Intelligent peak regulation method with optimized combination of energy storage and conventional peak regulation means

Abstract

The invention discloses an intelligent peak regulation method with optimized combination of energy storage and conventional peak regulation means, and the method comprises the following steps: building a holographic situation awareness layer: capturing the voltage, current phase angle and frequency deviation of a power grid, and monitoring the pressure fluctuation in a diesel engine cylinder; data transmission and preprocessing: encrypting data of the miniature phasor measurement unit and the quantum sensor by using a quantum key distribution technology; the self-adaptive decision engine is used for performing Nash equilibrium calculation by utilizing a federal reinforcement learning architecture, through a local Agent based on a power generation / energy storage unit state autonomous optimization control strategy and in combination with a digital twin platform, solving multi-target conflicts of economy, reliability and carbon emission, and generating a global optimal scheduling instruction; the execution control layer is used for tracking resonance points and dispersing harmonic energy spectrums; dynamic topology reconstruction: a harmonic source is positioned by using a digital twin platform, and the topology of the energy storage system is automatically switched according to the harmonic frequency and position; continuously optimizing the digital twin platform; and full-process closed-loop feedback and self-learning are carried out.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of offshore power supply, and particularly relates to an intelligent peak shaving method of optimized combination of energy storage and conventional peak shaving means. BACKGROUND

[0002] In the field of offshore platform power supply, with the increasing penetration of renewable energy and the increasing demand for power supply reliability under extreme conditions, the optimized combination of energy storage systems and conventional peak shaving equipment has become a key path to solve the contradiction between energy microgrid stability and economy on offshore platforms by building a multi-time scale energy coordination mechanism. Specifically, due to the space limitation, high salt spray corrosion environment and operation and maintenance cost constraints on offshore platforms, precise matching of the power generation side and the load side needs to be achieved under limited equipment capacity. Fast-response energy storage devices such as lithium-ion batteries and flywheel energy storage can effectively smooth out wind power / photovoltaic output fluctuations at the second to minute level, and cooperate with the output adjustment capacity of diesel generators at the hour level to form a cross-time scale power support system. At the same time, through deep learning of meteorological prediction data, historical load curves and equipment operating states by artificial intelligence algorithms, a multi-objective optimization model considering energy storage charging and discharging loss, generator climbing rate and fuel consumption is built to realize intelligent collaboration of energy storage system peak clipping and valley filling and conventional peak shaving power base load operation. Ultimately, while ensuring uninterrupted power supply of key loads on offshore platforms, diesel consumption and carbon emission intensity are significantly reduced, forming an intelligent energy dispatching paradigm suitable for complex marine conditions.

[0003] In the prior art, the mainstream paradigm of traditional power system control is centralized optimization, but in extreme environments such as offshore platforms, centralized optimization relies on global data aggregation and central calculation, resulting in a decision delay of hundreds of milliseconds, which cannot cope with instantaneous load mutations on offshore platforms. Therefore, an intelligent peak shaving method of optimized combination of energy storage and conventional peak shaving means is proposed. SUMMARY

[0004] The present application relates to the technical field of offshore power supply, and particularly relates to an intelligent peak shaving method of optimized combination of energy storage and conventional peak shaving means.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0006] An intelligent peak shaving method of optimized combination of energy storage and conventional peak shaving means, comprising the following steps:

[0007] Holographic situational awareness layer: Deploy micro phasor measurement units (PMUs) and quantum sensors to build a holographic situational awareness layer, use synchronous phasor measurement technology to capture grid voltage, current phase angle and frequency offset, and monitor diesel cylinder pressure fluctuations, install diamond NV color center quantum pressure sensors in the diesel cylinder to monitor combustion chamber pressure fluctuations with 0.1 MPa resolution, predict load mutations, and generate power quality fingerprints through edge computing nodes and trigger local self-healing control (such as isolating fault branches within 0.1 seconds);

[0008] Data transmission and preprocessing: Build a quantum communication network on the offshore platform, use quantum key distribution (QKD) technology to encrypt micro phasor measurement unit (PMU) and quantum sensor data, deploy lightweight filtering algorithms (Kalman filter) in local agents (each power generation / energy storage unit) to remove sensor noise and extract key features (such as load mutation amplitude, harmonic frequency), and after data compression, transmit to the global coordinator through 5G-Advanced link;

[0009] Adaptive decision engine: Use federated reinforcement learning (FRL) architecture, through local agents (each power generation / energy storage unit) to optimize control strategies based on the state of the power generation / energy storage unit, where the power generation unit agent dynamically adjusts the diesel engine output based on local state (SOC, oil pressure, temperature, etc.), the energy storage unit agent optimizes the charging and discharging strategy (such as charging to the upper limit of SOC during valley period) based on load prediction and electricity price signal, the harmonic control agent monitors THD in real time, triggers adaptive trap parameter adjustment, and combines with the digital twin platform to perform Nash equilibrium calculation, solving the multi-objective conflict of economy (minimizing diesel consumption), reliability (maximizing power supply continuity), and carbon emissions (maximizing renewable energy ratio), and generating global optimal scheduling instructions;

