Quick response control method and device based on millisecond-level data acquisition in energy storage system

Abstract

The invention discloses a quick response control method and device based on millisecond-level data acquisition in an energy storage system, and relates to the technical field of power electronics. According to the method, power grid state parameters are collected in a millisecond-level period through a distributed sensor network; validity verification and abnormal value elimination are carried out on the parameters; calculating a parameter change rate; inputting the preprocessed parameters and the change rate into a trained machine learning model to analyze the state of the power grid; determining a modal decision result of the power grid according to the analysis result; and outputting a power regulation instruction according to the modal decision result, wherein the power regulation instruction comprises active power, reactive power and harmonic compensation current. Millisecond-level data acquisition is adopted, power grid state changes can be rapidly adapted, and the electric energy quality is improved.

Description

Technical Field

[0001] This application relates to the field of power electronics technology, and in particular to a fast response control method and device based on millisecond-level data acquisition in an energy storage system. Background Technology

[0002] Existing traditional energy storage systems have significant limitations in data acquisition and processing, resulting in delays in responding to transient changes in the power grid. Specifically, these systems typically have a data acquisition cycle of at least 100 milliseconds, meaning they cannot capture transient changes in the power grid in real time, such as voltage drops or harmonic shifts. This response delay leads to lag in control commands, causing low-voltage mitigation response times to exceed 2 seconds, thus impacting the stability and reliability of the power grid.

[0003] Existing technologies disclose a method for millisecond-level data acquisition in energy storage systems and subsequent rapid power control to stabilize the power grid. However, this method lacks intelligent state analysis and multimodal decision-making capabilities based on machine learning models, making it unsuitable for complex power grid operating conditions. Other methods disclose the use of machine learning models to analyze power grid data and identify unstable states, but do not address the integration of millisecond-level data acquisition with multimodal collaborative control.

[0004] Therefore, there is an urgent need in this field for a rapid response control scheme for energy storage systems that can achieve millisecond-level data acquisition, intelligent state recognition, and multimodal collaborative control. Summary of the Invention

[0005] The purpose of this application is to provide a fast response control method and device based on millisecond-level data acquisition in an energy storage system. By adopting millisecond-level data acquisition, it can quickly adapt to changes in grid conditions and improve power quality.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] In a first aspect, this application provides a fast response control method for an energy storage system based on millisecond-level data acquisition, comprising:

[0008] Power grid status parameters are collected through a distributed sensor network at a set period; the set period is on the order of milliseconds.

[0009] The power grid status parameters are input into the local device manager to perform data validity verification and remove outliers, resulting in preprocessed power grid status parameters.

[0010] Calculate the rate of change of power grid state parameters based on the preprocessed power grid state parameters;

[0011] The preprocessed power grid state parameters and the rate of change of power grid state parameters are input into a trained machine learning model to analyze the current state of the power grid and obtain the analysis results; the analysis results include low voltage state, three-phase imbalance state, cross-regional coordination resource state, and communication anomaly state.

[0012] Based on the analysis results, the modal decision-making results of the power grid are determined; the modal decision-making results of the power grid include decentralized autonomous mode, centralized collaborative mode, and off-grid emergency mode;

[0013] Based on the modal decision results of the power grid, a power regulation command is output; the power regulation command includes active power, reactive power and harmonic compensation current.

[0014] Preferably, the distributed sensor network includes a BeiDou positioning module and a 4G / 5G communication module; the collected voltage, current, frequency and harmonic content are encrypted and transmitted using the national cryptographic SM4 algorithm, with an end-to-end delay of ≤2ms.

[0015] Preferably, the machine learning model includes: an input layer, a decision layer, and a switching condition; the input layer is used to normalize the input power grid state parameters and the rate of change of the power grid state parameters; the decision layer outputs the mode probability Pm through a three-layer GRU neural network, where m=1,2,3; the switching condition is to trigger mode switching when the mode probability Pm>0.8 and lasts for 3 sampling periods.

