IoT-based energy storage device management methods and systems

Abstract

This invention provides an IoT-based energy storage device management method and system, relating to the field of energy storage management technology. Based on real-time operating data of energy storage devices collected by an IoT platform, it determines energy dispatch target nodes and node interaction strategies through a blockchain network, generating decentralized dispatch instructions. Following these instructions, it connects to the edge computing unit of the target node, outputting network configuration parameters. Global status monitoring is performed through an end-edge-cloud collaborative architecture, acquiring data on physical layer charging / discharging anomalies and service layer communication anomalies. An anomaly handling strategy is generated based on an adaptive deep reinforcement learning model, recalculating the energy dispatch target node, forming a closed-loop control instruction, and feeding it back to the blockchain network. This invention avoids the processing delays easily caused by relying on master-slave blockchain structures in shared energy storage scenarios; it enhances adaptability to dynamic environments and improves decision-making effectiveness in complex power grid environments by automatically adjusting to cope with sudden changes in electricity prices or equipment anomalies.

Description

Technical Field

[0001] This invention relates to the field of energy storage management technology, and in particular to a method and system for managing energy storage devices based on the Internet of Things. Background Technology

[0002] As the global energy structure shifts towards cleaner and distributed energy, energy storage devices are increasingly being used in scenarios such as renewable energy consumption, grid peak shaving and valley filling, and user-side energy management. The Internet of Things (IoT) technology, with its real-time sensing of device status, multi-node data interconnection, and remote collaborative control, has achieved large-scale applications in industrial equipment management, smart homes, and intelligent transportation, providing technical support for the intelligent upgrading of traditional equipment management models. While existing IoT-based energy storage device management methods utilize Directed Acyclic Graphs (DAGs) to optimize the processing efficiency of concurrent blocks in blockchains, significantly reducing sorting complexity and redundant references, making them suitable for high-frequency energy storage scenarios, and combining IoT and blockchain technologies to ensure data trustworthiness and decentralized scheduling, and achieving multi-objective collaborative decision-making through deep reinforcement learning to improve the economy of energy allocation and equipment lifespan, existing methods rely on a master-slave blockchain structure. As the number of slave nodes increases, the master chain is prone to becoming a single point of failure, leading to processing delays. Deep reinforcement learning strategies can optimize multiple objectives, but training relies on historical data and lacks adaptability to dynamic environments. The fixed weight coefficients of the reward function cannot be automatically adjusted to cope with sudden changes in electricity prices or equipment malfunctions, resulting in a decline in the effectiveness of decision-making in complex power grid environments. The optimization of the initial topology graph mathematical model requires calculating the out-degree of each block. When the number of concurrent blocks increases, the processing steps become lengthy. These shortcomings lead to low energy allocation efficiency, high equipment lifespan loss, and difficulty in supporting cross-regional energy trading in shared energy storage scenarios.

[0003] In summary, existing energy storage equipment management models can no longer meet the demands of the new energy era for intelligent and refined equipment management. There is an urgent need for energy storage equipment management methods and systems based on the Internet of Things to at least address the aforementioned shortcomings. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for managing energy storage devices based on the Internet of Things (IoT) to solve the problems in the prior art. The specific technical solution is as follows:

[0005] This invention provides a method for managing energy storage devices based on the Internet of Things, comprising:

[0006] Step 1: Based on the real-time operating data of energy storage devices collected by the IoT platform, determine the target nodes for energy dispatch and the node interaction strategy through the blockchain network, and generate decentralized dispatch instructions;

[0007] Step 2: Based on the decentralized scheduling instructions, connect to the edge computing unit of the target node, generate a local monitoring network topology through dynamic sharding technology, and output network configuration parameters;

[0008] Step 3: Based on the network configuration parameters, perform global status monitoring through the end-edge-cloud collaborative computing architecture to obtain device anomaly data, including physical layer charging and discharging anomalies and service layer communication anomalies;

[0009] Step 4: Based on the equipment anomaly data, generate anomaly handling strategies through an adaptive deep reinforcement learning model. The model dynamically adjusts the weight coefficients of the reward function based on real-time environmental data.

[0010] Step 5: Based on the anomaly handling strategy, recalculate the energy scheduling target node through smart contracts, form closed-loop control instructions, and feed them back to the blockchain network.

[0011] Furthermore, the blockchain network is a decentralized architecture composed of multiple functionally heterogeneous parallel chains, including an energy storage status chain for storing device status and a transaction request chain for processing transaction requests. The chains communicate across chains through a distributed chain relay.

[0012] Furthermore, the node interaction strategy is an optimized scheduling scheme generated by a deep reinforcement learning DRL model based on an adaptive near-end strategy optimization PPO framework. Its elements include flow direction, power magnitude, time scale, and priority. The energy scheduling target node is recalculated through smart contracts and multi-objective optimization is performed using the non-dominated sorting genetic algorithm NSGA-II. The optimization objectives include system operation safety, maximizing economic benefits, satisfying grid ancillary service demand, and minimizing equipment lifespan loss.

