Load prediction-based voltage stabilizer dynamic voltage regulating system and method

Abstract

The application relates to the technical field of power system control, and particularly discloses a dynamic voltage regulating system and method of a voltage stabilizer based on load prediction, the method comprising: synchronously collecting load side data and energy side data in the power system; based on a prediction model integrating physical laws of the power grid, processing the load side data and the energy side data to generate a multi-dimensional state prediction result of the load and the energy. By introducing the deep prediction model integrating the physical laws of the power grid, the application can predict the future load and energy fluctuation state of the system in advance, realizes a fundamental change from passive response to active pre-control, and thus gains valuable response time for coping with power impact; secondly, the scheme creatively proposes a coordinated impact coefficient, can accurately quantify the coupling influence degree of fluctuation on both the supply side and the demand side, and generates a hierarchical and differentiated multi-resource coordinated voltage regulating strategy based on the same.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system control, in particular to a load prediction-based dynamic voltage regulator system and method. BACKGROUND

[0002] In the current new power system dominated by new energy, the proportion of renewable energy such as photovoltaic and wind power is increasing. Such energy has the characteristics of intermittency and volatility in nature, which leads to unprecedented uncertainty in energy supply side of the power system. At the same time, with the popularity of new forms of electricity such as electric vehicle charging piles and flexible industrial loads, the demand side of the power system also shows the characteristics of high dynamics and high randomness. The superposition of fluctuations on both supply and demand sides makes the balance of power grid power face serious challenges, easily causing frequent and severe fluctuations of node voltage of the power grid, seriously affecting power quality, and even threatening the safe and stable operation of critical equipment.

[0003] The prior art usually configures various regulation resources such as on-load voltage regulating transformers (or voltage regulators) and energy storage systems on the grid side. However, the management and control strategies of these regulation resources often have inherent conflicts and limitations:

[0004] On the one hand, traditional voltage regulation strategies are mostly passive response type, i.e. regulation action is started only after detecting voltage out-of-limit, which is difficult to cope with rapid power impact, resulting in poor voltage regulation effect;

[0005] On the other hand, there is a lack of effective coordination mechanism among various regulation units, which often operate independently and separately. For example, if only mechanical devices such as voltage regulators are relied on for frequent regulation, it will sharply accelerate the wear and tear of mechanical parts, shortening the service life of the equipment; and if the energy storage system is used indiscriminately to absorb all fluctuations, it will cause frequent charging and discharging of the battery, significantly increasing the cycle aging cost of the battery and reducing the overall operation economy of the system. SUMMARY

[0006] The present application aims to at least partially solve one of the technical problems in the related art. To this end, the purpose of the present application is to propose a load prediction-based dynamic voltage regulator system and method to enhance the robustness and power supply reliability of the power system.

[0007] To achieve the above-mentioned purpose, the first aspect of the present application proposes a load prediction-based dynamic voltage regulator method, applied to a power system including a voltage regulator, an energy storage system, a new energy generation unit and a grid interface, the method comprising the following steps:

[0008] S1, synchronously collecting load side data and energy side data in the power system;

[0009] S2, based on a prediction model integrating power grid physical laws, processing the load-side data and the energy-side data to generate a multi-dimensional state prediction result of the load and the energy; the multi-dimensional state prediction result comprises a load power prediction value, a load change rate prediction value, and an energy net output change amount prediction value;

[0010] S3, based on the multi-dimensional state prediction result, calculating a collaborative impact coefficient for quantifying the coupling effect between load demand fluctuation and energy supply fluctuation;

[0011] S4, taking a composite cost function as an objective function, constructing and solving a rolling optimization model to generate a collaborative voltage regulation strategy defining response priorities of different regulation units;

[0012] S5, executing the collaborative voltage regulation strategy to send control instructions to the voltage stabilizer, the energy storage system, the new energy generation unit, and the power grid interface.

[0013] To achieve the above object, the second aspect embodiment of the present application proposes a load prediction-based voltage stabilizer dynamic voltage regulation system, comprising:

[0014] a data acquisition module configured to synchronously acquire load-side data and energy-side data in a power system;

[0015] a state prediction module configured to process the load-side data and the energy-side data based on a prediction model integrating power grid physical laws to generate a multi-dimensional state prediction result of the load and the energy;

[0016] a decision and strategy generation module configured to calculate a collaborative impact coefficient based on the multi-dimensional state prediction result, and construct and solve a rolling optimization model taking a composite cost function as an objective function to generate a collaborative voltage regulation strategy defining response priorities of different regulation units;

[0017] a strategy execution module configured to send control instructions to the voltage stabilizer, the energy storage system, the new energy generation unit, and the power grid interface in the system according to the collaborative voltage regulation strategy.

[0018] To achieve the above object, the third aspect embodiment of the present application proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the load prediction-based voltage stabilizer dynamic voltage regulation method.

[0019] The load prediction-based voltage stabilizer dynamic voltage regulation system and method of the embodiment of the present application can predict the future load and energy fluctuation state of the system in advance by introducing a deep prediction model that integrates the physical law of the power grid, realizing a fundamental change from passive response to active pre-control, thereby gaining valuable response time for coping with power impact; secondly, the scheme creatively proposes a coordinated impact coefficient that can accurately quantify the coupling influence degree of fluctuations on both supply and demand sides, and based on this, generates a hierarchical and differentiated multi-resource coordinated voltage regulation strategy; the strategy can intelligently schedule the most suitable regulation unit according to the impact level, including the participation of voltage stabilizers, energy storage, new energy units and other units in regulation, avoiding misuse and abuse of regulation resources, and realizing precise control.

[0020] More importantly, by constructing and solving a composite cost function that comprehensively considers power quality, operating energy consumption and equipment life depreciation, the scheme can find the optimal control strategy that takes into account system stability, economy and reliability in a dynamically changing environment; in addition, the scheme also has fault monitoring and functional substitution fault-tolerant capabilities, and when a key regulation unit fails, it can instruct other units to temporarily take over its core functions, greatly enhancing the robustness and power supply reliability of the entire power system. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0022] Figure 2 is a load power prediction comparison curve diagram under different prediction models in the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0023] Figure 3 is a three-dimensional surface diagram of the coordinated impact coefficient varying with load and energy fluctuation in the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0024] Figure 4 is a node voltage response curve comparison diagram under rolling optimization in the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0025] Figure 5 is a regulation unit power response curve diagram under different impact levels in the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0026] Figure 6 is a comparison diagram of the influence of risk adaptive weight adjustment on voltage deviation and energy consumption in the load prediction-based voltage stabilizer dynamic voltage regulation method provided by the present application;

[0027] Figure 7is a system voltage fluctuation and energy storage takeover response curve schematic diagram when the voltage stabilizer fails in the voltage stabilizer dynamic voltage regulation method based on load prediction provided by the application;

[0028] Figure 8 is an implementation execution schematic diagram of the voltage stabilizer dynamic voltage regulation system based on load prediction provided by the application;

[0029] Figure 9 is a structural schematic diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0030] The embodiments of the application are described in detail below, examples of which are shown in the drawings, wherein the same or similar notations represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are intended to explain the application, and cannot be understood as a limitation of the application.

[0031] The voltage stabilizer dynamic voltage regulation system, method and electronic device based on load prediction of the embodiments of the application are described below with reference to the drawings.