[0016] Preferably, the decentralized autonomous mode is used to calculate the local compensation amount through a dynamic game algorithm; the centralized collaborative mode is used to generate virtual adjustable capacity and allocate power according to line vulnerability; the off-grid emergency mode is used to predict the off-grid probability based on Stacking ensemble learning, and to fuse load mutation gradient and communication packet loss rate; if the predicted probability is >90%, the diesel generator is started to operate in parallel.

[0017] Preferably, the power grid state parameters include voltage, current, frequency, and harmonic content; the rate of change of the power grid state parameters includes the rate of change of voltage and the rate of change of current.

[0018] Secondly, the present invention provides a fast response control device based on millisecond-level data acquisition in an energy storage system, comprising:

[0019] The parameter acquisition module is used to acquire power grid status parameters through a distributed sensor network at a set period; the set period is on the order of milliseconds.

[0020] The preprocessing module is used to input the power grid status parameters into the local device manager, perform data validity verification and remove outliers on the power grid status parameters, and obtain the preprocessed power grid status parameters.

[0021] The rate of change calculation module is used to calculate the rate of change of the power grid state parameters based on the preprocessed power grid state parameters.

[0022] The analysis module is used to input the preprocessed power grid state parameters and the rate of change of power grid state parameters into a trained machine learning model to analyze the current state of the power grid and obtain analysis results. The analysis results include low voltage status, three-phase imbalance status, cross-regional coordination resource status, and communication anomaly status.

[0023] The decision-making module is used to determine the modal decision results of the power grid based on the analysis results; the modal decision results of the power grid include decentralized autonomous mode, centralized collaborative mode, and off-grid emergency mode;

[0024] The instruction module is used to output power regulation instructions based on the modal decision results of the power grid; the power regulation instructions include active power, reactive power and harmonic compensation current.

[0025] Thirdly, the present invention provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a fast response control method based on millisecond-level data acquisition in an energy storage system as described in any one of the above-mentioned methods.

[0026] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a fast response control method based on millisecond-level data acquisition in an energy storage system as described above.

[0027] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0028] This application employs a millisecond-level data acquisition sensor network with an acquisition period of less than or equal to 5 milliseconds, uses SM4 encrypted transmission, and has an end-to-end latency of less than or equal to 2 milliseconds.

[0029] This application employs a three-modal flexible switching mechanism, with the GRU neural network decision layer combined with a probabilistic triggering mechanism, achieving a mode switching accuracy of over 95%.

[0030] This application adopts the autonomous management of the transformer area to achieve a second-level response. By using a dynamic game algorithm, the response time in low voltage management is less than or equal to 1 second.

[0031] This application uses virtual adjustable capacity aggregation and entropy weight method to allocate power, combined with vulnerability weight, so that the regional peak shaving response time is less than or equal to 1 minute.

[0032] This application adopts off-grid plug-and-play networking with Stacking integrated prediction and phase synchronization switching technology, and the off-grid switching time is less than or equal to 8 milliseconds.

[0033] This application adopts domestically produced security control using national cryptographic chips and optocoupler isolation drivers, achieving a network attack resistance capability that meets the Level 4 standard of information security protection. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is an application environment diagram of a fast response control method based on millisecond-level data acquisition in an energy storage system according to an embodiment of the present invention;

[0036] Figure 2 This is a flowchart illustrating a fast response control method based on millisecond-level data acquisition in an energy storage system, according to an embodiment of the present invention.

[0037] Figure 3 This is a logic diagram of a fast response control method based on millisecond-level data acquisition in an energy storage system according to an embodiment of the present invention;

[0038] Figure 4 This is a schematic diagram of millisecond-level acquisition and real-time response - three-phase low voltage provided in an embodiment of the present invention;

[0039] Figure 5 This is a schematic diagram of millisecond-level acquisition and real-time response - three-phase imbalance management provided in an embodiment of the present invention;

[0040] Figure 6 This is a schematic diagram of the functional modules of a fast response control device based on millisecond-level data acquisition in an energy storage system according to an embodiment of the present invention;

[0041] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] The fast response control method based on millisecond-level data acquisition in the energy storage system provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server.

