An active power distribution network power quality multi-objective comprehensive treatment optimization method and system

Abstract

The application discloses a kind of active power distribution network power quality multi-objective comprehensive treatment optimization method and system, belong to the field of power system automation control technology.This method establishes the comprehensive objective function including voltage deviation measure, three-phase unbalance measure, harmonic distortion measure and network loss measure;In slow time domain optimization layer, the optimal reference value of each treatment equipment and system level index pricing vector are calculated by price-response game iteration, and the reference value is filtered and limited to generate non-intrusive reference;In fast time domain coordination layer, the actual output of equipment is updated according to reference tracking item, constraint potential function gradient item and collaborative cost item, and the conflict between equipment is suppressed;After the end of the cycle, the multi-objective weight and equipment sensitivity matrix are adaptively updated based on statistics;Discrete decision is made for intelligent capacitor and phase changer.The application realizes the collaborative optimization treatment of multi-index, multi-device and multi-time scale.

Description

Technical Field

[0001] This invention relates to the field of power system automation control technology, specifically to a multi-objective comprehensive governance and optimization method and system for power quality in active distribution networks. It is particularly applicable to active distribution networks containing multiple governance devices such as static var generators (SVG), active power filters (APF), smart capacitor banks, and phase-switching switches. It can achieve coordinated optimization and governance of multiple power quality indicators such as voltage deviation, three-phase imbalance, harmonic distortion, and network loss, and meet the dynamic control requirements of power quality in distribution networks under different operating conditions. Background Technology

[0002] With the large-scale integration of distributed power sources (such as photovoltaic and wind power) and the widespread application of power electronic loads (such as charging piles and frequency converters), the operating characteristics of active distribution networks are becoming increasingly complex. Power quality problems are showing new characteristics of "multi-index coupling, multi-device coexistence, and multi-condition fluctuations"—voltage deviations frequently exceed the allowable range due to fluctuations in the output of distributed power sources, three-phase imbalances are exacerbated by random switching of single-phase loads, harmonic distortions are superimposed by the nonlinear characteristics of power electronic devices, and network losses change dynamically with the power flow distribution. Traditional governance models based on a single index and a single device are no longer suitable.

[0003] Currently, power quality management systems for distribution networks mainly employ two control modes: one is a centralized control mode, where the main control center collects operational data such as voltage, current, and harmonics from the entire power grid via a communication network. Based on a single optimization objective (such as maximizing voltage compliance or minimizing reactive power), it performs centralized calculations, generates control commands, and sends them to management devices such as SVG, APF, and capacitor banks. Each device independently executes the main control center's reference values. While this mode can achieve global data aggregation, it is limited by its single-objective optimization logic and cannot simultaneously balance the demands of multiple indicators such as voltage, harmonics, and network losses. For example, reducing reactive power compensation to lower network losses may lead to excessive voltage deviation; adjusting APF output to suppress harmonics may conflict with the reactive power regulation of the SVG.

[0004] Another type is the distributed independent control mode, where each governance device is equipped with a local control unit. It independently judges its operating status and performs compensation actions by monitoring the voltage and current waveforms of its own installation node. There is no information exchange between devices, and the system stability depends only on preset thresholds and local feedback. Although this mode has a fast response speed, it lacks a global coordination mechanism: multiple devices may adjust at the same node simultaneously (such as SVG and capacitor banks injecting reactive power into the same node), causing power oscillations or harmonic superposition; discrete devices such as smart capacitors and phase-changing switches are based on simple, fixed threshold actions that only reflect local information, which are prone to frequent switching (such as repeated switching of capacitor taps when the load fluctuates), which not only increases equipment wear but also causes fluctuations in the operating status of the distribution network.

[0005] Further analysis of the core shortcomings of existing technologies reveals the following: First, there is a lack of multi-indicator coordination. Existing solutions do not establish a unified weighted model for indicators such as voltage, harmonics, and network losses. The optimization process tends to favor a single indicator at the expense of other performance aspects, failing to meet the comprehensive power quality requirements of active distribution networks. Second, there is insufficient coordination between devices. The centralized mode relies on unified calculation by the master station, which is difficult to cope with the decoupling complexity as the number of devices increases. The distributed mode suffers from action conflicts due to the lack of information exchange. Neither mode has formed a cross-device coordination constraint mechanism. Third, there is poor time-domain adaptability. Existing solutions mostly adopt fixed-cycle control. The strategy fails to distinguish between the time characteristics of "slow time-domain strategy decisions (such as weight adjustment and equipment reference value optimization)" and "fast time-domain real-time response (such as dynamic compensation and boundary constraints)". The excessively long calculation cycle in the slow time domain will lead to a lag in dynamic response, while frequent adjustments in the fast time domain will reduce system stability. Fourth, the model has weak adaptability. The optimization algorithm mostly uses fixed parameter matrices (such as the sensitivity matrix of equipment to indicators). When the distribution network topology changes (such as line switching) or the load distribution is adjusted (such as the start and stop of industrial and commercial loads), the deviation between the model and the actual operation increases, the convergence of the optimization results decreases, and even control instability may occur.

[0006] To address the aforementioned issues, the industry urgently needs an active power distribution network power quality governance solution that can balance multi-indicator coordination, multi-device coordination, multi-time domain adaptation, and dynamic model updates. Through a hierarchical control architecture and adaptive optimization mechanism, it can achieve global optimal governance of multiple power quality objectives while ensuring system stability and the rationality of equipment actions. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a multi-objective comprehensive governance and optimization method and system for power quality in active distribution networks. By establishing a slow-time-domain price-response game optimization mechanism and a fast-time-domain constraint-adaptive coordinated control mechanism, combined with multi-index weight self-adjustment and online sensitivity updates, global coordinated governance under multi-objective conditions is achieved. This solves the problems of single objective, decentralized control, and fixed parameters in existing technologies, constructing a multi-timescale, hierarchical collaborative optimization framework that makes the governance process computable, constrainable, and scalable.

