Intelligent control method and system for outdoor totally-enclosed small high-voltage isolation load switch

Abstract

The invention discloses an intelligent control method and system for an outdoor totally-enclosed small high-voltage isolation load switch, and relates to the technical field of intelligent control systems. The method comprises the steps that a historical switch operation data sequence set of a target switch in a historical time zone is collected through a multi-source sensing monitoring array of an edge sensing layer, and the target switch is an outdoor totally-closed small high-voltage isolation load switch; predicting and acquiring a predicted switch operation data sequence set in a preset time zone according to the historical switch operation data sequence set by using a long-short-term memory network; based on the predicted switch operation data sequence set, taking minimization of line loss, fault outage minutes, voltage out-of-limit time and voltage sag times as multiple optimization targets, performing iterative optimization search on control parameters of a target switch, and determining an adaptive control strategy; and performing intelligent control on the target switch in the preset time zone according to the adaptive control strategy. The operation efficiency and the power supply reliability of the power distribution network in the remote area are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control systems, in particular to an intelligent control method and system for an outdoor fully-enclosed small high-voltage isolation load switch. BACKGROUND

[0002] With the intelligent construction of distribution networks extending to remote areas such as suburbs and rural areas, the operation stability, intelligent level and outdoor environmental adaptability of high-voltage isolation load switches for 10kV distribution networks are required to be higher. The outdoor fully-enclosed small high-voltage isolation load switch is still widely used due to its compact structure and moderate cost.

[0003] However, the traditional outdoor high-voltage isolation load switch still has technical problems: on the one hand, the device generally has the problems of transmission jamming and poor outdoor weather resistance, which is difficult to adapt to the complex natural environment in remote areas; on the other hand, it mostly relies on regular manual inspection and manual operation, which not only has high operation and maintenance cost, but also cannot avoid the risk of device failure in time, thereby resulting in insufficient operation efficiency and power supply reliability of the distribution network in remote areas. SUMMARY

[0004] The present application provides an intelligent control method and system for an outdoor fully-enclosed small high-voltage isolation load switch, aiming to solve the technical problem of insufficient operation efficiency and power supply reliability of the distribution network in remote areas in the prior art.

[0005] In view of the above problems, the present application provides an intelligent control method and system for an outdoor fully-enclosed small high-voltage isolation load switch.

[0006] In a first aspect, the present application provides an intelligent control method for an outdoor fully-enclosed small high-voltage isolation load switch, comprising: acquiring a historical switch operation data sequence set of a target switch in a historical time zone through a multi-source sensing monitoring array of an edge perception layer, wherein the target switch is an outdoor fully-enclosed small high-voltage isolation load switch; predicting and acquiring a predicted switch operation data sequence set in a preset time zone according to the historical switch operation data sequence set by using a long short-term memory network; based on the predicted switch operation data sequence set, iteratively optimizing and searching the control parameters of the target switch with the minimum line loss, fault outage minutes, voltage out-of-limit time and voltage sag frequency as multi-optimization targets, determining an adaptive control strategy, and intelligently controlling the target switch in the preset time zone according to the adaptive control strategy.

[0007] In a second aspect, the present application provides an intelligent control system for an outdoor fully-enclosed small high-voltage isolation load switch, comprising: The edge sensing acquisition module is used to acquire a set of historical switch operation data sequences of the target switch in the historical time zone through the multi-source sensor monitoring array of the edge sensing layer, wherein the target switch is an outdoor fully enclosed small high-voltage isolation load switch; The data prediction module is used to use a long short-term memory network to predict and obtain a set of predicted switch operation data sequences within a preset time zone based on the historical switch operation data sequence set. The multi-objective optimization control module is used to perform iterative optimization search on the control parameters of the target switch based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, to determine an adaptive control strategy, and to perform intelligent control on the target switch within the preset time zone according to the adaptive control strategy.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides an intelligent control method and system for outdoor fully enclosed small high-voltage disconnecting load switches. It accurately collects historical operating data sequences of the outdoor fully enclosed small high-voltage disconnecting load switches through a multi-source sensor monitoring array at the edge sensing layer, providing comprehensive and reliable data support for switch status analysis. Then, it utilizes a long short-time memory network to accurately predict the switch operating data in a preset time zone, enabling advance prediction of the switch's future operating status. Based on the predicted data, and with the goal of minimizing line losses, fault outage minutes, voltage over-limit time, and voltage sag frequency, iteratively optimizes control parameters and determines an adaptive control strategy, achieving dynamic intelligent control of the switch. This forms a complete closed loop of data acquisition, status prediction, and precise control, effectively improving the operating efficiency and power supply reliability of power distribution networks in remote areas. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0010] Figure 1 A flowchart illustrating the intelligent control method for an outdoor fully enclosed miniature high-voltage disconnector load switch provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the intelligent control system for outdoor fully enclosed small high-voltage disconnect load switch provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Edge sensing acquisition module 11, data prediction module 12, and multi-objective optimization control module 13. Detailed Implementation

[0011] This invention provides an intelligent control method and system for outdoor fully enclosed small high-voltage disconnect load switches, which is used to address the technical problems of insufficient operating efficiency and power supply reliability in existing power distribution networks in remote areas.

[0012] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this invention provides an intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switches, the method comprising: S100: Collect a set of historical switch operation data sequences of the target switch in the historical time zone through a multi-source sensor monitoring array of the edge sensing layer, wherein the target switch is an outdoor fully enclosed small high-voltage isolation load switch.

[0015] In this embodiment of the invention, a multi-source sensing monitoring array of the edge perception layer collects a set of historical switch operation data sequences within a historical time zone for the target switch, wherein the target switch is an outdoor fully enclosed small high-voltage disconnector switch. As a key device in suburban and rural 10kV distribution networks, the operating status of the outdoor fully enclosed small high-voltage disconnector switch directly determines the stability of the power supply. However, the outdoor environment in remote areas is complex, and traditional switch data acquisition suffers from three major drawbacks: first, the acquisition hardware is scattered, not integrated with the switch body, has poor weather resistance, and is easily interfered with; second, the monitoring indicators are singular, mostly focusing on basic electrical parameters such as line current and voltage, ignoring the switch's own health status and environmental influencing factors; and third, the data lacks temporal continuity, failing to form a complete historical operation data sequence, resulting in a lack of comprehensive and reliable data support for subsequent control strategy formulation. Therefore, this step integrates switch hardware structure design with multi-dimensional indicator monitoring to construct a data acquisition system that is hardware-integrated, indicator-comprehensive, and time-sequential, providing a high-quality data source for subsequent operation status prediction and multi-objective control parameter optimization based on long short-term memory networks.

[0016] Step S100 in the method provided in this embodiment of the invention includes: The outdoor fully enclosed miniature high-voltage isolating load switch includes a fully enclosed housing, a three-phase double-break structure, a three-phase linkage spring operating mechanism, an edge sensing layer, and an intelligent control unit. The three-phase double-break structure integrates the vacuum interrupter and the visible isolation break in series in each phase circuit. The three-phase linkage spring operating mechanism adopts a two-position spring mechanism, which drives the three-phase contacts to open and close synchronously through mechanical linkages. The edge sensing layer has a built-in multi-source sensor monitoring array.

[0017] Obtain the set of operation monitoring indicators for the target switch, wherein the set of operation monitoring indicators includes switch operation status, electrical parameters, health indicators and environmental parameters, the switch operation status includes opening and closing positions and opening and closing operation time, the electrical parameters include line current, power factor and three-phase voltage, the health indicators include contact temperature and contact resistance, and the environmental parameters include ambient temperature and ambient humidity; According to the preset data monitoring time interval and the set of operation monitoring indicators, the target switch is monitored in real time through the multi-source sensor monitoring array of the edge perception layer within the historical time zone to obtain the historical switch operation data sequence set.

[0018] First, the hardware structure of the target switch is defined. The target switch is an outdoor fully enclosed miniature high-voltage disconnector load switch. This outdoor fully enclosed miniature high-voltage disconnector load switch includes a fully enclosed housing, a three-phase double-break structure, a three-phase linkage spring operating mechanism, an edge sensing layer, and an intelligent control unit. The three-phase double-break structure integrates a vacuum interrupter and a visible isolation break in series within each phase circuit. The three-phase linkage spring operating mechanism uses a two-position spring mechanism, driving the three-phase contacts to open and close synchronously via mechanical linkages. The edge sensing layer incorporates a multi-source sensor monitoring array. The outdoor fully enclosed miniature high-voltage disconnector load switch refers to a miniaturized power device suitable for 10kV distribution networks, possessing the dual functions of circuit isolation and load current interruption, and its fully enclosed structure is adapted to complex outdoor environments.

