Self-adaptive multi-level station domain reactive power optimization control system and method based on intelligent network

Abstract

The invention discloses a self-adaptive multi-level station domain reactive power optimization control system and method based on an intelligent network, and the system comprises an AVC / VGC control system, a coordination controller device, and a local controller device, and the AVC / VGC control system serves as a core management and control center of the voltage and reactive power of a transformer substation. Data message interaction with the coordination controller device is carried out by adopting a publishing / subscribing mode, so that the real-time exchange of a control instruction and state information is ensured, and the voltage level and reactive power flow distribution of the whole station are monitored and adjusted in real time; the coordination controller device and the local controller device adopt a 4G / 5G wireless communication network as a transmission channel, and are used for bearing centralized scheduling responsibility of off-station reactive power resources and realizing optimal distribution and cooperative control of multi-station reactive power through real-time communication with a superior power grid scheduling system; and the local controller device is used for providing accurate reactive power local compensation and adjustment functions for local equipment of a photovoltaic power generation area, a power distribution area and an energy storage system.

Description

Technical Field

[0001] This invention relates to the field of power system automation operation and maintenance technology, and in particular to a smart network-based adaptive multi-level station reactive power optimization control system and method. Background Technology

[0002] With the accelerated construction of new power systems based on new energy sources, power supply systems with a high proportion of distributed photovoltaic (PV) power are gradually becoming the core of future conventional active power distribution. Against this backdrop, reactive power, voltage, and line loss levels, as key indicators for measuring the grid operation efficiency of new power systems, are becoming increasingly important. However, with the high proportion of centralized and distributed PV grid connections and the widespread application of power electronic devices such as converters, the potential reactive power output capacity of these devices provides abundant resources for multi-level joint compensation in substation reactive power compensation. However, current control strategies often neglect the reactive power support capacity of PV power plants and low-voltage distribution areas, leading to drawbacks such as slow speed, limited capacity, and poor economic efficiency during the compensation process. Therefore, adopting a multi-level, refined reactive power joint compensation technology targeting three levels—inside the substation, outside the PV power plant, and the distribution area—can deeply tap the reactive power support potential of PV power plants, significantly improving compensation efficiency and capacity, and demonstrating outstanding economic benefits.

[0003] Reactive power optimization, a key aspect of power system operation and management, aims to improve the system's power factor and effectively reduce reactive power losses during transmission through a series of scientific and effective technical measures, thereby enhancing the overall operating efficiency and stability of the power system. The core objective of this process is to optimize voltage quality, reduce power losses, and ultimately achieve a comprehensive improvement in system operating efficiency. Reactive power optimization strategies are diverse, mainly including the following: First, adjusting transformer taps to precisely control the transformer's output voltage and achieve reasonable optimization of reactive power distribution; second, switching capacitors on and off flexibly according to system needs to dynamically adjust the reactive power balance; third, using static var compensators (SVCs), which have rapid response capabilities and can precisely control the magnitude and phase of the current by adjusting the thyristor firing angle, thereby achieving dynamic compensation for the reactive power required by the system; and fourth, applying dynamic var compensators (SVGs), an advanced reactive power compensation technology that can automatically track changes in system reactive power and provide accurate reactive power compensation in real time, ensuring stable system operation. Currently, reactive power compensation schemes generally follow the principle of local compensation, combining centralized and decentralized compensation to reduce the flow of reactive power in the power grid and thus lower transmission losses. However, at the substation level, reactive power compensation strategies still primarily rely on tiered compensation using on-site reactive power devices, achieving dynamic adjustment of reactive power through switching capacitor banks and static var generators. It is worth noting that this compensation method lacks systematic reactive power planning, relying mainly on the experience and judgment of planners. This makes it difficult to meet the refined needs of actual power grid operation, resulting in unsatisfactory reactive power compensation effects and failing to fully utilize the reactive power compensation resources of photovoltaic power plants, energy storage systems, and substation areas.

[0004] The multi-level reactive power compensation strategy for substations, while inheriting and optimizing traditional reactive power compensation methods, innovatively plans and coordinates the output of reactive power compensation resources at various levels from multiple dimensions, including substation-level, external photovoltaic power stations, and transformer substations. This strategy achieves efficient and orderly reactive power compensation across the entire substation area by meticulously optimizing the selection and configuration of reactive power equipment within the substation, while simultaneously considering the collaborative compensation mechanisms of internal and external photovoltaic power stations and transformer substation reactive power compensation equipment. This process not only effectively stabilizes voltage levels but also significantly improves power quality and reduces energy losses during grid transmission. Furthermore, this strategy fully utilizes the reactive power regulation potential of new energy power plants, constructing a multi-level joint compensation system through the scientific and rational configuration of the capacity of reactive power compensation devices at all levels within the substation area to effectively address voltage fluctuations caused by grid load changes. These measures not only enhance the flexibility and stability of the power grid but also significantly improve power quality, providing strong support for the safe and economical operation of the power system. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides an intelligent network-based adaptive multi-level station-domain reactive power optimization control system and method, which aims to achieve multi-level refined reactive power compensation under grid voltage fluctuations and photovoltaic output changes.