[0045] In one exemplary embodiment, such as Figure 2 As shown, a fast response control method based on millisecond-level data acquisition is provided for an energy storage system. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 206. Wherein:

[0046] Step 201: Collect power grid state parameters through a distributed sensor network at a set period; the set period is in the millisecond range.

[0047] Specifically, the distributed sensor network includes a BeiDou positioning module (positioning accuracy ±1m) and a 4G / 5G communication module; the collected voltage, current, frequency, and harmonic content are encrypted and transmitted using the national cryptographic SM4 algorithm, with an end-to-end delay ≤2ms. The set period is a millisecond-level period, specifically 1-5 milliseconds, preferably 3 milliseconds.

[0048] Step 202: Input the power grid status parameters into the local device manager, perform data validity verification and remove outliers to obtain preprocessed power grid status parameters;

[0049] Specifically, the data validity verification includes range verification, spectrum analysis, and correlation testing. The range verification ensures that the voltage is between 180V and 260V and the frequency is between 49.5Hz and 50.5Hz. The outlier removal adopts an improved isolated forest algorithm, combined with a sliding time window to detect transient and steady-state anomalies.

[0050] Step 203: Calculate the rate of change of power grid state parameters based on the preprocessed power grid state parameters;

[0051] Specifically, the rate of change of the power grid state parameters includes instantaneous rate of change and trend rate of change. The instantaneous rate of change is calculated using the central difference method, and the trend rate of change is obtained by linear regression analysis of data from the past 10 sampling periods. The formula for calculating the instantaneous rate of change of voltage is: ΔV / Δt=(V_{t+1}-V_{t-1}) / (2Δt), ensuring a calculation accuracy of ±0.1% / s.

[0052] Step 204: Input the preprocessed power grid state parameters and the rate of change of power grid state parameters into the trained machine learning model to analyze the current state of the power grid and obtain the analysis results; the analysis results include low voltage state, three-phase imbalance state, cross-regional coordination resource state, and communication anomaly state;

[0053] Step 205: Based on the analysis results, determine the modal decision results of the power grid; the modal decision results of the power grid include decentralized autonomous mode, centralized collaborative mode, and off-grid emergency mode;

[0054] Step 206: Output a power regulation command based on the modal decision results of the power grid; the power regulation command includes active power, reactive power and harmonic compensation current.

[0055] Among them, such as Figure 3 As shown, when performing step 201, the specific steps can be as follows:

[0056] The distributed sensor network includes a BeiDou positioning module (positioning accuracy ±1m) and a 4G / 5G communication module; the collected voltage, current, frequency and harmonic content are encrypted and transmitted using the national cryptographic SM4 algorithm, with an end-to-end delay ≤2ms.

[0057] Among them, such as Figure 3 As shown, when performing step 202, the specific steps can be as follows:

[0058] After the power grid status parameters are input into the local device manager, the system first performs data validity checks on these parameters to confirm that each parameter value is within the expected reasonable range. This is to identify and handle any abnormal values ​​that may be caused by equipment failure, data transmission errors, or other reasons. Data validity checks may include, but are not limited to, checking the range, type, format, and logical relationships between parameter values.

[0059] After outliers are identified, they are removed to ensure the accuracy and reliability of subsequent analyses. The outlier removal process may involve setting thresholds; any parameter exceeding these thresholds will be considered anomaly and excluded. After data validity verification and outlier removal, the resulting power grid state parameters are the preprocessed data.

[0060] Specifically, when executing step 203, the following can be done:

[0061] The rate of change of the power grid state parameters includes the rate of change of voltage and the rate of change of current, wherein:

[0062] The formula for calculating the rate of change of voltage is:

[0063] ;

[0064] The formula for calculating the rate of change of current is:

[0065] ;

[0066] Where ΔV represents the voltage change, ΔI represents the current change, and Δt represents the time change.