[0019] The hardware of the target switch is the basic carrier for data acquisition. Its integrated structure provides hardware support for multi-source data acquisition. The specific structure and functions are as follows: The fully enclosed housing is made of corrosion-resistant aluminum alloy, achieving dustproof, waterproof, and condensation-proof properties, suitable for the humid and dusty outdoor environment in rural areas; the three-phase double-break structure connects the vacuum interrupter and the visible isolation break in series, allowing each phase circuit to independently complete the arc-extinguishing and isolation functions; the three-phase linkage spring operating mechanism drives the three-phase contacts to open and close synchronously through mechanical linkages, ensuring consistent opening and closing actions; the edge sensing layer is a built-in module, and the multi-source sensor monitoring array is the core functional component of this layer; the intelligent control unit is responsible for receiving and temporarily storing the acquired sensor data. For example, this embodiment uses the GW4-10 series outdoor high-voltage disconnect switch as a unified example object, where G represents disconnect switch, W represents outdoor type, 4 represents design serial number, and 10 represents rated voltage 10kV, which is a standard model for 10kV distribution networks. The GW4-10 series outdoor high-voltage disconnect switch has a fully enclosed housing with an IP67 sealing rating. The opening and closing synchronization error of the three-phase linkage spring operating mechanism is ≤2ms. The multi-source sensor monitoring array built into the edge sensing layer contains eight different types of sensing elements, which can simultaneously collect electrical, health, and environmental data, making it suitable for rural outdoor operating environments with large temperature differences and high humidity.

[0020] Secondly, the operation monitoring index set of the target switch is obtained. This set includes switch operating status, electrical parameters, health indicators, and environmental parameters. The switch operating status includes the opening and closing positions and operation times. The electrical parameters include line current, power factor, and three-phase voltage. The health indicators include contact temperature and contact resistance. The environmental parameters include ambient temperature and humidity. The operation monitoring index set is a comprehensive set of parameters characterizing the switch's operating status, covering four dimensions: switch condition, electrical performance, equipment health, and external environment. Based on switch operation requirements and distribution network maintenance priorities, a multi-dimensional and comprehensive operation monitoring index set is constructed to avoid data omissions and ensure the effectiveness of long-term data. The categories and specific contents of each indicator are as follows: switch operating status reflects switch operation; electrical parameters reflect line power supply performance; health indicators reflect the aging degree of major switch components; and environmental parameters reflect the long-term impact of the outdoor environment on the switch.

[0021] For example, for the GW4-10 series switches, the specific set of operational monitoring indicators is as follows: Switch operating status: open / close position (close is 1, open is 0), open / close operation time (accurate to milliseconds); Electrical parameters: line current (range 0-630A), power factor (range 0.8-1.0), three-phase voltage (range 10kV±10%); Health indicators: contact temperature (range -40℃-150℃), contact resistance (range 0-100mΩ); Environmental parameters: ambient temperature (range -40℃-85℃), ambient humidity (range 0-100%RH). This covers the key monitoring dimensions of the switch throughout its long-term operation lifecycle and can capture long-term trends such as contact aging and insulation performance degradation.

[0022] Furthermore, according to the preset data monitoring time interval and the set of operational monitoring indicators, within the historical time zone, the target switch is monitored in real time by a multi-source sensor monitoring array in the edge perception layer to obtain a historical switch operation data sequence set. The multi-source sensor monitoring array refers to a combined monitoring module integrating multiple sensing elements such as current sensors, temperature sensors, humidity sensors, and displacement sensors, built into the switch edge perception layer, capable of simultaneously collecting multi-dimensional operational data. The historical time zone refers to the time range used for collecting historical data; in this embodiment, it is uniformly set to the most recent three years, conforming to the power industry's requirements for historical data storage duration. The historical switch operation data sequence set refers to a continuous set of switch operation data arranged chronologically within the historical time zone. Each data point corresponds to full-dimensional indicator data for a monitoring time point, possessing the characteristics of not losing data during power outages and supporting remote access. According to the preset time interval and determined monitoring indicators, data is continuously collected by the multi-source sensor monitoring array in the edge perception layer and organized chronologically to form the historical switch operation data sequence set. The preset data monitoring interval needs to balance data density and storage pressure. An interval of 5 minutes per time can ensure data accuracy without causing storage overload. The historical time zone is selected from the last three years, which can fully cover a complete seasonal operating cycle and short-term aging process of the equipment. During the acquisition process, the sensor array converts analog signals into digital signals, which are sorted and stored by the intelligent control unit according to the timestamp, supporting subsequent remote retrieval.

[0023] For example, for the GW4-10 series switches, real-time monitoring is performed at a preset interval of 5 minutes within the historical time zone of the past three years. Taking the data collection start time as 00:00 on January 1, three years ago, the data collected at that time is: switch position 1 (closed state), switch operation time 20ms, line current 320A, power factor 0.92, phase A voltage 10.1kV / phase B voltage 10.0kV / phase C voltage 10.2kV, contact temperature 45℃, contact resistance 15mΩ, ambient temperature 25℃, and ambient humidity 60%RH. The latest set of data collected at 00:05 on the same date three years later is: switch position 1 (closed state), switch operation time 20ms, line current 320A, power factor 0.92, phase A voltage 10.1kV / phase B voltage 10.0kV / phase C voltage 10.2kV, contact temperature 45℃, contact resistance 15mΩ, ambient temperature 25℃, and ambient humidity 60%RH. The operating time is 21ms, the line current is 318A, the power factor is 0.91, the voltage of phase A is 10.0kV, the voltage of phase B is 9.9kV, the voltage of phase C is 10.1kV, the contact temperature is 47℃, the contact resistance is 17mΩ, the ambient temperature is 26℃, and the ambient humidity is 58%RH. By analogy, a historical switch operation data sequence set of 3×365×24×12=315360 data points is formed over three years. Each data point contains eleven specific indicators, which fully record the changes in the switch's operating status and environmental impact over the past three years.

[0024] In this embodiment of the invention, by embedding a multi-source sensing array into the edge sensing layer of the switch and combining it with a fully enclosed housing, the problems of poor weather resistance and susceptibility to interference of traditional distributed sensors are solved, making it suitable for complex outdoor environments in remote areas and for long-term operation. It covers specific indicators in four dimensions, and the collected data is comprehensive, accurate, and compliant. The resulting historical switch operation data sequence set has completeness, temporality, and long periodicity, which can not only reflect the short-term operating status of the equipment, but also capture long-term patterns such as seasonal load fluctuations and component aging. This provides a high-quality, high-density training data source for the operation status prediction of the long short-term memory network in the subsequent S200 step, avoiding deviations in prediction results due to missing data, distortion, or excessively short periods.

[0025] S200: Using a long short-term memory network, predict the set of switch operation data sequences within a preset time zone based on the set of historical switch operation data sequences.

[0026] In this embodiment of the invention, a Long Short-Term Memory (LSTM) network is used to predict and obtain a set of switch operation data sequences within a preset time zone based on the historical switch operation data sequence set. Distribution network operation and maintenance requires advance knowledge of the future operating status of switches to formulate predictive strategies. Therefore, this step, based on the historical data obtained in S100, combines data volatility analysis with the time-series prediction advantages of LSTM networks to construct a technical system for volatility perception, dynamic adaptation, and accurate prediction, ensuring that the predicted data accurately reflects the future operating trends of switches.

[0027] Step S200 in the method provided in this embodiment of the invention includes: Data volatility analysis is performed on several historical switch operation data sequences in the historical switch operation data sequence set to obtain several operation data variation coefficients, wherein the operation data variation coefficient is the ratio of the data standard deviation to the data mean in the historical switch operation data sequence; The switching operation fluctuation within the historical time zone is determined by weighted evaluation based on the coefficient of variation of the aforementioned operational data. Based on the fluctuation of the switch operation, the switch operation status prediction engine is invoked to predict the switch operation data sequence set within a preset time zone according to the historical switch operation data sequence set. The switch operation status prediction engine is constructed based on a long short-term memory network.

[0028] First, data volatility analysis is performed on several historical switch operation data sequences in the historical switch operation data sequence set to obtain several coefficients of variation (CVs). The CV is the ratio of the standard deviation to the mean of the data in the historical switch operation data sequence. The CV measures the degree of volatility of a single operation data sequence, and its calculation formula is: CV = Standard Deviation of Data / Data Mean The value ranges from ≥0, with larger values ​​indicating more volatile sequences. Volatility analysis is performed on the core indicator sequences that have the greatest impact on the switch's operating state, avoiding interference from irrelevant indicators. The volatility characteristics of a single indicator are quantified using the coefficient of variation.

