Capacitor switching control method based on industrial intelligent gateway

Abstract

The application belongs to the technical field of industrial intelligent gateway and relates to a capacitor switching control method based on an industrial intelligent gateway. The method collects voltage, current, power factor and capacitor operating state through an industrial intelligent gateway and a multi-protocol conversion unit, constructs a synchronous state sequence in an edge computing module according to a sampling time, a protocol refresh cycle and a state return time; performs timing prediction based on the sequence to obtain a power factor change trend, a reactive power demand direction and a prediction confidence; inputs a current deviation, a trend slope, a state of a capacity that has been put into operation and a switching response offset into a fuzzy PID algorithm to generate a candidate switching amount and timing, and forms an execution, a delayed confirmation or a conservative closed-loop control instruction through confidence gating; and corrects prediction and control parameters according to state window changes after switching. The application can reduce false switching, over-compensation and switching oscillation caused by data mismatch and short-time disturbance.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent gateway technology, and relates to a capacitor switching control method based on an industrial intelligent gateway. Background Technology

[0002] In power distribution and industrial power systems, capacitor switching control is commonly used for reactive power compensation and power factor regulation. Existing conventional solutions typically involve a field controller or low-voltage reactive power compensation device collecting grid voltage, current, power factor, and capacitor switching status. The controller generates switching commands based on preset power factor upper and lower limits, voltage protection boundaries, switching delay times, and capacitor bank capacity levels. When the power factor is detected to be below the activation limit and remains below it for a set time, the controller activates the corresponding capacitor bank; when the power factor is detected to be above the deactivation limit, the bus voltage is too high, or the compensation amount is too large, the controller deactivates some capacitor banks. To avoid frequent contactor operation, conventional control logic also sets switching interlocking times, cyclic switching sequences, and switching frequency limits. This type of solution is simple to implement and suitable for scenarios with slow load changes, a single data acquisition link, and stable compensation objects. However, its control is mainly based on current sampled values ​​and fixed delay judgments, and capacitor switching actions typically lag behind reactive load changes.

[0003] With the increasing variety of field devices, industrial smart gateways are increasingly being used for data aggregation and communication conversion in capacitor switching control. Conventional industrial smart gateways typically connect to power meters, reactive power compensation controllers, capacitor switching switches, and upper-level platforms via multiple communication protocols, converting data from different protocol formats into a unified data format and providing voltage, current, power factor, capacitor status, and alarm information to local control programs or remote platforms. Some solutions configure edge computing capabilities at the gateway end, enabling it to perform limit judgment, simple interlocking, and switching command forwarding locally, reducing communication latency that occurs when relying entirely on the upper-level platform. The focus of these solutions is usually on protocol adaptation, data forwarding, and remote monitoring; the control logic still largely relies on threshold judgment, fixed delays, or ordinary closed-loop regulation. For data from devices using different protocols, the gateway generally buffers and calculates data in the order of reception, rarely performing unified reconstruction of sampling time, protocol refresh cycle, and capacitor status return time.

[0004] Some existing capacitor switching control schemes incorporate predictive control, fuzzy control, or proportional-integral-derivative (PID) control to address the insufficient adaptability of fixed threshold control to load fluctuations. These schemes typically predict short-term reactive power demand based on historical power factor variation curves or adjust the switching stage based on the current power factor deviation and its rate of change. Other schemes combine fuzzy rules with PID control to provide smoother control inputs based on the magnitude and trend of the deviation. While these algorithms can reduce the rigidity of simple threshold control to some extent, they still suffer from unstable control input quality in industrial applications. Because voltage, current, power factor, and capacitor status may originate from different devices, data refresh cycles, communication delays, and status feedback times may differ, leading to potential asynchrony between the predicted sequence and the actual execution state. When the algorithm directly uses misaligned data as input, the predicted trend may reflect late samples or short-term disturbances rather than actual reactive power demand changes.

[0005] The main technical problem with existing technologies is that industrial smart gateways, when performing capacitor switching control, struggle to distinguish between real reactive power demand changes and transient disturbances or data mismatches under conditions of inconsistent timing of multi-protocol data, sampling jitter, and lag in capacitor state feedback. This leads to a mismatch between switching commands generated based on current values, ordinary predicted values, or conventional fuzzy control quantities and the actual grid state on site. The technical reason for this problem is that existing control processes typically lack closed-loop constraints between unified timescale reconstruction, prediction reliability determination, and post-switching feedback correction. There is no quality gating relationship between prediction results and control outputs, and the actual capacitor switching response does not correct subsequent predictions and control judgments in reverse. When the load exhibits short-cycle fluctuations or capacitor state return is delayed, the controller is prone to identifying transient deviations as persistent reactive power gaps, or performing switching based on outdated data even after the actual reactive power demand has changed, resulting in erroneous switching, delayed switching, overcompensation, or switching oscillations. Summary of the Invention

[0006] The purpose of this invention is to provide a capacitor switching control method based on an industrial intelligent gateway, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A capacitor switching control method based on an industrial smart gateway includes: collecting grid voltage, current, power factor, and capacitor operating status data through the multi-protocol conversion unit of the industrial smart gateway, and constructing a synchronous state sequence in the edge computing module based on the sampling time, protocol refresh cycle, and state return time; performing time-series prediction based on the synchronous state sequence to obtain the power factor change trend, reactive power demand direction, and prediction reliability; inputting the current power factor deviation, the power factor change trend, the reactive power demand direction, and the status of the already engaged capacitor capacity into a fuzzy PID algorithm based on time-series prediction to generate candidate switching quantities and candidate switching opportunities; performing gating judgment on the candidate switching quantities and candidate switching opportunities based on the prediction reliability to generate switching commands for execution, delayed confirmation, or conservative closed-loop control; and performing feedback correction on the time-series prediction and fuzzy PID control parameters based on the grid state changes before and after switching.

[0009] Preferably, constructing the synchronization state sequence includes: extracting the sampling time, reception time, protocol type, and capacitor status return time from data frames from different communication protocols; mapping voltage, current, power factor, and capacitor operating status data to a unified time axis according to the same control cycle; marking data frames exceeding the control cycle as late samples, and determining whether late samples enter the prediction input based on the state continuity of adjacent cycles; setting an unconfirmed flag for capacitor operating status data with missing status returns, so that the unconfirmed flag and the power grid measurement data participate together in the subsequent prediction reliability calculation.

