System and method for monitoring grid-connected performance of network construction type energy storage power station

Abstract

The invention discloses a system and a method for monitoring the grid-connected performance of a network-constructed energy storage power station, and belongs to the field of monitoring the grid-connected performance of the network-constructed energy storage power station, and the method comprises the following steps: collecting a power grid side parameter, an internal parameter of the energy storage power station and a dynamic response variable, and carrying out the signal preprocessing; steady-state performance prediction and optimization are realized through reinforcement learning; dynamic response optimization of virtual-physical cooperation is carried out; carrying out fault ride-through credible optimization on block chain evidence storage; electric energy quality optimization based on AI vision and graph network fusion; and performing energy management coordination analysis, and performing parameter adjustment on the network construction type energy storage power station according to an analysis result. According to the method, the execution parameters of the energy storage power station are adjusted by collecting and processing the core parameters and variables of grid connection of the network-forming type energy storage power station in real time, carrying out steady-state performance monitoring, dynamic response testing, fault ride-through capability verification, electric energy quality analysis and optimization and carrying out energy management coordination analysis and optimization.

Description

Technical Field

[0001] This invention relates to the field of grid-connected performance monitoring technology for grid-connected energy storage power stations, and in particular to a grid-connected performance monitoring system and method for grid-connected energy storage power stations. Background Technology

[0002] Grid-based energy storage power stations are energy storage systems that simulate the characteristics of synchronous generators using virtual synchronous generator technology. They can autonomously construct and maintain the voltage and frequency stability of the power grid, possessing the ability to actively support the grid. Their core lies in the voltage source characteristics of the energy storage converter (PCS), which can independently set the voltage amplitude and phase, operating without relying on external grid voltage and frequency signals. Grid-based energy storage can operate both grid-connected and off-grid, providing independent power supply, significantly improving the flexibility and stability of the power system.

[0003] As the proportion of renewable energy in the power grid system continues to increase, China's peak-shaving, frequency regulation, and voltage regulation capabilities are growing, significantly impacting grid flexibility. Grid-based energy storage power stations, with their voltage source characteristics, active control capabilities, and adaptability to multiple scenarios, have become a key technology in new power systems for addressing issues such as the volatility and low inertia of renewable energy sources.

[0004] However, there is currently a lack of complete and specific methods for monitoring the grid-connected performance of grid-connected energy storage power stations, to conduct overall monitoring of the grid-connected performance of grid-connected energy storage power stations, and to optimize parameters based on the monitoring data. Summary of the Invention

[0005] The purpose of this invention is to provide a grid-connected performance monitoring system and method for grid-connected energy storage power stations. By real-time acquisition and processing of the core parameters and variables of grid-connected energy storage power stations, steady-state performance monitoring, dynamic response testing, fault ride-through capability verification, power quality analysis and optimization are performed, and energy management coordination analysis and optimization are conducted, thereby adjusting the execution parameters of the energy storage power station.

[0006] To achieve the above objectives, this invention provides a method for monitoring the grid-connected performance of a grid-connected energy storage power station, comprising the following steps:

[0007] Step 1: Adaptive data acquisition and feature enhancement based on digital twin model. The acquired data includes grid-side parameters, internal parameters of the energy storage power station, and dynamic response variables during the grid connection process of the grid-type energy storage power station. Signal preprocessing is then performed on the acquired data.

[0008] Step 2: Perform steady-state performance monitoring, decompose the grid-side power components using a variational autoencoder, construct an Actor-Critic network to optimize power tracking parameters, and achieve steady-state performance prediction and optimization through reinforcement learning.

[0009] Step 3: Inject a composite disturbance of frequency step and broadband harmonics, calculate the response error using dynamic time warping and attention mechanism, and optimize the dynamic response by dynamically adjusting the virtual inertia coefficient through particle swarm optimization based on the phase margin and amplitude margin of the Bode plot.

[0010] Step 4: Construct an electromagnetic transient model to simulate a three-phase short circuit, calculate the dynamic reactive power support, use blockchain to record the protection action timestamps, and use smart contracts to automatically verify thresholds to perform blockchain-based fault-crossing reliability optimization.

[0011] Step 5: Optimize the grid topology by fusing AI vision with graph networks to reduce three-phase imbalance and optimize power quality;

[0012] Step 6: Establish a correlation model between SOC and maximum charge / discharge power, perform energy efficiency calculations, complete energy management coordination analysis, and adjust parameters of the grid-type energy storage power station based on the analysis results.