[0029] For example, from the eleven indicators collected by S100, four series are selected: line current, contact temperature, three-phase voltage, and ambient temperature, covering electrical parameters, health indicators, and environmental parameters. The mean and standard deviation of three years' data for each series are calculated, and then substituted into the formula for the coefficient of variation. For the three-year historical data of the GW4-10 series switch, the calculation is as follows: Line current series: mean Standard deviation coefficient of variation Contact temperature sequence: mean Standard deviation coefficient of variation Phase A voltage sequence: mean Standard deviation coefficient of variation Ambient temperature series: mean Standard deviation coefficient of variation The results showed that the A-phase voltage was most stable during the most drastic fluctuations in ambient temperature.

[0030] Secondly, the operational volatility of the switch within the historical time zone is determined by a weighted evaluation based on the coefficients of variation of the aforementioned operational data. Switch operational volatility refers to the weighted evaluation value of the coefficients of variation of all monitored indicators, reflecting the overall operational stability of the switch within the historical time zone. The value ranges from 0 to 1, where 0 represents no volatility and 1 represents extremely drastic volatility. Based on the weighted allocation coefficients of different indicators' influence on switch operation, the overall volatility is obtained through weighted summation, avoiding misjudgment of the overall state due to fluctuations in a single indicator.

[0031] For example, the analytic hierarchy process (AHP) is used to determine the weights of the indicators: health indicators (contact temperature) are directly related to equipment failure risk and have the highest weight, for example, 0.4; electrical parameters (line current, A-phase voltage) are related to power supply quality and have the next highest weight, for example, 0.25 and 0.2 respectively; environmental parameters (ambient temperature) are external influencing factors and have the lowest weight, for example, 0.15. Substituting the coefficients of variation of the four types of indicators for the GW4-10 series switch into the weighted summation: volatility = (0.0469 × 0.25) + (0.1667 × 0.4) + (0.0199 × 0.2) + (0.4545 × 0.15) ≈ 0.1506. For example, the grading standard is: 0-0.1 low volatility, 0.1-0.3 medium volatility, and above 0.3 high volatility. The operating volatility of this switch is medium, requiring a medium number of prediction modules to balance accuracy and efficiency.

[0032] Furthermore, based on the switch operation fluctuation, the switch operation state prediction engine is invoked to predict and obtain the predicted switch operation data sequence set within a preset time zone according to the historical switch operation data sequence set. The switch operation state prediction engine is constructed based on a long short-term memory network.

[0033] The process of invoking the switch operation state prediction engine based on the switch operation fluctuation includes: Based on the historical operation monitoring records of similar switches of the target switch, and constrained by the time span of the historical time zone, several sample switch operation data sequence sets are collected. The historical switch operation data sequence sets within the preset time zone of different sample switch operation data sequence sets are used as sample prediction switch operation data sequence sets to obtain several sample prediction switch operation data sequence sets. The historical time zone and the preset time zone have the same time span. The training dataset is constructed using the aforementioned sample switch running data sequence set and the aforementioned sample prediction switch running data sequence set, and a random selection strategy with replacement is adopted. K sample training sets are constructed using the training dataset, where K is an integer greater than or equal to 20. Using the sample switch operation data sequence set as input and the sample predicted switch operation data sequence set as supervision, the long short-term memory network is trained to converge using the K sample training sets to generate K switch operation state prediction plug-ins. Based on the principle of ensemble learning, the K switch operation state prediction plug-ins are integrated and fused according to the mean fusion strategy to generate a switch operation state prediction engine. The number of compatible plug-ins selected, J, is obtained by multiplying the ratio of the switch operation fluctuation to the historical maximum switch operation fluctuation within the historical time range by K and rounding it down, where J is greater than or equal to 3 and less than or equal to K. J prediction plugins are randomly selected from the K switch operation status prediction plugins of the switch operation status prediction engine, and switch operation data within the preset time zone are predicted based on the historical switch operation data sequence set.

[0034] First, based on the historical operation monitoring records of similar switches to the target switch, and constrained by the time span of the historical time zone, several sample switch operation data sequence sets are collected. The historical switch operation data sequence sets within a preset time zone are then used as sample prediction switch operation data sequence sets, resulting in several sample prediction switch operation data sequence sets. The historical time zone and the preset time zone have the same time span. Similar switches refer to high-voltage disconnect switches with the same model, deployment scenario, and similar operating environment as the target switch. The sample switch operation data sequence sets refer to the historical time zone operation data of similar switches, whose time span is completely consistent with the historical time zone of the target switch, and are used as input data for model training. The sample prediction switch operation data sequence sets refer to the actual operation data of similar switches within the preset time zone, and are used as supervision labels for model training.

[0035] For example, to ensure the time dimension of the sample data is fully aligned with the target switch, three hundred groups of GW4-10 series switches of the same type, all deployed in rural 10kV distribution networks, were selected. For each sample group, a three-year historical operating data sequence was collected as the sample input. Historical data for each seven-day period within the corresponding three years was collected as the sample supervision label, forming a one-to-one input-supervision sample pair. Three hundred switches of the same model in a rural distribution network of a certain county were selected as samples. For the first sample switch, historical data from January 1, 2021 to December 31, 2023 (three years) was collected as the sample input set. Simultaneously, historical data for all seven-day periods of the same switch, such as January 8-14, 2021, January 15-21, 2021, etc., were collected as the sample prediction set, ultimately forming three sets of three-year input + multiple seven-day supervision sample data pairs.

[0036] Secondly, the aforementioned sample switch-run data sequence sets and sample prediction switch-run data sequence sets are used as training datasets. A random selection strategy with replacement is employed to construct K sample training sets using these training datasets, where K is an integer greater than or equal to 20. The random selection strategy with replacement means that when randomly selecting samples from the training dataset to construct a new training set, the samples are returned to the original dataset after each selection, allowing for repeated selection and efficiently constructing multiple independent training sets. Constructing multiple independent training sets through sampling with replacement provides a diverse foundation for subsequent ensemble learning while reducing the risk of overfitting.

[0037] For example, 300 samples are randomly selected from 300 sets of samples. For example, the first sample is sample 1, sample 2, sample 3, ... sample 300, which is used to construct training set 1. The second sample is randomly selected again, which may include sample 3, sample 4, sample 5, ..., which is used to construct training set 2. And so on, a total of K=30 training sets are generated, each of which contains 300 input-supervised sample pairs.

[0038] Furthermore, using the sample switch operation data sequence set as input and the sample predicted switch operation data sequence set as supervision, the Long Short-Term Memory (LSTM) network is trained to convergence using the K sample training sets, generating K switch operation state prediction plugins. The LSTM network is an improved recurrent neural network that solves the gradient vanishing problem in long-sequence data training through a gating mechanism, making it suitable for predicting long-cycle time-series data of power equipment operation. Each prediction plugin refers to an independent LSTM model trained to convergence based on a single sample training set, possessing independent prediction capabilities and serving as a core component of the prediction engine. Training multiple LSTM sub-models based on independent training sets enhances the stability of ensemble prediction through model diversity.

[0039] For example, LSTM networks with identical structures are configured for 30 training sets. Each input layer has eleven neurons corresponding to eleven operational monitoring metrics, two hidden layers each have 64 neurons, and an output layer has eleven neurons. Using sample switch operation data sequences as input and sample predicted switch operation data sequences as supervision, the Adam optimizer is used for training until the loss function (MAE) converges below 0.05, generating 30 prediction plugins. Plugin one is trained using training set one, with the input being three years of indicator data for a sample switch and the supervision being the real data for the corresponding seven days. After one hundred training epochs, the loss function converges to 0.048. Similarly, plugin one, plugin two, ..., plugin thirty are trained using thirty training sets, each plugin possessing the ability to independently predict seven days of switch operation data.

[0040] Subsequently, based on the principle of ensemble learning, the K switch operation status prediction plugins are integrated and fused according to the mean fusion strategy to generate a switch operation status prediction engine. Ensemble learning combines the advantages of multiple plugins to solve the problem of random error in single-model prediction. Based on the majority rule principle of ensemble learning, the K prediction plugins are integrated into a prediction engine, and a mean fusion strategy is adopted. That is, for the same prediction time point, the arithmetic mean of the output results of all plugins is taken as the temporary prediction value to ensure the stability of the prediction results. For example, if test data of a certain type of switch is input into the prediction engine, plugin one predicts the line current at a certain time point as 350A, plugin two as 354A, ..., plugin 30 as 348A. After mean fusion, the predicted value at that time point is (350+354+...+348) / 30=351A, with an error of only 0.29% compared to the true value of 350A, which is lower than the average error of 1.5% for a single plugin.