[0006] To achieve the aforementioned objectives of the invention, the technical solution adopted to solve its technical problems is as follows: This invention discloses an adaptive multi-level station-domain reactive power optimization control system based on intelligent networks, comprising an AVC / VGC control system, a coordination controller device, and a local controller device, wherein: The AVC / VGC control system, as the core control center for substation voltage and reactive power, interacts with the coordinating controller device via a publish / subscribe mode for data message exchange. The communication medium is an RJ45 Ethernet interface, which is used to ensure the real-time exchange of control commands and status information and to monitor and adjust the voltage level and reactive power distribution of the entire station in real time. The coordination controller device and the local controller device use a 4G / 5G wireless communication network as the transmission channel to undertake the centralized scheduling responsibility of reactive power resources outside the station and realize the optimized allocation and coordinated control of reactive power of multiple stations through real-time communication with the upper power grid dispatching system. The local controller device is used to provide precise local reactive power compensation and regulation functions for local equipment in photovoltaic power generation areas, distribution areas, and energy storage systems.

[0007] This invention also discloses a reactive power optimization control method based on intelligent network adaptive multi-level station area, which uses the above-mentioned intelligent network adaptive multi-level station area reactive power optimization control system for control, including the following steps: Step S1: During reactive power optimization and adjustment, the new energy photovoltaic power station, new energy distribution station, and energy storage system outside the substation are treated as a virtual SVG to simplify and optimize the control objectives and meet the target requirements of the 17-zone diagram of reactive power optimization within the substation. Step S2: Reactive power control and the external energy photovoltaic power stations and distribution areas are uniformly controlled by the coordination controller for target strategy control. The coordination controller collects reactive power data of all external energy photovoltaic power stations and distribution areas through local control. Step S3: The local controller collects the operating data of each power generation device, energy storage system and dispatch control system of each new energy power station in its area, and transmits it to the coordination controller through the 4G / 5G network to monitor the operating status of each device and the power factor of the current new energy power station and energy storage system in real time. Step S4: In order to identify the primary topology model of the power grid, intelligent network topology analysis is adopted, and graph neural networks are used to identify the power grid topology and estimate its state in real time. Step S5: The control strategy of the coordinating controller adopts the real-time reactive power optimization and adjustment strategy of the intelligent network adaptive model. During network training, the reactive power data changes of each energy photovoltaic power station and distribution area are used as the network output, and the reactive power change value affecting the whole station is used as the network input value to complete the training of the initial model of the intelligent network. The reactive power control target is input into the network through reactive power data control target, and the reactive power control target of each energy photovoltaic power station and distribution area is calculated. Step S6: Establish an intelligent network adaptive model. Obtain relevant power supply parameters and reactive power change values ​​from the power grid side through the communication network. Set a step length m for each energy photovoltaic power station and transformer inverter to generate a dynamic model of reactive power change. Step S7: Treat all new energy photovoltaic power stations, new energy distribution stations, and energy storage systems outside the station as a virtual SVG. The reactive power control of the new energy photovoltaic power stations, new energy distribution stations, and energy storage systems outside the station is uniformly controlled by the regional coordinator. The control target is distributed to the local controllers in each region for reactive power compensation target control. Step S8: Data interaction between the plant VQC control and the regional coordinator control is carried out in a publish-subscribe mode. The regional coordinator control publishes the total reactive power and existing reactive power of each new energy photovoltaic power station and distribution area to the VQC in real time. The plant VQC publishes the reactive power control target to the regional coordinator control. Step S9: The reactive power control strategy uses a machine learning algorithm based on a decision tree network model to classify the target control. The classification targets include: main transformer tap changer control, capacitor switching control, and renewable energy reactive power control. A classification tree model is constructed. The advantage of using decision trees is that the training data is very fast and the network model is relatively stable. Step S10: For the new energy photovoltaic power station, new energy distribution station, and energy storage system contained in the virtual SVG, each is treated as a node. The photovoltaic power station, distribution station, and energy storage system are treated as independent nodes. Each node is set as a predictive reactive power compensation output in the smart network. Step S11: For the virtual SVG containing new energy photovoltaic power plants, new energy distribution substations, and energy storage systems, establish sensitivity indicators for the impact factors of each node's reactive power source on the reactive power of the entire substation. Optimize the timing and increment of the switching, and use the sensitivity index as an evaluation index for node adjustment; Step S12: The objective function for evaluating the overall system adjustment strategy is to construct a multi-objective optimization model with power factor deviation and voltage deviation as objectives, combined with the uncertainty of new energy output. Step S13: Inter-domain collaboration optimization strategy, which realizes information interaction and collaborative optimization between different levels through intelligent networks; Step S14: Adaptive dynamic adjustment strategy. Based on the system operating status and changes in the external environment, combined with the transformer voltage operation status in the substation, the reactive power compensation of the capacitor, and the virtual quantity outside the station, the control strategy is adaptively adjusted. The system operating status and changes in the external environment are monitored in real time. Based on preset rules and reinforcement learning models, it is determined whether the control strategy needs to be adjusted, and the optimization target weights and control parameters at each level are dynamically adjusted.

[0008] Furthermore, step S1 includes the following: For both internal and external control objectives of the substation, a multi-objective control strategy based on decision trees is adopted for target control. For off-site virtual SVG, the target strategy control is uniformly performed by the coordination controller, and the reactive power compensation strategy control is based on the intelligent network model.

[0009] Furthermore, step S4 includes the following: 1) Construct a power grid topology model; 2) Step S4, which trains a graph neural network based on historical data, includes the following: model; 3) Input grid measurement data in real time for topology identification and state estimation.