[0032] Embodiment one:

[0033] Figure 1 is a flowchart schematic diagram of the voltage stabilizer dynamic voltage regulation method based on load prediction of an embodiment of the application. As shown in Figure 1 , the embodiment elaborates a voltage stabilizer dynamic voltage regulation method based on load prediction. The method is applied to a power system including a voltage stabilizer, an energy storage system, a new energy generation unit and a grid interface, and the core is to realize precise coordination of multiple regulation resources through prospective prediction and collaborative optimization to cope with voltage stability problems caused by new energy and load fluctuations.

[0034] The method described in the embodiment specifically includes the following steps:

[0035] Step S1, synchronously collecting load side data and energy side data in the power system.

[0036] For example, in order to realize accurate prediction and control, the system needs to obtain key data reflecting the current running state of the system in real time. The data acquisition process adopts high-precision sensors and synchronous phasor measurement units to ensure that the load side and energy side data are strictly synchronized in time, and eliminate the prediction error caused by data time lag.

[0037] Optionally, the load side data specifically includes load power and load power change rate . Among them, the load power characterizes the total active power demand of the power system at a certain moment, and the load power change rate which reflects the dynamic trend of the load demand, and its value can be calculated by dividing the difference of load power of two consecutive sampling periods by the sampling time interval. This parameter is crucial for predicting the mutation of the load.

[0038] Optionally, the energy side data includes photovoltaic output , wind power output , state of charge of the energy storage system , and power supply of the power grid . The photovoltaic output and the wind power output directly reflect the real-time power generation capacity of the new energy power generation unit, which has significant intermittency and volatility. The state of charge of the energy storage system characterizes the current energy reserve level of the energy storage system and is a key state quantity determining the charging and discharging capacity of the energy storage system. The power supply of the power grid represents the power exchange between the system and the main grid, and a positive value usually indicates that power is drawn from the grid, and a negative value indicates that power is fed back to the grid. These data collectively constitute a complete portrait of the energy supply side.

[0039] In step S2, based on a prediction model that integrates the physical laws of the power grid, the load side data and the energy side data are processed to generate a multi-dimensional state prediction result of the load and the energy.

[0040] This step is the cornerstone of realizing the transition from passive response to active pre-control. Since the traditional data-driven prediction model only relies on the statistical laws of historical data, when encountering extreme or special operating conditions that are not covered by the training data, the prediction result may deviate significantly from the physical reality. To solve this problem, the present embodiment adopts a prediction model that deeply integrates the physical laws of the power grid, specifically a physical information neural network (PINN).

[0041] By way of example, the training process of the physical information neural network (PINN) is constrained by a loss function that includes a physical law residual loss . The expression of the loss function is as follows:

[0042]

[0043] wherein, the prediction accuracy loss based on the measured data, such as mean square error or mean absolute error, is used to ensure that the prediction output of the model is as consistent as possible with the historical true data; the physical law residual loss based on the Kirchhoff's law and the power flow equation, which embeds the physical laws of the power grid, such as power conservation and voltage-current relationship, as soft constraints into the training process of the neural network. is a weight factor of an adjustable physical constraint, used to balance the weight between data fitting accuracy and physical law satisfaction.

[0044] Optionally, by minimizing the loss function so that the prediction model not only fits the historical operation data with high accuracy, but more importantly, it can generalize to those physically possible but not present in the historical data. For example, when the system appears a new load and new energy combination mode, the pure data-driven model may fail, while the physical information neural network will give a more reasonable prediction result that meets the basic operation rules of the power grid, greatly enhancing the generalization ability and robustness of the model.

[0045] Optionally, after processing the input real-time and historical data sequence, the prediction model outputs a multi-dimensional state prediction result about the system's running state in the future period. The prediction result is not a single value, but a set containing multiple key physical quantities. Specifically, the multi-dimensional state prediction result at least includes a load power prediction value, a load power change rate prediction value, and an energy net output change prediction value.

[0046] The load power prediction value gives an estimate of the total load demand at the future time; the load power change rate prediction value predicts the speed and direction of load change at the future time; and the energy net output change prediction value integrates the power changes of photovoltaic, wind power, energy storage and grid interface, reflecting the overall trend of energy side output. These prediction results provide a forward-looking data basis for subsequent evaluation and decision-making.

[0047] As Figure 2 shows the curve of the real load power changing with time and the prediction results of the future load power by three different prediction models, where the black solid line represents the real load change, the red dashed line represents the prediction result of the traditional time series model, the blue dotted line represents the output of the LSTM model based on deep learning, and the green solid line corresponds to the physical information neural network model proposed in this patent that integrates the physical laws of the power grid.

[0048] It can be observed that the waveform of the red dashed line lags behind the real load curve as a whole, and there is a significant deviation at the inflection points of the load rising and falling, which shows that the pure time series method is difficult to capture the transient characteristics; although the blue dotted line is relatively close to the real curve in trend, there are still fluctuations and over-adjustment phenomena in the period of rapid load change, which shows that its prediction stability is insufficient when facing the coupling of new energy output and load disturbance.

[0049] The green solid line almost coincides with the black solid line, only with a slight deviation in the local disturbance area, and the overall waveform is smooth and continuous, which shows that the PINN model that combines Kirchhoff's law and power flow constraints can embed the physical laws of the power grid outside the data, thereby significantly improving the adaptability and prediction generalization ability of the model to abnormal operating conditions. Figure 2 The close-to-real-value and non-obvious delay characteristics of the medium green curve intuitively prove that the transition from passive response to active pre-control is realized through the physical information neural network, enabling the system to perceive the power fluctuation trend in advance and provide more accurate and reliable forward-looking data support for subsequent rolling optimization and coordinated voltage regulation.

[0050] Step S3, based on the multi-dimensional state prediction result, calculate the coordinated impact coefficient for quantifying the coupling effect between load demand fluctuation and energy supply fluctuation.

[0051] The core innovation of this step is to propose an index that can accurately quantify the impact intensity faced by the system, namely the coordinated impact coefficient . Traditional control strategies often only focus on the fluctuation of one side of the load or energy, ignoring the coupling amplification effect of both. This embodiment unifies the impact of double-sided fluctuations into a quantitative index through the coordinated impact coefficient.

[0052] For example, the calculation formula of the coordinated impact coefficient is as follows:

[0053]

[0054] Wherein, the absolute value of the load change rate prediction value in the formula is directly taken from the prediction result in step S2, which quantifies the future dynamic intensity of the load side; the absolute value of the energy net output change rate prediction value is calculated based on the energy net output change prediction value, which quantifies the future fluctuation intensity of the energy supply side; the energy reliability coefficient is a preset weight factor, the value of which can be set according to factors such as the penetration rate of new energy in the system and the reliability of weather prediction. For example, in a high-penetration photovoltaic system, if the weather forecast predicts that there will be rapidly moving clouds in the future, the value of at this moment should be increased accordingly to amplify the contribution of energy side fluctuation to the overall impact coefficient, because the unreliability of the energy side is higher at this moment.

[0055] Optionally, the coordinated impact coefficient The calculation essence is to fuse the two different sources and different nature disturbances of load demand fluctuation and energy supply fluctuation through a unified mathematical framework, and the calculation result is a non-negative scalar value. The greater the value, the more severe the imbalance impact on the system from both supply and demand sides in the future short time, and the higher the risk of voltage instability. This provides accurate and quantitative decision-making basis for subsequent differentiated and hierarchical control strategies.