[0041] Next, the ratio of the switch operation fluctuation to the historical maximum switch operation fluctuation within the historical time range is multiplied by K and rounded to obtain the number of compatible plug-ins selected, J, where J is greater than or equal to 3 and less than or equal to K. The historical maximum switch operation fluctuation refers to the upper limit of extreme fluctuation obtained by statistically analyzing the operation data of the target switch and similar switches over the past 3 years. In this embodiment, the value is taken as 0.8, corresponding to the operation state under extreme weather conditions. The number of plug-ins is dynamically matched to avoid the waste of computing power caused by running all plug-ins. The calculation formula is: Number of compatible plug-ins selected J = round[(target switch fluctuation / historical maximum fluctuation) × K], while forcibly constraining 3 ≤ J ≤ K to ensure prediction stability and adaptability. For example, if the target switch fluctuation is 0.1506, the historical maximum fluctuation is 0.8, and K = 30, substituting into the formula, we get J = round[(0.1506 / 0.8) × 30] = round[5.6475] = 6, which satisfies the constraint 3 ≤ 6 ≤ 30, that is, six plug-ins are selected to participate in the prediction, and only 20% of the computing power of the thirty plug-ins is activated.

[0042] Finally, J prediction plugins are randomly selected from the K switch operation status prediction plugins of the switch operation status prediction engine to predict switch operation data within the preset time zone based on the historical switch operation data sequence set. The random selection of J plugins for parallel prediction followed by fusion balances prediction accuracy and efficiency, improving prediction efficiency while reducing computational resource waste. For example, six plugins (5, 12, 18, 22, 25, and 29) are randomly selected, and historical data from the GW4-10 target switch over the past three years are input. The six plugins output twelve indicators, including line current and contact temperature, for the next seven days. For the contact temperature at 14:00 on the fifth day, the predicted values ​​from the six plugins are 53℃, 51℃, 54℃, 52℃, 53℃, and 52℃ respectively. After fusion of the average values, the final predicted value is 52.5℃, with an error of only 0.94% compared to the subsequent actual operating value of 53℃. Simultaneously, the computational power consumption is reduced by 80% compared to using all thirty plugins.

[0043] In this embodiment of the invention, based on long-term historical data of all-dimensional indicators collected by S100, operational characteristics are quantified through fluctuation analysis. Combining long short-term memory networks and ensemble learning, an appropriate number of prediction plugins are dynamically selected based on operational volatility. In low-to-medium volatility scenarios, only some plugins are activated, effectively improving prediction efficiency and reducing computational redundancy. In high-volatility scenarios, more plugins are added to ensure accuracy. Finally, all-dimensional prediction data for a preset time zone is output, providing a reliable basis for optimizing the control parameters of S300.

[0044] S300: Based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time and voltage sag count as multiple optimization objectives, iteratively optimize and search the control parameters of the target switch to determine the appropriate control strategy, and intelligently control the target switch within the preset time zone according to the appropriate control strategy.

[0045] Step S300 in the method provided in this embodiment of the invention includes: Obtain the control parameter adjustment space of the target switch, wherein the control parameters include protection parameters, operation parameters, running parameters and system parameters; Several initial control parameters are randomly generated within the control parameter adjustment space; The predicted switch operation data sequence set is fused with the several initial control parameters to generate several switch operation control schemes; Within the digital operation simulation space of the target switch, control simulations are performed according to the several switch operation control schemes, and several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit times, and several simulated voltage sags are output. With minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, several parameter fitnesss are determined by weighted evaluation based on several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit time, and several simulated voltage sag counts. The parameter fitnesss are negatively correlated with simulated line losses, simulated fault outage minutes, simulated voltage over-limit time, and simulated voltage sag counts. Using the control parameter adjustment space as the search space, based on the several initial control parameters and several parameter fitness, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy.

[0046] The protection parameters include instantaneous overcurrent action current threshold, definite-time overcurrent action current threshold, zero-sequence current action threshold, overcurrent protection action delay, and reclosing action delay. The operation parameters include target closing phase angle, target opening phase angle, and closing / opening operation speed. The running parameters include dynamic load current upper limit, alarm temperature threshold, and forced air-cooled start-up temperature threshold. The system parameters include reactive power compensation switching threshold and network reconfiguration logic status.

[0047] First, the control parameter adjustment space of the target switch is obtained. The control parameters include protection parameters, operational parameters, running parameters, and system parameters. The protection parameters include instantaneous overcurrent operating current threshold, definite-time overcurrent operating current threshold, zero-sequence current operating threshold, overcurrent protection operating delay, and reclosing operating delay. The operational parameters include the target closing phase angle, the target opening phase angle, and the closing / opening operation speed. The running parameters include the dynamic load current upper limit, alarm temperature threshold, and forced air-cooled start-up temperature threshold. The system parameters include the reactive power compensation switching threshold and the network reconfiguration logic state. The control parameter adjustment space refers to the legal value range defined for various control parameters of the target switch. This range is determined based on the switch hardware performance limits, power industry operation and maintenance standards, and the actual operating requirements of the distribution network. It serves as the boundary for subsequent parameter optimization searches, preventing parameter settings from exceeding the equipment's tolerance or failing to comply with industry standards.

[0048] Among them, protection parameters are core parameters used to realize the switch fault protection function and prevent equipment damage and the spread of line faults. They are the protective barrier for the safe operation of the switch. Specifically, they include: Instantaneous overcurrent operating current threshold: a set instantaneous overcurrent protection starting current standard. When the line current instantaneously exceeds this threshold, the switch will quickly trigger a trip to prevent damage to the equipment from sudden faults such as short circuits; Time-delay overcurrent operating current threshold: a current threshold set for continuous overcurrent faults. It is used in conjunction with the action delay. When the line current exceeds this threshold and continues for a set time, the switch will only trip to avoid false tripping. Zero-sequence current action threshold: The current threshold used to detect line grounding faults. When a single-phase grounding fault occurs in the line, the zero-sequence current will increase significantly. If it exceeds this threshold, the protection action will be triggered to ensure the safety of the distribution network grounding. Overcurrent protection action delay: After an overcurrent fault occurs, the switch delays the execution of the protection action. This can be adjusted according to the characteristics of the distribution network load to balance the need for rapid fault clearing and avoidance of accidental tripping due to instantaneous faults. Reclosing action delay: After the switch trips, the reclosing operation is delayed to allow time for the fault to clear itself, reduce unnecessary power outages, and avoid reclosing on permanent faults.

[0049] Among them, the operational parameters are the parameters that control the core actions of the switch's opening and closing, directly affecting the smoothness and accuracy of the switch operation. Specifically, they include: Target closing phase angle: the set voltage or current phase angle when the switch performs a closing operation. Selecting an appropriate phase angle for closing can reduce the closing inrush current, extend the service life of the switch contacts, and ensure line stability; Target opening phase angle: the set voltage or current phase angle when the switch performs an opening operation. Opening near the current zero-crossing point can reduce the difficulty of arc extinguishing and reduce the wear of the arc on the switch components; Closing and opening operation speed: the rate at which the switch contacts complete the closing or opening action. Too fast a speed can easily cause mechanical shock, while too slow a speed will prolong the arc extinguishing time. It is necessary to adapt the mechanical performance settings to match the spring operating mechanism.

[0050] Among them, the operational parameters are those that ensure the switch is in a safe and stable state during daily operation and provide timely warnings of abnormalities. Specifically, these include: Dynamic load current limit: The maximum allowable current value set according to real-time load changes in the distribution network, which can be dynamically adjusted with load fluctuations to avoid long-term overload operation of the switch; Alarm temperature threshold: The temperature warning value for the switch's core components (such as contacts). When the temperature reaches this threshold, an alarm signal is triggered to remind maintenance personnel to check the equipment status in a timely manner; Forced air-cooling start-up temperature threshold: The temperature standard for activating the switch's air-cooling heat dissipation device. When the component temperature exceeds this threshold, the air-cooling system is automatically activated to reduce the equipment temperature and prevent failures caused by overheating.

[0051] Among them, system parameters refer to parameters that adapt to the overall operation requirements of the distribution network and realize the coordinated optimization of switches and the distribution network. Specifically, they include: reactive power compensation switching threshold: the start and stop standard of reactive power compensation device set based on the line power factor. When the power factor is lower or higher than the threshold, the reactive power compensation device is automatically switched to reduce line loss and improve power supply quality; network reconfiguration logic state: setting the trigger conditions and execution logic for switches to participate in distribution network topology reconfiguration. When a fault occurs in the distribution network or the load is uneven, the switch acts according to the preset logic to assist in reconfiguring the distribution network structure and ensure power supply continuity.