[0010] Preferably, obtaining the prediction reliability includes: generating a short-time prediction sequence in the edge computing module based on the synchronous state sequence of multiple consecutive control cycles; comparing the short-time prediction sequence with the subsequently acquired measured power factor sequence to form a prediction residual sequence; constructing the prediction reliability based on the prediction residual sequence, the phase consistency of voltage and current within the same control cycle, the continuity of power factor change direction, and the integrity of capacitor operating state data; when the prediction residual sequence shows a continuous reverse offset, marking the corresponding prediction cycle as a disturbance cycle, and preventing the disturbance cycle from directly triggering early switching.

[0011] Preferably, the fuzzy PID algorithm based on time-series prediction includes: using the current power factor deviation, the slope of the power factor change trend, the direction of reactive power demand, the proportion of the capacity of the capacitors already in operation, and the response offset after the most recent switching as fuzzy input variables; determining the adjustment direction of the proportional, integral, and derivative control quantities based on the fuzzy input variables; matching the adjustment direction with the available capacitor switching capacity levels to obtain candidate switching quantities; and jointly determining the slope of the power factor change trend and the response offset after the most recent switching to obtain candidate switching timing.

[0012] Preferably, the gating decision based on the prediction confidence level includes: when the prediction confidence level meets the high confidence level condition and the reactive power demand direction is consistent within the continuous control cycle, the candidate switching amount is restricted to a small-step switching amount that matches the predicted reactive power gap; when the prediction confidence level meets the medium confidence level condition, the candidate switching amount is written into the delayed confirmation queue, and the power factor change trend is re-compared in the next control cycle to determine whether to execute; when the prediction confidence level meets the low confidence level condition or there is an unconfirmable capacitor status indicator, the participation of timing prediction in the candidate switching timing is suppressed, and conservative closed-loop control is executed according to the current power factor deviation.

[0013] Preferably, when reconstructing the state of the synchronization state sequence, a refresh cycle profile is established for data frames from different sources according to the protocol type; the timing offset of each data frame relative to the current control cycle is calculated based on the refresh cycle profile; interpolation alignment is performed on data frames whose timing offset is within the compensable range, and frozen reference is performed on data frames whose timing offset exceeds the compensable range; the interpolation alignment result, frozen reference result, and unconfirmed flag are written together into the quality label of the synchronization state sequence, so that the quality label enters the timing prediction process along with the voltage, current, power factor, and capacitor operating state data.

[0014] Preferably, before generating candidate switching quantities, the edge computing module performs correlation and discrimination between the predicted residual sequence and the response offset after the most recent switching; when the direction of change of the predicted residual sequence is consistent with the direction of response offset, the predicted residual sequence is classified as capacitor switching response deviation; when the direction of change of the predicted residual sequence is inconsistent with the direction of response offset, the predicted residual sequence is classified as external load disturbance; based on the classification results, the input weights of the power factor change trend slope and the current power factor deviation in the fuzzy PID algorithm are adjusted respectively, and the adjusted input weights are used for generating candidate switching quantities in the next control cycle.

[0015] Preferably, the delayed confirmation queue establishes queue entries based on candidate switching amounts, candidate switching timings, prediction reliability, and reactive power demand direction. In the next control cycle, the edge computing module compares the reactive power demand direction corresponding to the queue entry with the newly generated reactive power demand direction. When the two are consistent and there is no unconfirmed identifier in the capacitor operating status data, a switching instruction is generated according to the candidate switching amounts in the queue entry. When the two are inconsistent or a new unconfirmed identifier appears, the queue entry is deleted and conservative closed-loop control is re-executed.

[0016] Preferably, after any switching command is executed, the industrial intelligent gateway captures the state window before and after the switching; in the two state windows, it calculates the voltage change, current change, power factor change, reactive power demand direction change, and capacitor operating status change, respectively; it associates and records these changes with candidate switching amounts, candidate switching timings, quality labels, prediction confidence, and delayed confirmation queue status; when the associated records show that the power factor change direction in the post-switching state window is inconsistent with the predicted direction before the switching, it writes the corresponding control cycle into the feedback correction sample and reduces the prediction confidence under the same quality label in subsequent control cycles.

[0017] Preferably, the feedback correction samples are grouped in the edge computing module according to quality label type, prediction residual type, response offset type, and switching command source; for each group, the local error compensation amount of time series prediction, the input weight of the fuzzy PID algorithm, and the gating judgment condition are updated respectively; when the cloud platform issues the global reactive power optimization boundary, the industrial intelligent gateway performs consistency verification between the global reactive power optimization boundary and the updated gating judgment condition, and generates the local switching command for the next control cycle without changing the global reactive power optimization boundary.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] 1. This invention uses an industrial intelligent gateway to collect voltage, current, power factor, and capacitor operating status data via multiple protocols. A synchronous state sequence is constructed based on the sampling time, protocol refresh cycle, and state return time, ensuring a unified time reference for the data entering control calculations. Timing prediction generates reactive power demand direction and power factor change trends on the synchronous state sequence. Combined with prediction residuals, data integrity, phase consistency, and state confirmation, prediction reliability is formed, filtering out unreliable predictions caused by late samples, missing states, and short-term disturbances before control commands are generated. Fuzzy proportional-integral-derivative control no longer calculates the switching amount solely based on the current power factor deviation. Instead, it combines trend slope, reactive power demand direction, the proportion of already engaged capacitor capacity, and switching response offset to generate candidate switching amounts and timings. Prediction reliability gating distinguishes candidate switching actions into direct execution, delayed confirmation, or conservative closed-loop control. This allows reliable trends to drive early, small-step switching, moderately reliable trends to be reviewed in the next control cycle, and low-reliability data to prevent direct triggering of predicted switching. The state windows before and after the switching process form feedback correction samples, and update the timing prediction error compensation, fuzzy control input weights, and gating judgment conditions to ensure that subsequent switching commands continuously adapt to changes in the field state. This control chain can effectively solve the problem of inaccurate switching commands caused by multi-protocol data mismatch and disturbance misjudgment, reducing erroneous switching, overcompensation, and repeated switching.