[0008] This invention is achieved through the following technical solution: A multi-objective comprehensive management and optimization method for power quality in active power distribution networks, comprising:

[0009] S1: Establish a comprehensive objective function that includes voltage deviation measurement, three-phase imbalance measurement, harmonic distortion measurement and network loss measurement, and set voltage boundary, line current boundary, harmonic limit and equipment capacity boundary to form the operational feasible domain.

[0010] S2: Set up a slow time domain optimization layer and a fast time domain coordination layer, and determine the slow time domain time step and the fast time domain time step;

[0011] S3: Construct a system-level indicator pricing vector. For each governance device, construct a mapping term between the device's individual cost and sensitivity matrix to obtain an optimization model for the device's optimal slow time domain reference. Perform price-response iteration to obtain the device's optimal slow time domain reference and the updated system-level indicator pricing vector.

[0012] S4: Perform first-order filtering and feasible region limiting on the optimal slow time domain reference of the device to generate a non-intrusive reference;

[0013] S5: In the fast time-domain coordination layer, the fast time-domain actual output is updated based on the reference tracking term and the gradient term of the constraint potential function;

[0014] S6: Calculate the periodic statistics to obtain the voltage deviation periodic statistics, three-phase imbalance periodic statistics, harmonic distortion periodic statistics and network loss periodic statistics, and update the multi-objective weights and the sensitivity matrix based on the periodic statistics.

[0015] S7: Perform range enumeration and cost increment calculation on the smart capacitor to select the range, and perform neighborhood enumeration and cost increment calculation on the commutation switch to select the configuration;

[0016] S8: Repeat steps S3 to S7 according to the slow time domain time step and the fast time domain time step.

[0017] The execution price-response iteration in step S3 includes:

[0018] S3-1: Initialize the system-level indicator pricing vector;

[0019] S3-2: Within its own equipment capability boundary, each governance device minimizes the sum of its individual equipment cost and the sensitivity matrix mapping term to obtain the current device's optimal slow time domain reference.

[0020] S3-3: Based on the difference between the periodic statistics and the preset reference threshold, perform a non-negative projection subgradient update on the system-level indicator pricing vector;

[0021] S3-4: Repeat steps S3-2 and S3-3 until the change in the system-level indicator pricing vector and the change in the device's optimal slow time domain reference are both less than their respective set thresholds.

[0022] The generation of the non-invasive reference in step S4 includes:

[0023] A first-order discrete filter is performed on the optimal slow-time domain reference of the device to obtain the filtered reference.

[0024] The filtered reference is projected onto the operational feasible region defined by the device capability boundary and limited to generate the non-intrusive reference;

[0025] The first-order discrete filter is used to suppress step changes in the reference value, and the feasible region limiting is used to ensure that the non-intrusive reference does not exceed the actual operating boundary of the device.

[0026] The terms on which the update of the actual output in the fast time domain is based in step S5 also include a cooperative cost gradient term; the cooperative cost gradient term is obtained by calculating the L2 squared gradient of the difference between the outputs of paired governance devices in the index space, and is used to force the device outputs to adjust in the direction of consistency, so as to suppress the output conflicts of multiple devices on the same index.

[0027] The updating of multi-objective weights in step S6 includes:

[0028] Based on the deviation between the periodic statistics of each indicator and the corresponding reference threshold, the weights of each indicator are non-negatively truncated.

[0029] The non-negative truncated weights are normalized to obtain the updated multi-objective weights.

[0030] When the periodic statistic of a certain indicator deviates from its reference threshold, its weight increases in the next period.

[0031] The sensitivity matrix update in step S6 is performed using a recursive least squares algorithm. The recursive least squares algorithm uses incremental data composed of changes in device reference values ​​and index changes in the current slow time domain period to correct the historical sensitivity matrix estimate, so as to reduce the impact of model mismatch.

[0032] Step S7, which involves performing level enumeration and cost increment calculation on the smart capacitor to select a level, includes:

[0033] Within the set of selectable ranges for the smart capacitor, enumerate each candidate range.

[0034] Calculate the cost increment corresponding to each candidate level, wherein the cost increment is obtained by adding the change in the individual cost of the device to the change in the sensitivity matrix mapping term;

[0035] Select the candidate gear with the smallest cost increment as the gear instruction for the next cycle.

[0036] If the current gear position is held for less than the preset minimum dwell time, the current gear position will remain unchanged.

[0037] Step S7, which involves performing neighborhood enumeration and cost increment calculation on the commutation switch to select a configuration, includes:

[0038] Construct a finite neighborhood centered on the current configuration of the commutator switches, allowing for a maximum number of switches.

[0039] Calculate the cost increment for each candidate configuration within a finite neighborhood, whereby the cost increment is obtained by adding the individual device cost change to the change in the sensitivity matrix mapping term;

[0040] Select the candidate configuration with the smallest cost increment as the switch state for the next cycle;

[0041] If the number of switching actions within a preset time window exceeds the maximum allowed number, the action will be frozen, and the current configuration will be maintained.

[0042] This invention also provides a multi-objective integrated management and optimization system for power quality in active power distribution networks, comprising:

[0043] The main station module is used to execute steps S1, S2, S3, S6 and S7 of the multi-objective comprehensive governance and optimization method for power quality of the active distribution network.

[0044] A fast time-domain coordinator is used to execute steps S4 and S5 of the active distribution network power quality multi-objective integrated governance and optimization method.

[0045] The device-side agent module is connected to the governance device and is used to receive and send non-intrusive references from the fast time-domain coordinator to the static var generator and active power filter, and to receive and send discrete decisions from the master station module to the smart capacitor and commutation switch.

[0046] The data acquisition module is used to collect fast time-domain data and slow time-domain statistics of voltage, current, harmonics and equipment status, and to interact with the master station module and the fast time-domain coordinator.

[0047] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it can control a computing device to implement each step of the deep learning-based main and distribution network planned maintenance optimization method.

[0048] The present invention has the following technical effects:

[0049] The multi-objective comprehensive governance and optimization method and system for power quality in active distribution networks provided by this invention effectively overcomes the core defects of existing technologies and achieves significant technological progress by constructing a hierarchical collaborative and adaptive optimization technical framework. Its technical effects are specifically reflected in the following aspects:

[0050] First, it achieves dynamic and coordinated optimization of multiple power quality indicators, solving the problem of a single objective.