[0013] Preferably, in step one, the system state is predicted based on the real-time digital twin model, and the sampling frequency is dynamically adjusted using a Bayesian optimization algorithm, as shown in the following formula:

[0014] f s (t)=f base (11+λσ pred (t));

[0015] Among them, f s (t) is the adaptive sampling frequency at time t, f base σ is the reference sampling rate. pred (t) represents the state variance predicted by the digital twin model, and λ is the adjustment coefficient, λ = 0.5;

[0016] Enable 10kHz high-frequency sampling for key parameters and trigger the sampling window 5ms in advance through a real-time digital twin model;

[0017] The current harmonics are collected using a quantum magnetometer, and the time-frequency features are extracted using a Transformer network, as shown in the following formula:

[0018] X freq =Transformer(Wavele(v(t)));

[0019] Where v(t) is the voltage time-domain signal, and Wavele(·) is the wavelet transform operation;

[0020] Grid-side parameters include voltage amplitude, frequency, phase angle, active power, reactive power, harmonic components, and three-phase imbalance.

[0021] Internal parameters of an energy storage power station include battery SOC, battery SOH, charge and discharge efficiency, DC / AC inverter output characteristics, and the inertia coefficient and damping coefficient of the virtual synchronous machine (VSG).

[0022] Dynamic response variables include voltage frequency step response time, overshoot, oscillation decay time, low voltage ride-through (LVRT) and high voltage ride-through (HVRT) characteristics during faults.

[0023] Preferably, in step one, the signal preprocessing uses a low-pass filter to eliminate high-frequency noise, and the transfer function is:

[0024]

[0025] Where H(s) is the filter transfer function, s is the Laplace operator, n1 is the filter order, n1 = 2, and ζ k Let ω be the damping ratio of the k-th second-order element. c The cutoff angular frequency;

[0026] Multi-channel data synchronization is achieved based on GPS second pulse signals and IRIG-B codes, using the following formula:

[0027] t s =t GPS +Δt d -Δt o ;

[0028] Among them, t s For the synchronized timestamp, t GPS For the original GPS timestamp, Δt d For signal transmission delay, Δt o For device clock offset;

[0029]

[0030] Among them, t A→B Let t be the transmission time of the signal from device A to device B. B→A The signal transmission time from device B to device A;

[0031] The Lagrange interpolation method is used to unify the time base for signals with different sampling rates, as shown in the following formula:

[0032]

[0033] Where f(x) is the function value at the interpolation point, n2 is the number of interpolation base points, and y i Let x be the function value at the i-th base point. i Let x be the time coordinate of the i-th base point. i Let be the time coordinate of the j-th base point, and x be the time coordinate of the point to be interpolated.

[0034] Preferably, in step two, the power components on the grid side are decomposed using a variational self-encoder (VAE):

[0035] (P grid Q grid ) = VAE(P measured Q measured )+∈;

[0036] Among them, P grid Q grid P represents the decomposed active and reactive power on the grid side. measured Q measured Let ∈ be the total power measured, and ∈ be the prediction error;

[0037] Construct an Actor-Critic network to optimize power tracking parameters, with the reward function as follows:

[0038]

[0039] Where R is the cumulative reward value, P ref (t) represents the reference power command, P actual (t) represents the actual output power, SOC margin γ represents the SOC safety margin, and γ is the weighting coefficient, where γ = 0.1.

[0040] Preferably, in step three, a frequency step plus broadband harmonic composite disturbance is injected into the loop-in-the-loop PHIL system through power hardware, and the response error is calculated using dynamic time warping (DTW) and attention mechanisms.

[0041]

[0042] Among them, E dtw For dynamic time warping error, a i For the reference signal sequence point, b φ(i) For the actual response sequence points, dist is the Euclidean distance calculation function, and φ(i) is the time series mapping function;

[0043] Based on the phase margin and magnitude margin of the Bode diagram, the virtual inertia coefficient is dynamically adjusted through particle swarm optimization (PSO).

[0044] J new =J old +ω·ΔJ·sign(PM-45°);

[0045] Among them, J new J old These are the virtual inertia coefficients before and after the update, respectively. ω is the inertia weight, ω = 0.1, ΔJ is the inertia adjustment step size, and PM is the phase margin.

[0046] Preferably, in step four, an electromagnetic transient model is constructed to simulate a three-phase short circuit to a voltage of 0.1 pu, and the dynamic reactive power support is calculated based on the dq coordinate system.