[0056] As Figure 3 The variation law of the system synergistic impact coefficient under different load change rates and energy output change rates is shown in the form of a three-dimensional curved surface. The horizontal axis represents the power change rate on the load side, the vertical axis represents the output change rate on the energy side, and the vertical axis represents the synergistic impact coefficient formed by the coupling of the two. The color increases from blue to red, corresponding to the change of the impact intensity of the system from slight to severe.

[0057] It can be seen that when both sides fluctuate slightly, i.e. Figure 3 When the load fluctuation or energy fluctuation increases, the surface rapidly rises and presents a significant nonlinear amplification effect, and the color transitions to the yellow and orange-red regions, indicating that the supply and demand fluctuations will cause the system voltage deviation to increase significantly when superimposed.

[0058] Especially when the load change rate exceeds about 0.7 and the new energy output change rate is close to 1.0, the top region of the surface presents a deep red peak, representing a sharp rise in the synergistic impact coefficient, and the system will face the risk of short-time voltage instability or even out-of-limit. This trend is consistent with the coupling amplification effect between load demand fluctuation and energy supply fluctuation described above, verifying that the synergistic impact coefficient as a core quantitative indicator can accurately reflect the influence of future supply and demand coupling on voltage stability.

[0059] Figure 3 The continuous color gradient from cold to hot not only reveals the dynamic evolution process of the system state from steady state to high-risk state, but also proves that the quantification and stratification of impact levels are achieved through the coefficient, providing a clear basis for the subsequent hierarchical synergistic voltage regulation strategy in the implementation example.

[0060] Step S4, a rolling optimization model is constructed and solved with a composite cost function as the objective function, generating a synergistic voltage regulation strategy that defines the response priority of different regulation units.

[0061] This step is the decision-making center of the whole method, which receives the prediction results from step S2 and the impact coefficient from step S3, and generates the optimal control instruction. The core idea is to construct the voltage regulation problem as a rolling horizon optimization problem, and in each decision-making period, an optimal control sequence in a finite time domain is solved based on the latest system state prediction, and the first control quantity is applied to the system.

[0062] For example, when constructing the rolling optimization model, the objective function used is a composite cost function The design of this function takes into account the economy, safety and equipment life of the system, rather than a single voltage quality indicator. The expression of the composite cost function is as follows:

[0063]

[0064] wherein, is the predicted total energy consumption of the system, which includes network loss, energy storage charging and discharging loss, etc., minimizing this aims to improve the overall energy efficiency of the system; the absolute difference between the predicted node voltage and the target voltage directly reflects the key indicator of power quality, i.e. voltage stability, minimizing this is to ensure that the voltage is maintained within the qualified range; the depreciation cost of the cycle life of the energy storage system is a function related to the depth and rate of energy storage charging and discharging, which is introduced to quantify the device aging cost caused by energy storage action, and to avoid sacrificing the long-term operation economy due to excessive use of energy storage; , , are preset weight coefficients, which determine the trade-off relationship between energy consumption, voltage quality and energy storage life, which are sometimes in conflict with each other.

[0065] Optionally, by solving the rolling optimization model, a set of optimal control parameters can be dynamically output. This set of parameters constitutes the specific instruction set of the coordinated voltage regulation strategy, mainly including the voltage set value of the voltage stabilizer, the charging and discharging power instruction of the energy storage system and the power exchange plan value of the grid interface. For example, the optimization result may indicate that the voltage stabilizer will output voltage fine-tuning to a new set point, while instructing the energy storage system to charge at a certain power to absorb excess power, and adjusting the exchange power with the grid to the optimal value.

[0066] Optionally, the process of generating the coordinated voltage regulation strategy is further refined based on the coordinated impact coefficient calculated in step S3 to achieve fine allocation of control resources. Specifically, this process includes: dividing the calculated coordinated impact coefficient The preset cooperative control matrix defines the response priority of different regulating units under different impact levels. The core idea of the matrix is to intelligently and hierarchically call the most suitable regulating resource combination according to the severity level of the impact, so as to achieve the best balance between regulating cost and regulating effect.

[0067] For example, if the cooperative impact coefficient is in the low impact level, it indicates that the system is facing relatively gentle fluctuations. At this time, the cooperative control matrix will only activate the voltage stabilizer to perform high-frequency voltage fine tuning. As a mechanical or power electronic device, the voltage stabilizer has fast response speed and is suitable for fine voltage correction, and for small and frequent fluctuations, the mechanical wear or switching loss is within an acceptable range. By only enabling the voltage stabilizer, the precise control of power quality can be achieved while avoiding unnecessary consumption of life-sensitive devices such as energy storage systems.

[0068] Alternatively, if the cooperative impact coefficient is in the medium or high impact level, it indicates that the system is facing or will face a large power impact. At this time, if the voltage stabilizer is still used for regulation, it may not be able to suppress voltage fluctuations due to its limited regulating capacity or speed, and even may accelerate device damage. Therefore, the cooperative control matrix will activate the corresponding regulating units in a preset priority order, i.e., the voltage stabilizer adjustment first, the energy storage system charging and discharging second, and the new energy and grid power dispatching last.

[0069] As Figure 4 shown, the node voltage dynamic response process of the power system under different control strategies after suffering power impact is shown, the horizontal axis represents time, and the vertical axis represents the standard value of node voltage. The black dotted line is the target voltage reference value, the red solid line represents the voltage response curve of the traditional PID or fixed parameter control, and the green solid line represents the response result under the rolling optimization cooperative voltage regulation strategy based on load prediction proposed in this embodiment.

[0070] From Figure 4It can be seen that when the system is disturbed by a sudden load surge at about 2 seconds, the red curve shows a significant drop and repeated oscillation, with the lowest point once falling to about 0.86, and then experiencing multiple over-regulation and overshoot, and gradually stabilizing to the rated value after 10 seconds. This shows that the traditional method cannot compensate for the load impact in time under dynamic conditions due to fixed control parameters. The green curve responds quickly under the same disturbance conditions, with a voltage drop of only about 0.92, smooth waveform and no obvious oscillation, and returns to the target voltage at about 4 seconds, and then stabilizes at a slight fluctuation around 1.0. This response characteristic is completely consistent with the rolling optimization model described above, which targets a composite cost function, indicating that by combining load prediction results and coordinated impact coefficients, the system can dynamically adjust the control parameters of the voltage stabilizer, energy storage and grid interface in each decision cycle, achieving predictive compensation and multi-resource coordinated control.

[0071] Figure 4 The rapid convergence and small overshoot of the green curve compared to the red curve directly prove the significant advantages of the rolling optimization strategy in energy efficiency, response speed and steady-state accuracy. This difference not only reflects the forward-looking allocation ability of the optimization algorithm for impact energy, but also shows that the multi-objective weighting mechanism of the composite cost function effectively suppresses voltage oscillation and energy waste, balancing stability and economy.

[0072] Alternatively, under this strategy, the voltage stabilizer will still be used for preliminary and rapid voltage adjustment. Then, the energy storage system will be instructed to provide rapid power buffering to absorb the main power impact. Energy storage systems, especially battery storage with high-performance inverters, have millisecond-level response speed and can effectively smooth power deficits or surpluses, making them the core strength in dealing with medium and high impacts. Finally, for sustained and large-scale power imbalances, the output power of the new energy generation unit (such as active power reduction) or the power exchange value of the grid interface is adjusted to adjust the power baseline. This hierarchical and differentiated coordinated voltage regulation strategy ensures that the most suitable and economical combination of regulation resources is always used to undertake regulation tasks when dealing with different intensity impacts, achieving the optimal balance of system stability, economy and equipment life.