[0010] Execution control layer: Through energy storage converter (PCS) dynamically adjusts charging and discharging power, flywheel energy storage compensates for instantaneous power gap, and adaptive trap and chaotic injection PWM technology real-time tracks resonance point and disperses harmonic energy spectrum;

[0011] Dynamic topology reconfiguration: Use the digital twin platform to locate harmonic sources and automatically switch energy storage system topology according to harmonic frequency and location;

[0012] Continuous optimization of digital twin platform: Through the bidirectional mapping of digital twin model and physical system, verify the execution effect of scheduling instructions, and feedback the actual running data to the federated reinforcement learning (FRL) model to optimize the reward function, realizing the continuous optimization and closed-loop feedback of the digital twin platform;

[0013] Full-process closed-loop feedback and self-learning: Through key indicator monitoring and long short-term memory network (LSTM) prediction of future load / renewable energy output, combined with abnormal event library accumulation, system strategy iteration and self-learning are promoted, realizing full-process closed-loop feedback and continuous evolution.

[0014] The above further comprises:

[0015] Further, the specific steps of constructing the holographic situational awareness layer are:

[0016] Through micro phasor measurement unit deployment and synchronous phasor measurement: Real-time capture of grid voltage, current phase angle and frequency offset, use formula And Δt = Δθ / 2πΔf quantifies phase deviation and time synchronization error, combined with the light detection magnetic resonance principle of diamond NV color centers in quantum sensors, through the relationship f = f0+k·P to monitor diesel combustion chamber pressure fluctuations, predict load mutation in advance;

[0017] Edge computing node generates power quality fingerprint: The edge computing node analyzes the micro phasor measurement unit and quantum sensor data in real time to generate a power quality fingerprint;

[0018] Holographic perception data fusion and load mutation prediction: Fusion of micro phasor measurement unit phase angle deviation Δθ and quantum sensor pressure fluctuation ΔP, prediction of load mutation probability through support vector machine model, triggering flywheel energy storage compensation, diesel load rate adjustment and lithium battery discharge.

[0019] Further, the specific steps of data transmission and preprocessing are:

[0020] Build offshore platform quantum communication network: Use BB84 protocol to generate dynamically updated shared key, AES-256 encrypt micro phasor measurement unit voltage data and quantum sensor pressure data, deploy trust nodes between platforms to extend quantum communication distance;

[0021] Local Agent lightweight filtering algorithm: Deploy lightweight Kalman filtering algorithm in local Agent, model sensor dynamic characteristics through state equation and fuse quantum measurement values through observation equation, calculate Kalman gain, the formula of which is The state update formula is represented as Where P(k) is the modeled sensor dynamic characteristics, Z(k) is the quantum measurement value, and key features are extracted;

[0022] Data compression and 5G-Advanced transmission: Use Daubechies-4 wavelet basis for 3-level decomposition and compression, and configure subcarrier spacing and time slot format through 5G-Advanced link, and preferentially schedule peak regulation instruction data packets;

[0023] Global coordinator receives and decrypts: At the global coordinator end, through QKD key synchronization verification and data decryption, combined with wavelet reconstruction algorithm to restore the original signal time domain characteristics.

[0024] Further, the adaptive decision engine generates global optimal scheduling instructions through the deep integration of federal reinforcement learning and Nash equilibrium calculation, including the following steps:

[0025] Power generation unit Agent autonomous optimization: The power generation unit Agent uses quantum sensors to monitor the in-cylinder pressure, SOC and temperature parameters of the diesel engine in real time, combines the objective function of minimizing fuel consumption rate and low load penalty coefficient, and dynamically adjusts the fuel injection amount and turbocharging pressure;

[0026] Energy storage unit Agent autonomous optimization: Based on the load prediction model and time-of-use price signal, the energy storage unit Agent adopts a double-objective optimization function of maximizing arbitrage income and minimizing battery life loss to adjust the charge and discharge power instruction;

[0027] Harmonic management Agent autonomous optimization: The harmonic management Agent dynamically adjusts the filter frequency band and injects a chaotic sequence with Lyapunov exponent > 0 by real-time monitoring of total harmonic distortion and applying adaptive notch filter and chaotic injection PWM technology, so that the harmonic amplitude decreases;

[0028] Digital twin platform modeling: The global coordinator builds a diesel engine efficiency curve, a lithium battery life model, and a drilling load dynamic model in the digital twin platform to simulate a scenario where wind power output drops by 50% in the next 15 minutes;

[0029] Nash equilibrium solution: A multi-objective optimization function including fuel cost, power shortage risk, and carbon emissions and an improved particle swarm algorithm are used to solve the Nash equilibrium solution that meets the power balance constraint and the climbing rate limit;

[0030] Execution layer response: Through the execution layer response, the diesel engine electronic governor adjusts to the target output, the lithium battery PCS discharges, and the SOC is fed back to the global coordinator in real time.