[0047]

[0048] Among them, Q s V represents the total reactive power support. d V q These are the d-axis and q-axis voltage components, respectively. d I q These are the d-axis and q-axis current components, ΔQ. ML The reactive power compensation increment predicted by machine learning;

[0049] Blockchain records and protects action timestamps; smart contracts automatically verify thresholds.

[0050]

[0051] Where V is the real-time voltage, V n For the rated voltage, t delay To protect the action delay time.

[0052] Preferably, in step five, the THD spectrum is converted into a heatmap, and abnormal harmonics are detected using YOLOv8. The diagnostic model is as follows:

[0053] THD=YOLOv8(Wavelet(v(t)));

[0054] Wherein, THD is the result of harmonic anomaly detection, YOLOv8(·) is the YOLOv8 target detection model, and Wavelet(v(t)) is the wavelet transform result of the voltage signal;

[0055] Optimize grid topology based on graph neural network (GNN) to reduce three-phase imbalance:

[0056] U2% new =U2% old (1-αGNN(U2%) old ,topology));

[0057] Among them, U2% new 、U2% old These represent the three-phase imbalance before and after optimization, respectively. α is the optimization coefficient, α = 0.4. GNN(·) is the graph neural network model, and topology represents the power grid topology parameters.

[0058] Preferably, in step six, a correlation model between SOC and maximum charge / discharge power is established through polynomial fitting to verify the power limiting logic trigger accuracy when SOC ≤ 20%, with an error ≤ 3% of rated power. The formula is as follows:

[0059] P m (SOC)=aSOC 3 +bSOC 2 +cSOC+d;

[0060] Where SOC is the battery state of charge, ranging from 0% to 100%; a, b, c, and d are the polynomial fitting coefficients, and the goodness of fit R is... 2 ≥0.95;

[0061] The formula for calculating charge-discharge cycle efficiency is as follows:

[0062]

[0063] Among them, P discharge Let P be the discharge power at time point t. charge Let η(t) be the charging power at time t, and let η(t) be the cycle efficiency at time t, which reflects the energy conversion loss, and η≥90%.

[0064] Introducing carbon trading price signals C price (t), construct a multi-objective reinforcement learning model:

[0065]

[0066] Where π represents the charging / discharging strategy, E[·] represents the expectation operator, and C emission (t) represents the carbon emissions at time t, and ω1 and ω2 are weighting coefficients, where ω1 = 0.7 and ω2 = 0.3.

[0067] The optimized charging and discharging strategy is verified through digital twins before being distributed to the energy storage power station for parameter adjustment.

[0068] Δθ=RL(P ref SOC, C price );

[0069] Where Δθ is the grid-connected phase angle adjustment, RL(·) is the reinforcement learning decision function, and P ref For reference power command, C price This refers to the price of carbon trading.

[0070] This invention provides a system for monitoring the grid-connected performance of a grid-connected energy storage power station, comprising:

[0071] The data acquisition and preprocessing module acquires grid-side parameters, internal parameters of the energy storage power station, and dynamic response variables through a high-precision synchronous measurement device, a high-speed data acquisition card, and temperature and SOC sensors. It also eliminates switching noise through a low-pass filter, achieves multi-channel data synchronization based on a GPS clock, and completes signal preprocessing.

[0072] The steady-state performance optimization module uses a variational autoencoder to decompose the power components on the grid side, constructs an Actor-Critic network to optimize power tracking parameters, and achieves steady-state performance prediction and optimization through reinforcement learning.

[0073] The dynamic response optimization module injects a composite disturbance of frequency step and broadband harmonics, calculates the response error using dynamic time warping and attention mechanisms, and optimizes the dynamic response by dynamically adjusting the virtual inertia coefficient based on the phase margin and amplitude margin of the Bode diagram through particle swarm optimization.

[0074] The fault ride-through trusted optimization module constructs an electromagnetic transient model, simulates a three-phase short circuit, calculates dynamic reactive power support, uses blockchain to record protection action timestamps, and uses smart contracts to automatically verify thresholds to perform blockchain-based trusted optimization for fault ride-through.

[0075] The power quality analysis module optimizes the grid topology and reduces three-phase imbalance by using a fusion of AI vision and graph networks to improve power quality.

[0076] The energy management coordination analysis module establishes a correlation model between SOC and maximum charge / discharge power, calculates charge / discharge cycle efficiency, completes energy management coordination analysis, and adjusts the parameters of grid-type energy storage power stations based on comprehensive test results.

[0077] The central control and visualization module uses a time-series database to store raw data and calculation results, and displays voltage / frequency waveforms, power tracking curves, key parameters of harmonic spectrum, and generates compliance reports.