[0073] For example, in this embodiment, two impact coefficient thresholds are set, a low threshold and a high threshold These two thresholds divide future impacts into three levels and correspond to three different coordinated control strategies:

[0074] First level, small impact level: when the predicted ​At this time, the system determines that future fluctuations are very weak and belong to normal system background noise. Activating heavy-duty regulating equipment such as on-load tap changers (OLTC) or large-scale energy storage at this point would not only be costly but also cause unnecessary equipment damage. Therefore, the strategy of the hierarchical collaborative control matrix output is self-regulation of the source-grid-load-storage system. Specifically, it means not activating the main active voltage regulating equipment, but relying on the system's own damping characteristics and the autonomous response capabilities of each unit to smooth out these minor fluctuations. This includes: the impedance voltage drop effect of the grid lines themselves; the voltage sensitivity of some flexible loads (power naturally decreases when the voltage is slightly lower); the rapid frequency / voltage support provided by the droop control or virtual synchronous machine control of the renewable energy inverters; and the weak damping support that the energy storage system may provide in standby mode. This strategy minimizes active intervention, thereby reducing operating energy consumption and equipment depreciation to a minimum.

[0075] Level 2, Medium Impact: When predicted satisfy ≤ < When the system determines that future fluctuations have exceeded its self-regulating capabilities and require active intervention, but the intensity of the impact is still within the handling capacity of conventional regulating equipment, the control matrix output strategy is to activate the voltage regulator as the main regulating unit. This voltage regulator can be a traditional on-load tap changer (OLTC) or a static var compensator (SVC). The core of this strategy is to assign the primary task of voltage regulation to the voltage regulator. The central controller calculates the required adjustment level or compensation capacity of the voltage regulator based on the predicted voltage deviation and issues a pre-adjustment command in advance. Because the adjustment cost of such equipment is relatively low compared to energy storage, but the response speed is slow (e.g., OLTC level switching takes several seconds), it is very suitable for handling fluctuations of moderate amplitude and relatively mild changes.

[0076] Level 3, Severe Impact Level: When predicted ≥ When the system determines that a severe power impact will occur in the future, this impact can cause a large and rapid transient voltage drop or rise. At this time, if only the slow-responding voltage regulator is relied on, it can not be able to pull the voltage back to the safe range in time, causing the transient voltage to exceed the limit. Therefore, the strategy of the control matrix output is to start the voltage regulator and the energy storage system to cooperate in voltage regulation. Under this strategy, the energy storage system with extremely fast response (millisecond level) will play the role of a "transient impact absorber". It will quickly release or absorb power at the first time of the impact, quickly flatten the peak of the impact, and give the slow-responding voltage regulator valuable regulation time. Subsequently, the voltage regulator will calmly switch gears according to the voltage curve smoothed by the energy storage to adjust the system to a new steady-state operating point. This combination of fast and slow, long and short, cooperative strategy not only guarantees the voltage quality in the transient process, but also avoids the expensive energy storage system from bearing the steady-state regulation task for a long time, achieving a perfect combination of regulation performance and economy.

[0077] The two thresholds and are key parameters that can be adjusted online according to the safety requirements of the actual power grid, equipment cost and operation experience, giving the method high flexibility.

[0078] As Figure 5 shows the power response characteristics of various types of regulation units under low, medium and high impact levels. The vertical axis is the per-unit value of the output power, and the horizontal axis is time. The blue curve represents the response of the voltage regulator, the green curve represents the power change of the energy storage system, and the red curve represents the output adjustment of the new energy unit.

[0079] In the case of low impact level, it can be seen in the upper subgraph that the blue voltage regulator curve only oscillates slightly, and the power regulation peak is about 0.05, while the green and red curves are almost close to the zero axis, indicating that the system mainly relies on the voltage regulator to perform high-frequency voltage fine tuning to maintain power quality.

[0080] When the impact level rises to medium, as shown in the middle subgraph, the regulation amplitude of the blue voltage regulator curve increases significantly, and the green curve of the energy storage system begins to appear a phase-leading compensation action, with a power peak of about 0.05. The red curve of the new energy unit also shows slight dynamic participation, indicating that the system has automatically switched from single control to voltage regulator-energy storage cooperative mode, achieving dynamic coordination of flexible regulation and energy buffering.

[0081] Under high impact level, as shown in the lower subgraph, the green energy storage curve becomes dominant, with a power peak of about 0.25 and the shortest response time. The blue curve of the voltage regulator maintains moderate participation, and the red curve of the new energy unit appears significant power fluctuations to assist the system in returning to stability.

[0082] The hierarchy of this three-stage curve is completely consistent with the hierarchical response logic of the collaborative control matrix described in this application, which prioritizes voltage stabilizers, followed by energy storage, and finally new energy dispatch. Especially in high impact scenarios, the rapid and high amplitude response of the green curve shows that the energy storage system effectively suppresses voltage drop through transient power absorption, giving the slower responding voltage stabilizer time to adjust, thereby avoiding voltage out-of-limit and reducing equipment fatigue. The lagging adjustment of the red new energy curve reflects the economic orientation of the control strategy in this embodiment, that is, to let low-cost, slow-response units undertake steady-state regulation tasks.

[0083] These three curves prove the rationality and effectiveness of the hierarchical collaborative impact coefficient control in this embodiment: under different disturbance intensities, the system can intelligently select the most suitable combination of adjustment resources to achieve a multi-objective balance of control speed, economy, and equipment life, thereby forming a clear hierarchical and efficient dynamic voltage regulation system.

[0084] Step S5, execute the collaborative voltage regulation strategy and send control instructions to the voltage stabilizer, energy storage system, new energy generation unit, and grid interface.

[0085] This step is the final execution phase of decision-making. The strategy execution module converts the optimal control parameters generated in step S4 into specific control instructions that can be recognized by each adjustment unit, and issues them through the communication network.

[0086] For example, the instruction sent to the voltage stabilizer is usually a new voltage reference set value, and the local controller of the voltage stabilizer will adjust the OLTC or power electronic converter modulation strategy according to this set value. The instruction sent to the energy storage system is a specific charge and discharge power value, and the energy storage converter will control the battery charge and discharge process according to this power instruction. The instruction sent to the new energy generation unit may be an active power output limit or adjustment value, allowing it to participate in system voltage regulation. The instruction sent to the grid interface is the expected grid-connected power value, which is achieved through communication with the upper-level energy management system or the grid dispatch center.

[0087] Optionally, the entire system executes the process from step S1 to step S5 in a fixed control cycle, such as several times or tens of times per second. This combination of rolling optimization and closed-loop control allows the system to continuously perceive changes in the external environment and dynamically adjust the control strategy based on the latest prediction information, thereby maintaining an optimal or sub-optimal operating state at all times in a highly uncertain operating environment.

[0088] To sum up, the embodiment details a high-efficiency and economical dynamic voltage regulator method through the complete closed loop of steps S1 to S5. The method realizes accurate forward-looking prediction through a physical information neural network, quantifies system risks through a coordination impact coefficient, and generates hierarchical coordination strategies through rolling optimization of a composite cost function, thereby effectively improving the voltage stability, operation economy, and equipment service life of a power system with a high proportion of new energy power.

[0089] Embodiment Two

[0090] Based on the basic scheme of Embodiment One, this embodiment further introduces a mechanism for quantifying and actively managing prediction uncertainty, aiming to improve the decision robustness of the system in the face of highly random environments. The core of this embodiment is to convert the uncertainty information output by the prediction model into a quantifiable risk indicator, and dynamically adjust the weights of various factors in the optimization objective accordingly, so that the system can adaptively switch between risk aversion and energy efficiency optimization.