[0010] Furthermore, step S6 includes the following: To establish an intelligent network adaptive model, reactive power operation data of energy photovoltaic power plants and distribution areas are first collected. At the same time, reactive power changes of inverters in each energy photovoltaic power plant and distribution area are adjusted. All reactive power-related operation data are entered into the network. The power factor during operation is independently statistically adjusted. Power supply-related parameters and reactive power change values ​​are obtained from the power grid side through the communication network. Each energy photovoltaic power plant and distribution area inverter is incremented by a step size m. The overall system operation model is continuously run for a period of time to obtain the influence factor of each power generation device of each new energy plant on the reactive power optimization of the whole plant, and a dynamic model of power supply reactive power change is generated.

[0011] Furthermore, step S9 includes the following: For decision trees and classification trees, the optimal metric is called "impurity." The lower the impurity, the better the decision tree fits the training set. Impurity is calculated based on nodes, meaning each node has an impurity level. The impurity measure is evaluated using a combination of information entropy and the Gini coefficient. Information entropy is more sensitive to impurity and has the strongest penalty for it. The formula for calculating information entropy is: The formula for calculating the Gini coefficient is: Where t represents a given node, i represents any category of the label, and p(i|t) represents the proportion of label category i in node t; To comprehensively evaluate the information entropy and Gini coefficient for impurity, a combined value of the two is used here: The formula for calculating impurity is: Here, α is the balance coefficient, and model training is an incremental optimization system that aims to achieve a relatively optimal impurity value.

[0012] Furthermore, step S10 includes the following: The adjustment value for the reactive power compensation node is: in, This represents the reactive power allocation value of the i-th node at time t; This represents the real-time voltage of the i-th node; Represents the total reactive power demand at the regional level; Represents the voltage deviation in the region; This represents the dynamic weighting coefficient, which is adaptively adjusted by the neural network based on real-time operating conditions.

[0013] Furthermore, step S11 includes the following: Sensitivity optimization metrics: in, The adjustment amount represents the reactive power source j at node j. The variable representing the reactive power change of the substation is the adjustment quantity of the reactive power source j at node j. The representation is a time-sharing weighted matrix, which balances the adjustment needs of different regions and time zones for reactive power sources of equal priority.

[0014] Furthermore, step S12 includes the following: The objective evaluation function is: in, Represents power factor deviation. Represents voltage deviation. This represents the weighting coefficient.

[0015] Furthermore, step S13 includes the following: 1) Establish a communication mechanism between the regional local controller and the coordination controller to achieve real-time data exchange; 2) Based on the intelligent network adaptive multi-level station domain control algorithm strategy, the coordinated controller can coordinate collaborative decision-making among various levels. 3) Dynamically adjust the optimization objectives and control strategies at each level to achieve global optimization.

[0016] Furthermore, step S14 includes the following: Step S14_1: For the control objectives inside and outside the substation, a decision tree-based multi-objective control strategy is adopted. Each control is assigned a corresponding priority and influence factor. The voltage and reactive power inputs that need to be compensated are selected by the decision tree multi-objective model, and the control objectives are selected for target supplementary control. Step S14_2: When it is necessary to control the transformer tap position, the main transformer tap changer is mainly used to adjust the low-voltage side voltage of the transformer. When the tap position is increased, the voltage increases and the total reactive power increases, resulting in a decrease in the power factor; when the tap position is decreased, the voltage decreases and the total reactive power decreases, resulting in a larger power factor. When controlling the tap changer, the main transformer must first be allowed to adjust the tap position, then the main transformer tap adjustment time must be allowed, and finally the closed-loop operation must be checked. When all conditions are met, the tap changer is adjusted up or down. Otherwise, the current adjustment is terminated, the result is returned to the control node, the influencing factor is changed, and the process returns to step S14_1 to continue the next round of optimization adjustment. Step S14_3: When it is necessary to control the switching of capacitors, the switching of capacitors is mainly used to adjust the total reactive power through the transformer. When the capacitor bank is connected, the capacitive reactive power increases, the voltage increases, the total reactive power decreases, and the power factor increases. When the capacitor bank is disconnected, the capacitive reactive power decreases, the voltage decreases, the total reactive power increases, and the power factor decreases. When it is necessary to switch the capacitor, firstly, it is predicted that the capacitor can be switched, secondly, it is determined that the adjustment time of the main transformer is allowed, and finally, it is determined whether it is closed-loop operation or semi-closed-loop operation. When all conditions are met, the capacitor is switched. Otherwise, the current round of adjustment ends, the result is returned to the control node, the influencing factor is changed, and it returns to step S14_1 to continue the next round of optimization adjustment. Step S14_4: When it is necessary to adjust the reactive power of the virtual SVG, first determine whether the reactive power capacity of the photovoltaic power station or the distribution area meets the conditions. If the conditions are met, then the reactive power adjustment is carried out by adopting the real-time reactive power optimization and adjustment strategy based on the intelligent network adaptive model. Otherwise, the current round of adjustment is ended, the result is returned to the control node, the influencing factor is changed, and the process returns to step S14_1 to continue the next round of optimization and adjustment. Step S14_5: Real-time power grid data, historical data, and model information serve as the data foundation for the real-time reactive power optimization and adjustment strategy model of the intelligent network adaptive model. First, the real-time reactive power and load data of the power grid, along with the power grid model data, topology data, historical data, and historical strategy data, are input into the data structuring transformation module. Step S14_6: Data structuring transformation module. First, input the real-time reactive power, load, historical data, and power grid model; second, perform matrix exchange on the power grid model structure and node topology data matrices; third, import the real-time data and historical data into the transformation model; finally, obtain structured historical data and real-time data. Step S14_7: Establishment of intelligent network adaptive reactive power optimization model. First, the network weights are reset using historical optimization strategies to initialize the intelligent network adaptive model structure. Second, the network parameters are obtained through multiple iterations, and line particles are taken as intelligent network weights. Finally, network training begins, which is divided into test set and training set to obtain the input and output after training. Step S14_8: Real-time reactive power optimization adjustment strategy of intelligent network adaptive model. First, according to the structured real-time data obtained in step S14_6 and the model establishment obtained in step S14_7, input the reactive power optimization model to obtain the reactive power optimization strategy at the current time. Then, reactive power optimization adjustment is performed according to the strategy. After that, the stability effect of the power grid after adjustment is verified. Finally, the adjustment strategy evaluation output result is fed back to step S14_5 as the input for model iteration.