[0067] Among them, such as Figure 3 As shown, when performing steps 204-206, the specific steps can be as follows:

[0068] The machine learning model adopts an online incremental learning mechanism, updating the model parameters every 24 hours; the three-layer structure of the GRU neural network consists of 128 units, 64 units, and 32 units, respectively, and a Dropout rate of 0.2 is used to prevent overfitting; the modal probability Pm is normalized by the Softmax function to ensure that ∑Pm=1.

[0069] The decentralized autonomous mode calculates local compensation through a multi-agent dynamic game algorithm, taking into account the capacity constraints and response speed of each energy storage unit within the distribution area, to achieve Pareto optimality. The centralized collaborative mode uses the entropy weight-TOPSIS method to assess line vulnerability and combines virtual adjustable capacity to achieve fair and efficient power allocation. The off-grid emergency mode's Stacking ensemble learning includes three base learners: random forest, XGBoost, and LightGBM. The prediction results are fused through a logistic regression meta-learner, achieving a prediction accuracy of up to 96.3%.

[0070] The machine learning model includes: an input layer, a decision layer, and switching conditions;

[0071] The input layer is used to normalize the input power grid state parameters and the rate of change of the power grid state parameters; the decision layer outputs the mode probability through a three-layer GRU neural network. Where m=1,2,3; the switching condition is when the modal probability Mode switching is triggered when the value is >0.8 and lasts for 3 consecutive sampling periods.

[0072] The decentralized autonomous mode responds to low voltage and three-phase imbalance events within the transformer area (response time ≤ 1s) and is used to calculate local compensation through a dynamic game theory algorithm. The centralized collaborative mode responds to the upper-level peak-shaving instructions by aggregating cross-transformer area energy storage resources (response time ≤ 1min) and is used to generate virtual adjustable capacity and allocate power according to line vulnerability. The off-grid emergency mode predicts communication anomalies and initiates plug-and-play networking (switching time ≤ 10ms) and is used to predict off-grid probability based on Stacking ensemble learning, integrating load mutation gradients and communication packet loss rates. If the predicted probability is > 90%, diesel generators are started to operate in parallel.

[0073] The implementation of modal decision-making mainly relies on the following steps and techniques:

[0074] First, after millisecond-level data acquisition and edge preprocessing, the system obtains effective and pre-processed power grid state parameters and their rates of change. This data is then fed into a pre-trained machine learning model.

[0075] In the modal decision-making phase, machine learning models quickly analyze the current state and trends of the power grid based on the characteristics of the input data.

[0076] Based on the analysis results from the model, the system triggers a switch to one of the three modes:

[0077] Decentralized Autonomous Mode: When the model detects local events such as low voltage or three-phase imbalance within a distribution area, the system quickly switches to decentralized autonomous mode. In this mode, equipment within each distribution area can autonomously adjust its operating status to quickly respond to and resolve local problems. Due to the very short response time (≤1s), this helps reduce the impact of grid faults on users.

[0078] Centralized Coordination Mode: When the model identifies the need for cross-regional resource coordination to respond to higher-level peak-shaving instructions, the system switches to centralized coordination mode. In this mode, the system can aggregate energy storage resources from multiple regions for unified scheduling and optimized allocation. Due to the relatively long response time (≤1 minute), this allows the system sufficient time to evaluate and select the optimal resource allocation scheme.

[0079] Off-grid emergency mode: When the model predicts a potential communication anomaly, the system immediately switches to off-grid emergency mode. In this mode, the system can activate plug-and-play networking capabilities to ensure the power grid remains stable even if the main communication link is lost. Because the switching time is very short (≤10ms), this helps reduce the impact of communication failures on power grid stability.