[0091] Optionally, the method described in this embodiment, when specifically implemented, adds a risk assessment and strategy optimization link based on prediction uncertainty between step S2 and step S4 of Embodiment One.

[0092] Illustratively, in step S2, the prediction model used in this embodiment, such as the physical information neural network PINN, is further enhanced in function. It not only generates point estimates of multi-dimensional state prediction results of loads and energy, such as load power prediction values, load change rate prediction values, etc., but also further generates probability distributions of these prediction results. This means that for future load power, the model no longer gives a single numerical value, but a range of possible values and their corresponding probability density, for example, in the form of a probability distribution function or quantile. This probability distribution clearly reveals the degree of uncertainty of the prediction results; the more dispersed the distribution, the higher the uncertainty of the prediction.

[0093] Optionally, one specific implementation of generating a probability distribution is to parameterize a probability distribution, such as a Gaussian distribution, in the output layer of the physical information neural network PINN, i.e., to output both the mean μ and the variance σ² of the prediction value. The mean μ represents the most likely prediction value, while the variance σ² quantifies the uncertainty of the prediction. By performing corresponding probabilistic training on the network, such as using the negative log-likelihood as part of the loss function, the model can learn to evaluate its own prediction confidence while fitting the data.

[0094] Optionally, after obtaining the probability distribution of the multi-dimensional state prediction results, before entering step S4 and solving the rolling optimization model, this embodiment adds a key step: based on the statistical characteristics of the probability distribution, a randomness risk coefficient The randomness risk coefficient is used to accurately quantify the overall uncertainty level of the multi-dimensional state prediction result.

[0095] An effective way to calculate the randomness risk coefficient is to analyze the prediction interval width of the key state variable. For example, the difference between the 95th percentile and the 5th percentile of the load power prediction value at the next few key time points can be calculated, and this difference is integrated or averaged over the prediction time domain. The larger the difference range, the more uncertain the model is about the future load, and the higher the value of the corresponding randomness risk coefficient . Similarly, the same analysis can be performed on the energy net output prediction value, and the uncertainty measures on both sides of the load and energy are fused and finally normalized to a scalar between zero and one, i.e., the randomness risk coefficient .

[0096] Optionally, the calculation formula of the randomness risk coefficient may be specifically represented as: which is a function positively related to the prediction interval width. For example:

[0097]

[0098] wherein and represent the 95th percentile and the 5th percentile of a certain key state variable (such as load power) at time ; is a normalization coefficient, so that the value of ranges between zero and one. The system comprehensively considers the statistical characteristics of the prediction uncertainty of multiple state variables, so as to comprehensively and objectively reflect the overall risk level of the future operation state of the system.

[0099] Optionally, after calculating the randomness risk coefficient , the core innovation point of the present embodiment is that the weight coefficients , , , in the composite cost function are dynamically adjusted according to the numerical value of the risk coefficient. This makes the optimization objective of the system no longer fixed, but an intelligent objective that can evolve adaptively with the environmental risk level.

[0100] When the uncertainty is high, the system is more likely to deviate from the expected state, and the risk of voltage fluctuation and overcharge / overdischarge of the energy storage is higher. In this case, the control strategy should be more conservative and robust, i.e. more emphasis should be put on the safety and the life of the equipment. Therefore, the system will increase the weight coefficient for characterizing the voltage deviation and the depreciation of the energy storage life and . Increasing means that the optimization model will more severely punish the situation of voltage deviation from the target value, forcing the system to use more resources to maintain voltage stability. Increasing means that the model will cherish the cycle life of the energy storage system more, and avoid making aggressive charging and discharging decisions that may cause the energy storage to accelerate aging in the case of high uncertainty. Through this adjustment, the coordinated voltage regulation strategy output by the rolling optimization model will be explicitly biased towards risk aversion.

[0101] Alternatively, when the randomness risk coefficient decreases, it indicates that the prediction result is more reliable, the uncertainty is small, and the future state of the system is more predictable. In this relatively safe and certain environment, the control strategy can be more biased towards pursuing economic operation. Therefore, the system will appropriately reduce the weight coefficient for characterizing the total energy consumption of the system . Reducing means that the optimization model can tolerate slightly higher network loss or regulation energy consumption in exchange for more economical operation, such as reducing the purchase of electricity from the grid at a higher price, making more full use of new energy, or allowing the energy storage system to charge and discharge at a more economical time. This enables the coordinated voltage regulation strategy to be biased towards energy efficiency optimization at this moment.

[0102] Alternatively, the specific adjustment method of the weight coefficient can be a preset mapping table corresponding one-to-one to the value, or a continuous linear or nonlinear function. For example, it can be set as:

[0103]

[0104] wherein, is the basic weight; is the amplification coefficient. Through this dynamic weight mechanism, the embodiment successfully converts the abstract concept of prediction uncertainty into specific and executable control parameters in the optimization model.

[0105] As Figure 6The change trends of voltage deviation and energy consumption when the system adopts two control strategies of fixed weight and risk adaptive weight under different randomness risk coefficients are shown, the horizontal axis is the randomness risk coefficient, the left vertical axis represents the average voltage deviation of the node, and the right vertical axis represents the normalized value of the total energy consumption of the system. The red curve group corresponds to the voltage stability index, and the blue curve group corresponds to the energy consumption index. The solid line is the adaptive weight strategy proposed in the embodiment, and the dashed line is the traditional fixed weight strategy.

[0106] As can be seen from Figure 6 In the low risk interval (risk coefficient less than 0.3), the blue solid line is obviously lower than the blue dashed line, indicating that the adaptive strategy can effectively reduce the system energy consumption by about 30%, and the red solid line and the red dashed line are almost coincident, indicating that the voltage deviation remains the same level. At this stage, the system mainly operates economically.

[0107] When the risk coefficient rises to a medium-high level (0.5 to 0.8), the voltage deviation of the red solid line rises significantly less than that of the red dashed line, and finally maintains at about 0.08, while the voltage deviation under the fixed weight control exceeds 0.1, indicating that the method of the embodiment can actively increase the weight of voltage stability when uncertainty increases, so that the system maintains higher robustness; at the same time, the blue solid line rises slightly, reflecting that the system actively sacrifices part of the energy efficiency to ensure power quality when the risk rises, which embodies the adaptive characteristics of the dynamic adjustment of the weight parameter in the composite cost function.

[0108] When the risk coefficient approaches 1, the voltage deviation control effect is still smooth, and the energy consumption increases limitedly, indicating that the method of the embodiment still realizes a reasonable compromise between energy consumption and safety under high-risk working conditions.

[0109] The overall curve trend is completely consistent with the description of the embodiment: when the prediction result credibility decreases, the system improves the weight of voltage stability and energy storage protection through the risk adaptive weight adjustment mechanism, and the control strategy is shifted from energy efficiency optimization to risk avoidance, realizing dynamic rebalancing of the control target under different operating environments.

[0110] Figure 6 The difference between the red and blue solid lines and the dashed lines in the embodiment is the key advantage of the method compared with the traditional method: it not only realizes economic operation under low risk, but also maintains stable power supply under high risk, thereby achieving an optimal balance between economy and safety.