[0017] By employing the above technical solutions, this invention has the following advantages and positive effects compared with the prior art: This invention constructs a multi-level collaborative control architecture at the substation, regional, and global levels, and introduces a decision tree network model to achieve precise selection and dynamic control of reactive power compensation targets. The control target coordinates reactive power compensation output at each level from three perspectives: transformers and capacitors within the substation, and external systems and equipment. It innovatively constructs virtual SVG (Static Var Generator) reactive power control units for distributed new energy photovoltaic power plants, distribution substations, and energy storage systems outside the substation, achieving global optimization of reactive power resource allocation. Simultaneously, it integrates deep reinforcement learning algorithms, relying on an intelligent network to construct an adaptive reactive power optimization model, implementing refined optimization control for key target nodes. The regional coordinator, as the virtual SVG hub, is responsible for accurately distributing optimized control commands to local controllers in each region, ensuring seamless collaboration and efficient response between multiple levels. This invention successfully achieves intelligent and efficient management of power grid reactive power and voltage, significantly improving power supply quality and system stability, while greatly accelerating the response speed of reactive power compensation. Even under complex operating conditions such as grid voltage fluctuations and frequent changes in photovoltaic output, it can achieve multi-level joint and refined compensation of reactive power, providing strong technical support for the stable operation and efficient management of the power system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is an architecture diagram of an intelligent network-based adaptive multi-level station domain reactive power optimization control system according to the present invention. Figure 2 This is a flowchart of the multi-objective reactive power optimization and regulation strategy in this invention; Figure 3 This is a flowchart of the real-time reactive power optimization and adjustment strategy of the intelligent network adaptive model in this invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figure 1 As shown, this embodiment discloses an intelligent network-based adaptive multi-level station reactive power optimization control system, including an AVC / VGC control system, a coordination controller device, and a local controller device, wherein: The AVC / VGC control system, as the core control hub for substation voltage and reactive power, interacts with the coordination controller device using a publish / subscribe mode for data message exchange. The communication medium is an RJ45 Ethernet interface, which is used to ensure the real-time exchange of control commands and status information and to monitor and adjust the voltage level and reactive power distribution of the entire station in real time. Due to the long physical distance between the coordination controller device and the local controller device, a high-speed and stable 4G / 5G wireless communication network is used as the transmission channel. The communication protocol follows the standard IEC 60870-5-104 protocol. It is used to undertake the centralized scheduling responsibility of reactive power resources outside the station and realize the optimized allocation and coordinated control of reactive power of multiple stations through real-time communication with the upper power grid dispatching system. The local controller device is used to provide precise local reactive power compensation and regulation functions for local equipment such as photovoltaic power generation areas, distribution substations, and energy storage systems.

[0021] During reactive power optimization and regulation, a multi-objective control strategy based on decision trees is adopted for both internal and external substation control objectives. The new energy photovoltaic power stations, new energy distribution substations, and energy storage systems outside the substation are treated as a single virtual SVG, and the target strategy is uniformly controlled by the coordinating controller. A reactive power compensation strategy based on an intelligent network model is used to simplify and optimize the control objectives. The coordinating controller collects reactive power data from all external energy photovoltaic power stations and distribution substations through local control. Local controllers collect operational data from each generating unit, energy storage system, and dispatch control system of each new energy station within their jurisdiction, transmitting this data to the coordinating controller via 4G / 5G networks. This allows for real-time monitoring of the operational status of each device and the power factor of the current new energy station and energy storage system, comprehensively meeting the target requirements of the 17-zone reactive power optimization map within the substation. This fully utilizes the reactive power support capabilities of the new energy stations, achieving multi-level refined reactive power compensation under grid voltage fluctuations and photovoltaic output changes.

[0022] like Figure 2 and 3 As shown, this embodiment also discloses a reactive power optimization control method based on intelligent network adaptive multi-level station domain, which uses the above-mentioned intelligent network adaptive multi-level station domain reactive power optimization control system for control, including the following steps: Step S1: During reactive power optimization and adjustment, the new energy photovoltaic power station, new energy distribution station, and energy storage system outside the substation are treated as a virtual SVG to simplify and optimize the control objectives and meet the target requirements of the 17-zone diagram of reactive power optimization within the substation. Specifically, step S1 includes the following: For both internal and external control objectives of the substation, a multi-objective control strategy based on decision trees is adopted for target control. For off-site virtual SVG, the target strategy control is uniformly performed by the coordination controller, and the reactive power compensation strategy control is based on the intelligent network model.