[0080] In some embodiments, such as Figure 4As shown, a low voltage of <215V is detected before point A1. After filtering, this is determined to be the true and accurate value. The energy storage unit outputs active power in a ramp-up manner with a fixed step size. When the voltage fluctuates around the standard 220V, the energy storage unit outputs power accordingly to stabilize the voltage. When the load decreases (A2, A4) causing a sudden voltage surge >220V, the energy storage unit gradually reduces its power until the voltage stabilizes.

[0081] In some embodiments, such as Figure 5 As shown, this is the steady-state compensation effect of the energy storage unit output when three-phase imbalance occurs.

[0082] Based on the same inventive concept, this application also provides a fast response control device for millisecond-level data acquisition in an energy storage system for implementing the fast response control method based on millisecond-level data acquisition in the energy storage system described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the fast response control device for millisecond-level data acquisition in an energy storage system provided below can be found in the limitations of the fast response control method for millisecond-level data acquisition in an energy storage system described above, and will not be repeated here.

[0083] In one exemplary embodiment, such as Figure 6 As shown, a fast-response control device based on millisecond-level data acquisition is provided for an energy storage system, comprising:

[0084] The parameter acquisition module 601 is used to acquire power grid status parameters through a distributed sensor network at a set period; the set period is a millisecond-level period.

[0085] The preprocessing module 602 is used to input the power grid status parameters into the local device manager, perform data validity verification on the power grid status parameters and remove outliers to obtain preprocessed power grid status parameters.

[0086] The rate of change calculation module 603 is used to calculate the rate of change of the power grid state parameters based on the preprocessed power grid state parameters.

[0087] Analysis module 604 is used to input the preprocessed power grid state parameters and the rate of change of power grid state parameters into a trained machine learning model to analyze the current state of the power grid and obtain analysis results; the analysis results include low voltage state, three-phase imbalance state, cross-regional coordination resource state, and communication anomaly state.

[0088] The decision module 605 is used to determine the modal decision results of the power grid based on the analysis results; the modal decision results of the power grid include decentralized autonomous mode, centralized collaborative mode, and off-grid emergency mode;

[0089] The instruction module 606 is used to output power regulation instructions based on the modal decision results of the power grid; the power regulation instructions include active power, reactive power and harmonic compensation current.

[0090] As an optional implementation, the power grid state parameters include voltage, current, frequency, and harmonic content.

[0091] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores power regulation instructions. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a fast-response control method based on millisecond-level data acquisition in an energy storage system.

[0092] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0093] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0094] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0095] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0096] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0097] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0098] In summary, this application has the following technical effects:

[0099] 1) This application uses a millisecond-level data acquisition sensor network with an acquisition period of less than or equal to 5 milliseconds, uses SM4 encrypted transmission, and has an end-to-end delay of less than or equal to 2 milliseconds.

[0100] 2) This application adopts a three-modal flexible switching, and the GRU neural network decision layer is combined with a probability triggering mechanism, with a mode switching accuracy of more than 95%.

[0101] 3) This application adopts the autonomous management of the transformer area to achieve a second-level response. By using a dynamic game algorithm, the response time in low voltage management is less than or equal to 1 second.

[0102] 4) This application adopts virtual adjustable capacity aggregation and uses the entropy weight method to allocate power. Combined with vulnerability weight, the regional peak shaving response time is less than or equal to 1 minute.

[0103] 5) This application adopts off-grid plug-and-play networking with Stacking integrated prediction and phase synchronization switching technology, and the off-grid switching time is less than or equal to 8 milliseconds.

[0104] 6) This application adopts domestically produced security control using national cryptographic chips and optocoupler isolation drivers, and its anti-network attack capability reaches the Level 4 standard of information security protection.