[0052] For example, the control parameter adjustment range details for the GW4-10 series switches are as follows: Protection type: Instantaneous overcurrent operating current threshold (1.2-1.5 times rated current, i.e., 756-945A), definite-time overcurrent operating current threshold (1.1-1.3 times rated current, i.e., 693-819A), zero-sequence current operating threshold (5-10A, adapted for distribution network grounding fault detection), overcurrent protection operating delay (0.5-3s), reclosing operating delay (0.5-3s); Operation type: Target closing phase angle (0-360°, voltage zero crossing). Optimal values ​​for: 0° (optimal), target tripping phase angle (optimal 180° current zero crossing), closing and tripping operation speed (0.8-1.2m / s, adaptable to the mechanical limits of the spring mechanism); Operational: dynamic load current upper limit (630-800A, adjusted according to load fluctuations), alarm temperature threshold (60-75℃, below the contact withstand temperature of 80℃), forced air cooling start-up temperature (70-75℃); System: reactive power compensation switching threshold (power factor 0.85-0.95), network reconfiguration logic status (automatic switching when load deviation > 20%).

[0053] Secondly, several initial control parameters are randomly generated within the control parameter adjustment space. A uniform sampling method is used to generate five sets of parameters within the adjustment space, covering high, medium, and low value ranges, avoiding local optima caused by concentrated initial samples. For example, Group 1: Instantaneous overcurrent = 1.3 × 630 = 819 A, closing phase angle = 90°, alarm temperature = 65℃, reactive power compensation switching threshold = 0.9; Group 2: Instantaneous overcurrent = 1.5 × 630 = 945 A, closing phase angle = 0°, alarm temperature = 70℃, switching threshold = 0.85; Group 3: Instantaneous overcurrent = 1.2 × 630 = 756 A, closing phase angle = 180°, alarm temperature = 60℃, switching threshold = 0.95; Groups 4 and 5: These parameters are combined in combination to cover unexplored value scenarios.

[0054] Then, the predicted switch operation data sequence set is fused with the several initial control parameters to generate several switch operation control schemes. Each set of initial parameters is bound to the operating condition characteristics predicted by S200, clarifying the parameter execution logic for different time periods, so that the scheme has time-series adaptability. For example, the scheme generated after fusing the predicted data of the first set of parameters is as follows: Morning peak (6:00-9:00, predicted current 800A): instantaneous overcurrent threshold is maintained at 819A (slightly higher than the predicted peak), forced air cooling start-up temperature is reduced to 70℃ (pre-heating), and the dynamic load current upper limit is set to 800A; Rainy day (3rd day, humidity 85%): closing phase angle is fixed at 0° (voltage zero crossing point, reducing closing impact), reclosing delay is extended to 2s (allowing time for fault self-clearing); Nighttime off-peak (0:00-5:00, predicted current 300A): reactive power compensation switching threshold is raised to 0.95 (reducing invalid compensation), alarm temperature is raised to 70℃ (reducing false alarms).

[0055] Furthermore, within the digital operation simulation space of the target switch, control simulations are performed according to several switch operation control schemes, outputting several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit times, and several simulated voltage sags. The digital operation simulation space refers to a virtual environment calibrated with historical data, based on the physical structure of the target switch and the distribution network topology. It can reproduce switch actions and line responses under load fluctuations and temperature / humidity changes, outputting quantitative results. The control schemes are input into the digital operation simulation space, driving the switch operation according to the predicted timing sequence, reproducing load fluctuations and environmental changes, and outputting quantitative results for four types of targets. For example, five schemes simultaneously simulate seven days of operation: Scheme 1 simulation results: simulated line losses are at a moderate level; simulated fault outage minutes are 0 minutes; simulated voltage over-limit times are 10 minutes; simulated voltage sags occur 2 times. Scheme 2 simulation results: simulated line losses are at a low level; simulated fault outage minutes are 30 minutes; simulated voltage over-limit times are 30 minutes; simulated voltage sags occur 3 times.

[0056] Subsequently, with minimizing line losses, the number of minutes of power outages, voltage over-limit time, and the number of voltage sags as multiple optimization objectives, several parameter fitnesss were determined based on a weighted evaluation of several simulated line losses, simulated power outage minutes, simulated voltage over-limit time, and simulated voltage sags. The parameter fitness is negatively correlated with these parameters. Parameter fitness is a comprehensive score allocated according to maintenance priority, with a maximum score of 10 points, and is negatively correlated with the optimization objectives; for example, no power outages receive full marks, and lower losses result in higher scores.

[0057] For example, weights are assigned according to maintenance priorities, with power outages accounting for 40%, voltage sags for 25%, voltage exceedances for 20%, and line losses for 15%. Scores are calculated accordingly, with a maximum of 10 points. Group 1: No power outages occurred, receiving 4 points based on the 40% weighting for power outages; two voltage sags occurred, receiving 3 points based on the 25% weighting for voltage sags; a voltage exceedance lasted 10 minutes, receiving 1.8 points based on the 20% weighting for voltage exceedances; line losses were at a moderate level, receiving 1.2 points based on the 15% weighting for line losses. The scores from these four items are added together for a total of 10 points. Group 2: One power outage lasting 30 minutes will result in a deduction of 4 points (40% weight for power outage), resulting in 0 points for this item; three voltage dips will result in a 25% weight for voltage dips; a voltage over-limit time of 30 minutes will result in a 1% weight for voltage over-limits; and a low line loss level will result in a 1.5% weight for line loss. The total score for all four items is 4.5 points.

[0058] Furthermore, using the control parameter adjustment space as the optimization space, based on the several initial control parameters and several parameter fitness, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy.

[0059] Specifically, using the control parameter adjustment space as the optimization space, and based on the several initial control parameters and several parameter fitness values, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy, including: The initial control parameters are set as the initial solutions, and several initial solutions are arranged in descending order of parameter fitness to generate an initial solution sequence; The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining initial solutions other than the optimal and poor solutions are selected as inferior solutions, thus obtaining the optimal solution, multiple inferior solutions, and poor solutions; According to the preset optimization direction and the preset optimization step size, the multiple inferior solutions are updated and adjusted, and then reordered to obtain the updated solution sequence; After eliminating a preset number of solutions from the updated solution sequence, control parameters that did not appear during the optimization process are randomly selected within the control parameter adjustment space and supplemented with equivalent values ​​to generate an updated solution sequence. The preset number decreases as the number of optimization attempts increases. Based on the optimization mechanism of selecting optimization direction, adjusting inferior solutions, eliminating solutions, and supplementing solutions, iterative optimization is continued according to the updated solution sequence until the preset number of convergences is reached, and the optimal solution of the current updated solution sequence is output as the adaptation control strategy.

[0060] First, the initial control parameters are set as initial solutions. Several initial solutions are then arranged in descending order of parameter fitness to generate an initial solution sequence. An initial solution refers to the generated combination of initial control parameters; each initial solution corresponds to a set of control parameters and a corresponding parameter fitness. The initial solutions are sorted in descending order of fitness to determine the fitness level of each solution. For example, five initial solutions are sorted by fitness as follows: Group 1 (10 points), Group 4 (8.5 points), Group 3 (7.8 points), Group 2 (6.2 points), and Group 5 (4.5 points), forming the initial solution sequence: [Group 1, Group 4, Group 3, Group 1, Group 5].

[0061] Next, the first solution in the initial solution sequence is selected as the optimal solution, the last solution as the suboptimal solution, and the remaining initial solutions (excluding the optimal and suboptimal solutions) are selected as inferior solutions, resulting in optimal solutions, multiple suboptimal solutions, and suboptimal solutions. The first and last solutions of the sequence are extracted as optimal and suboptimal solutions, respectively, while the middle solutions are identified as suboptimal solutions, thus clarifying the adjustment targets and benchmarks. For example, the first solution group (group 1) is selected as the optimal solution; the last solution group (group 5) is selected as the suboptimal solution; and the remaining groups (groups 4, 3, and 2) are selected as suboptimal solutions.

[0062] Furthermore, the multiple inferior solutions are updated and adjusted according to the preset optimization direction and the preset optimization step size, and then reordered to obtain an updated solution sequence.

[0063] The method for setting the preset optimization direction includes: Obtain the fitness of the optimal solution parameters, the fitness of the poor solution parameters, and the fitness of the multiple inferior solutions parameters, and calculate the average fitness of the inferior solutions parameters. The deviation of the fitness values ​​of the optimal solution parameters and the mean fitness values ​​of the inferior solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the optimal solution. The deviation between the mean fitness of the inferior solution parameters and the fitness of the poor solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the poor solution. If the fitness deviation of the optimal solution is greater than or equal to the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the optimization strategy, wherein the optimization strategy is the direction that moves closer to the optimal solution; If the fitness deviation of the optimal solution is less than the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the deviation avoidance strategy, wherein the deviation avoidance strategy is the direction away from the poor solution.