[0031] Further, the dynamic adjustment of charge and discharge power of energy storage converter, flywheel energy storage compensation of instantaneous power gap, and adaptive notch filter and chaotic injection PWM technology for real-time tracking of resonance points and dispersion of harmonic energy spectrum, including the following steps:

[0032] Dynamic adjustment of charge and discharge power of energy storage converter: By dynamically adjusting the charge and discharge power of the energy storage converter, a PID controller is used to calculate the power instruction based on the grid frequency deviation or load prediction error in real time, and the power instruction calculation formula is expressed as P ref (t)=K pΔf(t)+K i ∫Δf(t)dt+K d dΔf(t) / dt, where K p , K i , K d It is a PID controller parameter and combines the power change rate limit and smoothing control technology to set the power change rate limit;

[0033] Flywheel energy storage compensates for instantaneous power gap: The flywheel energy storage system is used to compensate for the instantaneous power gap. The high-frequency power fluctuation component is separated by a low-pass filter to generate a flywheel power command and is limited by its maximum energy storage capacity constraint. The flywheel power command is generated as P FESS (t) = K·P HF (t), K is the gain coefficient, and the maximum energy storage capacity constraint is expressed as;

[0034] Adaptive notch filter and chaos injection PWM technology: An adaptive notch filter is deployed to track the grid resonant frequency in real time. A second-order damping filter is designed to suppress specific harmonic components. Combined with chaos injection PWM modulation technology, a chaotic signal is introduced into the traditional pulse-width modulation carrier to disperse the harmonic energy spectrum, converting concentrated harmonics into broadband noise.

[0035] Furthermore, the specific steps of using the digital twin platform to locate the harmonic source and automatically switch the energy storage system topology according to the harmonic frequency and location are as follows:

[0036] Digital twin platform modeling and data acquisition: The digital twin platform integrates real-time monitoring data and historical operating data from the power grid and energy storage system. Fourier transforms are used to perform harmonic analysis on voltage and current signals to extract harmonic frequency components and amplitudes. Total harmonic distortion is then calculated to quantify power quality. A digital twin model of the power grid and energy storage system is constructed to achieve virtual mapping.

[0037] Harmonic source location and topology switching decision-making: Based on the impedance method, the harmonic source position is located through harmonic impedance calculation and upstream and downstream impedance change comparison. In combination with the harmonic frequency characteristics and the coordination requirements of multiple harmonic sources, the energy storage system topology mode is dynamically matched and the connection matrix of modular energy storage units is optimized;

[0038] Topology switching execution and effect evaluation: Harmonic suppression and peak shaving strategies are implemented through topology switching of power electronic switches, and system performance is evaluated with THD reduction rate and peak shaving capacity as optimization targets.

[0039] Furthermore, the specific steps for continuous optimization of the digital twin platform are:

[0040] Bidirectional mapping and dispatch instruction verification: By building a digital twin model of the power grid-energy storage system, the state space and action space are defined. The state space includes load power, PV output, energy storage power, state of charge, frequency, and voltage. The action space includes energy storage charging and discharging, load shedding, and thermal power unit output adjustment. Bidirectional mapping verification is performed using the frequency and voltage errors between the actual operating data of the physical system and the model prediction results.

[0041] Federated reinforcement learning model training and optimization: The federated reinforcement learning framework is then used to divide the power grid into multiple regions. Local reinforcement learning agents are deployed in each region, and local strategies are trained using the Q-learning algorithm. Each regional agent uploads local Q network parameters to the central server, and the global model parameters are aggregated using the federated averaging algorithm. The reward function weights are dynamically adjusted based on feedback data from the physical system to achieve multi-objective collaborative optimization.

[0042] Closed-loop feedback: Deploy the optimized global strategy to the digital twin platform to generate new scheduling instructions, and continuously update the experience replay pool data through the execution results of the physical system.

[0043] Furthermore, the specific steps of predicting future load / renewable energy output through key indicator monitoring and long short-term memory networks, combined with the accumulation of abnormal event libraries, to promote system strategy iteration and self-learning are as follows:

[0044] Data Collection: Grid frequency, node voltage, energy storage system SOC, and renewable energy output are collected through a multi-source sensor network to calculate core peak-shaving performance indicators, including peak-shaving response speed, harmonic suppression rate, and energy storage utilization rate.

[0045] Long Short-Term Memory Network: Build a long short-term memory network model, use historical data series to predict future load and renewable energy output, and optimize model parameters by defining a mean square error loss function;

[0046] Abnormal event library construction and strategy matching: Build a feature vector library based on historical abnormal events, use a dynamic time warping algorithm to match the similarity between current events and historical cases, and extract the optimal response strategy;

[0047] Strategy iteration: Based on the prediction results and event matching results, an optimized peak-shaving strategy is generated. The peak-shaving strategy parameters are iteratively optimized through a genetic algorithm. The fitness function is constructed based on the energy storage utilization rate, harmonic suppression rate, and response time. The fitness function is expressed as Fitness = w1·η ESS +w2·(1-η harmonic )-w3·t resp , dynamically adjust the energy storage charging and discharging power and the ramp rate of conventional units.