[0111] In summary, the embodiment introduces probability prediction, randomness risk coefficient and dynamic weight adjustment mechanism, which endows the system of embodiment one with powerful risk management capability. It makes the system no longer mechanically execute optimization based on a single prediction value, but intelligently perceives the degree of certainty of the environment and flexibly adjusts its decision focus accordingly, so as to realize a higher level of adaptive balance between stability, economy and equipment longevity in the highly volatile new energy power system.

[0112] Embodiment three:

[0113] On the basis of the active prediction and optimization control system constructed in embodiment one and embodiment two, the embodiment further introduces a set of parallelly executed fault monitoring and function replacement fault-tolerant mechanism. The mechanism aims to deal with two types of core risks:

[0114] One is the decrease of decision reliability caused by the inaccuracy of the prediction model under special working conditions;

[0115] Two is the sudden failure of the key execution unit, especially the voltage stabilizer.

[0116] The core of the embodiment is that when the core function unit of the system fails or the prediction credibility is insufficient, the standby scheme can be started immediately, and the key control functions can be temporarily taken over by other healthy units, so as to greatly improve the robustness and power supply continuity of the entire power system.

[0117] Optionally, the method described in the embodiment includes an independent fault monitoring and fault-tolerant triggering process parallelly executed with steps S1 to S5. The process runs continuously in the background, and its monitoring and decision-making cycle can be the same as or shorter than the main control loop to ensure the timeliness of the response.

[0118] Illustratively, the primary task of fault monitoring is to evaluate the real-time prediction confidence of the system state. The specific implementation is to calculate the deviation between the multi-dimensional state prediction results generated in step S2 and the actual values of the corresponding physical quantities collected by the sensors deployed in the system. For example, the predicted load power is calculated by the difference value of the actual measured load power, and the prediction deviation of multiple state quantities is evaluated, for example, a normalized root mean square error or average absolute percentage error is calculated.

[0119] Optionally, based on the deviation, the system calculates a quantitative system state prediction confidence in real time. The confidence is a value between zero and one, and the higher the value, the better the prediction result matches the actual situation, and the more reliable the current output of the model. Its calculation can take into account the statistical characteristics of short-term historical deviations, for example, if the prediction error of the recent consecutive periods is lower than the preset threshold, the confidence is maintained at a high level; otherwise, if the prediction error continues to increase or appears a sudden peak, the confidence decreases accordingly.

[0120] Optionally, the system presets a confidence threshold. When the real-time calculated prediction confidence is lower than the preset threshold, it indicates that the prediction model may be temporarily disabled due to the current operating state exceeding its learning range or abnormal data. If the control instructions based on this unreliable prediction are continued to be used, it may lead to system operation risks. At this time, the system will trigger the function replacement fault-tolerant mechanism.

[0121] Optionally, another path of fault monitoring is to directly detect the hardware failure of any regulation unit. This can be judged by the self-diagnosis signal of the device, communication interruption alarm, or sustained abnormal deviation of the actuator feedback signal and the instruction value. For example, if a voltage regulation instruction is sent to the voltage regulator, but no corresponding change in the voltage set value is detected within the specified time, or over-temperature, over-current, and other fault codes are detected from the voltage regulator, it is determined that the voltage regulator has failed. Similarly, once such a failure is detected, the function replacement fault-tolerant mechanism will be triggered immediately.

[0122] Illustratively, the core of the function replacement fault-tolerant mechanism is that when the key regulation unit fails, the system can intelligently reconfigure the control resources and instruct other healthy units with corresponding potential to temporarily take over the core functions of the failed unit. A crucial application scenario is that when the voltage regulator fails, the system immediately instructs the inverter of the energy storage system to switch to a special high-frequency control mode to temporarily simulate the fast voltage adjustment function of the voltage regulator, thereby ensuring the power quality of the key load node.

[0123] Optionally, the feasibility of such function replacement is due to the high controllability and fast response capability of modern energy storage converters. Normally, the energy storage converter mainly executes power scheduling instructions, and its control bandwidth focuses on achieving accurate output of active and reactive power, which is a relatively low-frequency control target. In the fault-tolerant mode of the present embodiment, the control target is temporarily reconfigured, and in addition to completing the basic power scheduling task, an additional high-frequency control loop with the goal of fast and accurate voltage tracking is added.

[0124] Optionally, when the function replacement fault-tolerant mechanism is triggered, its specific implementation includes a refined dynamic resource management and control process. First, the system needs to calculate the dynamic regulation margin of the energy storage system in real time The dynamic regulation margin is used to accurately quantify the remaining control capability of the energy storage system inverter that can be used to perform high-frequency voltage fine-tuning under the premise of meeting the current main power scheduling task.

[0125] By way of example, the dynamic regulation margin The calculation needs to consider multiple constraints. The core is the difference between the apparent capacity limit of the energy storage converter and the current actual output power, while also considering the charge-discharge rate limit of the energy storage battery itself, the current state of charge SOC to avoid overcharging or overdischarging, and the heat dissipation capability of the converter itself. The system calculates a safe power and current range that can be used for additional voltage control tasks through a real-time updating algorithm, which comprehensively considers these constraints. The size of the range is quantified as the dynamic regulation margin The larger the margin value, the greater the potential of the energy storage system to temporarily assume the function of the voltage stabilizer.

[0126] Optionally, after obtaining the dynamic regulation margin , the system executes a specially designed dual-mode fusion control algorithm in the inverter control logic of the energy storage system. The essence of the algorithm is that it superimposes the low-frequency control signal originally used by the inverter to respond to the energy management system, which focuses on steady-state accuracy, with a newly generated high-frequency control signal specifically designed to quickly suppress voltage fluctuations in the control layer.

[0127] By way of example, the low-frequency control signal is derived from the instructions for energy storage charge and discharge power in the coordinated voltage regulation strategy generated in step S4, which changes relatively smoothly and aims to complete the system's energy balance and power baseline adjustment. The high-frequency control signal is generated based on the deviation of the key node voltage from the target voltage through a high-gain proportional-integral or more advanced controller, and its response speed is extremely fast, aiming to compensate for the voltage deviation within milliseconds.

[0128] Optionally, the amplitude of the high-frequency control signal is not arbitrarily set, but is strictly constrained by the dynamic regulation margin The system will set an amplitude limiting module to ensure that the total control instruction after superposition does not exceed the safe operating boundary of the dynamic regulation margin of the inverter. This design ensures that the energy storage system will not be damaged by overloading when performing its "temporary voltage stabilizer" function, achieving the maximum simulation of the original voltage stabilizer function within the safe operating boundary of the energy storage system.

[0129] Optionally, if it is found through real-time monitoring that the power impact of the system is further intensified, or the SOC of the energy storage system is in an extreme state, resulting in a calculated dynamic regulation margin Below the preset fault-tolerant lower limit, which indicates that the energy storage system has been unable to bear the voltage support task alone. At this time, the system will automatically enter the degraded operation mode. In this mode, the system will actively adjust the power supply strategy for non-critical loads, for example, according to the preset priority order, cut off part of the non-critical loads, reduce the total power demand of the system, and reduce the adjustment pressure, so as to preferentially guarantee the power supply quality and continuity of key loads such as hospitals and data centers. This is a strategy of sacrificing pawns to protect the king, which is the last guarantee for maintaining the minimum stable operation of the system in extreme failure conditions.

[0130] As Figure 7 The dynamic response process of the system voltage and the recovery effect after the energy storage system takes over the regulation are shown in the case of sudden failure of the voltage stabilizer. The horizontal axis is time, the vertical axis is the node voltage per unit, the black dashed line represents the rated target voltage, and the blue solid line represents the actual voltage change of the system.