[0020] 2. The quality tags, late sample markers, frozen reference results, and unconfirmed identifiers in the synchronization state sequence of this invention enable industrial intelligent gateways to retain temporal quality differences between data from different sources, avoiding the equivalent processing of data with large communication delays as real-time measurement data. The delayed confirmation queue associates and saves candidate switching quantities, candidate switching timings, prediction reliability, and reactive power demand direction, and re-compares the reactive power demand direction and capacitor status confirmation status in the next control cycle, providing a verification process for switching actions in unstable trends and reducing frequent capacitor actions caused by short-term limit violations. The association and discrimination between prediction residuals and switching response offsets enable the edge side to distinguish between capacitor switching response deviations and external load disturbances, and adjust the input weights of the power factor change trend slope and the current power factor deviation respectively, preventing the same type of deviation from repeatedly triggering the same switching judgment in subsequent control cycles. Consistency verification is performed between the global reactive power optimization boundary issued by the cloud and the gated judgment conditions updated locally by the gateway, ensuring that local millisecond-level switching control is executed without breaking global constraints. The above processing enables the capacitor switching control to maintain a relatively stable control logic under conditions of long-term operation, mixed access of protocol devices, and frequent load fluctuations. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of the capacitor switching control method based on an industrial intelligent gateway of the present invention;

[0022] Figure 2 This is a flowchart of the multi-protocol data time stamp reconstruction and synchronization state sequence generation process of the present invention;

[0023] Figure 3 This is a flowchart of the prediction reliability gating and delayed confirmation switching process of the present invention;

[0024] Figure 4 This is a flowchart of the switching feedback correction and global reactive power optimization boundary consistency verification of the present invention;

[0025] Figure 5 This is a synchronization state reconstruction diagram of the present invention;

[0026] Figure 6 The state quality contribution diagram of the present invention;

[0027] Figure 7 This is a graph showing the predictive residuals and trend continuity of the present invention.

[0028] Figure 8 This is a residual response correlation diagram of the present invention. Detailed Implementation

[0029] refer to Figure 1In one embodiment, the capacitor switching control method based on an industrial smart gateway is applied to a reactive power compensation control link consisting of an industrial smart gateway, an edge computing module, a multi-protocol conversion unit, a cloud platform, and a capacitor switching execution device. The industrial smart gateway receives grid voltage, current, power factor, and capacitor operating status data through the multi-protocol conversion unit. The edge computing module does not directly perform switching judgments according to the receiving order, but instead writes the sampling time, protocol refresh cycle, and capacitor status return time into a data buffer of the same control cycle, and generates a synchronous state sequence according to a unified time base. This synchronous state sequence serves as the common input for time-series prediction and a fuzzy PID algorithm based on time-series prediction. The time-series prediction outputs the power factor change trend, reactive power demand direction, and prediction reliability. The fuzzy PID algorithm, based on... The current power factor deviation, trend slope, reactive power demand direction, and the status of the capacitor capacity already in operation generate candidate switching quantities and candidate switching times. The prediction reliability gating logic then converts the candidate results into switching commands that can be executed directly, confirmed with delay, or controlled conservatively in a closed loop. After the switching action is completed, the industrial intelligent gateway intercepts the changes in the grid state before and after the switching and forms feedback correction samples. The feedback correction samples are written back to the timing prediction error compensation, fuzzy PID input weights, and gating judgment conditions. The cloud platform is used to issue global reactive power optimization boundaries and strategy versions. The millisecond-level switching commands are generated locally by the industrial intelligent gateway. The advantage of this embodiment is that it integrates multi-protocol data timing, prediction reliability gating, fuzzy PID control, and switching feedback correction into a closed loop, so that the switching commands do not depend on a single current value or an unfiltered prediction value.

[0030] refer to Figure 2 Specifically, after receiving data frames from different protocol sources, the industrial smart gateway establishes a data channel identifier, sampling time, reception time, protocol type, status return time, and control cycle identifier for each measurement data. Voltage, current, power factor, and capacitor operating status data are all mapped to a unified time axis according to the control cycle. When multiple measurement data exist within the same control cycle, the edge computing module selects the data with higher status quality that is adjacent to the center time of the control cycle as the master sample. When a data frame is between two adjacent control cycles, an aligned sample is formed through linear interpolation. When a data frame is later than the current control cycle and exceeds the compensable range, it is marked as a late sample and temporarily excluded from the prediction input of the current cycle. When the capacitor status return is missing, an unconfirmable identifier is set and written into the synchronous status sequence along with the measurement data. Each record in the synchronous status sequence contains a physical measurement value and a quality label, so that the reliability of subsequent predictions can simultaneously reflect power grid changes and data source reliability. The advantage of this embodiment is that it eliminates input misalignment caused by different protocol refresh cycles and status return delays.

[0031]

[0032] in, Indicates the data channel index. Indicates data channel At any moment Normalized measurement value, Indicates the first Alignment time of a unified control cycle and They respectively represent the locations Sampling time of the same data channel on both sides This represents the synchronization state value mapped to the unified control cycle. This indicates the positional ratio of the alignment time between two adjacent sampling times. For example, if the normalized value of the voltage channel is 0.94 at control cycle number 2 and 0.98 at control cycle number 4, and the unified alignment time is control cycle number 3, then... The calculation result represents the aligned input value of the voltage channel in a uniform control cycle.

[0033] refer to Figure 5 This figure is an engineering illustration of the synchronization state reconstruction, covering 1200 unified control cycles, ranging from CYC-030801 to CYC-032000. The horizontal axis represents the unified control cycle, and the vertical axis represents the normalized measurement value. The curve data is jointly generated by the gateway-side running frames, synchronization states, prediction records, gating decisions, and feedback samples. The current window has a maximum value of 1.0000, a minimum value of 0.3570, and an average value of 0.5941. The most recent gating result is a delayed confirmation, and the reactive power direction is required. There are 0 late samples and 0 unconfirmed states within the window, indicating that the channel timing quality directly affects the review priority of candidate controls. This figure focuses on observing how voltage, current, and power factor are organized into the same control input when the sampling times of multiple protocols are inconsistent. If frozen references or late samples increase, the curve will show a plateau segment or local deviation, which will weaken the subsequent reliability, and the switching action is more likely to enter delayed confirmation.

[0034] In this embodiment, the synchronization state sequence field and its processing meaning are shown in Table 1.