[0051] This invention establishes a unified weighted objective function encompassing voltage deviation, three-phase imbalance, harmonic distortion, and network loss, and designs an adaptive weight update mechanism based on the deviation between periodic statistics and a reference threshold. This overcomes the limitation of traditional schemes that optimize only a single indicator. When an indicator (such as voltage deviation) deviates from a safety threshold, its weight automatically increases, guiding the optimization process to allocate more control resources to governance devices highly coupled with that indicator, thereby achieving dynamic trade-offs and collaborative governance of multiple indicators. This avoids sacrificing the performance of other indicators in pursuit of the optimal performance of a single indicator, ensuring the overall optimal power quality of the distribution network.

[0052] Second, it achieved global coordinated control of multiple governance devices, resolving the issue of device conflicts.

[0053] This invention proposes a "slow-time domain price-response game" optimization mechanism. The master station transmits global marginal cost signals to devices through a system-level indicator pricing vector. Each device independently solves for the optimal reference based on its individual cost and the global pricing. A collaborative cost term based on a sensitivity matrix is ​​introduced in the fast time domain. This mechanism decouples complex centralized computation into distributed optimization, ensuring global optimization guidance while reducing computational complexity. Through pricing signals and collaborative costs, the actions of different devices (such as SVG and smart capacitors) are effectively coordinated, avoiding output superposition, mutual cancellation, or "preemption" phenomena caused by a lack of information interaction, significantly improving the stability of system operation.

[0054] Third, it achieves precise adaptation and control across multiple time scales, solving the problem of poor time-domain adaptability.

[0055] This invention employs a hierarchical architecture with a slow-time-domain optimization layer and a fast-time-domain coordination layer, defining differentiated time steps. The slow-time domain handles strategic decisions (such as reference value optimization and parameter updates), while the fast-time domain handles real-time response and constraint control. This architecture clearly distinguishes between optimization of slowly changing processes and control of rapidly changing processes. The "freeze window" mechanism in the slow-time domain ensures the stability of optimization instructions within a given time period, avoiding system oscillations caused by frequent instruction issuance; the rapid response in the fast-time domain effectively suppresses transient disturbances. This hierarchical temporal isolation enables the system to possess both the globality of optimization and the speed of control.

[0056] Fourth, the system model was made available for online self-updating, thus solving the problem of weak model adaptability.

[0057] This invention employs a recursive least squares algorithm, utilizing the equipment reference changes and index changes generated within each slow time-domain period to update the sensitivity matrix of equipment to system indicators online. When the operating point changes due to topology adjustments or load distribution variations in the distribution network, this mechanism can dynamically correct the linear approximation model, ensuring that the optimized model remains highly consistent with the actual operating state of the power grid. This significantly improves the convergence of the optimization algorithm and the accuracy of the control results, enhancing the system's robustness to changes in network structure and fluctuations in operating conditions.

[0058] Fifth, it achieves full-process safety constraints in the control process, ensuring the reliability of system operation.

[0059] This invention performs feasible region limiting during the generation of a non-intrusive reference; introduces a constraint potential function gradient term into the fast time-domain control law; and sets management parameters such as minimum dwell time and maximum number of actions for discrete devices (smart capacitors, commutation switches). These measures ensure that the output of the control equipment is always limited within the equipment capacity boundary and the grid safety boundary, fundamentally preventing system risks caused by equipment overload or control command overruns. Simultaneously, the action management of discrete devices effectively avoids frequent switching in a short period, reducing equipment wear and improving equipment lifespan and system operational stability. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0061] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0062] Example 1

[0063] Reference Figure 1 This embodiment of a multi-objective comprehensive governance and optimization method for power quality in active distribution networks includes:

[0064] S1: Establish a comprehensive objective function consisting of voltage deviation measurement, three-phase imbalance measurement, harmonic distortion measurement and network loss measurement, set voltage boundary, line current boundary, harmonic limit and equipment capacity boundary to form the feasible operating domain;

[0065] S2: Set up a slow time domain optimization layer and a fast time domain coordination layer, and determine the slow time domain time step and the fast time domain time step;

[0066] S3: Construct a system-level indicator pricing vector. For each governance device, construct a mapping term between the device's individual cost and sensitivity matrix to obtain an optimization model for the device's optimal slow time domain reference. Perform price-response iteration to obtain the device's optimal slow time domain reference and the system-level indicator pricing vector.

[0067] S4: Perform first-order filtering and feasible region limiting on the device's optimal slow time domain reference to generate a non-intrusive reference;

[0068] S5: In the fast time-domain coordination layer, the fast time-domain actual output is updated based on the reference tracking term and the gradient term of the constraint potential function;

[0069] S6: Calculate the periodic statistics to obtain the voltage deviation periodic statistics, three-phase imbalance periodic statistics, harmonic distortion periodic statistics and network loss periodic statistics, update the multi-objective weights and update the sensitivity matrix.

[0070] S7: Perform range enumeration and cost increment calculation for the smart capacitor, and select the range; perform neighborhood enumeration and cost increment calculation for the commutation switch, and select the configuration;

[0071] S8: Repeat steps S3 to S7 according to the slow time domain time step and the fast time domain time step.

[0072] Specifically, in one implementation, this method is applicable to active power distribution network systems containing static var generators, active power filters, smart capacitors, and commutation switches. The system connects the master station module, fast time-domain coordinator, and various governance devices via a communication network to achieve hierarchical calculation and command interaction.

[0073] Specifically, in step S1 of this embodiment, the voltage deviation measure is obtained by summing the squares of the differences between the voltage phasor of each node and each phase and the reference voltage; the three-phase imbalance measure is obtained by summing the squares of the negative sequence current and the zero sequence current of each node; the harmonic distortion measure is obtained by summing the harmonic current of each harmonic order with the squares according to the weighting coefficient; and the network loss measure is obtained by the periodic average value of the active power loss.