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

[0049] In this invention, a federated reinforcement learning architecture is used to autonomously optimize the control strategy based on the status of the power generation / energy storage unit through the local agent, and Nash equilibrium calculation is performed in combination with the digital twin platform to solve the multi-objective conflicts of economy, reliability, and carbon emissions, generate global optimal scheduling instructions, reduce local decision-making delays, and adapt to the instantaneous load mutations of the offshore platform. At the same time, the digital twin platform is used to locate the harmonic source, and the energy storage system topology is automatically switched according to the harmonic frequency and position. The topology reconstruction is linked with the federated reinforcement learning (FRL) strategy to automatically adjust the diesel engine output during harmonic control to avoid the power gap caused by filtering. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a step diagram of an intelligent peak-shaving method proposed by the present invention that optimizes the combination of energy storage and conventional peak-shaving means. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] See also Figure 1 As shown, the present invention is an intelligent peak-shaving method that optimizes the combination of energy storage and conventional peak-shaving means, comprising the following steps:

[0053] Holographic situational awareness layer construction: Deploy micro-phasor measurement units (PMUs) and quantum sensors to build a holographic situational awareness layer. Synchronized phasor measurement technology is used to capture grid voltage, current phase angle, and frequency offset, and to monitor diesel engine cylinder pressure fluctuations. Diamond NV color center quantum pressure sensors are installed in the diesel engine cylinder to monitor combustion chamber pressure fluctuations with a resolution of 0.1 MPa, predicting sudden load changes. At the same time, power quality fingerprints are generated through edge computing nodes and local self-healing control is triggered (e.g., isolating a faulty branch within 0.1 seconds).

[0054] Data transmission and preprocessing: Build a quantum communication network for offshore platforms, using quantum key distribution (QKD) technology to encrypt data from micro-phasor measurement units (PMUs) and quantum sensors. Deploy a lightweight filtering algorithm (Kalman filter) on local agents (each power generation / storage unit) to remove sensor noise and extract key features (such as load mutation amplitude and harmonic frequency). After data compression, transmit it to the global coordinator via a 5G-Advanced link.

[0055] Adaptive Decision Engine: Leveraging a federated reinforcement learning (FRL) architecture, local agents (each power generation / storage unit) autonomously optimize control strategies based on the power generation / storage unit status. The power generation unit agent dynamically adjusts diesel engine output based on local status (SOC, oil pressure, temperature, etc.). The energy storage unit agent optimizes charging and discharging strategies (such as charging to the upper SOC limit during off-peak hours) based on load forecasts and electricity price signals. The harmonic control agent monitors THD in real time, triggering adaptive notch filter parameter adjustments. Combined with the digital twin platform, Nash equilibrium calculations are performed to resolve multi-objective conflicts among economy (minimizing diesel consumption), reliability (maximizing power supply continuity), and carbon emissions (maximizing the proportion of renewable energy), generating globally optimal scheduling instructions.

[0056] Execution control layer: Dynamically adjusts charging and discharging power through the power storage converter (PCS), uses flywheel energy storage to compensate for instantaneous power gaps, and uses adaptive notch filters and chaos injection PWM technology to track resonance points and disperse harmonic energy spectra in real time.

[0057] Dynamic topology reconstruction: Use the digital twin platform to locate harmonic sources and automatically switch the energy storage system topology based on the harmonic frequency and location;

[0058] Continuous optimization of the digital twin platform: Through bidirectional mapping between the digital twin model and the physical system, the execution effect of scheduling instructions is verified, and actual operation data is fed back to the federated reinforcement learning (FRL) model to optimize the reward function, achieving continuous optimization and closed-loop feedback of the digital twin platform;

[0059] Full-process closed-loop feedback and self-learning: Through key indicator monitoring and long short-term memory networks (LSTMs), future load / renewable energy output is predicted. Combined with the accumulation of abnormal event libraries, system strategy iteration and self-learning are promoted, achieving full-process closed-loop feedback and continuous evolution.

[0060] In one embodiment, the specific steps of constructing the holographic situational awareness layer are:

[0061] Through the deployment of micro phasor measurement units (PMUs) and synchronized phasor measurements: real-time capture of grid voltage, current phase angle and frequency offset, using the formula The phase deviation and time synchronization error are quantified by Δt = Δθ / 2πΔf. Combining the optical detection magnetic resonance (ODMR) principle of diamond NV color centers in quantum sensors, the pressure fluctuations in the diesel engine combustion chamber are monitored with a resolution of 0.1 MPa through the relationship f = f0 + k·P, thus predicting load mutations in advance.

[0062] Edge computing node generates power quality fingerprint: The edge computing node performs real-time analysis on micro phasor measurement unit (PMU) and quantum sensor data to generate a power quality fingerprint (PQF) containing indicators such as voltage sag (Sag) and total harmonic distortion (THD);

[0063] Holographic perception data fusion and load mutation prediction: Fusion of micro phasor measurement unit (PMU) phase angle deviation Δθ and quantum sensor pressure fluctuation ΔP, prediction of load mutation probability through support vector machine (SVM) model, triggering of flywheel energy storage compensation, diesel engine load rate adjustment and lithium battery discharge and other emergency peak shaving actions.