[0035] Table 1 Synchronization State Sequence Fields and Their Processing Meanings

[0036] Data Channel Identifier Voltage, current, power factor, capacitor operating status Establish a channel index based on the source device and protocol type. Forming traceable predictive input sources Sampling time The moment the data is formed on the source side Compared with unified control cycle Determine if interpolation is aligned or late. Reception time The moment when the industrial smart gateway receives the data frame Joint judgment with protocol refresh cycle profile Distinguishing between communication delays and actual state changes State return time Feedback time after capacitor is connected or disconnected Related to the timing of the switching instruction Identify execution response offset Quality Label Interpolation alignment, frozen references, late samples, unconfirmed flag Write the synchronization state sequence Participating in prediction reliability and gating decision

[0037] refer to Figure 3In a further embodiment, the edge computing module establishes a refresh cycle profile for data frames from different protocol sources. The refresh cycle profile is jointly described by the arrival interval, sampling time interval, and status return interval of historical adjacent data frames. If a data channel maintains stable refresh over multiple control cycles, it is marked as a stable channel. If a data channel has consecutive late arrivals or missing returns, it is marked as a fluctuating channel. Interpolation alignment or direct referencing is performed on the data of stable channels, while frozen referencing and quality labels are attached to the data of fluctuating channels. Frozen referencing means retaining the most recently confirmed status when there is a lack of reliable new samples in the current control cycle, but at the same time reducing the status quality score. The capacitor operating status data adopts the confirmation priority principle. It is only considered as a confirmed status when the status return time can be associated with the switching command time. Otherwise, it enters the synchronization status sequence with an unconfirmed mark. The advantage of this embodiment is that it enables time series prediction to identify data quality differences and avoids mistaking late samples for real load changes.

[0038]

[0039] in, Indicates the first State quality value for each control cycle Indicates the completeness of measurement data. Indicates the degree of confirmation of the capacitor's condition. This indicates that the protocol refreshes the matching degree. This indicates that the voltage and current are in phase. , , and This indicates the calculated weight of the corresponding quality factor. This indicates a weighted sum, for example, taking... , , , and take , , , ,but The calculation result indicates that the state quality of the current control cycle is reduced due to the unconfirmed state of the capacitor.

[0040] refer to Figure 6This figure is an engineering illustration of the state quality contribution, covering 1200 unified control cycles, ranging from CYC-030801 to CYC-032000. The horizontal axis represents the unified control cycle, and the vertical axis represents the quality contribution; the curve data is jointly generated by the gateway-side running frames, synchronization status, prediction records, gating decisions, and feedback samples. The current window has a maximum value of 0.9993, a minimum value of 0.1851, and an average value of 0.3925. The most recent gating result is a delayed confirmation, and the reactive power direction requires input. There are 0 late samples and 0 unconfirmed states within the window, indicating that the channel timing quality directly affects the review priority of candidate controls. This figure is used to break down the contributions of measurement integrity, state confirmation, refresh matching, and phase consistency to state quality. A decrease in the total quality value usually means incomplete information on the access side; even if the power factor deviation is large, it is not advisable to directly rely on the prediction trend to trigger large actions.

[0041] Preferably, the timing prediction process is performed on a synchronous state sequence. The edge computing module takes the normalized power factor, voltage, current, and the state of the capacitors already in operation for multiple consecutive control cycles as input to form a short-time prediction sequence. The short-time prediction sequence includes the predicted power factor value and reactive power demand direction for several future control cycles. After the predicted value arrives at the subsequent measured value, the residual is compared with the measured power factor sequence to form a prediction residual sequence. If the prediction residual sequence shows a reverse shift within a consecutive control cycle, the edge computing module marks the corresponding control cycle as a disturbance cycle. The prediction result within the disturbance cycle does not directly trigger early switching, but enters the reliability gating for verification. The continuity of the power factor change direction is calculated by the sign change of adjacent values ​​in the prediction sequence. A high continuity indicates that the trend direction is stable, while a low continuity indicates that the predicted value is affected by short-time disturbances or data mismatch. The advantage of this embodiment is that the reliability of the prediction result itself is written into the control link, rather than using all prediction results equally for switching calculation.

[0042]

[0043] in, Indicates the first The prediction residual for each control period Indicates the first The measured normalized power factor value for each control cycle. This indicates the prediction obtained in the previous control cycle. Normalized power factor value for each control cycle Indicates the continuity of the predicted trend direction. This indicates the forecast span for participating in trend judgment. This represents a sign function where positive values ​​are 1, zero values ​​are 0, and negative values ​​are -1, for example... , hour If the differences for the next three predictions are all positive, then The calculation result indicates that the predicted value is higher than the measured value, but the trend continues to rise.

[0044] refer to Figure 7 This figure is an engineering illustration of the predicted residuals and trend continuity, covering 1200 unified control cycles, ranging from CYC-030801 to CYC-032000. The horizontal axis represents the unified control cycle, and the vertical axis represents the residuals and continuity. The curve data is jointly generated by the gateway-side running frames, synchronization status, prediction records, gating decisions, and feedback samples. The current window has a maximum value of 1.0000, a minimum value of -0.0690, and an average value of 0.2889. The most recent gating result is a delayed confirmation, and the reactive power direction requires input. There are 0 late samples and 0 unconfirmed states within the window, indicating that the channel timing quality directly affects the review priority of candidate controls. This figure also shows the prediction deviation and the stability of the trend direction. The continuous same sign of the residuals indicates a systematic deviation in the prediction, and the trend continuity being close to zero indicates short-term directional instability. The control strategy should increase the review ratio and reduce the tendency to act prematurely.

[0045] In a preferred embodiment, the prediction reliability is jointly formed by state quality, prediction residual, and trend direction continuity. The edge computing module normalizes and compares the prediction residual with an acceptable residual benchmark, uses state quality as the basis for data reliability, and trend direction continuity as the stability of reactive power demand direction. The three are multiplied to obtain the prediction reliability of the current control cycle. The prediction reliability is not directly equivalent to the switching command, but serves as a gating input for candidate switching quantities and candidate switching timing. When the prediction reliability is low, even if the candidate switching quantity is large, it will not be executed directly. When the prediction reliability is in an intermediate state, the candidate switching quantity enters the delayed confirmation queue. When the prediction reliability is high and the reactive power demand direction is continuous and consistent, the edge computing module allows for early small-step switching. The advantage of this embodiment is that by constraining the prediction results through the source of quality, prediction error, and trend stability, prediction misjudgments caused by transient disturbances are reduced.