[0131] As Figure 7 can be seen, in the first 5 seconds, the system is in the normal operation stage, the blue curve is almost coincident with the black dashed line, and the voltage is stably maintained at 1.0 with slight fluctuations, indicating that the voltage stabilizer is operating well. When the time reaches about 5 seconds, the voltage stabilizer fails, the voltage drops significantly and is accompanied by a short shock, with a minimum of about 0.88, which corresponds to the situation described above that the confidence level drops sharply to trigger the fault-tolerant mechanism.

[0132] Subsequently, at about 8 seconds, the energy storage system is instructed to switch to the high-frequency control mode, quickly taking over the voltage stabilization function. The blue curve quickly rises in a few seconds and tends to approach the target value, with smooth waveform and no obvious overshoot, indicating that the energy storage inverter successfully performs the high-frequency voltage fine-tuning task. The time constant of the recovery stage is significantly smaller than the mechanical adjustment time of the voltage stabilizer, verifying the fast dynamic response capability of the energy storage system in the fault-tolerant mode.

[0133] The significant turning point in the curve before and after 8 seconds reflects the immediacy of the fault-tolerant control logic of the embodiment: when the system detects that the prediction confidence is below the threshold or detects an abnormality in the voltage stabilizer, the control center immediately calculates the dynamic adjustment margin of the energy storage and activates its high-frequency control loop within a safe range to simulate the voltage stabilization function, thereby avoiding sustained voltage drops or large fluctuations. The voltage curve stabilizes at 1.0 after 10 seconds, with a fluctuation amplitude of less than 0.02, indicating that the system after takeover not only restores the steady-state voltage level, but also achieves high power quality.

[0134] This Figure 7 The clear time sequence change directly proves that the embodiment method has the fault-tolerant self-healing property in the core device failure scenario: when the main regulation unit fails, the energy storage system can seamlessly take over within a few seconds, making the voltage recover smoothly and stably, thereby greatly improving the robustness and power supply continuity of the power system under abnormal working conditions.

[0135] In summary, the embodiment introduces parallel fault monitoring, prediction confidence evaluation and functional replacement fault-tolerant mechanism, which gives the intelligent voltage regulation system composed of the foregoing embodiments high toughness and survivability. It not only optimizes operation under normal conditions, but also realizes smooth switching and degradation self-healing of control function under abnormal conditions such as predicted failure or device failure, ensures the serious commitment to the reliability of power supply to key loads under any conditions, and greatly improves the basic safety level of the entire power system.

[0136] Embodiment four:

[0137] As shown in Figure 8 , corresponding to the method embodiment, the application also proposes a load prediction-based dynamic voltage regulator system, which comprises:

[0138] a data acquisition module for synchronously acquiring load-side data and energy-side data in a power system;

[0139] a state prediction module for processing the load-side data and the energy-side data based on a prediction model that integrates power grid physical laws, to generate multi-dimensional state prediction results of load and energy;

[0140] a decision and strategy generation module for calculating a coordinated impact coefficient based on the multi-dimensional state prediction results , and constructing and solving a rolling optimization model with a composite cost function as the objective function to generate a coordinated voltage regulation strategy that defines the response priority of different regulation units;

[0141] a strategy execution module for sending control instructions to voltage regulators, energy storage systems, new energy generation units and power grid interfaces in the system according to the coordinated voltage regulation strategy.

[0142] When the system is running, first, the data acquisition module starts working. This module synchronously acquires load-side data and energy-side data through a high-precision sensor network deployed in the power system. The load-side data includes load power and load power change rate ; the energy-side data includes photovoltaic output , wind power output , state of charge of energy storage system and grid power supply . The collected data is sent to the state prediction module in real time after preprocessing and synchronization alignment.

[0143] Then, the state prediction module receives real-time data from the data acquisition module. The module is built-in a prediction model that integrates the physical laws of the power grid, specifically a trained physical information neural network (PINN). This model processes the input real-time and historical data sequences and outputs multi-dimensional state prediction results about the future operating state of the system, including load power prediction values, load change rate prediction values, and energy net output change prediction values. The prediction results of this module provide key forward-looking information for subsequent decision-making.

[0144] Then, the decision and strategy generation module is activated. This module first calculates the coordinated impact coefficient based on the multi-dimensional state prediction results provided by the state prediction module, which accurately quantifies the coupling influence degree between load demand fluctuations and energy supply fluctuations. Subsequently, the module constructs and solves a rolling optimization model with a composite cost function as the objective function. During the solving process, the module considers the economy, safety, and equipment life of the system, dynamically outputs a set of optimal control parameters, and generates a coordinated voltage regulation strategy that defines the response priorities of different regulation units. According to the level of the coordinated impact coefficient, the strategy intelligently decides whether to activate only the voltage stabilizer for fine-tuning or to activate the energy storage system, new energy unit, and grid interface step by step or in parallel.

[0145] Finally, the strategy execution module receives the coordinated voltage regulation strategy from the decision and strategy generation module. The module converts the strategy into specific control instructions that each regulation unit can recognize and sends them to the voltage stabilizer, energy storage system, new energy generation unit, and grid interface through a high-speed communication network. The instruction content includes the voltage set value of the voltage stabilizer, the charge and discharge power of the energy storage system, the output power adjustment amount of the new energy unit, and the power exchange value of the grid interface.

[0146] The entire system executes the above process in a fixed control cycle, forming a complete closed loop from data perception, state prediction, intelligent decision-making, to precise execution. This closed-loop control enables the system to continuously adapt to changes in the external environment, maintaining voltage stability, economic operation, and equipment safety of the power system under highly uncertain operating conditions.

[0147] Embodiment Five:

[0148] Corresponding to the above embodiments, the present application also provides an electronic device.

[0149] As Figure 9As shown is a structural schematic diagram of an electronic device in the present application, the electronic device 200 comprises a processor 201 and a memory 203. Wherein, the processor 201 and the memory 203 are connected, such as connected through a bus 202. Optionally, the electronic device 200 can further comprise a transceiver 204. It needs to be explained that, in actual application, the transceiver 204 is not limited to one, and the structure of the electronic device 200 does not constitute a limitation to the embodiments of the present application.

[0150] The processor 201 can be a CPU, a general processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can realize or execute various exemplary logic blocks, modules and circuits described in combination with the disclosure. The processor 201 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0151] The bus 202 can comprise a channel for transmitting information between the above-mentioned components. The bus 202 can be a PCI bus or an EISA bus, etc. The bus 202 can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 9 In the present application, only one thick line is used to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0152] The memory 203 is used for storing a computer program corresponding to the load prediction based voltage stabilizer dynamic voltage regulation method of the above-mentioned embodiments of the present application, which is controlled and executed by the processor 201. The processor 201 is used for executing the computer program stored in the memory 203 to realize the content shown in the foregoing method embodiments.

[0153] Wherein, the electronic device 200 comprises but is not limited to: notebook computers, PAD (tablet computers) and other mobile terminals, and fixed terminals such as desktop computers and the like. Figure 9 The electronic device 200 shown is only an example, and should not bring any limitation to the function and use range of the embodiments of the present application.

[0154] It is to be appreciated that the above description and the examples that follow are intended to be illustrative only and that changes can be made to the description, either functionally or chronologically, as well as changes being made concerning which specific steps are disclosed as a preferred embodiment. Accordingly, the disclosure is not intended to be limited to the described or illustrated examples. It is therefore contemplated to cover by the claims any and all modifications coming within the scope of the present application.