[0064] In one embodiment, the specific steps of data transmission and preprocessing are as follows:

[0065] Building offshore platform quantum communication network: Using BB84 protocol to generate dynamically updated 256-bit shared key, AES-256 encryption of micro phasor measurement unit (PMU) voltage data and quantum sensor pressure data, deployment of trust nodes between platforms to extend quantum communication distance;

[0066] Local Agent lightweight filtering algorithm: Deploying a lightweight Kalman filter algorithm in the local Agent, modeling sensor dynamic characteristics through a state equation and fusing quantum measurement values through an observation equation, calculating Kalman gain to achieve noise suppression (signal-to-noise ratio improved from 20 dB to 35 dB), and the formula for calculating the Kalman gain is The state update formula is represented as Where P(k) is the modeled sensor dynamic characteristics, Z(k) is the quantum measurement value, and key features such as load mutation amplitude and harmonic frequency are accurately extracted;

[0067] Data compression and 5G-Advanced transmission: 3-level decomposition and compression using Daubechies-4 wavelet basis (compression ratio CR = 8.2), and through 5G-Advanced link configuration of 30 kHz subcarrier spacing and DDDSU time slot format, preferential scheduling of peak shaving instruction data packets;

[0068] Global coordinator receives and decrypts: At the global coordinator end, through QKD key synchronization verification (SHA-256 hash comparison) and data decryption (D_AES-256), combining wavelet reconstruction algorithm to restore original signal time domain characteristics.

[0069] In one embodiment, the adaptive decision engine generates globally optimal scheduling instructions through deep integration of federated reinforcement learning and Nash equilibrium calculation, including the following steps:

[0070] Power generation unit agent autonomous optimization: The power generation unit agent uses quantum sensors to monitor the diesel engine's in-cylinder pressure, SOC, and temperature parameters in real time. Based on the objective function of minimizing fuel consumption and a low-load penalty factor, it dynamically adjusts the fuel injection rate and turbocharging pressure. For example, when the SOC is 40% and the in-cylinder pressure drops suddenly, the fuel injection rate is reduced to 60% of the rated value, reducing the fuel consumption rate from 220g / kWh to 205g / kWh.

[0071] Energy storage unit agent autonomous optimization: Based on the load forecast model and time-of-use electricity price signals, the energy storage unit agent uses a dual-objective optimization function to maximize arbitrage returns and minimize battery life loss to adjust charging and discharging power instructions;

[0072] Harmonic Suppression Agent autonomous optimization: The harmonic suppression agent monitors the total harmonic distortion (THD) in real time and applies adaptive notch filters and chaos injection PWM technology to dynamically adjust the filter band and inject a chaotic sequence with a Lyapunov exponent greater than 0, thereby reducing the harmonic amplitude.

[0073] Digital twin platform modeling: The global coordinator constructs diesel engine efficiency curves, lithium battery life models, and drilling load dynamic models within the digital twin platform, simulating a scenario where wind power output suddenly drops by 50% in the next 15 minutes.

[0074] Nash equilibrium solution: A multi-objective optimization function that includes fuel costs, power shortage risk, and carbon emissions, along with an improved particle swarm optimization algorithm, is used to find a Nash equilibrium solution that satisfies power balance constraints and ramp rate limits. For example, when the wind power gap is 0.6 MW, a strategy of increasing diesel engine output to 0.8 MW, discharging lithium batteries by 0.3 MW, and starting gas turbines by 0.5 MW is used. This strategy reduces total costs by 8% and carbon emissions by 12%.

[0075] Execution layer response: Through the execution layer response, the diesel engine electronic speed governor is adjusted to the target output, the lithium battery PCS is discharged, and the SOC is fed back to the global coordinator in real time.

[0076] In one embodiment, the method of dynamically adjusting the charge and discharge power through a power storage converter (PCS), compensating for the instantaneous power gap through flywheel energy storage, and tracking the resonance point in real time and dispersing the harmonic energy spectrum through adaptive notch filter and chaos injection PWM technology includes the following steps:

[0077] Dynamically adjust the charge and discharge power of the energy storage converter (PCS): By dynamically adjusting the charge and discharge power of the energy storage converter (PCS), a PID controller is used to calculate the power command in real time based on the grid frequency deviation or load forecast error. The power command calculation formula is expressed as P ref (t) = K p Δf(t)+K i∫Δf(t)dt+K d dΔf(t) / dt, where K p , K i , K d The power rate of change is set as a PID controller parameter, and combined with power rate of change limiting and smoothing control technology, the power rate of change limit is set to ensure that the PCS can accurately respond to power demand while avoiding the impact of frequent start and stop on equipment, thereby achieving rapid dynamic balance and stable operation of the power grid;

[0078] Flywheel energy storage compensates for instantaneous power gap: The flywheel energy storage system (FESS) is used to compensate for the instantaneous power gap. The high-frequency power fluctuation component is separated by a low-pass filter to generate a flywheel power command subject to its maximum energy storage capacity constraint. The flywheel power command is generated as P FESS (t) = K·P HF (t), K is the gain coefficient, and the maximum energy storage capacity constraint is expressed as follows: the flywheel's millisecond-level response characteristics are used to quickly absorb or release energy, effectively smoothing out instantaneous power shocks in scenarios such as sudden drops in photovoltaic output or sudden increases in load;

[0079] Adaptive notch filter and chaos injection PWM technology: An adaptive notch filter is deployed to track the grid resonant frequency in real time. A second-order damping filter is designed to suppress specific harmonic components. Combined with chaos injection PWM modulation technology, a chaotic signal is introduced into the traditional pulse-width modulation carrier to disperse the harmonic energy spectrum, converting concentrated harmonics into broadband noise. This significantly reduces total harmonic distortion and improves power quality.