[0064] First, obtain the fitness of the optimal solution, the fitness of the poor solution, and the fitness of the multiple inferior solutions, and calculate the mean fitness of the inferior solutions. The mean fitness of the inferior solutions is the sum of the fitness values ​​of all inferior solutions divided by the number of inferior solutions, reflecting the average performance of the medium fitness parameters. Extract the fitness values ​​of the optimal solution, the poor solution, and all inferior solutions, and calculate the mean fitness of the inferior solutions by arithmetic mean to define the benchmark for medium fitness level. For example, obtain the fitness of each solution: optimal solution fitness = 10 points, poor solution fitness = 4.5 points, and the fitness of the three inferior solutions are 8.5 points, 7.8 points, and 6.2 points, respectively. Calculate the mean fitness of the inferior solutions = (8.5 + 7.8 + 6.2) / 3 = 7.5 points.

[0065] Secondly, the deviation between the fitness scores of the optimal and inferior solutions is calculated, and the absolute value of the deviation is taken as the fitness deviation of the optimal solution. The fitness deviation of the optimal solution refers to the absolute difference between the fitness score of the optimal solution and the mean fitness score of the inferior solutions, quantifying the gap between the optimal solution and the average fitness level. The fitness deviation of the optimal solution is obtained by subtracting the mean fitness score of the inferior solutions from the fitness score of the optimal solution, and taking the absolute value, which measures the extent of the advantage of the optimal solution. For example, the fitness deviation of the optimal solution = |fitness score of the optimal solution - mean fitness score of the inferior solutions| = |10 - 7.5| = 2.5 points, indicating that the optimal solution has an advantage of 2.5 points higher than the average fitness level.

[0066] Furthermore, the deviation between the mean fitness of inferior solutions and the fitness of poor solutions is calculated, and the absolute value of the deviation is taken as the fitness deviation of poor solutions. The fitness deviation of poor solutions refers to the absolute difference between the mean fitness of inferior solutions and the fitness of poor solutions, quantifying the gap between poor solutions and the moderate fitness level. The fitness deviation of poor solutions is obtained by subtracting the fitness of poor solutions from the mean fitness of inferior solutions and taking the absolute value, which measures the extent of the inferiority of the poor solutions. For example, the fitness deviation of poor solutions = |mean fitness of inferior solutions - fitness of poor solutions| = |7.5 - 4.5| = 3.0 points, indicating that the poor solutions are 3.0 points lower in fitness than the moderate fitness level.

[0067] Furthermore, if the fitness deviation of the optimal solution is greater than or equal to the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the optimization-oriented strategy, where the optimization-oriented strategy is the direction closer to the optimal solution. The optimization-oriented strategy refers to the parameter value range in which the optimization direction points towards the optimal solution, improving fitness by adjusting the poor solution to move closer to the optimal solution. By comparing the two deviations, if the deviation of the optimal solution is greater than or equal to the deviation of the poor solution, the optimization-oriented strategy is selected.

[0068] Secondly, if the fitness deviation of the optimal solution is less than the fitness deviation of the inferior solution, then the optimization direction for this optimization is set as the deviation avoidance strategy, whereby the deviation avoidance strategy is the direction away from the inferior solution. The deviation avoidance strategy refers to the optimization direction deviating from the parameter value range of the inferior solution, expanding the effective search range by avoiding inefficient parameter combinations of the inferior solution. Compare the two deviations; if the optimal solution deviation < the inferior solution deviation, the deviation avoidance strategy is selected. For example, if the fitness deviation of the optimal solution is 2.5 points < the fitness deviation of the inferior solution is 3.0 points, it indicates that the inefficient parameter combination of the inferior solution has a greater interference with the optimization, so the deviation avoidance strategy is selected. According to the preset optimization step size, focus on adjusting the parameters of the inferior solution that are close to the inferior solution and move away from the unreasonable values ​​of the inferior solution. The instantaneous overcurrent of Group 4 is adjusted from 780A to 810A; the closing phase angle of Group 3 is adjusted from 120° to 30°; and the alarm temperature of Group 2 is adjusted from 60℃ to 65℃, expanding the effective search range by avoiding the shortcomings of the inferior solution.

[0069] Based on this, the multiple inferior solutions are updated and adjusted according to the preset optimization direction and the preset optimization step size, and then reordered to obtain an updated solution sequence. For example, the three groups of inferior solutions after adjustment are input into the simulation space for re-evaluation to obtain new fitness scores: Group 4 (9.2 points), Group 3 (8.8 points), and Group 2 (8.0 points). Combining the original excellent solutions and inferior solutions, the updated solution sequence is generated by sorting them from largest to smallest fitness score: [Group 1 (10 points), Group 4 (9.2 points), Group 3 (8.8 points), Group 2 (8.0 points), Group 5 (4.5 points)].

[0070] Further, after eliminating a predetermined number of solutions from the updated solution sequence, control parameters that did not appear during the optimization process are randomly selected from the control parameter adjustment space for equivalent supplementation, generating an updated solution sequence. The predetermined number decreases as the number of optimization attempts increases. The predetermined number refers to the number of solutions to be eliminated in each iteration, decreasing with the number of optimization attempts. Initially, inefficient solutions are eliminated to expand the search range, while later eliminations are reduced to stabilize the optimal solution. Equivalent supplementation means that the number of new solutions added equal to the number of eliminated solutions remains constant, ensuring a stable optimization sample size. Unappeared control parameters refer to combinations of control parameters that have never been included in any initial or updated solution and are strictly within the control parameter adjustment space to avoid redundant searches. An updated solution sequence is a new solution sequence formed after adjusting, reordering, eliminating, and supplementing inferior solutions; it serves as the base sample for the next iteration.

[0071] For example, a preset number is set according to the rule that the more optimization attempts, the fewer eliminations are made. This is the first iteration, and the preset elimination number is two. In the updated solution sequence, the two solutions with the worst fit at the end are eliminated, namely Group 2 (8.0 points) and Group 5 (4.5 points). After eliminating two solutions by adding new control parameters that have not appeared, two new combinations of control parameters that have never appeared are added to maintain the scale of five solutions: Supplementary Solution 1: instantaneous overcurrent 810A, closing phase angle 0°, alarm temperature 65℃; Supplementary Solution 2: instantaneous overcurrent 820A, closing phase angle 60°, alarm temperature 70℃. An updated solution sequence is generated: [Group 1 (10 points), Group 4 (9.2 points), Group 3 (8.8 points), Supplementary Solution 1, Supplementary Solution 2].

[0072] Finally, based on the optimization mechanism of selecting the optimization direction, adjusting inferior solutions, eliminating solutions, and supplementing solutions, iterative optimization continues according to the updated solution sequence until a preset convergence number is reached. The optimal solution of the current updated solution sequence is then output as the adaptation control strategy. The preset convergence number is a set number of consecutive iterations to ensure stability. When the fitness fluctuation of the optimal solution in consecutive iterations does not exceed a preset threshold, the optimization stops when the cumulative number of iterations reaches this value. The optimization direction may be different each time, set based on the actual state to maximize the optimization convergence speed. For example, the preset convergence number is three, and the fitness fluctuation threshold is 0.5 points. The system iterates cyclically using a closed-loop mechanism, focusing on avoiding inefficient parameters of the solution: Second iteration: The optimization direction remains the avoidance strategy. Supplementary solution one and supplementary solution two are adjusted, the last solution (supplementary solution two, fitness score 8.2) is eliminated, and a new set of solutions (supplementary solution three: instantaneous overcurrent 815A, closing phase angle 0°, alarm temperature 65℃) is added. The fitness of the optimal solution group one remains at 10 points. Third iteration: No elimination or addition. After adjustment, the fitness of supplementary solution three increases to 9.1 points, and the fitness of group one remains stable at 10 points. The fluctuation of the optimal solution is ≤0.5 points for three consecutive iterations, meeting the preset convergence count. The output is the adaptation control strategy: The current optimal solution is group one, and its core parameters are the final adaptation control strategy: instantaneous overcurrent 819A, closing phase angle 0°, alarm temperature 65℃.

[0073] In this embodiment of the invention, based on predicted switch operation data, the focus is on multi-objective optimization requirements. A dynamically adaptable optimization direction selection mechanism precisely balances the need to improve efficiency by seeking the best solution while avoiding errors and expanding the search range. Inferior solution updates and adjustments, combined with gradient step size, effectively correct the shortcomings of moderately adaptable parameters. Dynamically decreasing the number of eliminated solutions, coupled with the addition of equivalent new solutions, avoids inefficient solutions consuming computational power and introduces new parameter combinations to prevent the optimization from getting stuck in local optima. A closed-loop iterative mechanism, combined with a preset convergence standard, ensures the optimization process is efficient and stable, avoiding excessive iteration and wasted resources. The final output adaptive control strategy, verified through multiple rounds of simulation, can accurately adapt to the complex future operating conditions of the switch, reduce power outages and voltage anomalies, reduce line losses, achieve intelligent control without manual intervention, and is suitable for unattended distribution network scenarios in remote areas, effectively improving power supply stability and operating efficiency.