[0074] Specifically, firstly, regarding the voltage deviation index, the voltage amplitude is sampled at each phase of each node to obtain the node... The separation voltage phasor and based on the system reference voltage As a benchmark, the cumulative sum of the squares of the differences between the voltage phasor amplitude and the reference voltage amplitude is calculated. Voltage deviation measurement. The definition is as follows:

[0075] ;

[0076] in, Represents a set of nodes. Represents the set of distinct phases. Represents a node Parting The voltage phasor, For nodes Parting The voltage phasor amplitude, This indicates the reference voltage amplitude.

[0077] Secondly, regarding the three-phase imbalance index, the negative sequence current is obtained by performing symmetrical component decomposition on the node current signal. With zero-sequence current Three-phase imbalance measurement Defined as:

[0078] ;

[0079] in, For nodes negative sequence current, For nodes Zero-sequence current. Three-phase imbalance measurement. It reflects the total energy of the negative sequence and zero sequence components in the node current.

[0080] For harmonic distortion measurement, each harmonic component is extracted from the node current signal, and a set of harmonic orders is defined. Define a node for the highest harmonic order. The The amplitude of the subharmonic current is And set a weighting coefficient for each harmonic order. Harmonic distortion vector The definition is as follows:

[0081] ;

[0082] in, For the set of harmonic orders, Represents a node The Second harmonic current For the first Weighting coefficients for subharmonics This reflects the relative level of attention given to different harmonic orders.

[0083] For network loss measurement, the active power loss of each line segment is calculated based on the line current, and the line set is defined as follows: ,line The resistance is The line current is Network loss measurement The definition is as follows:

[0084] ;

[0085] in, Represents a set of routes. For the line The resistance, This represents the line current. Network loss metric indicates the total active power loss across all lines within a given period.

[0086] After measuring each individual indicator, a comprehensive objective function is constructed. Defined as:

[0087] ;

[0088] in, These are the weights of the four indicators. The weights satisfy:

[0089] ;

[0090] The weight of indicator X, These are indices for voltage deviation measurement, three-phase imbalance measurement, harmonic distortion vector, and network loss measurement, respectively.

[0091] In step S1, the feasible region is defined: four types of boundary conditions are set to ensure that the optimization process does not exceed the safe operating range of the distribution network.

[0092] Voltage boundary: Based on power grid operation standards, set the allowable voltage fluctuation range for each node (e.g., ±5% of the rated voltage).

[0093] Line current boundary: Set the upper limit of the safe current carrying capacity of the line according to the line material and cross-section to avoid overcurrent and heat generation;

[0094] Harmonic limits: Based on power quality standards, allowable current content for each harmonic is set (e.g., the third harmonic should not exceed 5% of the rated current).

[0095] Equipment capacity boundaries: Based on the parameters of the treatment equipment, set the maximum compensation capacity of SVG / APF, the range of intelligent capacitor levels, and the operating limits of the commutation switch, etc.

[0096] The aforementioned boundaries together constitute the feasible region for the operation of the distribution network, constraining the range of values ​​for subsequent optimization decisions.

[0097] In step S2, a slow time-domain optimization layer and a fast time-domain coordination layer are set up, and an inter-layer interaction mechanism is defined. The slow time-domain optimization layer is deployed in the main station module and is responsible for global optimization calculations (such as solving equipment reference values ​​and updating pricing vectors) and dynamic parameter adjustments (such as weight updates and sensitivity matrix updates), adapting to the slow-changing characteristics of the distribution network operation (such as load peak / valley switching and smooth fluctuations in distributed power output). The fast time-domain coordination layer is deployed in the fast time-domain coordinator and is responsible for real-time response and local adjustments (such as dynamic compensation output and boundary constraint control), adapting to the rapid response requirements of instantaneous disturbances (such as load mutations and harmonic impacts). The slow time-domain time step is defined as... The fast time step is ,satisfy Slow time-domain time steps are used for optimization calculations and parameter updates, while fast time-domain time steps are used for real-time execution and local adjustments.

[0098] Furthermore, step S3 includes:

[0099] S3-1, Set the initial value of the system-level indicator pricing vector in the main station;

[0100] S3-2, for each governance device within its capacity boundary, minimize the sum of the individual device cost and the sensitivity matrix mapping terms to obtain the optimal slow time domain reference for the device;

[0101] S3-3, the main station performs a non-negative projection subgradient update on the system-level indicator pricing vector based on the difference between the periodic statistics and the reference threshold;

[0102] S3-4, Repeat steps S3-2 and S3-3 until the pricing vector change threshold and the equipment reference value change threshold are met.

[0103] The acquisition and online updating of the sensitivity matrix includes: applying a small perturbation to the device's slow-time domain reference at the current operating point, collecting changes in multiple indicators, and using the least squares method to solve for the linear mapping coefficients to form the sensitivity matrix. Sensitivity Matrix Obtained from a linear approximation model of the current operating point. Assume the device reference value from the previous cycle is... The equipment reference value for this period is The corresponding change in the indicator vector is The sensitivity matrix is ​​updated using a recursive least squares method:

[0104] ;

[0105] in, Let be the sensitivity matrix of device d after the k-th slow time-domain period update. Let be the sensitivity matrix of device d after the (k-1)th slow time-domain period update. Let be the change in the reference value of device d during the k-th slow time-domain period. Let be the periodic statistics vector of the four power quality indicators within the k-th slow time domain period. for transpose, ,matrix Let be the gain matrix, satisfying:

[0106] ;

[0107] ;

[0108] in, Let be the covariance matrix of device d after the k-th slow time-domain period update. Let be the covariance matrix of device d after the (k-1)th slow time-domain period update, 𝜆 𝑔 This is the forgetting factor, with a value range of (0,1).

[0109] Specifically, in step S3, the price-response iteration (slow-time price-response game optimization) takes as input multiple index statistics, sensitivity matrix, and equipment capability boundaries within the period; the output is the optimal slow-time reference for each equipment and the system-level index pricing vector. The entire optimization process includes four main steps: game initialization, individual equipment solution, pricing vector update, and convergence determination.