[0023] Step S2: Reactive power control and the external energy photovoltaic power stations and distribution areas are uniformly controlled by the coordination controller for target strategy control. The coordination controller collects reactive power data of all external energy photovoltaic power stations and distribution areas through local control. Step S3: The local controller collects the operating data of each power generation device, energy storage system and dispatch control system of each new energy power station in its area, and transmits it to the coordination controller through the 4G / 5G network to monitor the operating status of each device and the power factor of the current new energy power station and energy storage system in real time. To prevent data from being tampered with or leaked, the data packets are transmitted in an encrypted manner. Step S4: In order to identify the primary topology model of the power grid, intelligent network topology analysis is adopted, and graph neural networks are used to identify the power grid topology and estimate its state in real time. Specifically, step S4 includes the following: 1) Construct a power grid topology model; 2) Step S4, which trains a graph neural network based on historical data, includes the following: model; 3) Input grid measurement data in real time for topology identification and state estimation.

[0024] Step S5: The control strategy of the coordinating controller adopts the real-time reactive power optimization and adjustment strategy of the intelligent network adaptive model. During network training, the reactive power data changes of each energy photovoltaic power station and distribution area are used as the network output, and the reactive power change value affecting the whole station is used as the network input value to complete the training of the initial model of the intelligent network. The reactive power control target is input into the network through reactive power data control target, and the reactive power control target of each energy photovoltaic power station and distribution area is calculated. Step S6: Establish an intelligent network adaptive model. Obtain relevant power supply parameters and reactive power change values ​​from the power grid side through the communication network. Set a step length m for each energy photovoltaic power station and transformer inverter to generate a dynamic model of reactive power change. Specifically, step S6 includes the following: To establish an intelligent network adaptive model, reactive power operation data of energy photovoltaic power plants and distribution areas are first collected. At the same time, reactive power changes of inverters in each energy photovoltaic power plant and distribution area are adjusted. All reactive power-related operation data are entered into the network. The power factor during operation is independently statistically adjusted. Power supply-related parameters and reactive power change values ​​are obtained from the power grid side through the communication network. Each energy photovoltaic power plant and distribution area inverter is incremented by a step size m. The overall system operation model is continuously run for a period of time to obtain the influence factor of each power generation device of each new energy plant on the reactive power optimization of the whole plant, and a dynamic model of power supply reactive power change is generated.

[0025] Step S7: Treat all new energy photovoltaic power stations, new energy distribution stations, and energy storage systems outside the station as a virtual SVG. The reactive power control of the new energy photovoltaic power stations, new energy distribution stations, and energy storage systems outside the station is uniformly controlled by the regional coordinator. The control target is distributed to the local controllers in each region for reactive power compensation target control. Step S8: Data interaction between the plant VQC control and the regional coordinator control is carried out in a publish-subscribe mode. The regional coordinator control publishes the total reactive power and existing reactive power of each new energy photovoltaic power station and distribution area to the VQC in real time. The plant VQC publishes the reactive power control target to the regional coordinator control. Step S9: The reactive power control strategy uses a machine learning algorithm based on a decision tree network model to classify the target control. The classification targets include: main transformer tap changer control, capacitor switching control, and renewable energy reactive power control. A classification tree model is constructed. The advantage of using decision trees is that the training data is very fast and the network model is relatively stable. Specifically, step S9 includes the following: For decision trees, the optimal metric is called "impurity." Lower impurity indicates a better fit to the training set. Impurity is calculated at the node level; each node has an impurity value. The impurity measure is a combination of entropy and Gini impurity. Entropy is more sensitive to impurity and has the strongest penalty for it. The formula for calculating information entropy is: The formula for calculating the Gini coefficient is: Where t represents a given node, i represents any category of the label, and p(i|t) represents the proportion of label category i in node t; To comprehensively evaluate the information entropy and Gini coefficient for impurity, a combined value of the two is used here: The formula for calculating impurity is: Here, α is the balance coefficient, and model training is an incremental optimization system that aims to achieve a relatively optimal impurity value.

[0026] Step S10: For the new energy photovoltaic power station, new energy distribution station, and energy storage system contained in the virtual SVG, each is treated as a node. The photovoltaic power station, distribution station, and energy storage system are treated as independent nodes. Each node is set as a predictive reactive power compensation output in the smart network. Specifically, step S10 includes the following: The adjustment value for the reactive power compensation node is: in, This represents the reactive power allocation value of the i-th node at time t; This represents the real-time voltage of the i-th node; Represents the total reactive power demand at the regional level; Represents the voltage deviation in the region; This represents the dynamic weighting coefficient, which is adaptively adjusted by the neural network based on real-time operating conditions.

[0027] Step S11: For the virtual SVG containing new energy photovoltaic power plants, new energy distribution substations, and energy storage systems, establish sensitivity indicators for the impact factors of each node's reactive power source on the reactive power of the entire substation. Optimize the timing and increment of the switching, and use the sensitivity index as an evaluation index for node adjustment; Specifically, step S11 includes the following: Sensitivity optimization metrics: in, The adjustment amount represents the reactive power source j at node j. The variable representing the reactive power change of the substation is the adjustment quantity of the reactive power source j at node j. The representation is a time-sharing weighted matrix, which balances the adjustment needs of different regions and time zones for reactive power sources of equal priority.

[0028] Step S12: The objective function for evaluating the overall system adjustment strategy is to construct a multi-objective optimization model with power factor deviation and voltage deviation as objectives, combined with the uncertainty of new energy output. Specifically, step S12 includes the following: The objective evaluation function is: in, Represents power factor deviation. Represents voltage deviation. This represents the weighting coefficient.