[0078] Therefore, the grid-connected performance monitoring system and method for grid-connected energy storage power stations described above have the following beneficial effects:

[0079] This invention provides a reliable data foundation for subsequent analysis by collecting grid-side parameters, internal parameters of energy storage power stations, and dynamic response variables. Through real-time acquisition and processing of core parameters and variables of grid-connected energy storage power stations, it performs steady-state performance monitoring, dynamic response testing, fault ride-through capability verification, power quality analysis and optimization, and energy management coordination analysis and optimization. This allows for adjustments to the execution parameters of the energy storage power station, providing systematic technical support for the efficient, safe, and reliable grid connection of grid-connected energy storage power stations. It is a key tool for supporting the construction of new power systems.

[0080] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0081] Figure 1 This is a flowchart of the method steps in an embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0084] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, such that a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0085] Similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0086] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed when in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0087] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0088] Example

[0089] like Figure 1 As shown, the grid-connected performance monitoring method for a grid-connected energy storage power station according to the present invention includes the following steps:

[0090] Step 1: Adaptive data acquisition and feature enhancement based on digital twin model. The acquired data includes grid-side parameters, internal parameters of the energy storage power station, and dynamic response variables during the grid connection process of the grid-connected energy storage power station. Signal preprocessing is then performed on the acquired data.

[0091] Grid-side parameters include voltage amplitude, frequency, phase angle, active power, reactive power, harmonic components, and three-phase imbalance. Internal parameters of the energy storage power station include battery SOC, battery SOH, charge / discharge efficiency, DC / AC inverter output characteristics, and the inertia coefficient and damping coefficient of the virtual synchronous machine (VSG). Dynamic response variables include voltage-frequency step response time, overshoot, oscillation decay time, and low-voltage ride-through (LVRT) and high-voltage ride-through (HVRT) characteristics during faults.

[0092] The system state is predicted based on a real-time digital twin model, and the sampling frequency is dynamically adjusted using a Bayesian optimization algorithm, as shown in the following formula:

[0093] f s (t)=f base (1+λσ pred (t));

[0094] Among them, f s (t) is the adaptive sampling frequency (Hz) at time t, f base =1kHz is the reference sampling rate, σ pred (t) represents the state variance predicted by the digital twin model, and λ = 0.5 is the adjustment coefficient;

[0095] For key parameters (such as voltage during LVRT), 10kHz high-frequency sampling is enabled, and the sampling window is triggered 5ms in advance through a real-time digital twin model.

[0096] A quantum magnetometer was used to collect current harmonics (accuracy improved to 0.001% THD), and time-frequency features were extracted using a Transformer network, as shown in the following formula:

[0097] X freq =Transformer(Wavele(v(t)));

[0098] Where v(t) is the voltage time-domain signal, and Wavele(·) is the wavelet transform operation.

[0099] Signal preprocessing uses a low-pass filter to eliminate high-frequency noise; the transfer function is:

[0100]

[0101] Where H(s) is the filter transfer function (complex frequency domain), s is the Laplace operator, n1 is the filter order (here n1 = 2, corresponding to a fourth-order filter), and ζ k Let ω be the damping ratio of the k-th second-order element. c ω is the cutoff angular frequency. c =2π×500Hz, suppressing switching noise ≥10kHz, passband ripple ≤0.5dB;

[0102] Multi-channel data synchronization is achieved based on GPS pulse-per-second (1PPS) signals and IRIG-B codes, with a synchronization error ≤1μs. The formula is as follows:

[0103] t s =t GPS +Δt d -Δt o ;

[0104] Among them, t s For the synchronized timestamp (s), t GPS For the original GPS timestamp (s), Δt d Let Δt be the signal transmission delay (s). o The device clock offset (s);

[0105]

[0106] Among them, t A→B Let t be the transmission time (s) of the signal from device A to device B. B→A Let be the transmission time (s) of the signal from device B to device A;

[0107] For signals with different sampling rates (e.g., voltage 10kHz, SOC 1Hz), the Lagrange interpolation method is used to unify the time base, with an interpolation error ≤0.1%, as shown in the following formula:

[0108]

[0109] Where f(x) is the function value (voltage, current, etc.) at the interpolation point, n2 is the number of interpolation base points, and y i x is the function value (original sampled value) at the i-th base point. i Let x be the time coordinate (s) of the i-th base point. i Let x be the time coordinate (s) of the j-th base point, and let x be the time coordinate (s) of the point to be interpolated.