[0105] To further improve the control performance, safety, and reliability of the present invention, one or more of the following optimization techniques may be adopted based on the above specific embodiments:

[0106] Data acquisition security optimization: The distributed sensor network uses the national cryptographic SM4 algorithm for encrypted data transmission and combines it with the Beidou positioning module to realize spatiotemporal calibration of the acquired data, ensuring the credibility and integrity of the data source; the sensor node adopts a dual-core heterogeneous architecture, with the data acquisition core and the communication core physically isolated to prevent side-channel attacks.

[0107] Machine learning model structure optimization: A three-layer GRU neural network is adopted as the core of the decision layer. Compared with traditional RNN or simple neural networks, GRU network can better capture the temporal characteristics of power grid state parameters and improve the accuracy of state identification. The model training adopts a transfer learning strategy, which is first pre-trained on the standard IEEE node system and then fine-tuned in specific application scenarios.

[0108] Multimodal switching condition optimization: Set a switching condition of mode probability Pm>0.8 for 3 consecutive sampling periods to avoid false switching caused by instantaneous fluctuations and improve system stability; at the same time, introduce a hysteresis comparator mechanism to prevent frequent mode oscillation switching.

[0109] Stacking ensemble learning optimization for off-grid emergency response: Heterogeneous base learners (including random forest, XGBoost and LightGBM) are used to predict off-grid probability. Weighted fusion of meta-learners improves prediction accuracy to over 95%. The ensemble learning model uses dynamic weight adjustment, adaptively updating the weights of each base learner based on historical prediction accuracy.

[0110] Real-time performance optimization: A hardware accelerator is embedded in the local device manager to accelerate the GRU neural network inference process with dedicated hardware, reducing the single inference time from 15ms to less than 2ms, ensuring that the system meets the millisecond-level response requirements; in-memory computing technology is used to reduce the data transmission latency between memory and processor.

[0111] Fault tolerance mechanism optimization: Establish a three-level fault tolerance system: multi-source data fusion technology is adopted at the sensor level, hot backup switching is implemented at the node level, and multiple protection strategies are set at the system level to ensure that the failure of a single component does not affect the overall function of the system.

[0112] The data validity verification adopts a three-level verification mechanism of "range verification - spectrum analysis - correlation test". The range verification limits the voltage to 180V-260V and the frequency to 49.5Hz-50.5Hz. Outlier removal adopts an improved isolated forest algorithm combined with a 50ms sliding time window.

[0113] The three-level verification mechanism clearly defines the acceptable range of parameters. The improved isolated forest algorithm, combined with the sliding window, not only eliminates transient anomalies but also identifies steady-state anomalies, solving the problem of easy omissions in single verification and ensuring that the reliability of the preprocessed data is ≥99%, providing reliable input for subsequent analysis.

[0114] The BeiDou positioning module supports sub-meter level positioning (accuracy ±1m) and is spatiotemporally bound to the collected data. The 4G / 5G communication module adopts a dual-link redundancy design to ensure uninterrupted data transmission.

[0115] Sub-meter level positioning enables spatiotemporal binding of data, and dual-link redundancy design avoids communication interruption, solving the pain points of difficult traceability and easy disconnection of distributed data collection, and improving the stability and traceability of data transmission.

[0116] The three-layer GRU neural network has 128, 64, and 32 units respectively, and the Dropout rate is set to 0.2. The input layer uses Min-Max normalization to map the data to the [0,1] interval.

[0117] By clarifying the GRU network structure parameters and normalization method, the Dropout rate suppresses overfitting, and normalization eliminates the influence of dimensions, the problems of unstable model training and poor generalization ability are solved, thereby improving the accuracy of modal probability output.

[0118] The line vulnerability is calculated using the entropy weight-TOPSIS method, which integrates three indicators: line load rate, voltage deviation, and harmonic distortion rate. The virtual adjustable capacity aggregation error is ≤3%.

[0119] Entropy weight-TOPSIS method quantifies line vulnerability to ensure more reasonable power allocation. Virtual adjustable capacity aggregation error control solves the problems of uneven resource allocation and low aggregation accuracy across distribution areas, and improves peak shaving response efficiency.

