A composite coordinated control system and method for new energy power generation

Abstract

The application discloses a new energy power generation composite coordinated control system and method, relates to the technical field of power system control, and comprises a multi-source heterogeneous data sensing and fusion module, a multi-time scale collaborative optimization decision module, an adaptive hierarchical control instruction distribution and reconstruction module, a local rapid response and virtual inertia simulation execution module and a closed-loop performance evaluation and parameter self-tuning module; the method comprises the following steps: collecting and fusing data to generate a data set; performing collaborative optimization calculation to generate a control instruction set; dynamically selecting an instruction distribution strategy and distributing the instruction; a control algorithm rapidly responds to a power instruction, and inertia support and damping characteristics are simulated and provided; the actual response of the system is continuously monitored, and key control parameters are adjusted. The application can improve the dynamic stability and regulation accuracy of the system, make up for the problem of the decreased anti-disturbance ability of the high-proportion new energy power grid, and greatly improve the utilization efficiency of adjustable resources and the overall operation economy of the system.

Description

Technical Field

[0001] This invention relates to the field of power system control technology, and in particular to a composite coordinated control system and method for new energy power generation. Background Technology

[0002] As the global energy structure accelerates its transformation towards cleaner and lower-carbon energy, the penetration rate of new energy power generation technologies in power systems continues to increase, with wind and solar energy becoming important components of modern power grids. New energy power generation possesses inherent characteristics such as intermittency, volatility, and weak inertia, posing severe challenges to the frequency stability, voltage regulation, and power balance capabilities of power systems when integrated into large-scale grids. Especially in areas with a high proportion of new energy integrated into the regional power grid, the traditional model relying on synchronous generators for inertia support and frequency regulation faces immense pressure, leading to increasingly prominent problems of decreased grid disturbance immunity and deteriorating transient stability.

[0003] In the process of building a new power system, large-scale power grid security defense systems and intelligent dispatch systems have become core supports for ensuring the reliability and resilience of power supply. However, existing dispatch and control systems are mostly designed based on the characteristics of traditional power sources, making it difficult to adapt to the coordinated regulation needs of multiple time scales and resource types brought about by the high penetration of new energy sources. On the one hand, new energy power generation units are mainly based on inverter interfaces and lack physical rotational inertia, making it difficult to provide effective frequency response and damping support for the system. On the other hand, wind power, photovoltaics, energy storage, and other heterogeneous resources are highly dispersed in time and space, with diverse control objectives. Traditional centralized or decentralized control architectures often struggle to balance global optimization and local rapid response, leading to frequent problems such as control hierarchy fragmentation, regulation dead zones, or command conflicts.

[0004] Furthermore, with the expansion of inter-regional power transmission, especially the commissioning of 750 kV and above AC transmission systems, the dynamic coupling of the power grid has become tighter, and local disturbances are more likely to trigger global frequency and power oscillations. Existing coordinated control methods are typically highly model-dependent and lack adaptability. When faced with complex operating conditions such as strong fluctuations in renewable energy output, communication delays, or switching of power grid operating modes, their control performance is prone to degradation, making it difficult to meet the requirements of highly resilient power grids for rapid response and adaptive reconfiguration. Therefore, how to construct a composite coordinated control system that can deeply integrate multi-source information, achieve multi-timescale coordination from second to hour, and possess intelligent learning and adaptive adjustment capabilities has become a key technical direction for improving the overall efficiency of intelligent dispatching systems and strengthening the security defense system of large-scale power grids. This is of great significance for ensuring the safe and stable operation of high-proportion renewable energy power grids and efficient energy consumption. Summary of the Invention

[0005] The purpose of this invention is to provide a composite coordinated control system and method for new energy power generation, in order to solve the problems in the existing new energy power generation control, such as hierarchical fragmentation leading to dead zones and over-adjustment, lack of multi-source coordination mechanism reducing the utilization rate of adjustable resources, strong model dependence resulting in performance degradation under non-ideal operating conditions such as sudden output changes and communication delays, and insufficient adaptability making it difficult to cope with dynamic switching of grid operation modes.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] On the one hand, a composite coordinated control system for new energy power generation includes the following components:

[0008] The multi-source heterogeneous data sensing and fusion module collects and fuses multi-dimensional operation data and status information from wind power, photovoltaics, energy storage and the grid side in real time, and generates a system panoramic dynamic sensing dataset with a unified spatiotemporal benchmark.

[0009] The multi-timescale collaborative optimization decision module is connected to the multi-source heterogeneous data perception and fusion module. Based on the system's panoramic dynamic perception dataset, it performs collaborative optimization calculations in parallel at three timescales: second, minute, and hour, generating a composite control instruction set that includes primary frequency regulation instructions, secondary frequency regulation instructions, and energy storage scheduling instructions.

[0010] The adaptive hierarchical control command distribution and reconfiguration module is connected to the multi-timescale collaborative optimization decision module. It receives the composite control command set and adaptively selects a centralized, distributed, or hybrid command distribution strategy according to the current power grid operation mode, communication status, and equipment availability. It dynamically distributes the control commands to each new energy power generation unit and energy storage system.

[0011] The local fast response and virtual inertia simulation execution module is deployed in the local controller of each new energy power generation unit and energy storage system. It receives and executes instructions from the adaptive hierarchical control instruction distribution and reconfiguration module, generates power regulation signals locally quickly, and simulates and provides inertia support and damping characteristics through virtual synchronous generator control algorithm.