[0110] Step two involves steady-state performance monitoring. A variational autoencoder is used to decompose the grid-side power components, and an Actor-Critic network is constructed to optimize power tracking parameters. Reinforcement learning is then used to predict and optimize steady-state performance. The specific steps are as follows:

[0111] Decomposing grid-side power components using a variational self-encoder (VAE):

[0112] (P grid Q grid ) = VAE(P measured Q measured )+∈;

[0113] Among them, P grid Q grid P represents the decomposed active and reactive power on the grid side (kW / kvar). measured Q measured The measured total power (kW / kvar) is denoted by ∈, where ∈ represents the prediction error (≤0.5% of rated power).

[0114] Construct an Actor-Critic network to optimize power tracking parameters, with the reward function as follows:

[0115]

[0116] Where R is the cumulative reward value, P ref (t) represents the reference power command (kW), P actual (t) represents the actual output power (kW), and SOC. margin =1-SOC / SOC margin The SOC safety margin is represented by γ = 0.1, which is the weighting coefficient.

[0117] Step 3: Inject a composite perturbation of frequency step and broadband harmonics. Calculate the response error using dynamic time warping and attention mechanisms. Based on the Bode plot phase margin and amplitude margin, dynamically adjust the virtual inertia coefficient through particle swarm optimization to optimize the dynamic response. The specific steps are as follows:

[0118] By injecting a frequency step and broadband harmonic composite disturbance into the power-in-loop PHIL system using power hardware, the response error is calculated using dynamic time warping (DTW) and an attention mechanism.

[0119]

[0120] Among them, E dtw For dynamic time warping error, a i For the reference signal sequence point, b φ(i) For the actual response sequence points, dist is the Euclidean distance calculation function, and φ(i) is the time series mapping function;

[0121] Based on the Bode plot phase margin (≥50°) and magnitude margin (≥8dB), the virtual inertia coefficient is dynamically adjusted through particle swarm optimization (PSO).

[0122] J new =J old +ω·ΔJ·sign(PM-45°);

[0123] Among them, J new J old The virtual inertia coefficients before and after the update (kg·m) are respectively. 2 ), ω is the inertia weight, ω=0.1, ΔJ is the inertia adjustment step size (0.01kg·m). 2 ), where PM is the phase margin (°).

[0124] Step four involves constructing an electromagnetic transient model to simulate a three-phase short circuit, calculating dynamic reactive power support, using blockchain to record protection action timestamps, and employing smart contracts to automatically verify thresholds for blockchain-based fault-crossing reliability optimization. The specific steps are as follows:

[0125] An electromagnetic transient model was constructed to simulate a three-phase short circuit to a voltage of 0.1 pu (lasting 200 ms), and the dynamic reactive power support was calculated based on the dq coordinate system.

[0126]

[0127] Among them, Q s V represents the total reactive power support (kvar). d V q These are the d-axis and q-axis voltage components (kV), respectively. d I q Let d and q be the current components (kV) of the d-axis and q-axis, respectively, and ΔQ. ML The reactive power compensation increment (kvar) predicted by machine learning;

[0128] Blockchain records and protects action timestamps; smart contracts automatically verify thresholds.

[0129]

[0130] Where V is the real-time voltage (kV), V n For rated voltage (kV), t delay The protection action delay time (ms).

[0131] Step 5: Calculate the three-phase unbalance by using Fast Fourier Transform (FFT) to detect harmonics and interharmonics, and calculate the proportion of negative sequence components to complete the power quality analysis. The specific steps are as follows:

[0132] The THD spectrum is converted into a heatmap, and abnormal harmonics (such as the 13th and 5th harmonics) are detected using YOLOv8. The diagnostic model is as follows:

[0133] THD=YOLOv8(Wavelet(v(t)));

[0134] Where THD is the harmonic anomaly detection result (binary value), YOLOv8(·) is the YOLOv8 target detection model, and Wavelet(v(t)) is the wavelet transform result of the voltage signal;

[0135] Optimize grid topology based on graph neural network (GNN) to reduce three-phase imbalance:

[0136] U2% new =U2% old (1-αGNN(U2%) old ,topology));

[0137] Among them, U2% new 、U2% old The values ​​are the three-phase imbalance (%) before and after optimization, respectively. α is the optimization coefficient, α = 0.4. GNN(·) is the graph neural network model, and topology is the power grid topology parameter.