[0120] The voltage change rate is additionally calculated as a trend change rate. The data from the past 10 sampling periods are analyzed by linear regression. The current change rate is calculated using the central difference method to improve the accuracy, with an error ≤0.1% / s.

[0121] The application of trend change rate and central difference method can capture both instantaneous fluctuations and long-term trends, improve the accuracy of change rate calculation, and solve the problem that calculating only instantaneous change rate may easily overlook potential power grid risks.

[0122] The preprocessing module incorporates a hardware acceleration unit, achieving a total processing time of ≤1ms for three-level verification and outlier removal, and supports parallel processing of multiple sensor data streams. The hardware acceleration unit shortens preprocessing time, enables parallel processing of multiple data streams, solves the problem of high latency in processing massive amounts of sensor data, and ensures that the overall timing requirements for millisecond-level response are met.

[0123] The analysis module integrates an online incremental learning unit, which fine-tunes the model parameters every 24 hours based on newly collected data, achieving a modal recognition accuracy of ≥95%. Online incremental learning allows the model to adapt to changes in power grid operating conditions without retraining, solving the problem of poor adaptability after model solidification and maintaining a high recognition accuracy over the long term.

[0124] The decision module is equipped with mode switching hysteresis protection, with an interval of ≥500ms between two adjacent mode switches to avoid system oscillations caused by frequent switching. Hysteresis protection avoids frequent mode switching, prevents system oscillations, solves the problem of false switching caused by instantaneous fluctuations in the power grid, and improves the stability of system operation.

[0125] The processor employs a dual-core heterogeneous architecture, with one core dedicated to data acquisition and preprocessing, and the other core handling model inference and modal decision-making, ensuring parallel and efficient operation. This dual-core heterogeneous architecture enables task parallelism, avoids excessive load on a single core, resolves the issue of data processing and decision inference competing for resources, and guarantees that timing targets are met at each stage.

[0126] The computer-readable storage medium uses the national cryptographic algorithm SM4 to encrypt and store power grid status data and model parameters, supporting hierarchical control of data access permissions to prevent data leakage. Encrypted storage and access control comply with Level 4 requirements of the Information Security Protection System, addressing data storage security risks and preventing the leakage of sensitive parameters and unauthorized access.

[0127] The data validity verification also includes spectral analysis and correlation testing to ensure the consistency of parameters in the frequency and time domains.

[0128] The distributed sensor network also includes a temperature sensor for monitoring ambient temperature and performing comprehensive correction in conjunction with power grid status parameters.

[0129] The machine learning model also employs an online incremental learning mechanism, automatically updating model parameters every 24 hours to adapt to dynamic changes in the power grid.

[0130] The mode switching conditions also include a hysteresis comparator mechanism to prevent frequent switching caused by instantaneous fluctuations and improve system stability.

[0131] The rate of change of the power grid state parameters also includes the rate of change of frequency and the rate of change of harmonics, which are calculated using the formulas Δf / Δt and ΔH / Δt to comprehensively capture the transient characteristics of the power grid.

[0132] The device also includes a fault-tolerant module for implementing a three-level fault-tolerant system: sensor-level multi-source data fusion, node-level hot backup switching, and system-level multi-protection strategy.

[0133] The GRU neural network also incorporates a hardware accelerator, enabling a single inference time of ≤2ms and ensuring millisecond-level response.

[0134] The computer equipment described also features a domestically produced security chip for implementing national cryptographic algorithms and optical isolation, achieving a network attack resistance level of Level 4. The decision module also integrates phase synchronization switching technology to ensure a smooth transition of voltage and frequency during off-grid switching.