[0110] The first step is to initialize the system-level indicator pricing vector in the main site module. The pricing vector is defined as follows:

[0111] ;

[0112] in, For pricing vector, These represent the pricing parameters corresponding to voltage deviation, three-phase imbalance, harmonic distortion, and network loss, respectively. The initial pricing vector can be set to the zero vector or the termination value of the previous cycle.

[0113] The second step involves each governance device, upon receiving the current pricing vector, calculating its optimal reference value based on its individual cost function.

[0114] Define the device set as Any device Individual cost function It consists of the device's operating losses, control costs, or action penalties. This is combined with the individual cost function of the equipment. and sensitivity matrix Construct the individual objective function for each device:

[0115] ;

[0116] in, For the individual objective function value of the device, The reference value for the equipment is a decision variable for the equipment, and its set of values ​​is as follows: equipment The sensitivity matrix for multiple system indicators is denoted as follows: , dimension ,in For the number of indicators For equipment The dimension of the decision variables. This represents a linear approximation of the impact of device output on multiple indicators. This represents the cost additive of system pricing on device behavior. Solving the optimization problem that minimizes the individual objective function value of each device yields the device's optimal slow-time domain reference.

[0117] ;

[0118] in For the device's optimal slow time domain reference;

[0119] Each device independently solves the above optimization problem, and the calculated optimal slow time domain reference for the device is returned to the main station module.

[0120] Third, the main station module updates the pricing vector based on the system's operational results. Let the indicator vector composed of periodic statistics be:

[0121] ;

[0122] in, This is a periodic statistic from the previous freeze window. These represent the average values ​​of voltage deviation, three-phase imbalance, harmonic distortion, and network loss, respectively. The superscript T indicates matrix transpose.

[0123] The corresponding reference threshold vector is:

[0124] ;

[0125] in, As the reference threshold vector, These are the average values ​​of reference thresholds for voltage deviation, three-phase imbalance, harmonic distortion, and network loss, respectively.

[0126] The main station module updates the pricing vector using the projective subgradient method:

[0127] ;

[0128] in, For the first The pricing vector for the next iteration. For the first The pricing vector for the next iteration. It is a non-negative projection operator. For the first Periodic statistics for each iteration. For the first The step size of each iteration. The projection operation ensures that the pricing vector components remain non-negative.

[0129] Fourth step, perform convergence determination: Repeat steps two and three until the change in the pricing vector between two consecutive iterations satisfies:

[0130] ;

[0131] And the changes in the reference values ​​of each device satisfy the following:

[0132] ;

[0133] in, The threshold for the change in the pricing vector. For the first The device reference value for the next iteration. For the first +1 iteration of device reference value, If the threshold value for the change in the device reference value is reached, the iteration terminates, and the final pricing vector and the slow-time domain optimal reference for each device are output.

[0134] Furthermore, in step S4, the non-intrusive reference generation includes: performing a first-order discrete filter on the device's optimal slow time-domain reference to obtain a filtered reference; and performing a limited projection of the filtered reference onto the device's capability boundary to obtain a non-intrusive reference.

[0135] First, within each fast time step, the device's optimal slow time reference is determined. First-order filtering is performed to smooth abrupt changes in the reference value and avoid transient impacts on the equipment. The formula is as follows:

[0136] ;

[0137] in, For equipment At any moment The filter reference, For equipment At any moment The filter reference, The following coefficient controls the filtering speed.

[0138] The filtered reference undergoes feasible region limiting calculations to ensure that the device reference value does not exceed the actual operating range of the device, thus preventing overload or malfunction. The operating boundary is The feasible region limiting operation is defined as follows:

[0139] ;

[0140] in, This serves as the lower bound for equipment decision variables (such as the minimum capacitive compensation amount of SVG). This represents the upper limit of the equipment decision variables (such as the maximum inductive compensation amount of SVG). This is a non-intrusive reference. Non-intrusive references do not alter the internal control logic of the device; they only ensure operational safety through external limiting and can be directly input into fast time-domain control loops. The resulting... Fast time-domain control law as a non-intrusive reference input.

[0141] In step S5, within the fast time-domain coordination layer, the actual output in the fast time domain is updated based on the reference tracking term and the gradient term of the constraint potential function. Non-intrusive reference generation, constraint adaptive control law, and cooperative cost step-up update together constitute the fast time-domain coordination control mechanism. This mechanism has a clear input-output relationship, a time scheduling structure, and an implementable mathematical description, enabling the formation of an independent fast time-domain closed-loop control between the master station and the device controller.

[0142] In the fast time-domain control stage, the actual output of the equipment Update the adaptive control law according to the following constraints:

[0143] ;

[0144] in, The actual fast time-domain output of device d at time t. The actual fast time-domain output of device d at time t+1. For reference tracking gain, To constrain control gain, For reference tracking deviation, Let be the constraint potential function. , This represents the gradient of the potential function. The constraint potential function increases as the device output approaches its boundary, and its gradient term generates a counter-regulatory adjustment to prevent the device output from exceeding the boundary. This control law does not change the internal control structure of the device; it achieves non-intrusive constraint maintenance only by superimposing constraint terms.

[0145] Furthermore, in step S5, to avoid mutual cancellation or superposition of output interference from different governance devices within the same index space, the fast time-domain coordination layer introduces a cooperative cost term for inter-device coordination. The linear mapping matrix from device to index is defined as follows: It consists of the sensitivity of each device's output to a multi-index space. (Cooperation cost) Defined as:

[0146] ';

[0147] in, For a set of paired devices, respectively equipment and equipment The fast time-domain output. This coordination cost term measures the consistency of the outputs of different devices in the metric space. In each fast time-domain period, the fast time-domain coordination layer calculates the coordination cost gradient:

[0148] ;

[0149] in, For the collaborative cost function The actual fast time-domain output of device d gradient vector, for transpose,

[0150] and with a fixed step size Perform a gradient step update on the device's actual fast time-domain output:

[0151] ;

[0152] This update is executed in conjunction with constraint-adaptive control to form multi-device collaborative control.