[0012] The closed-loop performance evaluation and parameter self-tuning module is connected to the multi-source heterogeneous data sensing and fusion module and the multi-timescale collaborative optimization decision module. It continuously monitors the actual control effect of system frequency, voltage and power balance, evaluates control deviation based on preset performance indicators, and dynamically adjusts the key control parameters in the multi-timescale collaborative optimization decision module using reinforcement learning algorithms.

[0013] The multi-source heterogeneous data sensing and fusion module includes a data acquisition unit, a data preprocessing unit, and a spatiotemporal alignment fusion unit. The data acquisition unit synchronously collects frequency deviations through sensors and communication interfaces deployed at wind farms, photovoltaic power plants, energy storage power plants, and key nodes of the power grid. The data includes voltage amplitude U, active power P, reactive power Q, equipment operating status flags, and meteorological forecast data. The data preprocessing unit performs outlier detection and removal, noise filtering, and data integrity verification on the collected raw data. Outlier detection employs a local outlier factor algorithm, setting the number of neighboring points. Standard deviation multiple The spatiotemporal alignment and fusion unit maps preprocessed data with different sampling periods and delays to the same spatiotemporal coordinate system using a timestamp-based interpolation algorithm and a spatial association model based on a geographic information system, forming the system's panoramic dynamic perception dataset. The dataset's data structure is a time-series data matrix containing timestamps, device IDs, geographic location coordinates, and multi-dimensional state vectors.

[0014] Furthermore, the multi-timescale collaborative optimization decision-making module internally operates three optimization decision-making sub-units in parallel. The second-level optimization decision-making sub-unit response frequency deviation... Its core is a primary frequency modulation controller, which employs a composite control law combining improved virtual inertial control and droop control, and its output power command... Calculated by the following formula:

[0015]

[0016] in, For virtual inertia coefficients, The droop coefficient is... are integral coefficients, and , , The initial values ​​are adjusted online based on the rated capacity of the new energy units and the equivalent inertia requirements of the system. The minute-level optimization decision-making subunit, based on regional control deviation and ultra-short-term power prediction errors, employs a distributed model predictive control algorithm to minimize regional power balance and regulation costs within the next 5-15 minutes, continuously solving for the optimal power allocation sequence of each controllable resource. The hour-level optimization decision-making subunit, based on day-ahead and intraday load forecasts, new energy output forecasts, and energy storage state of charge, constructs a mixed-integer linear programming model with a 24-hour cycle, aiming at optimal system operating economy and minimum energy storage lifespan loss, to solve for the optimal charging and discharging plan of the energy storage system and the start-up and shutdown plans of traditional units such as thermal power plants.

[0017] Furthermore, the adaptive hierarchical control command distribution and reconfiguration module incorporates a communication topology sensor and an operation mode discriminator. The communication topology sensor monitors the communication link quality with each substation controller in real time, quantifying and evaluating communication latency τ and packet loss rate ρ. The operation mode discriminator determines in real time whether the system is in grid-connected mode, islanded mode, or transition mode based on the grid connection point switch status, main grid voltage phase, and local load conditions. When communication quality is excellent and the system is in grid-connected mode, the module adopts a centralized distribution strategy, with the master station directly sending precise control commands to each substation. When a communication latency τ > 200ms or a packet loss rate ρ > 5% is detected, the module switches to a distributed distribution strategy, broadcasting optimization targets and constraints only to the substations, which then perform distributed collaborative calculations based on local information. When the system enters islanded mode, the module triggers control structure reconfiguration, dynamically transferring the dominant control in the original area to an energy storage unit with black-start capability, and activating a preset islanded operation control parameter set.

[0018] The core of the local fast response and virtual inertia simulation execution module is the virtual synchronous generator control algorithm. This algorithm adds a virtual rotor motion equation and a virtual excitation regulator to the outer loop of the dual closed-loop control of the new energy power generation unit. The virtual rotor motion equation simulates the rotor inertia J and damping D of the synchronous generator, and its expression is:

[0019]

[0020] in, For virtual rotor angular velocity, The rated angular velocity, For virtual mechanical torque, The virtual electromagnetic torque is calculated by measuring the voltage and current at the grid connection point. This module dynamically adjusts the virtual mechanical torque based on the received power command ΔP. The reference value is obtained, and the inverter output power is made to track ΔP through power loop control. At the same time, the frequency and phase of the inverter output voltage exhibit inertial response characteristics through the above equation, providing the grid with equivalent inertial support and primary frequency regulation capability.

[0021] Furthermore, the closed-loop performance evaluation and parameter self-tuning module uses the root mean square value of system frequency deviation, voltage qualification rate, and control command execution rate as core performance indicators. This module incorporates a deep deterministic policy gradient agent, whose state space 's' is a vector containing the aforementioned performance indicators, new energy penetration rate, and load volatility, and whose action space 'a' represents the key control parameters (such as...) in the multi-timescale collaborative optimization decision-making module. , , The adjustment amount is determined by the reward function r, which is designed as the negative of the weighted sum of performance indicators. Through continuous interaction with the system environment, the agent learns the optimal parameter adjustment strategy under different operating conditions, achieving adaptive tuning of control parameters and thus overcoming the control performance degradation caused by model inaccuracies or environmental changes.