[0046]

[0047] in, Indicates the first The predictive reliability of each control cycle Indicates the state quality value. This represents the absolute value of the predicted residual. This represents the residual benchmark used for normalizing the residuals. This indicates that the operation takes the smaller value. The absolute value representing the predictive continuity of the trend direction, for example , , , ,but The calculation results indicate that even when the state quality is high and the trend is continuous, the prediction residual will still reduce the prediction confidence.

[0048] Specifically, the fuzzy PID algorithm based on time-series prediction takes the current power factor deviation, the slope of the power factor change trend, the direction of reactive power demand, the proportion of already connected capacitor capacity, and the response offset after the most recent switching as inputs. The current power factor deviation describes the degree of deviation between the measured power factor and the target power factor range. The slope of the power factor change trend describes the direction and magnitude of the predicted sequence change in the future control cycle. The direction of reactive power demand describes the control direction in which the system needs to connect or disconnect capacitors. The proportion of already connected capacitor capacity is used to limit the available space for further connection or disconnection. The response offset after the most recent switching is used to reflect the difference between the executed action and the grid response. The fuzzy inference rule generates the adjustment direction of proportional, integral, and derivative control quantities based on the above inputs. The proportional control quantity corresponds to the current deviation response, the integral control quantity corresponds to the continuous deviation accumulation, and the derivative control quantity corresponds to the trend change suppression. The candidate switching quantity is obtained by matching the fuzzy PID output with the set of switchable capacity. The candidate switching timing is obtained by jointly judging the trend slope and the response offset. The advantage of this embodiment is that the predicted trend, current error, and actual switching response jointly participate in the switching calculation.

[0049]

[0050] in, Indicates the first Normalized control output for each control cycle Indicates the current power factor deviation. This indicates the slope of the power factor change trend. This indicates the response offset after the most recent switch. , and These represent the proportional, integral, and derivative control coefficients of the fuzzy rule output, respectively. Indicates recent The cumulative deviation over each control cycle This represents the change in deviation between adjacent control cycles, for example... , , , , , ,but The calculation result indicates that the current deviation, cumulative deviation, and deviation change together form the normalized switching control output.

[0051] Furthermore, the generation of candidate switching quantities does not employ a simple rounding to a fixed level. Instead, the normalized control output, the predicted reactive power gap scale, and the set of switchable capacitor capacities are all input into the level mapping function. The set of switchable capacitor capacities consists of capacitor capacity levels currently in an operational or operational state, with the operational direction corresponding to a positive level, the operational direction corresponding to a negative level, and the operational state corresponding to a zero level. The level mapping function selects the level with the smallest distance from the calculated value and without violating the operational capacity state. If the candidate switching quantity conflicts with the prediction confidence gating result, the gating result is used to generate the switching command. The candidate switching timing is determined by a combination of trend slope and response offset. If the trend slope continuously points to the expansion of the reactive power gap and the response offset is within a stable range, the candidate switching timing is allowed to be advanced to the current control cycle. If the response offset direction is opposite to the trend slope, it enters the delayed confirmation queue. The advantage of this embodiment is that both the switching capacity and the switching timing are constrained by the predicted trend and the actual execution response.

[0052]

[0053] in, Indicates the first Candidate shift positions for each control cycle This represents the set of capacitor switching positions that can be selected in the current control cycle. Indicates from Select the setting that is closest to the calculated value within the parentheses. This indicates the normalized scale corresponding to the predicted reactive power deficit. Indicates normalized control output, for example , , ,but , The calculation result indicates that the candidate throwing gear is two normalized gears in the throwing direction.

[0054] In a further embodiment, the prediction confidence gating decision consists of three control states: high confidence, medium confidence, and low confidence. These three control states are not determined in isolation by a fixed single threshold, but are jointly determined by prediction confidence, the continuity of reactive power demand direction, capacitor status confirmation, and quality label. The high confidence state requires that the prediction confidence is within the directly controllable range and that the reactive power demand direction is consistent within the continuous control cycle. The output small step switching amount must not exceed the candidate level corresponding to the predicted reactive power gap. The medium confidence state writes the candidate switching amount and candidate switching timing into the delayed confirmation queue, and determines whether to execute after re-comparing the reactive power demand direction in the next control cycle. In the low confidence state or when there is an unconfirmed flag, the participation of timing prediction in the candidate switching timing is suppressed, and conservative closed-loop control is executed according to the current power factor deviation. The advantage of this embodiment is that the reliable trend, the unconfirmed trend, and the unreliable trend enter different control paths respectively.

[0055] In this embodiment, the prediction confidence gating state and control processing are shown in Table 2.

[0056] Table 2 Prediction Confidence Gating Status and Control Process

[0057] Highly reliable state The forecast reliability is within a directly controllable range, the reactive power demand direction is continuous and consistent, and the capacitor status has been confirmed. Allow small step candidate cuts to enter the current control cycle Medium-credible state The prediction's reliability is within the verification range, and the trend direction requires further observation. Candidate pitching quantities and candidate pitching timings are entered into the delayed confirmation queue. Low Trust State The prediction's reliability falls within a conservative range, or there are unconfirmable indicators. Suppressing premature switching and entering conservative closed-loop control Disturbance period The predicted residuals show a continuous backward shift, or the quality labels indicate that late samples dominate. Pause candidate early submission and re-collect state for the next cycle

[0058] In a preferred embodiment, the delayed confirmation queue stores candidate switching quantities, candidate switching timings, prediction reliability, reactive power demand direction, quality tags, and capacitor status confirmation in the form of queue items. After a queue item enters the next control cycle, the edge computing module compares the reactive power demand direction recorded in the queue item with the newly generated reactive power demand direction. At the same time, it checks whether there is an unconfirmable capacitor operating status indicator in the new control cycle. If the two directions are consistent and there is no unconfirmable indicator, a switching instruction is generated according to the candidate switching quantities in the queue item. If the directions are inconsistent or a new unconfirmable indicator appears, the queue item is deleted and the conservative closed-loop control is re-entered. If there are multiple queue items in the same control cycle, they are merged according to the queue generation time and the consistency of reactive power demand direction. The merged candidate switching quantities must not exceed the current set of switchable capacities. The advantage of this embodiment is that switching actions in unstable trends are reviewed over time, reducing repeated switching caused by short-term overruns.

