which the load fluctuation prediction model is fully trained using the latest complete dataset accumulated at fixed time intervals.

Images