[0155] It should be understood that aspects of the application can be implemented in hardware, software, firmware or a combination thereof. In the above embodiment, the various steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or their combinations can be used: a discrete logic circuit having logic gates for implementing logic functions on data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0156] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in an appropriate manner.

[0157] Furthermore, the terms "first", "second", etc. are used only for descriptive purposes and do not connote or imply relative importance or a quantity of the indicated technical features. Thus, a feature defined with "first", "second", etc. can include at least one of the features implicitly or explicitly. In the description of the present application, the meaning of "a plurality" is at least two, for example, two, three, etc., unless otherwise specifically defined.

[0158] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that changes, modifications, substitutions and variations can be made by those skilled in the art without departing from the scope of the present application.

Claims

1. A load prediction based dynamic voltage regulation method for voltage regulator, characterized in that, The method is applied to a power system including a voltage stabilizer, an energy storage system, a new energy power generation unit and a grid interface, and the method comprises: S1, synchronously collecting load side data and energy side data in the power system; S2, processing the load side data and the energy side data based on a prediction model that integrates power grid physical laws, to generate a multi-dimensional state prediction result of the load and the energy; S3、based on the multi-dimensional state prediction result, calculate the synergistic impact coefficient for quantifying the coupling effect between load demand fluctuation and energy supply fluctuation; S4, with a composite cost function A rolling optimization model is constructed and solved with the composite cost function as the objective function, to generate a coordinated pressure regulation strategy that defines the response priority of different regulating units. The process of generating the coordinated voltage regulation strategy is refined based on the coordinated impact coefficient, and comprises: said synergic impact coefficient is mapped to a preset synergic control matrix defining the priority of response of the different conditioning units; If the coordinated impact coefficient is at a low impact level, the coordinated control matrix only activates the voltage stabilizer to perform high-frequency voltage fine tuning, so as to realize accurate control of power quality; If the coordinated impact coefficient is at a medium impact or high impact level, the coordinated control matrix activates corresponding regulation units in a priority order of voltage stabilizer adjustment-energy storage system charging and discharging-new energy and grid power scheduling, in a step-by-step or parallel manner; S5, executing the coordinated voltage regulation strategy, and sending control instructions to the voltage stabilizer, the energy storage system, the new energy power generation unit and the grid interface.

2. The method of claim 1, wherein, The load side data includes load power and load power change rate ; The energy side data comprises photovoltaic power output , wind power output , state of charge of an energy storage system and grid supply power .

3. The method of claim 1, wherein, The synergistic impact coefficient The calculation formula is: ； wherein, is an absolute value of a predicted value of a change rate of the energy net output, is an absolute value of a predicted value of a change rate of the load power, is a preset energy reliability coefficient.

4. The method of claim 1, wherein, The composite cost function used in constructing the rolling optimization model in step S4 The expression is: ； wherein, is the predicted system total energy consumption, is the predicted node voltage, is the target voltage, is the cycle life depreciation cost of the energy storage system, , , is the preset weight coefficient; By solving the rolling optimization model, a set of optimal control parameters are dynamically output, including voltage set value of the voltage stabilizer, charging and discharging power of the energy storage system and power exchange value of the grid interface.

5. The method of claim 4, wherein, The method further comprises: In the step S2, the prediction model further generates a probability distribution of the multi-dimensional state prediction result of the load and the energy; and before solving the rolling optimization model in the step S4, a randomness risk coefficient is calculated based on statistical features of the probability distribution , the randomness risk coefficient being used to quantify the uncertainty degree of the multi-dimensional state prediction result; Further, according to the randomness risk coefficient , the weight coefficient in the compound cost function is dynamically adjusted , ; wherein, when the random risk coefficient is increased, the weight coefficient for characterizing the voltage deviation and the energy storage life depreciation is increased , so that the coordinated voltage regulation strategy output by the rolling optimization model is biased towards risk aversion; decreasing the weight coefficient when the random risk coefficient decreases to make the coordinated pressure regulating strategy favor the energy efficiency optimization.

6. The method of claim 1, wherein, The prediction model used in step S2, which integrates the physical laws of the power grid, is a Physical Information Neural Network (PINN). Its training process uses a residual loss mechanism that incorporates physical laws. loss function Apply constraints, the loss function The expression is: ； wherein, is a loss of prediction accuracy based on measured data, is a loss of physical law residual based on Kirchhoff's law and flow equations, is a weight factor of physical constraints.

7. The method of claim 1, wherein, The method further comprises a fault monitoring and fault tolerance triggering step executed in parallel with the steps S1 to S5, and the step comprises: By comparing the deviation between the multi-dimensional state prediction result generated in the step S2 and the real-time collected value, the prediction confidence of the system state is calculated in real time; If the prediction confidence is lower than a preset threshold value, or any regulation unit is detected to be faulty, a functional replacement fault tolerance mechanism is triggered; The functional replacement fault tolerance mechanism comprises: when the voltage stabilizer is faulty, instructing the inverter of the energy storage system to switch to a high-frequency control mode, to temporarily simulate the voltage adjustment function of the voltage stabilizer, and to guarantee the power quality of critical loads.

8. The method of claim 7, wherein, When the functional replacement fault tolerance mechanism is triggered, the specific implementation mode comprises: calculating, in real time, a dynamic regulation margin of the energy storage system , the dynamic regulation margin quantifying a remaining capability of the energy storage system inverter to perform high-frequency voltage fine tuning under current primary power dispatch tasks In the inverter control logic of the energy storage system, a dual-mode fusion control algorithm is executed, which superimposes a low-frequency control signal for main power scheduling and a high-frequency control signal for voltage fine tuning; wherein the amplitude of the high frequency control signal is constrained by the dynamic adjustment margin to achieve maximization of the simulation of the function of the voltage regulator within the safe operating boundaries of the energy storage system. If the dynamic adjustment margin is lower than a preset fault tolerance lower limit, the system enters a degraded operation mode, and actively adjusts the power supply strategy for non-critical loads.

9. A load prediction based dynamic voltage regulator system, characterized in that, Comprise: A data collection module for synchronously collecting load side data and energy side data in the power system; A state prediction module for processing the load side data and the energy side data based on a prediction model that integrates power grid physical laws, to generate a multi-dimensional state prediction result of the load and the energy; A decision and strategy generation module is configured to calculate a collaborative impact coefficient based on the multi-dimensional state prediction result , and to construct and solve a rolling optimization model with a composite cost function as an objective function, to generate a collaborative voltage regulation strategy defining response priorities of different regulation units, wherein the collaborative impact coefficient is used to quantify the coupling effect between load demand fluctuation and energy supply fluctuation; and the decision and strategy generation module maps the collaborative impact coefficient to a preset collaborative control matrix defining response priorities of different regulation units when generating the collaborative voltage regulation strategy; if the collaborative impact coefficient is at a low impact level, only the voltage stabilizer is activated to perform high-frequency voltage fine-tuning; if the collaborative impact coefficient is at a medium or high impact level, corresponding regulation units are activated in a step-by-step or parallel manner according to the priority order of voltage stabilizer adjustment-energy storage system charging / discharging-new energy and grid power dispatching. A strategy execution module for sending control instructions to the voltage stabilizer, the energy storage system, the new energy power generation unit and the grid interface in the system according to the coordinated voltage regulation strategy.

Images