[0059] Furthermore, before generating candidate switching values, the edge computing module correlates and distinguishes the predicted residual sequence and the response offset after the most recent switching. The response offset is formed by the difference between the power factor change direction after the switching action and the candidate switching direction. If the change direction of the predicted residual sequence is consistent with the response offset direction, the predicted residual is classified as capacitor switching response deviation, indicating that the actual state change generated by the previous switching action was not fully absorbed by the prediction model. If the change direction of the predicted residual sequence is inconsistent with the response offset direction, the predicted residual is classified as external load disturbance, indicating that the prediction error mainly comes from load change rather than the capacitor action itself. After classification, the input weights of the trend slope and the current power factor deviation in the fuzzy PID algorithm are adjusted respectively. The prediction weight of the trend slope is reduced and the participation of the response offset is increased for the response deviation type residual. The participation of the current power factor deviation is increased and premature switching is suppressed for the external disturbance type residual. The advantage of this embodiment is that the same predicted residual enters different correction paths according to the difference in the source.

[0060]

[0061] in, Indicates the first The residual response correlation value for each control cycle. Indicates the length of the historical window used for association determination. Represents the historical forecast residuals. Indicates the historical response offset. This represents the mean of historical forecast residuals. The numerator represents the mean of the historical response offset, the denominator represents the co-variance of the residuals and the response offset, and the denominator represents the normalized product of their fluctuation scales. For example, if the historical prediction residuals are 0.02, 0.03, and 0.04, and the historical response offsets are 0.01, 0.015, and 0.02, then the two change in the same direction and are calculated as follows: The calculation results indicate that the predicted residual and the switching response offset remain in the same direction within this window.

[0062] refer to Figure 8This figure is an engineering diagram illustrating the residual response, covering 1200 unified control cycles, ranging from CYC-030801 to CYC-032000. The horizontal axis represents the unified control cycle, and the vertical axis represents correlation and offset. The curve data is jointly generated from gateway-side running frames, synchronization status, prediction records, gating decisions, and feedback samples. The current window has a maximum value of 0.4610, a minimum value of -0.6902, and an average value of -0.0461. The most recent gating result was a delayed confirmation, and the reactive power direction indicates that it needs to be activated. There are 0 late samples and 0 unconfirmed states within the window, indicating that the channel timing quality directly affects the review priority of candidate controls. This figure is used to determine whether the residual change and the response offset after switching are in the same direction. When they are in the same direction, most deviations come from insufficient capacitor action response; when they are not in the same direction, it is more likely to be due to external load disturbances, and the system will suppress premature action and increase the participation of the current deviation.

[0063] refer to Figure 4 In a further embodiment, the industrial intelligent gateway captures a pre-switching state window before the switching command is executed. The pre-switching state window includes voltage changes, current changes, power factor changes, reactive power demand direction changes, capacitor operating status changes, and quality tags for several consecutive control cycles before the switching command is generated. After the switching action is completed and a status return is received, a post-switching state window is captured. The post-switching state window includes similar data for several consecutive control cycles after the switching command is executed. The edge computing module calculates the power factor change direction, reactive power demand direction change, and capacitor status changes in the two state windows respectively, and associates and records the calculation results with candidate switching amounts, candidate switching timings, prediction confidence, quality tags, and delayed confirmation queue status. If the power factor change direction in the post-switching state window is inconsistent with the predicted direction before the switching, the corresponding control cycle is written into the feedback correction sample. If the two are consistent but the change magnitude deviates from the prediction result, the corresponding control cycle is written into the error compensation sample. The advantage of this embodiment is that the actual response after the switching can correct subsequent control calculations in reverse.

[0064]

[0065] in, Indicates the first Local prediction error compensation amount per control cycle This represents the updated local prediction error compensation amount. Indicates the compensation update step size. Indicates the confidence level of the prediction. This represents the measured change in power factor within the status window after switching. This represents the predicted change in power factor before the switching process. This represents the difference between the actual response and the predicted response, for example... , , , ,but The calculation result indicates that the prediction model reduces the predicted compensation amount corresponding to the same switching response in subsequent control cycles.

[0066] Specifically, the feedback correction samples are grouped in the edge computing module according to quality label type, prediction residual type, response offset type, and switching command source. The quality label type is used to distinguish between interpolation alignment, frozen references, late samples, and unconfirmed identifiers. The prediction residual type is used to distinguish between in-direction residuals, reverse residuals, and fluctuating residuals. The response offset type is used to distinguish between insufficient switching response, response delay, and inconsistent direction. The switching command source is used to distinguish between direct execution, delayed confirmation, and conservative closed-loop control. After grouping, the local error compensation amount of the time series prediction, the input weight of the fuzzy PID algorithm, and the gating decision conditions are updated respectively. If inconsistency in direction after switching occurs multiple times under a certain quality label, the prediction confidence contribution corresponding to that quality label is reduced. If a certain type of response offset is continuously associated with the lag in capacitor state return, the review priority of the delayed confirmation queue is increased. If the response corresponding to a certain type of switching command source is relatively stable, its gating decision conditions remain unchanged. The advantage of this embodiment is that the switching feedback is processed according to the cause, avoiding the use of the same correction method for all errors.

[0067] In this embodiment, the feedback correction sample grouping and update objects are shown in Table 3.

[0068] Table 3 Feedback Correction Sample Groups and Update Targets

[0069] Quality label type Interpolation alignment, frozen references, late samples, unconfirmed flag State quality contribution to prediction confidence Predicting residual types Same-direction residuals, opposite-direction residuals, and fluctuating residuals Local error compensation for time series forecasting Response offset type Insufficient response, delayed response, inconsistent direction Fuzzy PID input weights and candidate switching timing Source of throwing instructions Direct execution, delayed confirmation, conservative closed-loop control Gating criteria and queue processing logic

[0070] In a preferred embodiment, the cloud platform forms a global reactive power optimization boundary based on the statistical status reported by multiple industrial intelligent gateways. The global reactive power optimization boundary includes the target power factor range, the total allowable compensation, the branch control priority range, and the conflict relationship of prohibiting simultaneous actions. After receiving the global reactive power optimization boundary, the industrial intelligent gateway does not treat it as a direct switching command. Instead, it performs a consistency check between the global reactive power optimization boundary and the locally updated gating judgment conditions. If the local candidate switching quantity is within the allowable range of the global reactive power optimization boundary and does not trigger the conflict relationship of prohibiting simultaneous actions, a switching command is generated within the local control cycle. If the local candidate switching quantity exceeds the total allowable compensation or is inconsistent with the conflict relationship, the candidate switching quantity is reduced or enters the delayed confirmation queue. The status changes of surrounding electrical equipment are interconnected through the industrial intelligent gateway and written into the local statistical status, so that the global boundary and local gating can operate based on the same data caliber. The advantage of this embodiment is that while maintaining the rapid local response of the gateway, the switching action does not deviate from the global reactive power scheduling constraints.