[0138] Step six involves establishing a correlation model between SOC and maximum charge / discharge power, calculating energy efficiency, completing energy management coordination analysis, and adjusting parameters for the grid-type energy storage power station based on the analysis results. The specific steps are as follows:

[0139] A correlation model between State of Charge (SOC) and maximum charge / discharge power was established using polynomial fitting. The accuracy of the power limiting logic triggering was verified when SOC ≤ 20%, with an error ≤ 3% of rated power. The formula is as follows:

[0140] P m (SOC)=aSOC 3 +bSOC 2 +cSOC+d;

[0141] Where SOC is the battery's state of charge, ranging from 0% to 100%; a, b, c, and d are polynomial fitting coefficients, characterizing the nonlinear relationship between SOC and maximum charge / discharge power, with goodness of fit R0. 2 ≥0.95.

[0142] The formula for calculating charge-discharge cycle efficiency is as follows:

[0143]

[0144] Among them, P discharge Let P be the discharge power at time point t. charge Let η(t) be the charging power at time t, and let η(t) be the cycle efficiency (%) at time t, which reflects the energy conversion loss, and η≥90%.

[0145] Introducing carbon trading price signals C price (t), construct a multi-objective reinforcement learning model:

[0146]

[0147] Where π represents the charging / discharging strategy, E[·] represents the expectation operator, and C emission (t) represents the carbon emissions at time t (kgCO2 / kWh), and ω1 and ω2 are weighting coefficients, ω1 = 0.7 and ω2 = 0.3;

[0148] The optimized charging and discharging strategy is verified through digital twins before being distributed to the energy storage power station for parameter adjustment.

[0149] Δθ=RL(P ref SOC, C price );

[0150] Where Δθ is the grid-connected phase angle adjustment (°), RL(·) is the reinforcement learning decision function, and P ref For the reference power command (kW), C price The price is for carbon trading (RMB / kgCO2).

[0151] like Figure 2 As shown, the system for monitoring the grid-connected performance of a grid-connected energy storage power station according to the present invention includes:

[0152] Data acquisition and preprocessing module: It acquires grid-side parameters, internal parameters of energy storage power station, and dynamic response variables through high-precision synchronous measurement device, high-speed data acquisition card, temperature and SOC sensors, and eliminates switching noise through second-order Butterworth low-pass filter, realizes multi-channel data synchronization based on GPS clock, and completes signal preprocessing.

[0153] The steady-state performance optimization module uses a variational autoencoder to decompose the power components on the grid side, constructs an Actor-Critic network to optimize power tracking parameters, and achieves steady-state performance prediction and optimization through reinforcement learning.

[0154] The dynamic response optimization module injects a composite disturbance of frequency step and broadband harmonics, calculates the response error using dynamic time warping and attention mechanisms, and optimizes the dynamic response by dynamically adjusting the virtual inertia coefficient based on the phase margin and amplitude margin of the Bode diagram through particle swarm optimization.

[0155] The fault ride-through trusted optimization module constructs an electromagnetic transient model, simulates a three-phase short circuit, calculates dynamic reactive power support, uses blockchain to record protection action timestamps, and uses smart contracts to automatically verify thresholds to perform blockchain-based trusted optimization for fault ride-through.

[0156] The power quality analysis module optimizes the grid topology and reduces three-phase imbalance by using a fusion of AI vision and graph networks to improve power quality.

[0157] Energy Management Coordination Analysis Module: Establishes a correlation model between SOC and maximum charge / discharge power, calculates charge / discharge cycle efficiency, completes energy management coordination analysis, and adjusts grid-type energy storage power station parameters based on comprehensive test results;

[0158] Central control and visualization module: Uses time series database (InfluxDB) to store raw data and calculation results, and displays voltage / frequency waveforms, power tracking curves, key parameters of harmonic spectrum, and generates compliance reports (compliant with IEC62933, GB / T 36547 and other standards).

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring the grid-connected performance of a grid-connected energy storage power station, characterized in that: Includes the following steps: Step 1: Adaptive data acquisition and feature enhancement based on digital twin model. The acquired data includes grid-side parameters, internal parameters of the energy storage power station, and dynamic response variables during the grid connection process of the grid-type energy storage power station. Signal preprocessing is then performed on the acquired data. Step 2: Perform steady-state performance monitoring, decompose the grid-side power components using a variational autoencoder, construct an Actor-Critic network to optimize power tracking parameters, and achieve steady-state performance prediction and optimization through reinforcement learning. Step 3: Inject a composite disturbance of frequency step and broadband harmonics, calculate the response error using dynamic time warping and attention mechanism, and optimize the dynamic response by dynamically adjusting the virtual inertia coefficient through particle swarm optimization based on the phase margin and amplitude margin of the Bode plot. Step 4: Construct an electromagnetic transient model to simulate a three-phase short circuit, calculate the dynamic reactive power support, use blockchain to record the protection action timestamps, and use smart contracts to automatically verify thresholds to perform blockchain-based fault-crossing reliability optimization. Step 5: Optimize the grid topology by fusing AI vision with graph networks to reduce three-phase imbalance and optimize power quality; Step 6: Establish a correlation model between SOC and maximum charge / discharge power, perform energy efficiency calculations, complete energy management coordination analysis, and adjust parameters of the grid-type energy storage power station based on the analysis results.

2. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 1, characterized in that: In step one, the system state is predicted based on the real-time digital twin model, and the sampling frequency is dynamically adjusted using a Bayesian optimization algorithm, as shown in the following formula: f s (t)=f base (1+λσ pred (t)); Among them, f s (t) is the adaptive sampling frequency at time t, f base σ is the reference sampling rate. pred (t) represents the state variance predicted by the digital twin model, and λ is the adjustment coefficient, λ = 0.5; Enable 10kHz high-frequency sampling for key parameters and trigger the sampling window 5ms in advance through a real-time digital twin model; The current harmonics are collected using a quantum magnetometer, and the time-frequency features are extracted using a Transformer network, as shown in the following formula: X freq =Transformer(Wavele(v(t))); Where v(t) is the voltage time-domain signal, and Wavele(·) is the wavelet transform operation; Grid-side parameters include voltage amplitude, frequency, phase angle, active power, reactive power, harmonic components, and three-phase imbalance. Internal parameters of an energy storage power station include battery SOC, battery SOH, charge and discharge efficiency, DC / AC inverter output characteristics, and the inertia coefficient and damping coefficient of the virtual synchronous machine (VSG). Dynamic response variables include voltage frequency step response time, overshoot, oscillation decay time, low voltage ride-through (LVRT) and high voltage ride-through (HVRT) characteristics during faults.

3. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 1, characterized in that: In step one, signal preprocessing uses a low-pass filter to eliminate high-frequency noise, and the transfer function is: Where H(s) is the filter transfer function, s is the Laplace operator, n1 is the filter order, n1 = 2, and ζ k Let ω be the damping ratio of the k-th second-order element. c The cutoff angular frequency; Multi-channel data synchronization is achieved based on GPS second pulse signals and IRIG-B codes, using the following formula: t s =t GPS +Δt d -Δt o ; Among them, t s For the synchronized timestamp, t GPS For the original GPS timestamp, Δt d For signal transmission delay, Δt o For device clock offset; Among them, t A→B Let t be the transmission time of the signal from device A to device B. B→A The signal transmission time from device B to device A; The Lagrange interpolation method is used to unify the time base for signals with different sampling rates, as shown in the following formula: Where f(x) is the function value at the interpolation point, n2 is the number of interpolation base points, and y i Let x be the function value at the i-th base point. i Let x be the time coordinate of the i-th base point. i Let be the time coordinate of the j-th base point, and x be the time coordinate of the point to be interpolated.

4. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 1, characterized in that: In step two, the power components on the grid side are decomposed using a variational autoencoder (VAE): (P grid ,Q grid )=VAE(P measured ,Q measured )+∈; Among them, P grid Q grid P represents the decomposed active and reactive power on the grid side. measured Q measured Let ∈ be the total power measured, and ∈ be the prediction error; Construct an Actor-Critic network to optimize power tracking parameters, with the reward function as follows: Where R is the cumulative reward value, P ref (t) represents the reference power command, P actual (t) represents the actual output power, SOC margin γ represents the SOC safety margin, and γ is the weighting coefficient, where γ = 0.

1.

5. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 4, characterized in that: In step three, a frequency step and broadband harmonic composite disturbance is injected into the PHIL system via power hardware, and the response error is calculated using dynamic time warping (DTW) and attention mechanisms. Among them, E dtw For dynamic time warping error, a i For the reference signal sequence point, b φ(i) For the actual response sequence points, dist is the Euclidean distance calculation function, and φ(i) is the time series mapping function; Based on the phase margin and magnitude margin of the Bode diagram, the virtual inertia coefficient is dynamically adjusted through particle swarm optimization (PSO). J new =J old +ω·ΔJ·sign(PM-45°); Among them, J new J old These are the virtual inertia coefficients before and after the update, respectively. ω is the inertia weight, ω = 0.1, ΔJ is the inertia adjustment step size, and PM is the phase margin.

6. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 5, characterized in that: In step four, an electromagnetic transient model is constructed to simulate a three-phase short circuit to a voltage of 0.1 pu, and the dynamic reactive power support is calculated based on the dq coordinate system. Among them, Q s V represents the total reactive power support. d V q These are the d-axis and q-axis voltage components, respectively. d I q These are the d-axis and q-axis current components, ΔQ. ML The reactive power compensation increment predicted by machine learning; Blockchain records and protects action timestamps; smart contracts automatically verify thresholds. Where V is the real-time voltage, V n For the rated voltage, t delay To protect the action delay time.

7. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 6, characterized in that: In step five, the THD spectrum is converted into a heatmap, and abnormal harmonics are detected using YOLOv8. The diagnostic model is as follows: THD=YOLOv8(Wavelet(v(t))); Wherein, THD is the result of harmonic anomaly detection, YOLOv8(·) is the YOLOv8 target detection model, and Wavelet(v(t)) is the wavelet transform result of the voltage signal; Optimize grid topology based on graph neural network (GNN) to reduce three-phase imbalance: U2% new =U2% old (1-αGNN(U2% old ,toplogy)); Among them, U2% new 、U2% old These represent the three-phase imbalance before and after optimization, respectively. α is the optimization coefficient, α = 0.

4. GNN(·) is the graph neural network model, and topology represents the power grid topology parameters.

8. The grid-connected performance monitoring method for grid-connected energy storage power stations according to claim 7, characterized in that: In step six, a correlation model between SOC and maximum charge / discharge power is established through polynomial fitting to verify the power limiting logic trigger accuracy when SOC ≤ 20%, with an error ≤ 3% of rated power. The formula is as follows: P m (SOC)=aSOC 3 +bSOC 2 +cSOC+d; Where SOC is the battery state of charge, ranging from 0% to 100%; a, b, c, and d are the polynomial fitting coefficients, and the goodness of fit R is... 2 ≥0.95; The formula for calculating charge-discharge cycle efficiency is as follows: Among them, P discharge Let P be the discharge power at time point t. charge Let η(t) be the charging power at time t, and let η(t) be the cycle efficiency at time t, which reflects the energy conversion loss, and η≥90%. Introducing carbon trading price signals C price (t), construct a multi-objective reinforcement learning model: Where π represents the charging / discharging strategy, E[·] represents the expectation operator, and C emission (t) represents the carbon emissions at time t, and ω1 and ω2 are weighting coefficients, where ω1 = 0.7 and ω2 = 0.

3. The optimized charging and discharging strategy is verified through digital twins before being distributed to the energy storage power station for parameter adjustment. Δθ=RL(P ref ,SOC,C price ); Where Δθ is the grid-connected phase angle adjustment, RL(·) is the reinforcement learning decision function, and P ref For reference power command, C price This refers to the price of carbon trading.

9. A system for monitoring the grid-connected performance of a grid-connected energy storage power station according to any one of claims 1-8, characterized in that: include The data acquisition and preprocessing module acquires grid-side parameters, internal parameters of the energy storage power station, and dynamic response variables through a high-precision synchronous measurement device, a high-speed data acquisition card, and temperature and SOC sensors. It also eliminates switching noise through a low-pass filter, achieves multi-channel data synchronization based on a GPS clock, and completes signal preprocessing. The steady-state performance optimization module uses a variational autoencoder to decompose the power components on the grid side, constructs an Actor-Critic network to optimize power tracking parameters, and achieves steady-state performance prediction and optimization through reinforcement learning. The dynamic response optimization module injects a composite disturbance of frequency step and broadband harmonics, calculates the response error using dynamic time warping and attention mechanisms, and optimizes the dynamic response by dynamically adjusting the virtual inertia coefficient based on the phase margin and amplitude margin of the Bode diagram through particle swarm optimization. The fault ride-through trusted optimization module constructs an electromagnetic transient model, simulates a three-phase short circuit, calculates dynamic reactive power support, uses blockchain to record protection action timestamps, and uses smart contracts to automatically verify thresholds to perform blockchain-based trusted optimization for fault ride-through. The power quality analysis module optimizes the grid topology and reduces three-phase imbalance by using a fusion of AI vision and graph networks to improve power quality. The energy management coordination analysis module establishes a correlation model between SOC and maximum charge / discharge power, calculates charge / discharge cycle efficiency, completes energy management coordination analysis, and adjusts the parameters of grid-type energy storage power stations based on comprehensive test results. The central control and visualization module uses a time-series database to store raw data and calculation results, and displays voltage / frequency waveforms, power tracking curves, key parameters of harmonic spectrum, and generates compliance reports.

Images