[0074] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides an intelligent control method and system for outdoor fully enclosed small high-voltage disconnect load switches. It collects indicators covering all dimensions of switch operation, laying a solid data foundation for subsequent processes and ensuring data integrity and compliance. Combining long short-term memory networks and ensemble learning, it dynamically adapts the number of prediction plugins, improving prediction efficiency and reducing computational redundancy while ensuring the accuracy of core indicator predictions. This provides a reliable reference for future operating conditions in control strategy formulation. Guided by multi-objective optimization, it generates highly adaptable control strategies through dynamic selection of optimization directions, precise adjustment of inferior solutions, dynamic elimination of supplementary solutions, and closed-loop iterative convergence, effectively improving the operating efficiency and power supply reliability of distribution networks in remote areas.

[0075] Example 2, as Figure 2 As shown, this invention provides an intelligent control system for a fully enclosed outdoor miniature high-voltage disconnecting load switch, the system comprising: The edge sensing acquisition module 11 is used to acquire a set of historical switch operation data sequences of the target switch in the historical time zone through the multi-source sensing monitoring array of the edge sensing layer, wherein the target switch is an outdoor fully enclosed small high-voltage isolation load switch; Data prediction module 12 is used to use a long short-term memory network to predict and obtain a set of predicted switch operation data sequences within a preset time zone based on the historical switch operation data sequence set; The multi-objective optimization control module 13 is used to perform iterative optimization search on the control parameters of the target switch based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time and voltage sag number as multiple optimization objectives, to determine the appropriate control strategy, and to perform intelligent control on the target switch in the preset time zone according to the appropriate control strategy.

[0076] In one embodiment, the edge sensing acquisition module 11 is further configured to: The outdoor fully enclosed miniature high-voltage isolating load switch includes a fully enclosed housing, a three-phase double-break structure, a three-phase linkage spring operating mechanism, an edge sensing layer, and an intelligent control unit. The three-phase double-break structure integrates the vacuum interrupter and the visible isolation break in series in each phase circuit. The three-phase linkage spring operating mechanism adopts a two-position spring mechanism, which drives the three-phase contacts to open and close synchronously through mechanical linkages. The edge sensing layer has a built-in multi-source sensor monitoring array.

[0077] Obtain the set of operation monitoring indicators for the target switch, wherein the set of operation monitoring indicators includes switch operation status, electrical parameters, health indicators and environmental parameters, the switch operation status includes opening and closing positions and opening and closing operation time, the electrical parameters include line current, power factor and three-phase voltage, the health indicators include contact temperature and contact resistance, and the environmental parameters include ambient temperature and ambient humidity; According to the preset data monitoring time interval and the set of operation monitoring indicators, the target switch is monitored in real time through the multi-source sensor monitoring array of the edge perception layer within the historical time zone to obtain the historical switch operation data sequence set.

[0078] In one embodiment, the data prediction module 12 is further configured to: Data volatility analysis is performed on several historical switch operation data sequences in the historical switch operation data sequence set to obtain several operation data variation coefficients, wherein the operation data variation coefficient is the ratio of the data standard deviation to the data mean in the historical switch operation data sequence; The switching operation fluctuation within the historical time zone is determined by weighted evaluation based on the coefficient of variation of the aforementioned operational data. Based on the fluctuation of the switch operation, the switch operation status prediction engine is invoked to predict the switch operation data sequence set within a preset time zone according to the historical switch operation data sequence set. The switch operation status prediction engine is constructed based on a long short-term memory network.

[0079] The process of invoking the switch operation state prediction engine based on the switch operation fluctuation includes: Based on the historical operation monitoring records of similar switches of the target switch, and constrained by the time span of the historical time zone, several sample switch operation data sequence sets are collected. The historical switch operation data sequence sets within the preset time zone of different sample switch operation data sequence sets are used as sample prediction switch operation data sequence sets to obtain several sample prediction switch operation data sequence sets. The historical time zone and the preset time zone have the same time span. The training dataset is constructed using the aforementioned sample switch running data sequence set and the aforementioned sample prediction switch running data sequence set, and a random selection strategy with replacement is adopted. K sample training sets are constructed using the training dataset, where K is an integer greater than or equal to 20. Using the sample switch operation data sequence set as input and the sample predicted switch operation data sequence set as supervision, the long short-term memory network is trained to converge using the K sample training sets to generate K switch operation state prediction plug-ins. Based on the principle of ensemble learning, the K switch operation state prediction plug-ins are integrated and fused according to the mean fusion strategy to generate a switch operation state prediction engine. The number of compatible plug-ins selected, J, is obtained by multiplying the ratio of the switch operation fluctuation to the historical maximum switch operation fluctuation within the historical time range by K and rounding it down, where J is greater than or equal to 3 and less than or equal to K. J prediction plugins are randomly selected from the K switch operation status prediction plugins of the switch operation status prediction engine, and switch operation data within the preset time zone are predicted based on the historical switch operation data sequence set.

[0080] In one embodiment, the multi-objective optimization control module 13 is further configured to: Obtain the control parameter adjustment space of the target switch, wherein the control parameters include protection parameters, operation parameters, running parameters and system parameters; Several initial control parameters are randomly generated within the control parameter adjustment space; The predicted switch operation data sequence set is fused with the several initial control parameters to generate several switch operation control schemes; Within the digital operation simulation space of the target switch, control simulations are performed according to the several switch operation control schemes, and several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit times, and several simulated voltage sags are output. With minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, several parameter fitnesss are determined by weighted evaluation based on several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit time, and several simulated voltage sag counts. The parameter fitnesss are negatively correlated with simulated line losses, simulated fault outage minutes, simulated voltage over-limit time, and simulated voltage sag counts. Using the control parameter adjustment space as the search space, based on the several initial control parameters and several parameter fitness, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy.

[0081] The protection parameters include instantaneous overcurrent action current threshold, definite-time overcurrent action current threshold, zero-sequence current action threshold, overcurrent protection action delay, and reclosing action delay. The operation parameters include target closing phase angle, target opening phase angle, and closing / opening operation speed. The running parameters include dynamic load current upper limit, alarm temperature threshold, and forced air-cooled start-up temperature threshold. The system parameters include reactive power compensation switching threshold and network reconfiguration logic status.

[0082] Specifically, using the control parameter adjustment space as the optimization space, and based on the several initial control parameters and several parameter fitness values, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy, including: The initial control parameters are set as the initial solutions, and several initial solutions are arranged in descending order of parameter fitness to generate an initial solution sequence; The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining initial solutions other than the optimal and poor solutions are selected as inferior solutions, thus obtaining the optimal solution, multiple inferior solutions, and poor solutions; According to the preset optimization direction and the preset optimization step size, the multiple inferior solutions are updated and adjusted, and then reordered to obtain the updated solution sequence; After eliminating a preset number of solutions from the updated solution sequence, control parameters that did not appear during the optimization process are randomly selected within the control parameter adjustment space and supplemented with equivalent values ​​to generate an updated solution sequence. The preset number decreases as the number of optimization attempts increases. Based on the optimization mechanism of selecting optimization direction, adjusting inferior solutions, eliminating solutions, and supplementing solutions, iterative optimization is continued according to the updated solution sequence until the preset number of convergences is reached, and the optimal solution of the current updated solution sequence is output as the adaptation control strategy.

[0083] The method for setting the preset optimization direction includes: Obtain the fitness of the optimal solution parameters, the fitness of the poor solution parameters, and the fitness of the multiple inferior solutions parameters, and calculate the average fitness of the inferior solutions parameters. The deviation of the fitness values ​​of the optimal solution parameters and the mean fitness values ​​of the inferior solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the optimal solution. The deviation between the mean fitness of the inferior solution parameters and the fitness of the poor solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the poor solution. If the fitness deviation of the optimal solution is greater than or equal to the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the optimization strategy, wherein the optimization strategy is the direction that moves closer to the optimal solution; If the fitness deviation of the optimal solution is less than the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the deviation avoidance strategy, wherein the deviation avoidance strategy is the direction away from the poor solution.