[0153] Within each fast time domain cycle, the fast time domain coordination layer receives real-time data from the devices via the communication interface, updates the status variables of each device, and synchronously sends new control references. The devices receive non-intrusive reference values ​​through the reference interface and upload output measurement values ​​through the status interface. All communication is performed within fixed time intervals to ensure timing consistency between the slow and fast time domains.

[0154] The update of the fast time domain actual output includes: at each fast time domain time, first calculate the reference tracking deviation, then calculate the gradient of the constraint potential function, and perform an iterative update of the fast time domain actual output according to the given gain to ensure that the output tracks the reference and does not go out of bounds.

[0155] Collaborative cost calculation and gradient update refinement: A linear mapping matrix from device to index is set for paired governance devices, a linear combination of paired governance devices is calculated, the L2 norm squared weighted sum of the linear combination is calculated as the collaborative cost, and finally a step update is performed on the collaborative cost gradient to achieve conflict suppression between devices.

[0156] The fast time-domain coordinated control and non-intrusive reference generation method of this invention is used to perform rapid dynamic adjustments on the control equipment based on the slow time-domain optimization results, so as to maintain the operating state within the equipment capability boundary and realize coordinated control among multiple devices. This method is executed by a fast time-domain coordination layer, with inputs including the slow time-domain optimal reference value, sensitivity matrix, and equipment constraint parameters, and outputs including fast time-domain control commands and the actual fast time-domain output of the equipment.

[0157] In step S6, the slow time-domain optimization layer sends freeze window parameters and a non-intrusive reference update cycle to the fast time-domain coordination layer every cycle. During the freeze window, the fast time-domain coordination layer keeps the slow time-domain reference unchanged and only performs the fast time-domain update step. At the end of the freeze window, the system records the fast time-domain average output value and metric statistics.

[0158] At the start of each slow time domain period, the master station module sends the freeze window length T to the fast time domain coordination layer. freeze The length of the frozen window is defined as:

[0159] ;

[0160] in, To increase the number of time-domain iterations, This is the fast time-domain time step. During the freeze window, the fast time-domain coordination layer keeps the slow time-domain reference value constant, performing only constraint adaptive control and cooperative cost updates. At the end of the freeze window, the fast time-domain coordination layer summarizes the voltage, current, harmonics, and device state variables within that window, and calculates the periodic statistics:

[0161] ;

[0162] ;

[0163] Obtain the voltage deviation period statistics Three-phase imbalance periodic statistics Harmonic distortion period statistics and network loss period statistics The calculation results are sent to the slow time-domain optimization layer via the communication link for use in the next cycle of multi-objective weight updates and game optimization.

[0164] After the freeze window ends, the main station module triggers a slow time-domain optimization process, updating the pricing vector, weight vector, and device reference values. The operational states of all discrete devices are written to the device-side storage module at the end of the cycle, serving as the initial state for the next cycle. Simultaneously, the fast time-domain coordination layer resets the non-intrusive reference filter state r. d (t) to avoid reference drift across cycles.

[0165] In step S6, updating the multi-objective weights refers to dynamically adjusting the weight factors of the four indicators based on the deviation between the periodic statistics and the indicator reference thresholds, ensuring that the optimization objective is tilted towards the indicators with larger deviations.

[0166] The periodic statistics of each indicator are compared with the reference threshold to form a deviation. The weight of each indicator is updated by non-negative truncation and normalization during operation based on the indicator deviation to achieve dynamic adjustment.

[0167] In one implementation, the weights of each indicator are updated adaptively over a periodic period. Let the period be... The periodic statistics of the four indicators (voltage deviation measure, three-phase imbalance measure, harmonic distortion measure, and network loss measure) at any given time are: The corresponding reference thresholds are respectively Define the deviation:

[0168] ;

[0169] in, The deviation of index X is the amount of deviation, where X represents any one of the following: voltage deviation measure, three-phase imbalance measure, harmonic distortion measure, and network loss measure. Let X be the deviation of the index X at time k. Let X be the reference threshold for the index X at time k.

[0170] For each metric, perform a temporary weight update:

[0171] ;

[0172] in, Let X be the weight of the index at time k. For the temporary weight of the updated indicator X, The step size coefficient controls the magnitude of a single weight adjustment; max(0,⋅) is the non-negative truncation operator.

[0173] Next, all temporary weights are normalized to obtain the updated weight vector:

[0174] ;

[0175] Updated weight vector The calculation of the comprehensive objective function for the next cycle begins.

[0176] To ensure the stability of weight adjustments, a maximum step size constraint can be set:

[0177] ;

[0178] in, This is the upper limit constant for the step size.

[0179] At the end of each slow time-domain period, the sensitivity matrix is ​​updated once using the recursive least squares method. The sensitivity matrix is ​​automatically updated based on the measurement data within each slow time-domain period to maintain its linear validity near the operating point.

[0180] The slow time-domain optimization layer sets a freeze window within each slow time-domain cycle, while the fast time-domain coordination layer maintains a non-intrusive reference and performs fast time-domain updates within the freeze window. At the end of the freeze window, the voltage deviation cycle statistics, three-phase imbalance cycle statistics, harmonic distortion cycle statistics, and network loss cycle statistics are calculated, and multi-objective weight updates and sensitivity matrix updates are performed online based on these statistics.

[0181] In step S7, for the smart capacitor, the cost increment of the gear is calculated within the set of selectable gears, and the gear with the smallest cost increment is selected; for the commutation switch, the cost increment of the candidate configuration is calculated within a finite neighborhood and the candidate configuration with the smallest cost increment is selected; the cost increment of the gear and the cost increment of the configuration are both obtained by adding the individual cost change of the device and the change of the sensitivity matrix mapping term.