[0029] Step S13: Inter-domain collaboration optimization strategy, which realizes information interaction and collaborative optimization between different levels through intelligent networks; Specifically, step S13 includes the following: 1) Establish a communication mechanism between the regional local controller and the coordination controller to achieve real-time data exchange; 2) Based on the intelligent network adaptive multi-level station domain control algorithm strategy, the coordinated controller can coordinate collaborative decision-making among various levels. 3) Dynamically adjust the optimization objectives and control strategies at each level to achieve global optimization.

[0030] Step S14: Adaptive dynamic adjustment strategy. Based on the system operating status and changes in the external environment, combined with the transformer voltage operation status in the substation, the reactive power compensation of the capacitor, and the virtual quantity outside the station, the control strategy is adaptively adjusted. The system operating status and changes in the external environment are monitored in real time. Based on preset rules and reinforcement learning models, it is determined whether the control strategy needs to be adjusted, and the optimization target weights and control parameters at each level are dynamically adjusted.

[0031] Specifically, step S14 includes the following: Step S14_1: For the control objectives inside and outside the substation, a decision tree-based multi-objective control strategy is adopted. Each control is assigned a corresponding priority and influence factor. The voltage and reactive power inputs that need to be compensated are selected by the decision tree multi-objective model, and the control objectives are selected for target supplementary control. Step S14_2: When it is necessary to control the transformer tap position, the main transformer tap changer is mainly used to adjust the low-voltage side voltage of the transformer. When the tap position is increased, the voltage increases and the total reactive power increases, resulting in a decrease in the power factor; when the tap position is decreased, the voltage decreases and the total reactive power decreases, resulting in a larger power factor. When controlling the tap changer, the main transformer must first be allowed to adjust the tap position, then the main transformer tap adjustment time must be allowed, and finally the closed-loop operation must be checked. When all conditions are met, the tap changer is adjusted up or down. Otherwise, the current adjustment is terminated, the result is returned to the control node, the influencing factor is changed, and the process returns to step S14_1 to continue the next round of optimization adjustment. Step S14_3: When it is necessary to control the switching of capacitors, the switching of capacitors is mainly used to adjust the total reactive power (i.e., power factor cos) through the transformer. When the capacitor bank is connected, the capacitive reactive power increases, the voltage increases, the total reactive power decreases, resulting in a larger power factor. When the capacitor bank is disconnected, the capacitive reactive power decreases, the voltage decreases, the total reactive power increases, resulting in a smaller power factor. When it is necessary to switch the capacitor, firstly, it is predicted that the capacitor can be switched, secondly, it is determined that the adjustment time of the main transformer is allowed, and finally, it is determined whether it is closed-loop operation or semi-closed-loop operation. When all conditions are met, the capacitor is switched; otherwise, the current round of adjustment ends, the result is returned to the control node, the influencing factor is changed, and it returns to step S14_1 to continue the next round of optimization adjustment. Step S14_4: When it is necessary to adjust the reactive power of the virtual SVG, first determine whether the reactive power capacity of the photovoltaic power station or distribution area meets the conditions. If the conditions are met, then the reactive power adjustment is carried out using a real-time reactive power optimization and adjustment strategy based on an intelligent network adaptive model. The specific process is as follows: Figure 3 As shown, otherwise the current adjustment ends, the result is returned to the control node, the influencing factor is changed, and the process returns to step S14_1 to continue the next round of optimization adjustment; Step S14_5: Real-time power grid data, historical data, and model information serve as the data foundation for the real-time reactive power optimization and adjustment strategy model of the intelligent network adaptive model. First, the real-time reactive power and load data of the power grid, along with the power grid model data, topology data, historical data, and historical strategy data, are input into the data structuring transformation module. Step S14_6: Data structuring transformation module. First, input the real-time reactive power, load, historical data, and power grid model; second, perform matrix exchange on the power grid model structure and node topology data matrices; third, import the real-time data and historical data into the transformation model; finally, obtain structured historical data and real-time data. Step S14_7: Establishment of intelligent network adaptive reactive power optimization model. First, the network weights are reset using historical optimization strategies to initialize the intelligent network adaptive model structure. Second, the network parameters are obtained through multiple iterations, and line particles are taken as intelligent network weights. Finally, network training begins, which is divided into test set and training set to obtain the input and output after training. Step S14_8: Real-time reactive power optimization adjustment strategy of intelligent network adaptive model. First, according to the structured real-time data obtained in step S14_6 and the model establishment obtained in step S14_7, input the reactive power optimization model to obtain the reactive power optimization strategy at the current time. Then, reactive power optimization adjustment is performed according to the strategy. After that, the stability effect of the power grid after adjustment is verified. Finally, the adjustment strategy evaluation output result is fed back to step S14_5 as the input for model iteration. The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A self-adaptive multi-level zone reactive power optimization control system based on intelligent network, characterized in that, The AVC / VGC control system, the coordination controller device, and the on-site controller device are provided. The AVC / VGC control system is used as the core management center of the voltage and reactive power of the substation, and adopts the publish / subscribe mode to exchange data messages with the coordination controller device, and the communication medium is an RJ45 Ethernet interface, which is used to ensure the real-time exchange of control commands and state information and is responsible for real-time monitoring and adjustment of the voltage level and reactive power flow distribution of the whole station. The coordination controller device and the on-site controller device adopt a 4G / 5G wireless communication network as a transmission channel, which is used to undertake the centralized scheduling responsibility of the external reactive power resource and realize the optimal allocation and collaborative control of the multi-site reactive power through real-time communication with the superior grid dispatching system. The on-site controller device is used to provide accurate local reactive power compensation and adjustment function for the local equipment of the photovoltaic power generation area, the distribution area and the energy storage system.