[0080] In one embodiment, the specific steps of using the digital twin platform to locate the harmonic source and automatically switch the energy storage system topology according to the harmonic frequency and location are as follows:

[0081] Digital twin platform modeling and data acquisition: The digital twin platform integrates real-time monitoring data and historical operating data from the power grid and energy storage system. Fourier transforms are used to perform harmonic analysis on voltage and current signals to extract harmonic frequency components and amplitudes. Total harmonic distortion is then calculated to quantify power quality. A digital twin model of the power grid and energy storage system is constructed to achieve virtual mapping.

[0082] Harmonic source location and topology switching decision-making: Based on the impedance method, the harmonic source is located through harmonic impedance calculation and upstream and downstream impedance change comparison. Combined with the harmonic frequency characteristics (low-frequency harmonics use parallel mode, high-frequency harmonics use series mode) and the coordination requirements of multiple harmonic sources, the energy storage system topology is dynamically matched and the connection matrix of modular energy storage units is optimized to achieve efficient configuration of energy storage resources.

[0083] Topology switching execution and effect evaluation: Harmonic suppression and peak shaving strategies are implemented through topology switching of power electronic switches (IGBTs), and system performance is evaluated with THD reduction rate and peak shaving capacity as optimization targets.

[0084] In one embodiment, the specific steps of continuous optimization of the digital twin platform are:

[0085] Bidirectional mapping and dispatch instruction verification: By building a digital twin model of the power grid-energy storage system, the state space and action space are defined. The state space includes load power, PV output, energy storage power, state of charge, frequency, and voltage. The action space includes energy storage charging and discharging, load shedding, and thermal power unit output adjustment. Bidirectional mapping verification is performed using the frequency and voltage errors between the actual operating data of the physical system and the model prediction results.

[0086] Federated Reinforcement Learning (FRL) model training and optimization: The power grid is then divided into multiple regions using a federated reinforcement learning framework. Local reinforcement learning agents are deployed in each region, and local strategies are trained using the Q-learning algorithm. Each regional agent uploads local Q network parameters to a central server, and the global model parameters are aggregated using the Federated Averaging Algorithm (FedAvg). The reward function weights are dynamically adjusted based on feedback from the physical system (such as peak-shaving costs and harmonic suppression effects), achieving multi-objective collaborative optimization.

[0087] Closed-loop feedback: Deploy the optimized global strategy to the digital twin platform to generate new scheduling instructions, and continuously update the experience replay pool data through the execution results of the physical system.

[0088] In one embodiment, the specific steps of predicting future load / renewable energy output through key indicator monitoring and long short-term memory network (LSTM), combined with the accumulation of abnormal event database, to promote system strategy iteration and self-learning are as follows:

[0089] Data Collection: Grid frequency, node voltage, energy storage system SOC, and renewable energy output are collected through a multi-source sensor network to calculate core peak-shaving performance indicators, including peak-shaving response speed, harmonic suppression rate, and energy storage utilization rate.

[0090] Long Short-Term Memory (LSTM) Network: Build a LSTM model to use historical data sequences to predict future load and renewable energy output, and optimize model parameters by defining a mean square error (MSE) loss function.

[0091] Abnormal event library construction and strategy matching: Based on historical abnormal events, a feature vector library is constructed. The dynamic time warping (DTW) algorithm is used to match the similarity between current events and historical cases, extracting the optimal response strategy. This significantly improves the system's response speed to emergencies and the accuracy of its strategies.

[0092] Strategy iteration: Based on the prediction results and event matching results, an optimized peak-shaving strategy is generated. The peak-shaving strategy parameters are iteratively optimized through a genetic algorithm (GA). The fitness function is constructed based on the energy storage utilization rate, harmonic suppression rate, and response time. The fitness function is expressed as Fitness = w1·η ESS +w2·(1-η harmonic )-w3·t resp , dynamically adjust the energy storage charging and discharging power and the ramp rate of conventional units.