[0182] Step S7-1: Perform discrete decision-making for the range of the smart capacitor. Let the first... Within a slow time-domain period, the current setting of the smart capacitor is Its selectable gear set is N is the total number of gears, and for each candidate gear... Calculate candidate gears Incremental cost:

[0183] ;

[0184] in, Represents the individual cost function of the smart capacitor For system-level indicator pricing vectors, This is the sensitivity matrix corresponding to the smart capacitor. Individual cost function Transpose of;

[0185] Select the gear with the smallest incremental cost from the smart capacitor gears as the gear command for the next cycle:

[0186] ;

[0187] in The next gear will be selected if the current gear holding time does not reach the minimum dwell time. If so, the current gear will remain unchanged.

[0188] Step S7-2 involves performing a neighborhood search and discrete decision on the topological configuration of the commutator switches. The commutator switch configuration vector at time k is defined as follows: The value space is a discrete set. In the Within the period, construct a finite neighborhood centered on the current configuration:

[0189] ;

[0190] in, Configure vectors for commutation switches A finite neighborhood, Configure the maximum number of switches that can be switched simultaneously. Configure each candidate commutator switch. Calculate the cost increment:

[0191] ;

[0192] in, Configure candidate commutator switches The cost increment, Individual cost function for commutation switches This is the corresponding sensitivity matrix. The candidate commutator switch configuration with the smallest cost increment is selected as the switching state for the next cycle.

[0193] ;

[0194] in, For the first Cyclic commutation switch configuration vector.

[0195] If the number of switching attempts exceeds the maximum allowed number N within the preset time window. max,s If so, the action is frozen, keeping the current commutator configuration unchanged.

[0196] In step S8, an inter-layer loop is executed. The slow time domain and the fast time domain alternate. During the slow time domain cycle, the reference is frozen, while the fast time domain continuously performs control updates. After the cycle ends, the slow time domain recalculates the optimization variables and updates the parameters.

[0197] In this embodiment, the combined mechanism of discrete decision-making and frozen window ensures that the actions of discrete devices and the control of continuous devices remain synchronized in time, and that the states of each device transition smoothly during the cycle. The discrete decision-making and frozen window control method is used to perform decision calculations for governance devices with discrete states in the slow time domain cycle, and to perform time management of the reference signals and data sampling of the fast time domain coordination layer.

[0198] Example 2

[0199] Furthermore, an active power distribution network power quality multi-objective integrated management and optimization system includes:

[0200] The system consists of a main station module, a fast time-domain coordinator, a device-side agent module, and a data acquisition module.

[0201] The main station module stores program instructions and runs them on the processor, executing steps S1, S2, S3, S6 and S7 of Embodiment 1.

[0202] The fast time-domain coordinator stores program instructions and runs them on the processor, executing steps S4 and S5 of Example 1;

[0203] The equipment-side agent module establishes reference and status interfaces with the governance equipment, receives non-intrusive references and sends them to the static var generator and active power filter, and receives discrete decisions and sends them to the smart capacitor and commutation switch.

[0204] The data acquisition module collects fast time-domain data and slow time-domain statistics of voltage, current, harmonics and equipment status, and interacts with the master station module and the fast time-domain coordinator.

[0205] The active power distribution network power quality multi-objective integrated management and optimization system provided by this invention includes a master station module, a fast time-domain coordinator, an equipment-side agent module, and a data acquisition module. These modules are interconnected via communication links to form a centralized-hierarchical collaborative control system.

[0206] The master station module is the central computing unit of the system, comprising a processor, memory, and communication interfaces. The processor performs multi-indicator calculations, price-response game optimization, multi-objective weight updates, and sensitivity matrix recursive updates. The memory stores operating parameters, device information tables, and optimization programs. The master station module's communication interface is used for data exchange with the fast time-domain coordinator and device-side agent modules.

[0207] The fast time-domain coordinator includes a local control processor and a high-speed communication interface. The control processor performs non-intrusive reference filtering, amplitude limiting, constrained adaptive control laws, and cooperative cost calculations. The high-speed communication interface maintains data synchronization with the master module, receives the slow time-domain optimal reference and frozen window parameters, and sends fast time-domain control commands to the device-side agent module within each fast time-domain time step.

[0208] The equipment-side agent module is deployed at the site of each treatment device and connects to the static var generator, active power filter, smart capacitor, and commutator. This module includes a reference interface, a status interface, and a data buffer unit. The reference interface receives non-intrusive reference values ​​from the fast time-domain coordinator; the status interface collects the device's output voltage, current, and status parameters and uploads them via the communication link. The data buffer unit temporarily stores fast time-domain sampled data during the communication cycle for periodic statistical use by the upper-layer module.

[0209] The data acquisition module includes a voltage sampling device, a current sampling device, a harmonic analysis unit, and a synchronization clock. The voltage and current sampling devices are connected to the distribution network nodes and lines to acquire three-phase voltage and current signals in real time. The harmonic analysis unit extracts each harmonic component and generates harmonic current data using a Fast Fourier Transform algorithm. The synchronization clock provides a unified time reference for all sampling channels, ensuring data alignment.

[0210] The system's communication architecture employs a bidirectional link. The master station module is bidirectionally connected to the fast time-domain coordinator via a communication bus or fiber optic channel; the fast time-domain coordinator communicates with multiple device-side agent modules via Ethernet or fieldbus. Each device-side agent module connects to its corresponding governance device using a standardized control interface. The communication link supports data exchange at fixed time intervals, ensuring timing consistency between slow and fast time-domain control tasks.

[0211] During operation, the data acquisition module continuously collects node voltage, current, and harmonic data and transmits them to the master station module. The master station module performs multi-index metric calculations and game-theoretic optimization in each slow time-domain cycle to generate the optimal reference for the equipment. The fast time-domain coordinator receives this reference within a frozen window and performs non-intrusive reference filtering, limiting, and constraint adaptive control according to the fast time-domain time steps. The equipment-side agent module drives the governance equipment to perform corresponding actions according to instructions and uploads the operating status in real time. At the end of the cycle, the master station module updates the multi-objective weight and sensitivity matrix based on statistical results, forming a new optimization cycle.

[0212] The hardware components of this system can be implemented on a centralized server platform or deployed in a network among distributed control nodes. The master station module and the fast time-domain coordinator can be integrated into the same control server or set up independently on different computing nodes. The data acquisition module and the device-side agent module can be integrated through an edge control unit.