2. A method for adaptive multi-level zone reactive power optimization control based on intelligent network, characterized in that, The intelligent network adaptive multi-level station reactive power optimization control system is used for control, and the steps are as follows: Step S1: During the reactive power optimization adjustment, the new energy photovoltaic power station, the new energy area station and the energy storage system outside the station are regarded as a virtual SVG, the optimization control target is simplified, and the whole station reactive power optimization 17 area diagram is satisfied; Step S2: The reactive power control and the energy photovoltaic power station and the area outside the station are controlled by the coordination controller, and the reactive power data of all the energy photovoltaic power stations and the areas outside the station are collected by the on-site control; Step S3: The on-site controller collects the operation data of each power generation equipment, the energy storage system and the dispatching control system of each new energy field station in the region, and transmits the data to the coordination controller through the 4G / 5G network, so as to monitor the operation state of each equipment and the power factor of the current new energy field station and the energy storage system; Step S4: In order to identify the power grid topology model, intelligent network topology analysis is adopted, and the graph neural network is used to realize real-time identification and state estimation of the power grid topology; Step S5: The intelligent network adaptive model is used to realize real-time reactive power optimization adjustment strategy, the network training is completed by changing the reactive power data of each energy photovoltaic power station and the area outside the station as the network output, and the influence on the whole station reactive power change value is taken as the network input value, so as to complete the training of the initial model of the intelligent network, the network is input by the reactive power data control target, and the reactive power control target of each energy photovoltaic power station and the area outside the station is calculated and obtained; Step S6: The intelligent network adaptive model is established, the power supply related parameters and the reactive power change value of the power supply grid side are obtained through the communication network, the step length m of each energy photovoltaic power station and the area outside the station is increased, and the power supply reactive power change dynamic model is generated; Step S7: The new energy photovoltaic power station, the new energy area station and the energy storage system outside the station are regarded as a virtual SVG, the reactive power control of the new energy photovoltaic power station, the new energy area station and the energy storage system outside the station is controlled by the regional coordinator, and the control target is distributed to the on-site controller of each region for reactive power compensation target control. Step S8: Data interaction between plant VQC control and regional coordinator control adopts a publish-subscribe mode, and the regional coordinator control publishes the total reactive power and existing reactive power in each new energy photovoltaic power station and the area to the VQC in real time. The plant VQC publishes the reactive power control target to the regional coordinator control; Step S9: The reactive power strategy control strategy is based on a decision tree network model and uses a machine learning algorithm to classify the target control. The classification targets include: main transformer tap control, capacitor switching control, new energy reactive power control, and a classification tree model is constructed. The advantage of using a decision tree is that the training data is very fast, and the network model is relatively stable; Step S10: For the virtual SVG containing new energy photovoltaic power stations, new energy area stations, and energy storage systems, each acts as a node photovoltaic power station, area station, and energy storage system as an independent node. Each node in the intelligent network is set as a predicted reactive power compensation output; Step S11: the virtual SVG includes a new energy photovoltaic power station, a new energy transformer station, and an energy storage system. A sensitivity index is established for the influence factor of each node reactive power source on the entire transformer station reactive power , the optimization of switching timing and increment, and the sensitivity index as a node adjustment evaluation index. Step S12: The system overall adjustment strategy evaluates the target function with power factor deviation and voltage deviation as the target, and constructs a multi-objective optimization model combined with new energy output uncertainty; Step S13: Inter-domain collaborative optimization strategy, through intelligent network to realize the information interaction and collaborative optimization between each level; Step S14: Self-adaptive dynamic adjustment strategy, according to the system running state and external environment change, combined with the transformer voltage operation situation in the substation, the capacitor reactive power compensation amount, and the virtual quantity outside the station, the control strategy is adjusted adaptively, the system running state and external environment change are monitored in real time, based on the preset rules and reinforcement learning model, it is judged whether the control strategy needs to be adjusted, and the optimization target weight and control parameter of each level are dynamically adjusted.

3. The self-adaptive multi-level substation area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S1 includes the following content: For the control target inside and outside the substation, a multi-objective control strategy based on decision tree is adopted for target control; For the virtual SVG outside the station, the target strategy control is controlled by the coordination controller, and the reactive power compensation strategy control is controlled based on the intelligent network model.

4. The self-adaptive multi-level substation area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S4 includes the following content: 1) Construct a power grid topology graph model; 2) Train the graph neural network based on historical data Step S4 Including the following content: model; 3) Real-time input of power grid measurement data for topology identification and state estimation.

5. The self-adaptive multi-level substation area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S6 includes the following content: An intelligent network adaptive model is established. First, the energy photovoltaic power station and the area reactive power operation data are collected, and the reactive power of each energy photovoltaic power station and the area inverter is adjusted. All the reactive power related operation data are recorded into the network, the power factor during operation is independently adjusted, the power supply related parameters and reactive power change value on the power supply side are obtained through the communication network, and the step length m of each energy photovoltaic power station and the area inverter is increased. The overall system operation model is continuously running for a period of time, the influence factor of each power generation equipment of each new energy station on the overall station reactive power optimization is obtained, and a power supply reactive power change dynamic model is generated.