[0022] On the other hand, a composite coordinated control method for new energy power generation includes the following specific steps:

[0023] Step S110: Through the sensor network deployed in the source, network, and storage links, frequency, voltage, power, equipment status and meteorological data are collected synchronously, and the collected heterogeneous data are preprocessed and spatiotemporally aligned and fused to construct a panoramic dynamic perception dataset of the system.

[0024] Step S120: Based on the panoramic dynamic perception dataset of the system, perform collaborative optimization calculations at three time scales: second, minute, and hour. The second-level calculation generates primary frequency regulation power commands, the minute-level calculation generates secondary frequency regulation and power balance commands, and the hour-level calculation generates energy storage scheduling and economic operation commands, which together form a composite control command set.

[0025] Step S130: Based on the real-time communication network status and power grid operation mode, dynamically select the instruction distribution strategy and distribute the corresponding instructions in the composite control instruction set to the local controllers of each new energy power generation unit and energy storage system through centralized, distributed or hybrid paths.

[0026] In step S140, each local controller, based on the received instructions, quickly calculates and outputs a power regulation signal through a virtual synchronous generator control algorithm to drive the converter to perform power control, while simultaneously simulating and providing inertial response and damping characteristics.

[0027] Step S150: Continuously monitor the actual frequency, voltage, and power response of the system after step S140 is executed, calculate the control deviation from the expected target, and dynamically adjust the key control parameters of various optimization algorithms in step S120 based on reinforcement learning algorithm, according to historical control effects and current operating conditions, to form closed-loop adaptive optimization.

[0028] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0029] This invention constructs a multi-timescale parallel collaborative optimization decision-making architecture, which realizes the organic integration and dynamic coupling of second-level primary frequency regulation, minute-level secondary frequency regulation and hour-level economic scheduling. It effectively eliminates the dead zone and over-adjustment phenomenon caused by hierarchical fragmentation in traditional control, and improves the dynamic stability and regulation accuracy of the system under continuous disturbances.

[0030] This invention designs an adaptive instruction distribution and reconfiguration mechanism with communication topology awareness and operation mode discrimination capabilities, enabling the system to dynamically switch between the high precision of centralized control and the strong robustness of distributed control, significantly enhancing the system's survivability and adaptability under non-ideal operating conditions such as communication constraints, faults, or islanding.

[0031] This invention deeply integrates a virtual synchronous generator control algorithm into the local controller of the new energy power generation unit, enabling it not only to respond quickly to the power command from the upper level, but also to autonomously simulate and provide inertia support and damping, effectively compensating for the problem of reduced anti-disturbance capability caused by the lack of physical rotational inertia in a high proportion of new energy power grids.

[0032] This invention introduces a closed-loop performance evaluation and parameter self-tuning module based on reinforcement learning, enabling the system to break free from dependence on precise mathematical models and autonomously learn and optimize control parameters through data-driven methods. This significantly improves the robustness and adaptability of the control system in the face of uncertainties such as strong fluctuations in new energy output and equipment parameter drift.

[0033] This invention provides a highly consistent panoramic state perception for upper-level optimization decisions through deep fusion of multi-source heterogeneous data and unified spatiotemporal benchmark processing. Combined with a hierarchical and partitioned collaborative mechanism, it achieves a unified approach to global optimization and rapid local response for various types of heterogeneous resources such as wind power, photovoltaics, and energy storage, significantly improving the utilization efficiency of adjustable resources and the overall economic efficiency of the system. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the overall technical architecture of the composite coordinated control system for new energy power generation proposed in this invention;

[0035] Figure 2 This is a schematic diagram of the core principle framework of the multi-timescale collaborative optimization decision-making module proposed in this invention;

[0036] Figure 3 This is a schematic diagram illustrating the specific steps of the composite coordinated control method for new energy power generation proposed in this invention. Detailed Implementation

[0037] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0038] Example 1

[0039] In a new regional power system comprising large-scale wind farms, photovoltaic power plants, centralized energy storage power plants, and distributed energy storage units, the composite coordinated control system and method for new energy power generation provided by this invention have begun comprehensive deployment and operation to achieve frequency stability, voltage support, and economical operation under a high proportion of new energy grid connection. See also Figure 1 The overall architecture of this system includes a multi-source heterogeneous data perception and fusion module, a multi-timescale collaborative optimization decision-making module, an adaptive hierarchical control command distribution and reconfiguration module, a local rapid response and virtual inertia simulation execution module, and a closed-loop performance evaluation and parameter self-tuning module. These modules work together to form a complete control chain from panoramic perception, collaborative decision-making, intelligent distribution, rapid execution to closed-loop optimization.