[0213] Example 3

[0214] In the third embodiment of the present invention, based on the same inventive concept, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the deep learning-based main distribution network planned maintenance optimization method of the above embodiments.

[0215] It should be understood that the deep learning-based main distribution network planned maintenance optimization method of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system.

[0216] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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.

Claims

1. A multi-objective comprehensive management and optimization method for power quality in active power distribution networks, characterized in that, include: S1: Establish a comprehensive objective function that includes voltage deviation measurement, three-phase imbalance measurement, harmonic distortion measurement and network loss measurement, and set voltage boundary, line current boundary, harmonic limit and equipment capacity boundary to form the operational feasible domain. S2: Set up a slow time domain optimization layer and a fast time domain coordination layer, and determine the slow time domain time step and the fast time domain time step; S3: Construct a system-level indicator pricing vector. For each governance device, construct a mapping term between the device's individual cost and sensitivity matrix to obtain an optimization model for the device's optimal slow time domain reference. Perform price-response iteration to obtain the device's optimal slow time domain reference and the updated system-level indicator pricing vector. S4: Perform first-order filtering and feasible region limiting on the optimal slow time domain reference of the device to generate a non-intrusive reference; S5: In the fast time-domain coordination layer, the fast time-domain actual output is updated based on the reference tracking term and the gradient term of the constraint potential function; S6: Calculate the periodic statistics to obtain the voltage deviation periodic statistics, three-phase imbalance periodic statistics, harmonic distortion periodic statistics and network loss periodic statistics, and update the multi-objective weights and the sensitivity matrix based on the periodic statistics. S7: Perform range enumeration and cost increment calculation on the smart capacitor to select the range, and perform neighborhood enumeration and cost increment calculation on the commutation switch to select the configuration; S8: Repeat steps S3 to S7 according to the slow time domain time step and the fast time domain time step.

2. The method according to claim 1, characterized in that, The execution price-response iteration in step S3 includes: S3-1: Initialize the system-level indicator pricing vector; S3-2: Within its own equipment capability boundary, each governance device minimizes the sum of its individual equipment cost and the sensitivity matrix mapping term to obtain the current device's optimal slow time domain reference. S3-3: Based on the difference between the periodic statistics and the preset reference threshold, perform a non-negative projection subgradient update on the system-level indicator pricing vector; S3-4: Repeat steps S3-2 and S3-3 until the change in the system-level indicator pricing vector and the change in the device's optimal slow time domain reference are both less than their respective set thresholds.

3. The method according to claim 1, characterized in that, The generation of the non-invasive reference in step S4 includes: A first-order discrete filter is performed on the optimal slow-time domain reference of the device to obtain the filtered reference. The filtered reference is projected onto the operational feasible region defined by the device capability boundary and limited to generate the non-intrusive reference; The first-order discrete filter is used to suppress step changes in the reference value, and the feasible region limiting is used to ensure that the non-intrusive reference does not exceed the actual operating boundary of the device.

4. The method according to claim 1, characterized in that, The terms on which the update of the actual output in the fast time domain is based in step S5 also include a cooperative cost gradient term; the cooperative cost gradient term is obtained by calculating the L2 squared gradient of the difference between the outputs of paired governance devices in the index space, and is used to force the device outputs to adjust in the direction of consistency, so as to suppress the output conflicts of multiple devices on the same index.

5. The method according to claim 1, characterized in that, The updating of multi-objective weights in step S6 includes: Based on the deviation between the periodic statistics of each indicator and the corresponding reference threshold, the weights of each indicator are non-negatively truncated. The non-negative truncated weights are normalized to obtain the updated multi-objective weights. When the periodic statistic of a certain indicator deviates from its reference threshold, its weight increases in the next period.

6. The method according to claim 1, characterized in that, The sensitivity matrix update in step S6 is performed using a recursive least squares algorithm. The recursive least squares algorithm uses incremental data composed of changes in device reference values ​​and index changes in the current slow time domain period to correct the historical sensitivity matrix estimate, so as to reduce the impact of model mismatch.

7. The method according to claim 1, characterized in that, Step S7, which involves performing level enumeration and cost increment calculation on the smart capacitor to select a level, includes: Within the set of selectable ranges for the smart capacitor, enumerate each candidate range. Calculate the cost increment corresponding to each candidate level, wherein the cost increment is obtained by adding the change in the individual cost of the device to the change in the sensitivity matrix mapping term; Select the candidate gear with the smallest cost increment as the gear instruction for the next cycle. If the current gear position is held for less than the preset minimum dwell time, the current gear position will remain unchanged.

8. The method according to claim 1, characterized in that, Step S7, which involves performing neighborhood enumeration and cost increment calculation on the commutation switch to select a configuration, includes: Construct a finite neighborhood centered on the current configuration of the commutator switches, allowing for a maximum number of switches. Calculate the cost increment for each candidate configuration within a finite neighborhood, whereby the cost increment is obtained by adding the individual device cost change to the change in the sensitivity matrix mapping term; Select the candidate configuration with the smallest cost increment as the switch state for the next cycle; If the number of switching actions within a preset time window exceeds the maximum allowed number, the action will be frozen, and the current configuration will be maintained.

9. A multi-objective integrated management and optimization system for power quality in an active power distribution network, characterized in that, include: The main station module is used to execute steps S1, S2, S3, S6 and S7 of the method as described in any one of claims 1 to 8; A fast time-domain coordinator for performing steps S4 and S5 of the method as described in any one of claims 1 to 8; The device-side agent module is connected to the governance device and is used to receive and send non-intrusive references from the fast time-domain coordinator to the static var generator and active power filter, and to receive and send discrete decisions from the master station module to the smart capacitor and commutation switch. The data acquisition module is used to collect fast time-domain data and slow time-domain statistics of voltage, current, harmonics and equipment status, and to interact with the master station module and the fast time-domain coordinator.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it can control the computing device to implement each step of the multi-objective integrated governance and optimization method for power quality of active distribution networks as described in any one of claims 1 to 8.

Images