6. The self-adaptive multi-level zone reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S9 includes the following content: The best indicator for decision tree classification tree is called "impurity", the lower the impurity, the better the fitting of the decision tree to the training set, the impurity is calculated based on the node, that is, each node will have an impurity, the impurity value is evaluated by information entropy and gini coefficient, information entropy is more sensitive to impurity and has the strongest penalty on impurity: The information entropy value calculation formula is: The Gini coefficient calculation formula is: Where t represents the given node, i represents any classification of the label, and p(i|t) represents the proportion of label classification i in node t. In order to evaluate the information entropy and gini coefficient of impurity, the comprehensive value of the two values is taken here: The impurity calculation formula is: Where α is the balance coefficient, the model training is an incremental optimization system, and the impurity value is relatively optimal.

7. The self-adapting multi-level substation-area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S10 includes the following contents: The node reactive power supplement node adjustment value is: wherein, represents the reactive power allocation value of the i-th node at time t; represents the real-time voltage of the i-th node; represents the total reactive power demand of the regional layer; represents the regional voltage deviation; represents the dynamic weight coefficient, which is self-adaptively adjusted by the neural network according to the real-time working condition.

8. The self-adapting multi-level substation-area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S11 includes the following contents: Sensitivity optimization index: wherein, the adjustment amount of the representative node reactive power source j, the adjustment amount of the representative node reactive power source j, representing a time-weighted matrix, for the same priority reactive power source, balancing the adjustment demand of different time zones in different regions.

9. The self-adapting multi-level substation-area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S12 Includes the following contents: The objective evaluation function is: wherein, represents a power factor deviation, represents a voltage deviation, represents a weight coefficient.

10. The self-adapting multi-level zone reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S13 includes the following contents: 1) Establish the communication mechanism of regional local controller and coordination controller to realize real-time data exchange; 2) Based on intelligent network adaptive multi-level station area control algorithm strategy, realize the unified coordination of collaborative decision between each level by coordination controller; 3) Dynamically adjust the optimization target and control strategy of each level to realize global optimization.

11. The self-adapting multi-level substation-area reactive power optimization control method based on intelligent network according to claim 2, characterized in that, Step S14 includes the following contents: Step S14_1: For substation in-station and out-station control target, adopt multi-objective control strategy based on decision tree, each control is allocated with corresponding priority and influence factor, input the voltage and reactive power to be compensated into the multi-objective model of decision tree, select the control target, and carry out target compensation control; Step S14_2: When the transformer tap needs to be controlled, the main transformer tap is mainly used to adjust the voltage at the low voltage side of the transformer. When the tap is raised, the voltage is raised, and the total reactive power is increased, resulting in a decrease in power factor. When the tap is lowered, the voltage is reduced, and the total reactive power is reduced, resulting in an increase in power factor. When controlling the tap, the main transformer must first allow the tap to be adjusted, then the main transformer must allow the tap to be adjusted, and finally the closed-loop operation must be judged. When all conditions are met, the tap is raised or lowered, otherwise the round of adjustment is ended, the result is returned to the control node, the influence factor is changed, and the step S14_1 is returned to continue the next round of optimization adjustment; Step S14_3: When the capacitor needs to be controlled, the capacitor is mainly used to adjust the total reactive power through the transformer. When the capacitor bank is put into operation, the capacitive reactive power is increased, the voltage is increased, and the total reactive power is decreased, resulting in an increase in power factor. When the capacitor bank is removed, the capacitive reactive power is reduced, the voltage is reduced, and the total reactive power is increased, resulting in a decrease in power factor. When the capacitor needs to be switched, the capacitor must first be pre-judged, then the main transformer must be judged, and finally the closed-loop operation or semi-closed-loop operation must be judged. When all conditions are met, the capacitor is switched, otherwise the round of adjustment is ended, the result is returned to the control node, the influence factor is changed, and the step S14_1 is returned to continue the next round of optimization adjustment; Step S14_4: When the virtual SVG reactive power needs to be adjusted, first determine whether the photovoltaic power station or the area reactive power capacity meets the conditions. If it meets the adjustment, the real-time reactive power optimization adjustment strategy based on the intelligent network adaptive model is used for reactive power adjustment, otherwise the adjustment is ended, the result is returned to the control node, the influence factor is changed, and the optimization adjustment of the next round is continued. Step S14_5: The real-time data, historical data and model information of the power grid are used as the data basis of the intelligent network adaptive model real-time reactive power optimization adjustment strategy model. First, the real-time reactive power and load data of the power grid, the power grid model data, the topology data, the historical data and the historical strategy data are input into the data structure conversion module. Step S14_6: The data structure conversion module first inputs the real-time reactive power, load, historical data and power grid model. Then, the matrix exchange is performed on the power grid model structure and node topology structure data matrix. The real-time data and historical data are imported into the conversion model. Finally, the structured historical data and real-time data are obtained. Step S14_7: The intelligent network adaptive reactive power optimization model is established. First, the historical optimization strategy is used to reset the network weight and initialize the intelligent network adaptive model structure. Then, the network parameters are obtained by multiple iterations, and the line particles are used as the intelligent network weight. Finally, the network training is started, and the training set and the test set are divided, and the input and output after training are obtained. Step S14_8: The intelligent network adaptive model real-time reactive power optimization adjustment strategy is used. First, the structured real-time data are obtained according to step S14_6, and the model is established according to step S14_7. The reactive power optimization model is input, and the current time reactive power optimization strategy is obtained. Then, the reactive power optimization adjustment is performed according to the strategy, and the stability of the power grid after adjustment is verified. Finally, the adjustment strategy evaluation output result is fed back to step S14_5 as the model iteration input.

Images