[0084] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0086] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A method for intelligent control of outdoor fully enclosed miniature high-voltage disconnecting load switches, characterized in that: The methods include: The target switch is a small, fully enclosed outdoor high-voltage isolating load switch. The target switch is a set of historical switch operation data sequences in the historical time zone collected by the multi-source sensor monitoring array of the edge sensing layer. Using a long short-term memory network, a predicted set of switch operation data sequences within a preset time zone is obtained based on the historical switch operation data sequence set. Based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, the control parameters of the target switch are iteratively optimized and searched to determine an adaptive control strategy. The target switch is then intelligently controlled within the preset time zone according to the adaptive control strategy.

2. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnector load switches according to claim 1, characterized in that, The outdoor fully enclosed miniature high-voltage disconnect load switch includes a fully enclosed housing, a three-phase double-break structure, a three-phase linkage spring operating mechanism, an edge sensing layer, and an intelligent control unit. The three-phase double-break structure integrates the vacuum interrupter and the visible isolation break in series in each phase circuit. The three-phase linkage spring operating mechanism adopts a two-position spring mechanism, which drives the three-phase contacts to open and close synchronously through mechanical linkages. The edge sensing layer has a built-in multi-source sensor monitoring array.

3. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 1, characterized in that, The target switch's historical switch operation data sequence set within the historical time zone is collected through a multi-source sensing monitoring array at the edge sensing layer, including: Obtain the set of operation monitoring indicators for the target switch, wherein the set of operation monitoring indicators includes switch operation status, electrical parameters, health indicators and environmental parameters, the switch operation status includes opening and closing positions and opening and closing operation time, the electrical parameters include line current, power factor and three-phase voltage, the health indicators include contact temperature and contact resistance, and the environmental parameters include ambient temperature and ambient humidity; According to the preset data monitoring time interval and the set of operation monitoring indicators, the target switch is monitored in real time through the multi-source sensor monitoring array of the edge perception layer within the historical time zone to obtain the historical switch operation data sequence set.

4. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 1, characterized in that, Using a Long Short-Term Memory (LSTM) network, a predicted set of switch operation data sequences within a preset time zone is obtained based on the historical switch operation data sequence set, including: Data volatility analysis is performed on several historical switch operation data sequences in the historical switch operation data sequence set to obtain several operation data variation coefficients, wherein the operation data variation coefficient is the ratio of the data standard deviation to the data mean in the historical switch operation data sequence; The switching operation fluctuation within the historical time zone is determined by weighted evaluation based on the coefficient of variation of the aforementioned operational data. Based on the fluctuation of the switch operation, the switch operation status prediction engine is invoked to predict the switch operation data sequence set within a preset time zone according to the historical switch operation data sequence set. The switch operation status prediction engine is constructed based on a long short-term memory network.

5. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 4, characterized in that, The switch operation fluctuation is used to invoke the switch operation state prediction engine, including: Based on the historical operation monitoring records of similar switches of the target switch, and constrained by the time span of the historical time zone, several sample switch operation data sequence sets are collected. The historical switch operation data sequence sets within the preset time zone of different sample switch operation data sequence sets are used as sample prediction switch operation data sequence sets to obtain several sample prediction switch operation data sequence sets. The historical time zone and the preset time zone have the same time span. The training dataset is constructed using the aforementioned sample switch running data sequence set and the aforementioned sample prediction switch running data sequence set, and a random selection strategy with replacement is adopted. K sample training sets are constructed using the training dataset, where K is an integer greater than or equal to 20. Using the sample switch operation data sequence set as input and the sample predicted switch operation data sequence set as supervision, the long short-term memory network is trained to converge using the K sample training sets to generate K switch operation state prediction plug-ins. Based on the principle of ensemble learning, the K switch operation state prediction plug-ins are integrated and fused according to the mean fusion strategy to generate a switch operation state prediction engine. The number of compatible plug-ins selected, J, is obtained by multiplying the ratio of the switch operation fluctuation to the historical maximum switch operation fluctuation within the historical time range by K and rounding it down, where J is greater than or equal to 3 and less than or equal to K. J prediction plugins are randomly selected from the K switch operation status prediction plugins of the switch operation status prediction engine, and switch operation data within the preset time zone are predicted based on the historical switch operation data sequence set.

6. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 1, characterized in that, Based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, the control parameters of the target switch are iteratively optimized and searched to determine the appropriate control strategy, including: Obtain the control parameter adjustment space of the target switch, wherein the control parameters include protection parameters, operation parameters, running parameters and system parameters; Several initial control parameters are randomly generated within the control parameter adjustment space; The predicted switch operation data sequence set is fused with the several initial control parameters to generate several switch operation control schemes; Within the digital operation simulation space of the target switch, control simulations are performed according to the several switch operation control schemes, and several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit times, and several simulated voltage sags are output. With minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, several parameter fitnesss are determined by weighted evaluation based on several simulated line losses, several simulated fault outage minutes, several simulated voltage over-limit time, and several simulated voltage sag counts. The parameter fitnesss are negatively correlated with simulated line losses, simulated fault outage minutes, simulated voltage over-limit time, and simulated voltage sag counts. Using the control parameter adjustment space as the search space, based on the several initial control parameters and several parameter fitness, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy.

7. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 6, characterized in that, The protection parameters include instantaneous overcurrent action current threshold, definite-time overcurrent action current threshold, zero-sequence current action threshold, overcurrent protection action delay, and reclosing action delay. The operation parameters include target closing phase angle, target opening phase angle, and closing / opening operation speed. The running parameters include dynamic load current upper limit, alarm temperature threshold, and forced air-cooled start-up temperature threshold. The system parameters include reactive power compensation switching threshold and network reconfiguration logic status.

8. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 6, characterized in that, Using the control parameter adjustment space as the search space, and based on the several initial control parameters and several parameter fitness values, the control parameters of the target switch are iteratively optimized and searched to determine the suitable control strategy, including: The initial control parameters are set as the initial solutions, and several initial solutions are arranged in descending order of parameter fitness to generate an initial solution sequence; The first solution of the initial solution sequence is selected as the optimal solution, the last solution is selected as the poor solution, and the remaining initial solutions other than the optimal and poor solutions are selected as inferior solutions, thus obtaining the optimal solution, multiple inferior solutions, and poor solutions; According to the preset optimization direction and the preset optimization step size, the multiple inferior solutions are updated and adjusted, and then reordered to obtain the updated solution sequence; After eliminating a preset number of solutions from the updated solution sequence, control parameters that did not appear during the optimization process are randomly selected within the control parameter adjustment space and supplemented with equivalent values ​​to generate an updated solution sequence. The preset number decreases as the number of optimization attempts increases. Based on the optimization mechanism of selecting optimization direction, adjusting inferior solutions, eliminating solutions, and supplementing solutions, iterative optimization is continued according to the updated solution sequence until the preset number of convergences is reached, and the optimal solution of the current updated solution sequence is output as the adaptation control strategy.

9. The intelligent control method for outdoor fully enclosed miniature high-voltage disconnecting load switch according to claim 8, characterized in that, The method for setting the preset optimization direction includes: Obtain the fitness of the optimal solution parameters, the fitness of the poor solution parameters, and the fitness of the multiple inferior solutions parameters, and calculate the average fitness of the inferior solutions parameters. The deviation of the fitness values ​​of the optimal solution parameters and the mean fitness values ​​of the inferior solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the optimal solution. The deviation between the mean fitness of the inferior solution parameters and the fitness of the poor solution parameters is calculated, and the absolute value of the deviation is taken as the fitness deviation of the poor solution. If the fitness deviation of the optimal solution is greater than or equal to the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the optimization strategy, wherein the optimization strategy is the direction that moves closer to the optimal solution; If the fitness deviation of the optimal solution is less than the fitness deviation of the poor solution, then the optimization direction for this optimization is set as the deviation avoidance strategy, wherein the deviation avoidance strategy is the direction away from the poor solution.

10. An intelligent control system for a fully enclosed outdoor miniature high-voltage disconnector load switch, characterized in that: For implementing the intelligent control method of the outdoor fully enclosed miniature high-voltage disconnector load switch according to any one of claims 1-9, the system comprises: The edge sensing acquisition module is used to acquire a set of historical switch operation data sequences of the target switch in the historical time zone through the multi-source sensor monitoring array of the edge sensing layer, wherein the target switch is an outdoor fully enclosed small high-voltage isolation load switch; The data prediction module is used to use a long short-term memory network to predict and obtain a set of predicted switch operation data sequences within a preset time zone based on the historical switch operation data sequence set. The multi-objective optimization control module is used to perform iterative optimization search on the control parameters of the target switch based on the predicted switch operation data sequence set, with minimizing line loss, fault outage minutes, voltage over-limit time, and voltage sag count as multiple optimization objectives, to determine an adaptive control strategy, and to perform intelligent control on the target switch within the preset time zone according to the adaptive control strategy.

Images