[0040] The multi-source heterogeneous data sensing and fusion module, acting as the system's "sensory nerves," is responsible for constructing a unified, accurate, and real-time panoramic view of the system's status. This module includes a data acquisition unit, a data preprocessing unit, and a spatiotemporal alignment and fusion unit. The data acquisition unit simultaneously collects frequency deviation, voltage amplitude, three-phase active power, three-phase reactive power, equipment switching status, protection signals, ambient temperature, light intensity, wind speed and direction, and ultra-short-term forecast data from meteorological departments through various types of sensors and standardized communication interfaces deployed at wind turbine nacelles, DC and AC sides of photovoltaic inverters, energy storage converters, grid connection points, and grid dispatching sides. All collected data is accompanied by timestamps accurate to milliseconds and unique equipment identifiers. The data preprocessing unit performs the first round of cleaning on the massive influx of raw data every second. For continuous measurement data such as frequency deviation and power, an outlier detection mechanism based on the local outlier factor algorithm is adopted, setting the number of neighborhood points to 50 and using 2.0 times the standard deviation as the threshold to automatically identify and remove outliers caused by momentary sensor failures or electromagnetic interference. Simultaneously, a Kalman filter is applied to filter noise in the data, smoothing out random fluctuations. For discrete signals such as equipment status, data integrity verification is performed, confirming the normality of the communication link through a heartbeat message mechanism, and marking missing data. The spatiotemporal alignment and fusion unit is crucial for data unification. Due to the varying sampling periods of data sources such as wind farms and photovoltaic power plants, and the different transmission delays in communication networks, this unit uses timestamps synchronized based on a high-precision network time protocol to perform linear interpolation or spline interpolation on the preprocessed data, unifying all data to a fixed sampling period under the system's master clock. In the spatial dimension, a spatial association model based on a geographic information system is used to bind each data point to its corresponding physical equipment's geographical coordinates, ultimately generating a panoramic dynamic perception dataset for the system. This dataset is a multi-dimensional time-series data matrix, with each data entry containing a unified timestamp, a globally unique equipment identifier, latitude and longitude coordinates, and a numerical vector composed of frequency, voltage, power, and status flags, providing a data foundation with a completely consistent spatiotemporal reference for upper-level decision-making.

[0041] The multi-timescale collaborative optimization decision-making module is the system's "intelligent hub," performing collaborative optimization calculations in parallel across three timescales: second, minute, and hour, based on the aforementioned panoramic dynamic perception dataset. (See also...) Figure 2 This module internally operates three parallel optimization decision subunits. They share data input but differ in their optimization objectives, time windows, and controlled objects, ultimately merging their outputs into a single composite control command set. The second-level optimization decision subunit specifically responds to rapid frequency fluctuations caused by sudden load changes or a sharp drop in renewable energy power. Its core is a primary frequency modulation controller, employing a composite control law combining improved virtual inertial control and droop control. When a system frequency deviation is detected, the controller not only adjusts the power proportionally based on the current deviation value but also introduces a derivative term of the frequency change rate to simulate the inertial response and eliminates steady-state errors through an integral term. Its output primary frequency modulation power command is calculated using the following formula:

[0042]

[0043] in, The virtual inertia coefficient is used to simulate the rotor inertia time constant of the synchronous generator. Its initial value is set according to the ratio of the system's equivalent inertia deficit to the total capacity of the new energy units as evaluated online. The droop coefficient determines the power regulation amount corresponding to a unit frequency deviation, and the initial value is allocated according to the frequency regulation capacity of each unit. The integral coefficient is used to eliminate steady-state frequency deviation. This subunit performs rapid calculations with a period of 100 milliseconds, and the output commands directly apply to wind turbines, photovoltaic inverters, and energy storage converters with rapid adjustment capabilities. The minute-level optimization decision subunit focuses on eliminating regional control deviations and mitigating ultra-short-term power forecast errors. It employs a distributed model predictive control algorithm with a forecast time domain of 5 to 15 minutes and a control cycle of 1 minute. The algorithm uses regional power balance and minimizing regulation costs as its objective function, with constraints including upper and lower limits for power regulation and ramp rate limits for each renewable energy plant and energy storage system. At the beginning of each control cycle, the minute-level optimization decision subunit receives the latest ultra-short-term wind and solar power forecasts and load forecasts, and continuously solves for the optimal power allocation sequence of each controllable resource over multiple future time periods, generating secondary frequency regulation and power balance commands. It mainly coordinates the active power reservation control of wind farms and the minute-level charging and discharging of energy storage systems. The hour-level optimization decision subunit focuses on economic operation and energy storage scheduling on a 24 / 7 scale. Based on day-ahead and intraday load forecast curves, renewable energy output forecast curves, and the current state of charge of the energy storage system, a mixed-integer linear programming model with a 24-hour period is constructed. The objective function comprehensively considers the cost of electricity purchase, the cost of wind and solar curtailment penalties, the cost of energy storage cycle life loss, and the start-up and shutdown costs of traditional generating units. Decision variables include the hourly charging and discharging power plan of the energy storage system, the start-up and shutdown status of the thermal power units, and their output plans. The model is solved at a fixed time each day and distributes the optimal charging and discharging plan for the energy storage, guiding it to charge during off-peak hours when electricity prices are low and discharge during peak hours or when renewable energy output is insufficient, thereby achieving optimal overall system economics.