[0135] The storage medium also stores historical power grid data samples for training and validating machine learning models, thereby improving prediction accuracy.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fast response control method based on millisecond-level data acquisition in an energy storage system, characterized in that, include: Power grid status parameters are collected at millisecond intervals using a distributed sensor network. The power grid status parameters are input into the local device manager for data validity verification and outlier removal to obtain the preprocessed power grid status parameters. Calculate the rate of change of power grid state parameters based on the preprocessed power grid state parameters; The preprocessed power grid state parameters and the rate of change of power grid state parameters are input into a trained machine learning model to analyze the current power grid state and obtain analysis results. The analysis results include low voltage state, three-phase imbalance state, cross-regional coordinated resource state, and communication anomaly state. Based on the analysis results, the modal decision results of the power grid are determined, including decentralized autonomous mode, centralized collaborative mode, and off-grid emergency mode; Based on the modal decision results of the power grid, a power regulation command is output, which includes active power, reactive power and harmonic compensation current.

2. The method according to claim 1, characterized in that, The distributed sensor network includes a BeiDou positioning module and a 4G / 5G communication module. The collected power grid status parameters are encrypted and transmitted using the national cryptographic SM4 algorithm, with an end-to-end delay of ≤2ms. The millisecond period is 1-5 milliseconds.

3. The method according to claim 1, characterized in that, The machine learning model includes an input layer, a decision layer, and switching conditions; The input layer is used to normalize the input power grid state parameters and the rate of change of the power grid state parameters; the decision layer outputs the modal probability Pm through a three-layer GRU neural network, where m=1,2,3; The switching condition is triggered when the modal probability Pm > 0.8 and lasts for 3 sampling periods.

4. The method according to claim 1, characterized in that: The decentralized autonomous mode calculates the local compensation amount through a dynamic game algorithm, with a response time of ≤1s; The centralized collaborative mode generates virtual adjustable capacity and allocates power according to line vulnerability, with a response time of ≤1min; The off-grid emergency mode is based on Stacking ensemble learning to predict the off-grid probability, and integrates load mutation gradient and communication packet loss rate. If the predicted probability is >90%, the diesel generator is started to operate in parallel, and the switching time is ≤10ms.

5. The method according to claim 1, characterized in that, The power grid state parameters include voltage, current, frequency, and harmonic content; the power grid state parameter change rate includes voltage change rate and current change rate, calculated using the formulas ΔV / Δt and ΔI / Δt, where ΔV represents the voltage change, ΔI represents the current change, and Δt represents the time change.

6. A fast-response control device based on millisecond-level data acquisition in an energy storage system, characterized in that, include: The parameter acquisition module is used to acquire power grid status parameters at millisecond intervals through a distributed sensor network. The preprocessing module is used to input the power grid status parameters into the local device manager for data validity verification and outlier removal, so as to obtain the preprocessed power grid status parameters. The rate of change calculation module is used to calculate the rate of change of the power grid state parameters based on the preprocessed power grid state parameters. The analysis module is used to input the preprocessed power grid state parameters and the rate of change of power grid state parameters into a trained machine learning model to analyze the current power grid state and obtain the analysis results. The decision module is used to determine the modal decision results of the power grid based on the analysis results; The instruction module is used to output power regulation instructions based on the modal decision results of the power grid. The power regulation instructions include active power, reactive power and harmonic compensation current.

7. The apparatus according to claim 6, characterized in that, The decision layer of the machine learning model in the analysis module adopts a three-layer GRU neural network structure to output the modal probability Pm, where m=1,2,3, corresponding to the decentralized autonomous mode, the centralized collaborative mode, and the off-grid emergency mode, respectively.

8. The apparatus according to claim 6, characterized in that, The decision module is configured as follows: When the modal probability P1 > 0.8 and lasts for 3 sampling periods, switch to decentralized autonomous mode and calculate the local compensation amount through dynamic game algorithm; When the modal probability P2 > 0.8 and lasts for 3 sampling periods, switch to centralized cooperative mode, generate virtual adjustable capacity and allocate power according to line vulnerability; When the modal probability P3 > 0.8 and lasts for 3 consecutive sampling periods, switch to the off-grid emergency mode and predict the off-grid probability based on Stacking ensemble learning.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images