[0093] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent peak-shaving method that optimizes the combination of energy storage and conventional peak-shaving means, characterized in that: The following steps are involved: Holographic situational awareness layer construction: Deploy micro-phasor measurement units and quantum sensors to build a holographic situational awareness layer. Synchronized phasor measurement technology is used to capture grid voltage, current phase angle, and frequency offset. It also monitors diesel engine cylinder pressure fluctuations and predicts sudden load changes. Edge computing nodes are used to generate power quality fingerprints and trigger local self-healing control. Data transmission and preprocessing: Build a quantum communication network for offshore platforms, use quantum key distribution technology to encrypt data from micro-phasor measurement units and quantum sensors, deploy lightweight filtering algorithms on local agents to remove sensor noise and extract key features, and then compress the data and transmit it to the global coordinator via 5G-Advanced links. Adaptive Decision Engine: Leveraging a federated reinforcement learning architecture, the local agent autonomously optimizes control strategies based on the state of power generation and storage units. Combined with the digital twin platform, it performs Nash equilibrium calculations to resolve conflicts among multiple objectives, including economic efficiency, reliability, and carbon emissions, and generates globally optimal dispatch instructions. Execution control layer: Dynamically adjust the charging and discharging power through the energy storage converter, compensate for the instantaneous power gap through flywheel energy storage, and use adaptive notch filter and chaos injection PWM technology to track the resonance point in real time and disperse the harmonic energy spectrum; Dynamic topology reconstruction: Use the digital twin platform to locate harmonic sources and automatically switch the energy storage system topology based on the harmonic frequency and location; Continuous optimization of the digital twin platform: Through bidirectional mapping between the digital twin model and the physical system, the execution effect of scheduling instructions is verified, and actual operation data is fed back to the federated reinforcement learning model to optimize the reward function; Full-process closed-loop feedback and self-learning: Through key indicator monitoring and long-short-term memory networks, future load / renewable energy output is predicted, combined with the accumulation of abnormal event libraries to promote system strategy iteration and self-learning.

2. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The specific steps of constructing the holographic situational awareness layer are as follows: Through the deployment of micro-phasor measurement units and synchronized phasor measurement: real-time capture of grid voltage, current phase angle and frequency offset, using the formula The phase deviation and time synchronization error are quantified by Δt=Δθ / 2πΔf. Combining the optical detection magnetic resonance principle of diamond NV color centers in quantum sensors, the pressure fluctuation of the diesel engine combustion chamber is monitored through the relationship f=f0+k·P, and load mutations are predicted in advance. Edge computing nodes generate power quality fingerprints: Edge computing nodes analyze micro-phasor measurement units and quantum sensor data in real time to generate power quality fingerprints; Holographic perception data fusion and load mutation prediction: By integrating the phase angle deviation Δθ of the micro-phasor measurement unit and the pressure fluctuation ΔP of the quantum sensor, the load mutation probability is predicted through the support vector machine model, triggering flywheel energy storage compensation, diesel engine load rate adjustment and lithium battery discharge.

3. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The specific steps of data transmission and preprocessing are: Build a quantum communication network for offshore platforms: Use the BB84 protocol to generate dynamically updated shared keys, perform AES-256 encryption on the voltage data of micro-phasor measurement units and the pressure data of quantum sensors, deploy trusted nodes between platforms, and extend the quantum communication distance. Local Agent Lightweight Filtering Algorithm: A lightweight Kalman filter algorithm is deployed on the local agent. The sensor dynamic characteristics are modeled by the state equation and the quantum measurement values ​​are integrated with the observation equation to calculate the Kalman gain. The Kalman gain calculation formula is expressed as The state update formula is expressed as Where P(k) is the dynamic characteristics of the modeled sensor, Z(k) is the quantum measurement value, and the key features are extracted; Data compression and 5G-Advanced transmission: A Daubechies-4 wavelet basis is used for three-level decomposition and compression. Subcarrier spacing and time slot formats are configured over 5G-Advanced links, prioritizing peak-shaving command packets. Global coordinator reception and decryption: QKD key synchronization verification and data decryption are performed on the global coordinator, and the original signal time domain characteristics are restored in combination with the wavelet reconstruction algorithm.

4. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The adaptive decision engine generates global optimal scheduling instructions through the deep integration of federated reinforcement learning and Nash equilibrium calculation, including the following steps: Power generation unit agent autonomous optimization: The power generation unit agent uses quantum sensors to monitor the diesel engine's in-cylinder pressure, SOC, and temperature parameters in real time. It dynamically adjusts the fuel injection rate and turbocharging pressure based on the objective function of minimizing fuel consumption rate and low-load penalty coefficient. Energy storage unit agent autonomous optimization: Based on the load forecast model and time-of-use electricity price signals, the energy storage unit agent uses a dual-objective optimization function to maximize arbitrage returns and minimize battery life loss to adjust charging and discharging power instructions; Harmonic Suppression Agent autonomous optimization: The harmonic suppression agent monitors the total harmonic distortion rate in real time and applies adaptive notch filters and chaos injection PWM technology to dynamically adjust the filter band and inject a chaotic sequence with a Lyapunov exponent greater than 0, thereby reducing the harmonic amplitude. Digital twin platform modeling: The global coordinator constructs diesel engine efficiency curves, lithium battery life models, and drilling load dynamic models within the digital twin platform, simulating a scenario where wind power output suddenly drops by 50% in the next 15 minutes. Nash equilibrium solution: A multi-objective optimization function that includes fuel costs, power shortage risks, and carbon emissions, along with an improved particle swarm optimization algorithm, is used to find a Nash equilibrium solution that satisfies power balance constraints and ramp rate limits. Execution layer response: Through the execution layer response, the diesel engine electronic speed governor is adjusted to the target output, the lithium battery PCS is discharged, and the SOC is fed back to the global coordinator in real time.

5. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The method of dynamically adjusting the charging and discharging power through the energy storage converter, compensating the instantaneous power gap through flywheel energy storage, and tracking the resonance point in real time and dispersing the harmonic energy spectrum through the adaptive notch filter and chaos injection PWM technology includes the following steps: Dynamically adjust the charge and discharge power of the energy storage converter: By dynamically adjusting the charge and discharge power of the energy storage converter, a PID controller is used to calculate the power command in real time based on the grid frequency deviation or load forecast error. The power command calculation formula is expressed as P ref (t) = K p Δf(t)+K i ∫Δf(t)dt+K d dΔf(t) / dt, where K p , K i , K d It is a PID controller parameter and combines the power change rate limit and smoothing control technology to set the power change rate limit; Flywheel energy storage compensates for instantaneous power gap: The flywheel energy storage system is used to compensate for the instantaneous power gap. The high-frequency power fluctuation component is separated by a low-pass filter to generate a flywheel power command and is limited by its maximum energy storage capacity constraint. The flywheel power command is generated as P FESS (t) = K·P HF (t), K is the gain coefficient, and the maximum energy storage capacity constraint is expressed as; Adaptive notch filter and chaos injection PWM technology: An adaptive notch filter is deployed to track the grid resonant frequency in real time. A second-order damping filter is designed to suppress specific harmonic components. Combined with chaos injection PWM modulation technology, a chaotic signal is introduced into the traditional pulse-width modulation carrier to disperse the harmonic energy spectrum, converting concentrated harmonics into broadband noise.

6. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The specific steps of using the digital twin platform to locate the harmonic source and automatically switch the energy storage system topology according to the harmonic frequency and location are as follows: Digital twin platform modeling and data acquisition: The digital twin platform integrates real-time monitoring data and historical operating data from the power grid and energy storage system. Fourier transforms are used to perform harmonic analysis on voltage and current signals to extract harmonic frequency components and amplitudes. Total harmonic distortion is then calculated to quantify power quality. A digital twin model of the power grid and energy storage system is constructed to achieve virtual mapping. Harmonic source location and topology switching decision-making: Based on the impedance method, the harmonic source position is located through harmonic impedance calculation and upstream and downstream impedance change comparison. In combination with the harmonic frequency characteristics and the coordination requirements of multiple harmonic sources, the energy storage system topology mode is dynamically matched and the connection matrix of modular energy storage units is optimized; Topology switching execution and effect evaluation: Harmonic suppression and peak shaving strategies are implemented through topology switching of power electronic switches, and system performance is evaluated with THD reduction rate and peak shaving capacity as optimization targets.

7. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: Specific steps for continuous optimization of the digital twin platform: Bidirectional mapping and dispatch instruction verification: By building a digital twin model of the power grid-energy storage system, the state space and action space are defined. The state space includes load power, PV output, energy storage power, state of charge, frequency, and voltage. The action space includes energy storage charging and discharging, load shedding, and thermal power unit output adjustment. Bidirectional mapping verification is performed using the frequency and voltage errors between the actual operating data of the physical system and the model prediction results. Federated reinforcement learning model training and optimization: The federated reinforcement learning framework is then used to divide the power grid into multiple regions. Local reinforcement learning agents are deployed in each region, and local strategies are trained using the Q-learning algorithm. Each regional agent uploads local Q network parameters to the central server, and the global model parameters are aggregated using the federated averaging algorithm. The reward function weights are dynamically adjusted based on feedback data from the physical system to achieve multi-objective collaborative optimization. Closed-loop feedback: Deploy the optimized global strategy to the digital twin platform to generate new scheduling instructions, and continuously update the experience replay pool data through the execution results of the physical system.

8. The intelligent peak-shaving method of optimizing the combination of energy storage and conventional peak-shaving means according to claim 1, characterized in that: The specific steps for promoting system strategy iteration and self-learning by monitoring key indicators and using long-short-term memory networks to predict future load / renewable energy output, combined with the accumulation of abnormal event libraries, are as follows: Data Collection: Grid frequency, node voltage, energy storage system SOC, and renewable energy output are collected through a multi-source sensor network to calculate core peak-shaving performance indicators, including peak-shaving response speed, harmonic suppression rate, and energy storage utilization rate. Long Short-Term Memory Network: Build a long short-term memory network model, use historical data series to predict future load and renewable energy output, and optimize model parameters by defining a mean square error loss function; Abnormal event library construction and strategy matching: Build a feature vector library based on historical abnormal events, use a dynamic time warping algorithm to match the similarity between current events and historical cases, and extract the optimal response strategy; Strategy iteration: Based on the prediction results and event matching results, an optimized peak-shaving strategy is generated. The peak-shaving strategy parameters are iteratively optimized through a genetic algorithm. The fitness function is constructed based on the energy storage utilization rate, harmonic suppression rate, and response time. The fitness function is expressed as Fitness = w1·η ESS +w2·(1-η harmonic )-w3·t resp , dynamically adjust the energy storage charging and discharging power and the ramp rate of conventional units.

Images