[0044] The adaptive hierarchical control command distribution and reconfiguration module acts as the system's "intelligent scheduler," responsible for efficiently and reliably transmitting composite control command sets to the execution terminals. This module incorporates a communication topology sensor and an operation mode discriminator. The communication topology sensor continuously monitors all communication links between the master station and the local controllers of each new energy power generation unit and energy storage system, quantitatively evaluating the communication latency and packet loss rate of each link in real time. The operation mode discriminator comprehensively analyzes the opening and closing status of the circuit breaker at the grid connection point, the main grid voltage phase angle synchronization, and the switching status of important local loads, accurately determining in real time whether the system is currently in grid-connected mode, islanded mode, or a transient mode transitioning from grid-connected to islanded. Based on this real-time information, the module dynamically selects the command distribution strategy. When communication quality is excellent and the system is operating stably in grid-connected mode, a centralized distribution strategy is adopted, with the master station directly and accurately distributing second-level and minute-level optimized commands to the corresponding substation controllers. When the communication topology sensor detects that the latency of a communication link in a certain area exceeds 200 milliseconds or the packet loss rate is higher than 5%, the module automatically switches that area to a distributed distribution strategy. Under this strategy, the master station no longer issues specific command values, but instead broadcasts the current optimization objective function, global constraints, and necessary boundary information to all substations within the region. Each substation, based on local measurement information and received global information, runs distributed collaborative algorithms such as consensus algorithms to autonomously negotiate and calculate local control commands that satisfy the global objective, thus maintaining collaborative control capabilities even when communication is limited. When the operating mode discriminator determines that the system has entered islanding mode, the module immediately triggers a control structure reconfiguration process. This process first identifies energy storage units within the islanding region that possess black-start capability, dynamically transferring regional dominant control to these units, which then act as temporary "master stations." Simultaneously, the module loads a preset islanding operation control parameter set. This parameter set optimizes parameters such as the virtual inertia coefficient and droop coefficient, taking into account the characteristics of islanded systems—low inertia and significant load fluctuations—and adjusts the optimization objective to prioritize frequency and voltage stability within the island.

[0045] The local fast response and virtual inertia simulation execution module is the system's "end-point actuator," deployed in the local controller of each wind turbine converter, photovoltaic inverter, and energy storage converter. The core of this module is a deeply integrated virtual synchronous generator control algorithm. This algorithm adds a virtual rotor motion equation and a virtual excitation regulator to the outer layer of the traditional dual-closed-loop control. The virtual rotor motion equation is used to simulate the mechanical inertia and damping characteristics of the synchronous generator, and its expression is:

[0046]

[0047] in, This is a virtual moment of inertia, the value of which is related to the rated capacity of the unit and is used to provide the inertial time constant; This is a virtual damping coefficient used to suppress power oscillations; This represents the virtual rotor angular velocity. Rated angular velocity; This is a virtual mechanical torque, the reference value of which is derived from the received power command from the superior unit; The virtual electromagnetic torque is calculated by real-time measurement of the grid connection point voltage and current. When the local controller receives a power regulation command from the adaptive hierarchical control command distribution and reconfiguration module, it first converts it into an incremental reference for the virtual mechanical torque. Subsequently, through the inner power loop and current loop control, the power semiconductor switching devices of the converter are driven, enabling the inverter output power to quickly and accurately track the command value. At the same time, the virtual rotor motion equation dynamically adjusts the virtual rotor angular velocity based on the real-time difference between the electromagnetic torque and the mechanical torque, thereby affecting the phase and frequency of the inverter output voltage through the phase-locked loop mechanism. This makes the new energy power generation unit not only a passive power source, but also actively exhibit inertial response characteristics similar to a synchronous generator: when the grid frequency decreases, the virtual rotor releases the stored kinetic energy, instantaneously increasing the electromagnetic power output; when the frequency increases, it absorbs power, thus providing valuable equivalent inertial support and primary frequency regulation capability for the grid.

[0048] The closed-loop performance evaluation and parameter self-tuning module is the system's "learning optimization engine," ensuring the control system can continuously adapt to environmental changes. This module uses the root mean square value of the system frequency deviation, the grid connection point voltage qualification rate, and the actual execution rate of control commands as core performance indicators to continuously monitor control effectiveness. The module incorporates a reinforcement learning agent based on a deep deterministic policy gradient algorithm. The agent's state space is a multi-dimensional vector, containing the current performance indicator values, the system's real-time new energy penetration rate, load fluctuation rate, and the operating status of key equipment. The action space represents the adjustment amounts of key control parameters in the multi-timescale collaborative optimization decision-making module, such as the virtual inertia coefficient, droop coefficient, and integral coefficient in second-level optimization, and the weighting coefficient of the cost function in minute-level optimization. The reward function is designed as the negative value of the weighted sum of the above performance indicators; that is, the better the control effect, the higher the reward value. The agent interacts with the environment at a fixed learning cycle: it observes the current system state and outputs a set of parameter adjustment actions; after the system applies the new parameters and runs for a period of time, it generates a new state and reward; the agent updates its internal neural network parameters based on the experience of this four-tuple of state, action, reward, and new state. Through continuous learning from massive amounts of data, the intelligent agent can autonomously discover the optimal parameter configuration strategy under different operating conditions. For example, during the afternoon when the output of new energy sources fluctuates wildly, the agent may automatically increase the virtual inertia coefficient to enhance the system's anti-disturbance capability; while at night when the load is stable, it may optimize the economic weight to reduce operating costs. This data-driven approach frees the system from dependence on precise mathematical models and manual parameter tuning, enabling adaptive and self-optimization of control parameters and effectively addressing uncertainties such as equipment aging and model mismatch.