[0071] Furthermore, conservative closed-loop control is not an independent control method detached from the prediction link. Instead, it restricts candidate switching commands when prediction confidence is insufficient, quality labels are abnormal, or capacitor status is unconfirmed. In conservative closed-loop control, the edge computing module generates a small range of candidate switching actions based only on the current power factor deviation, the capacity status of the already engaged capacitors, and the confirmation status of the capacitors. It treats the prediction trend as an observation variable rather than a trigger variable. If the prediction confidence recovers within a continuous control cycle and the reactive power demand direction remains consistent, the conservative closed-loop control exits and re-enters the prediction confidence gating process. If the capacitor status is still unconfirmed, the frozen reference is maintained and status return is collected again. The switching records generated by conservative closed-loop control are also written into the feedback correction sample and grouped according to the source of the switching command. The advantage of this embodiment is that it maintains the closed-loop constraint of switching control when data quality deteriorates, avoiding low-confidence predictions from directly driving capacitor actions.

[0072] In one embodiment, the industrial intelligent gateway employs a pairing mechanism between the switching command time and the status return time for capacitor operating status data. The pairing includes candidate switching amounts, actual status returns, power factor changes before and after the return. If the same capacitor operating status does not receive confirmation after the switching command, the edge computing module does not include the corresponding capacity in the confirmed input capacity, but instead includes it in the unconfirmed capacity and sets an unconfirmed flag. When the unconfirmed capacity participates in the calculation of the available switching capacity set, it is subject to gating restrictions. If the subsequent status return is consistent with the switching command, the unconfirmed capacity is converted into a confirmed input capacity. If the subsequent status return is inconsistent with the switching command, the corresponding control cycle is written to a response offset type with inconsistent direction. The confirmation result of the capacitor operating status simultaneously affects the prediction reliability, candidate switching amounts, and delayed confirmation queue. The advantage of this embodiment is that the delayed capacitor status feedback will not be mistakenly equated with a completed action.

[0073] Preferably, the synchronous state sequence, prediction confidence, fuzzy PID control output, gating judgment state, delay confirmation queue, and feedback correction sample are associated and stored in the industrial intelligent gateway according to the same control cycle identifier. Any switching command can be traced back to the corresponding original data channel, quality label, prediction residual, trend direction, candidate switching amount, candidate switching timing, and post-switching state window. When the edge computing module reads the historical associated records in the next control cycle, it only uses the feedback correction result that matches the current quality label type and residual type, and does not mix the correction results under different data quality conditions. If the cloud platform issues a new global reactive power optimization boundary, the local associated records retain the original control cycle identifier and re-perform consistency verification under the new boundary to avoid switching logic breakage during strategy version switching. The advantage of this embodiment is that it keeps the data processing, control output, and feedback correction in a traceable and consistent chain.

[0074] In a preferred embodiment, the edge computing module calculates the slope of the power factor change trend using a predictive sequence difference method. If the predicted sequence changes in the same direction within a continuous control cycle, the trend slope is fed into the fuzzy PID algorithm to participate in the calculation of candidate switching quantities. If the predicted sequence changes in alternating directions, the trend slope is only used as an observation variable in the delayed confirmation queue. The current power factor deviation is still used as the main input of conservative closed-loop control. The prediction residual does not participate in early switching during the disturbance cycle, but participates in local error compensation updates. As a result, the industrial intelligent gateway can distinguish between continuous reactive power gaps and transient offsets when the load fluctuates in a short cycle. The switching action is not directly amplified due to exceeding the limit in a single control cycle. The advantage of this embodiment is that the degree of control participation of the trend input changes with the prediction stability, reducing switching oscillations.

[0075] Specifically, in the calculation of the state window before and after the switching, the edge computing module uses the same control cycle identifier to differentiate the changes in voltage, current, power factor, and capacitor operating status. The direction of power factor change is used to determine whether the switching result is consistent with the predicted direction. The current change is used to help determine whether the load disturbance is dominant. The change in reactive power demand direction is used to determine whether the candidate switching quantity is still necessary. The change in capacitor operating status is used to confirm whether the switching action has been completed. The quality label is used to determine whether the data in the window can be used as a feedback correction sample. If the late sample in the window is dominant or the capacitor status is unconfirmable, the corresponding record is not used to adjust the fuzzy PID input weight, but only to reduce the prediction credibility contribution under the same quality label. If the data quality in the window meets the feedback conditions, the corresponding record enters the error compensation and gating judgment condition update process. The advantage of this embodiment is that it prevents low-quality state windows from causing erroneous corrections to control parameters.

[0076] Furthermore, when the industrial intelligent gateway generates the switching command for the next control cycle locally, it organizes the control logic in the following order: synchronization state sequence, prediction confidence, fuzzy PID candidate calculation, gating decision, queue verification, global reactive power optimization boundary consistency verification, and feedback correction reading. This order does not represent a simple serial function superposition, but rather the output of each link serves as a constraint condition for the subsequent links. The synchronization state sequence constrains the prediction input, the prediction confidence constrains whether the fuzzy PID candidate result is allowed to execute, the queue verification constrains whether the confidence trend continues to exist, the global reactive power optimization boundary constrains the local switching range, and the feedback correction constrains the error compensation and weight value for the next control cycle. If any link has an unconfirmed flag, the control flow will fall back to conservative closed-loop control or delay the confirmation queue. The advantage of this embodiment is that it forms a closed-loop control path oriented towards multi-protocol data mismatch, prediction disturbance, and switching response offset.