[0049] Based on the above system, the composite coordinated control method for new energy power generation of the present invention is executed according to the following steps:

[0050] Step S110 involves synchronously collecting system frequency deviation, voltage amplitude, active and reactive power, equipment switching status signals, and meteorological data such as wind speed and solar irradiance through sensor networks and communication interfaces deployed at key nodes in wind farms, photovoltaic power plants, energy storage systems, and power grids. The collected raw heterogeneous data undergoes outlier detection and removal, noise filtering, and integrity verification. Outlier detection employs a local outlier factor algorithm, setting the number of neighboring points to 50 and a threshold of 2.0 times the standard deviation. Subsequently, using a timestamp-based interpolation algorithm and a geographic information system spatial association model, all preprocessed data is uniformly mapped to the same millisecond-level time base and a unified geographic coordinate system, ultimately constructing a system panoramic dynamic perception dataset containing timestamps, equipment identifiers, geographic location coordinates, and multi-dimensional state vectors.

[0051] Step S120: Based on the system panoramic dynamic perception dataset generated in step S110, parallel collaborative optimization calculation threads at three time scales—second-level, minute-level, and hour-level—are initiated. The second-level calculation thread runs with a 100-millisecond cycle, employing a composite control law to calculate and generate primary frequency regulation power commands to suppress frequency abrupt changes based on real-time frequency deviations and their derivative and integral values. The minute-level calculation thread runs with a 1-minute cycle, employing a distributed model predictive control algorithm to achieve regional power balance and minimize regulation costs within the next 5 to 15 minutes, continuously solving and generating secondary frequency regulation and power balance commands to eliminate regional control deviations. The hourly calculation thread runs with an hourly cycle, constructing a mixed-integer linear programming model based on all-day forecast data, aiming for optimal 24-hour system operating economy and minimum energy storage lifespan loss, solving and generating optimal charging / discharging plans for the energy storage system and start-up / shutdown plans for traditional units. These three types of commands together constitute a composite control command set.

[0052] In step S130, the adaptive hierarchical control command distribution and reconfiguration module monitors the communication network status and grid operation mode in real time. Based on communication link latency, packet loss rate, grid connection point status, and main grid synchronization, it dynamically selects the command distribution strategy. When communication is excellent and the system is in grid-connected mode, a centralized strategy is adopted, directly distributing the composite control commands generated in step S120 to each new energy source and energy storage local controller. When a communication delay exceeding 200 milliseconds or a packet loss rate exceeding 5% is detected, the system switches to a distributed strategy, broadcasting optimization objectives and constraints only to substations, which then collaboratively calculate local commands. When the system enters islanded mode, control reconfiguration is triggered, transferring control to the black-start energy storage unit and activating the islanded dedicated parameter set to ensure reliable command distribution and system survival under abnormal operating conditions.

[0053] In step S140, the local fast response and virtual inertia simulation execution module in the local controller of each new energy power generation unit and energy storage system starts working according to the instructions distributed in step S130. This module converts the received power command into a virtual mechanical torque reference value and simulates the inertia and damping characteristics of a synchronous generator through the virtual rotor motion equation in the virtual synchronous generator control algorithm. The algorithm affects the phase and frequency of the inverter output voltage by adjusting the virtual rotor angular velocity, and while quickly and accurately tracking the power command, it autonomously provides equivalent inertia support and damping for the grid, performing primary frequency regulation.

[0054] In step S150, the closed-loop performance evaluation and parameter self-tuning module continuously monitors the actual frequency response, voltage changes, and power balance of the system after step S140, calculating their deviation from the expected control target. The module's built-in reinforcement learning agent uses the root mean square value of frequency deviation and voltage compliance rate as performance indicators to construct a state, action, and reward model. Through continuous interaction with the environment, the agent learns the optimal strategy for adjusting key control parameters of the second-level and minute-level optimization algorithms in step S120 under different new energy penetration rates and load fluctuation conditions, and applies these adjustments in real time, thus forming a closed-loop adaptive optimization process that continuously improves the robustness and overall performance of the system control.

[0055] Example 2

[0056] In another embodiment, the present invention is applied to a city distribution network microgrid cluster primarily composed of distributed photovoltaic and residential energy storage. This scenario is characterized by dispersed resources, small individual unit capacity, a large number of units, and complex and variable communication conditions. This embodiment focuses on illustrating the adaptive coordination mechanism of the system in a highly distributed scenario.

[0057] The multi-source heterogeneous data sensing and fusion module faces the challenge of accessing massive amounts of terminal data. The data acquisition unit collects data such as power generation, power consumption, energy storage state of charge, and grid connection voltage for each household through smart meters, built-in communication modules in photovoltaic inverters, and energy storage controllers. The data preprocessing unit uses a combination of edge computing and cloud computing to perform preliminary abnormal data filtering and compression locally on the household inverter before uploading it to the regional aggregator. The spatiotemporal alignment and fusion unit, located on the aggregator side, uses a distribution network topology model to spatially aggregate tens of thousands of user data points according to their feeders and transformers, forming a regional panoramic dynamic sensing dataset with distribution transformers as nodes, significantly reducing data dimensionality and transmission processing pressure.

[0058] The architecture of the multi-timescale collaborative optimization decision-making module has been adaptively adjusted. The second-level optimization decision-making subunit mainly relies on the local virtual synchronous generator control algorithm of each household energy storage inverter to provide fast frequency support, while regional centralized optimization focuses on the minute and hour levels. The minute-level optimization decision-making subunit adopts a fully distributed collaborative algorithm, where each distribution transformer node acts as an agent, exchanging limited boundary information only with neighboring nodes. Through iterative calculations, it collaboratively optimizes the power balance of the entire distribution network and suppresses voltage exceedances caused by the simultaneous start-up and shutdown of a large number of photovoltaic systems. The hour-level optimization decision-making subunit, based on the overall load and power generation forecast of the cluster, optimizes the charging and discharging plans of the energy storage system on a community-by-community basis to achieve peak-valley arbitrage and improve the photovoltaic self-consumption rate.