Claims

1. A capacitor switching control method based on an industrial intelligent gateway, characterized in that, include: The multi-protocol conversion unit of the industrial smart gateway collects grid voltage, current, power factor and capacitor operating status data, and constructs a synchronous status sequence in the edge computing module based on the sampling time, protocol refresh cycle and status return time. Based on the synchronization state sequence, time-series prediction is performed to obtain the power factor change trend, reactive power demand direction, and prediction reliability. The current power factor deviation, the power factor change trend, the reactive power demand direction, and the status of the capacitor capacity already in operation are input into a fuzzy PID algorithm based on time-series prediction to generate candidate switching quantities and candidate switching times. Based on the prediction confidence level, the candidate switching quantity and candidate switching timing are gating and determined to generate switching instructions that can be executed, delayed confirmation, or conservative closed-loop control. Feedback correction is performed on the timing prediction and fuzzy PID control parameters based on the changes in the power grid state before and after the switching.

2. The capacitor switching control method based on an industrial intelligent gateway according to claim 1, characterized in that, Constructing the synchronization state sequence includes: Extract the sampling time, reception time, protocol type, and capacitor status return time from data frames from different communication protocols; The voltage, current, power factor, and capacitor operating status data are mapped to a unified time axis according to the same control cycle; Data frames that exceed the control period are marked as late samples, and the state continuity of adjacent periods is used to determine whether late samples enter the prediction input. An unconfirmable flag is set for capacitor operating status data returned in missing states, so that the unconfirmable flag and the power grid measurement data can be used together in the subsequent prediction reliability calculation.

3. The capacitor switching control method based on an industrial intelligent gateway according to claim 1, characterized in that, Obtaining the prediction confidence level includes: In the edge computing module, a short-time prediction sequence is generated based on the synchronization state sequence of multiple consecutive control cycles; The short-time prediction sequence is compared with the subsequently acquired measured power factor sequence to form a prediction residual sequence; The prediction reliability is constructed based on the predicted residual sequence, the phase consistency of voltage and current within the same control cycle, the continuity of power factor change direction, and the completeness of capacitor operating status data. When the predicted residual sequence shows a continuous reverse shift, the corresponding prediction period is marked as a perturbation period, and the perturbation period is prevented from directly triggering early switching.

4. The capacitor switching control method based on an industrial intelligent gateway according to claim 3, characterized in that, The fuzzy PID algorithm based on time-series prediction includes: The current power factor deviation, the slope of the power factor change trend, the direction of reactive power demand, the proportion of the capacity of the capacitors already in operation, and the response offset after the most recent switching are used as fuzzy input variables. The adjustment direction of the proportional, integral, and derivative control quantities is determined based on the fuzzy input variables. Match the adjustment direction with the switchable capacity level of the capacitor to obtain the candidate switching amount; The candidate switching timing is obtained by jointly determining the slope of the power factor change trend and the response offset after the most recent switching.

5. The capacitor switching control method based on an industrial intelligent gateway according to claim 1, characterized in that, Gating decisions based on the predicted confidence level include: When the prediction confidence meets the high confidence condition and the reactive power demand direction is consistent within the continuous control cycle, the candidate switching amount is restricted to a small step switching amount that matches the predicted reactive power gap. When the prediction confidence meets the medium confidence condition, the candidate switching amount is written into the delayed confirmation queue, and the power factor change trend is re-compared in the next control cycle to determine whether to execute. When the prediction confidence meets the low confidence condition or there is an unconfirmable capacitor state indicator, the participation of timing prediction in the candidate switching timing is suppressed, and conservative closed-loop control is performed according to the current power factor deviation.

6. The capacitor switching control method based on an industrial intelligent gateway according to claim 5, characterized in that, When reconstructing the state of the synchronization state sequence, refresh cycle profiles are established according to data frames from different sources based on protocol type; Calculate the timing offset of each data frame relative to the current control cycle based on the refresh cycle profile; For data frames whose timing offset is within the compensable range, perform interpolation alignment; for data frames whose timing offset exceeds the compensable range, perform frozen references. The interpolation alignment results, frozen reference results, and unverifiable flags are written together into the quality tag of the synchronization state sequence, so that the quality tag enters the timing prediction process along with the voltage, current, power factor, and capacitor operating state data.

7. The capacitor switching control method based on an industrial intelligent gateway according to claim 4, characterized in that, Before generating candidate cutoff values, the edge computing module performs correlation and discrimination between the predicted residual sequence and the response offset after the most recent cutoff. When the direction of change of the predicted residual sequence is consistent with the direction of response offset, the predicted residual sequence is classified into capacitor switching response deviation; When the direction of change of the predicted residual sequence is inconsistent with the direction of response offset, the predicted residual sequence is classified as an external load disturbance. Based on the classification results, the input weights of the power factor change trend slope and the current power factor deviation in the fuzzy PID algorithm are adjusted respectively, and the adjusted input weights are used to generate candidate switching quantities for the next control cycle.

8. The capacitor switching control method based on an industrial intelligent gateway according to claim 6, characterized in that, The delayed confirmation queue is established according to the candidate switching quantity, candidate switching timing, prediction confidence, and reactive power demand direction. In the next control cycle, the edge computing module will perform a consistency comparison between the reactive power demand direction corresponding to the queue item and the newly generated reactive power demand direction; When the two are consistent and there are no unconfirmable identifiers in the capacitor operating status data, a switching instruction is generated according to the candidate switching quantity in the queue item. When the two are inconsistent or a new unverifiable identifier appears, delete the queue item and re-execute conservative closed-loop control.

9. The capacitor switching control method based on an industrial intelligent gateway according to any one of claims 8, characterized in that, After the switching command is executed, the industrial intelligent gateway captures the status window before and after the switching. Calculate voltage changes, current changes, power factor changes, reactive power demand direction changes, and capacitor operating status changes in two state windows respectively; The changes are associated with candidate deployment quantity, candidate deployment timing, quality label, prediction confidence, and delayed confirmation queue status and recorded accordingly. When the associated record shows that the direction of power factor change in the status window after switching is inconsistent with the predicted direction before switching, the corresponding control cycle is written into the feedback correction sample, and the prediction confidence under the same quality label is reduced in subsequent control cycles.

10. The capacitor switching control method based on an industrial intelligent gateway according to claim 9, characterized in that, The feedback correction samples are grouped in the edge computing module according to quality label type, prediction residual type, response offset type, and switching instruction source; For each group, update the local error compensation amount of the time series prediction, the input weights of the fuzzy PID algorithm, and the gating decision conditions respectively; When the cloud platform issues the global reactive power optimization boundary, the industrial intelligent gateway performs a consistency check between the global reactive power optimization boundary and the updated gating judgment conditions, and generates a local switching instruction for the next control cycle without changing the global reactive power optimization boundary.

Images