[0059] In this scenario, the adaptive hierarchical control command distribution and reconfiguration module primarily employs a distributed and hybrid strategy. Since establishing low-latency communication directly with all user terminals is uneconomical and unreliable, the module deploys multiple regional agents at the distribution network level. The master station distributes optimization targets to the regional agents, which then determine whether to control subordinate terminals via broadcast targets or direct commands based on local communication conditions. When communication in a certain area is completely interrupted, the user energy storage devices within that area autonomously switch to a preset islanded operation mode based on local voltage and frequency measurements to maintain power supply to critical loads.

[0060] The local fast response and virtual inertia simulation execution module is implemented in the residential photovoltaic-storage integrated machine. Even without explicit instructions from the superior, the local fast response and virtual inertia simulation execution module can autonomously provide a small inertia response and primary frequency regulation based on the slight changes in the grid frequency measured locally, through the virtual rotor motion equation. Thousands of such distributed devices working together can aggregate into a considerable system inertia supplement.

[0061] The learning process of the closed-loop performance evaluation and parameter self-tuning modules is also more hierarchical. Regional agents first evaluate the control performance of their own region and fine-tune local parameters. The master station then evaluates the global performance and learns how to optimize the target weights and constraints distributed to each regional agent. This hierarchical learning mechanism adapts to the characteristics of distributed systems while ensuring the convergence of global optimization.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A composite coordinated control system for new energy power generation, characterized in that, include: The multi-source heterogeneous data sensing and fusion module collects and fuses multi-dimensional operation data and status information from wind power, photovoltaics, energy storage and the grid side in real time, and generates a system panoramic dynamic sensing dataset with a unified spatiotemporal benchmark. The multi-timescale collaborative optimization decision module is connected to the multi-source heterogeneous data perception and fusion module. Based on the system's panoramic dynamic perception dataset, it performs collaborative optimization calculations in parallel at three timescales: second, minute, and hour, generating a composite control instruction set that includes primary frequency regulation instructions, secondary frequency regulation instructions, and energy storage scheduling instructions. The adaptive hierarchical control command distribution and reconfiguration module is connected to the multi-timescale collaborative optimization decision module. It receives the composite control command set and adaptively selects a centralized, distributed, or hybrid command distribution strategy according to the current power grid operation mode, communication status, and equipment availability. It dynamically distributes the control commands to each new energy power generation unit and energy storage system. The local fast response and virtual inertia simulation execution module is deployed in the local controller of each new energy power generation unit and energy storage system. It receives and executes instructions from the adaptive hierarchical control instruction distribution and reconfiguration module, generates power regulation signals locally quickly, and simulates and provides inertia support and damping characteristics through virtual synchronous generator control algorithm. The closed-loop performance evaluation and parameter self-tuning module is connected to the multi-source heterogeneous data sensing and fusion module and the multi-timescale collaborative optimization decision module. It continuously monitors the actual control effect of system frequency, voltage and power balance, evaluates control deviation based on preset performance indicators, and dynamically adjusts the key control parameters in the multi-timescale collaborative optimization decision module using reinforcement learning algorithms.

2. The composite coordinated control system for new energy power generation according to claim 1, characterized in that, The multi-source heterogeneous data sensing and fusion module includes a data acquisition unit, a data preprocessing unit, and a spatiotemporal alignment fusion unit: The data acquisition unit synchronously collects frequency deviation, voltage amplitude, active power, reactive power, equipment operating status flags, and meteorological forecast data through sensors and communication interfaces deployed in wind farms, photovoltaic power stations, energy storage power stations, and key nodes of the power grid. The data preprocessing unit performs outlier detection and removal, noise filtering, and data integrity verification on the collected raw data. The spatiotemporal alignment and fusion unit maps preprocessed data with different sampling periods and delays to the same spatiotemporal coordinate system through a timestamp-based interpolation algorithm and a spatial association model based on a geographic information system, forming a system panoramic dynamic perception dataset. The data structure of the system panoramic dynamic perception dataset is a time-series data matrix containing timestamps, device IDs, geographic location coordinates, and multi-dimensional state vectors.

3. The composite coordinated control system for new energy power generation according to claim 2, characterized in that: The outlier detection mechanism based on the local outlier factor algorithm used in the data preprocessing unit has a neighborhood number of 50 and a threshold of 2.0 times the standard deviation. It automatically identifies and removes outliers caused by sensor momentary failures or electromagnetic interference.

4. The composite coordinated control system for new energy power generation according to claim 1, characterized in that, The multi-timescale collaborative optimization decision-making module internally operates three optimization decision-making sub-units in parallel: The second-level optimization decision subunit response frequency deviation is based on a primary frequency modulation controller, which employs a composite control law combining improved virtual inertial control and droop control. The minute-level optimization decision-making subunit, based on regional control deviation and ultra-short-term power prediction error, adopts a distributed model predictive control algorithm to solve the optimal power allocation sequence of each controllable resource in a rolling manner, with the goal of minimizing regional power balance and regulation cost within the next 5-15 minutes. The hourly optimization decision-making subunit is based on day-ahead and intraday load forecasts, new energy output forecasts, and energy storage charge status. With a 24-hour cycle, it constructs a mixed integer linear programming model with the objectives of optimizing system operation economy and minimizing energy storage life loss, and solves the optimal charging and discharging plan of the energy storage system and the start-up and shutdown plan of traditional units such as thermal power.

5. The composite coordinated control system for new energy power generation according to claim 4, characterized in that: The minute-level optimization decision-making subunit uses a distributed model predictive control algorithm with a prediction time domain of 5 to 15 minutes and a control cycle of 1 minute. At the beginning of each control cycle, it receives the latest ultra-short-term wind and solar power prediction and load prediction data, and solves the optimal power allocation sequence of each controllable resource in multiple future time periods.

6. The composite coordinated control system for new energy power generation according to claim 1, characterized in that: The adaptive hierarchical control command distribution and reconstruction module has a built-in communication topology sensor and an operation mode discriminator. The communication topology sensor monitors the quality of the communication link with each substation controller in real time and quantitatively evaluates the communication latency τ and packet loss rate ρ. The operating mode discriminator determines in real time whether the system is in grid-connected mode, islanded mode, or transition mode based on the grid connection point switch status, main grid voltage phase, and local load conditions. When the communication quality is good and the system is in grid-connected mode, the adaptive hierarchical control command distribution and reconfiguration module adopts a centralized distribution strategy, with the master station directly sending precise control commands to each substation. When a communication delay τ > 200ms or a packet loss rate ρ > 5% is detected, the adaptive hierarchical control command distribution and reconfiguration module switches to a distributed distribution strategy, broadcasting optimization objectives and constraints only to the substations, which then perform distributed collaborative calculations based on local information. When the composite coordinated control system for new energy power generation enters islanded mode, the adaptive hierarchical control command distribution and reconfiguration module triggers control structure reconfiguration, dynamically transferring the dominant control in the original area to an energy storage unit with black-start capability, and activating a preset islanded operation control parameter set.

7. The composite coordinated control system for new energy power generation according to claim 1, characterized in that: The core of the local fast response and virtual inertia simulation execution module is the virtual synchronous generator control algorithm. The virtual synchronous generator control algorithm adds a virtual rotor motion equation and a virtual excitation regulator on top of the double closed-loop control outer loop of the new energy power generation unit. The local fast response and virtual inertia simulation execution module dynamically adjusts the reference value of the virtual mechanical torque according to the received power command ΔP, and makes the inverter output power track ΔP through power loop control. At the same time, the frequency and phase of the inverter output voltage exhibit inertial response characteristics through the above equations.

8. The composite coordinated control system for new energy power generation according to claim 1, characterized in that: The closed-loop performance evaluation and parameter self-tuning module uses the root mean square value of system frequency deviation, voltage qualification rate, and control command execution rate as core performance indicators. This module incorporates a deep deterministic strategy gradient agent, whose state space 's' is a vector containing the aforementioned performance indicators, new energy penetration rate, and load fluctuation rate. Its action space 'a' represents the adjustment amounts of key control parameters in the multi-timescale collaborative optimization decision-making module, and the reward function 'r' is designed as the negative value of the weighted sum of performance indicators. Through continuous interaction with the system environment, the agent learns the optimal parameter adjustment strategy under different operating conditions, achieving adaptive tuning of the control parameters.

9. A composite coordinated control method for new energy power generation, characterized in that, Includes the following steps: S110, through a sensor network deployed in the source, network, and storage links, synchronously collects frequency, voltage, power, equipment status and meteorological data, and preprocesses and fuseds the collected heterogeneous data in a spatiotemporal alignment to construct a panoramic dynamic perception dataset for the system. S120, based on the panoramic dynamic perception dataset of the system, performs collaborative optimization calculations at three time scales: second, minute, and hour. The second-level calculation generates primary frequency regulation power commands, the minute-level calculation generates secondary frequency regulation and power balance commands, and the hour-level calculation generates energy storage scheduling and economic operation commands, which together form a composite control command set. S130, based on the real-time communication network status and power grid operation mode, dynamically select the instruction distribution strategy and distribute the corresponding instructions in the composite control instruction set to the local controllers of each new energy power generation unit and energy storage system through centralized, distributed or hybrid paths. S140, in each local controller, according to the received instructions, quickly calculates and outputs power regulation signals through the virtual synchronous generator control algorithm to drive the converter to perform power control, while simulating and providing inertial response and damping characteristics; S150: Continuously monitor the actual frequency, voltage, and power response of the system after the execution of step S140, calculate the control deviation from the expected target, and based on the reinforcement learning algorithm, dynamically adjust the key control parameters of various optimization algorithms in step S120 according to the historical control effect and the current operating conditions to form a closed-loop adaptive optimization.

10. The composite coordinated control method for new energy power generation according to claim 9, characterized in that, In step S120, the minute-level collaborative optimization calculation adopts a distributed model predictive control algorithm with the goal of minimizing regional power balance and regulation costs within the next 5-15 minutes; the hour-level collaborative optimization calculation constructs a mixed integer linear programming model with a period of 24 hours with the goal of optimizing system operation economy and minimizing energy storage life loss.

Images