[0013] Furthermore, the generation of the local monitoring network topology through dynamic fragmentation technology includes:

[0014] The granularity of the sharding is dynamically adjusted based on the power transmission parameters and time scale in the node interaction strategy.

[0015] When the power transmission parameters are greater than 100kW, the granularity of the slicing is set to 5 nodes / slice;

[0016] When the power transmission parameter is less than or equal to 100kW, the granularity of the sharding is set to 10 nodes / shard.

[0017] The time scale serves as the update cycle for the sharded topology.

[0018] Furthermore, the generation of the local monitoring network topology also includes constructing a digital twin verification model to simulate the expected operating state of energy storage devices under scheduling commands. This model generates the expected data stream by solving electrochemical-thermodynamic equations. The global state monitoring via an edge-cloud collaborative computing architecture includes:

[0019] Raw data is collected through monitoring agents deployed locally on the energy storage devices;

[0020] Real-time preprocessing and first-level anomaly detection are performed through edge computing units;

[0021] Deep analysis and second-level anomaly detection are performed through a cloud platform, and machine learning algorithms are applied to perform pattern recognition and root cause analysis on abnormal data.

[0022] Furthermore, the anomaly detection employs a dynamic threshold comparison algorithm, which adaptively adjusts the deviation tolerance based on the current operating status of the equipment and environmental conditions.

[0023] Furthermore, the electrochemical equation adopts the Butler-Volmer electrode reaction equation, the thermodynamic equation adopts the lumped parameter thermal equilibrium equation, the equation is solved using the Euler numerical method, and the simulation step size is consistent with the time scale.

[0024] Furthermore, the adaptive deep reinforcement learning model dynamically adjusts the weight coefficients of the reward function through a weight generation network under the meta-learning framework, which adopts a model-independent meta-learning (MAML) structure.

[0025] The present invention also relates to a management system for the aforementioned Internet of Things-based energy storage device management method, comprising:

[0026] The node determination module is used to determine the target nodes for energy dispatch and node interaction strategies through the blockchain network based on the real-time operating data of energy storage devices collected by the IoT platform, and to generate decentralized dispatch instructions.

[0027] The local monitoring network generation module is used to connect with the edge computing units of the target nodes according to decentralized scheduling instructions, generate the local monitoring network topology through dynamic sharding technology, and output network configuration parameters.

[0028] The monitoring module is used to perform global status monitoring based on network configuration parameters through an edge-cloud collaborative computing architecture, and to obtain abnormal device data, including physical layer charging and discharging abnormalities and service layer communication abnormalities.

[0029] The rescheduling module is used to generate anomaly handling strategies based on abnormal equipment data through an adaptive deep reinforcement learning model, and recalculate the energy scheduling target node through a smart contract according to the anomaly handling strategies, forming a closed-loop control command and feeding it back to the blockchain network.

[0030] Furthermore, the rescheduling module includes: verifying the digital signature of the closed-loop control command and executing the updated node interaction strategy; if the first and second anomalies occur, calling the backup node pool (node ​​selection criteria: SOH≥90%, no anomaly records, distance≤5km) to recalculate the scheduling target; if two consecutive anomalies occur, triggering equipment maintenance commands and activating the regional backup energy storage system.

[0031] Furthermore, an electronic device includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the IoT-based energy storage device management method.

[0032] 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

[0033] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0034] Figure 1 This is a schematic diagram of an IoT-based energy storage device management method in an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of an IoT-based energy storage device management system in an embodiment of the present invention. Detailed Implementation

[0036] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. This embodiment provides a method for managing energy storage devices based on the Internet of Things, such as... Figure 1 As shown, it includes:

[0037] Step 1: Based on the real-time operating data of energy storage devices collected by the IoT platform, determine the target nodes for energy dispatch and the node interaction strategy through the blockchain network, and generate decentralized dispatch instructions;

[0038] Among them, the real-time operating data is high-frequency status information of energy storage devices collected through the end-edge-cloud collaborative computing architecture, including but not limited to state of charge (SOC), charging and discharging power, health indicators, local electricity price signals and grid frequency deviation;

[0039] The blockchain network is a decentralized architecture consisting of multiple parallel chains with heterogeneous functions, including an energy storage status chain specifically for storing the status of the equipment and a transaction request chain specifically for processing transaction requests. The chains communicate across chains through a distributed chain relay.

[0040] Among them, the target nodes for energy dispatch are specific nodes that need to interact with each other in cross-regional energy transactions, which are automatically negotiated and matched according to smart contracts. These include: surplus energy supply nodes (such as wind farms and photovoltaic power stations), energy demand nodes (such as shared energy storage power stations), and frequency regulation ancillary service nodes.

[0041] The node interaction strategy is an optimized scheduling scheme generated based on an adaptive DRL model. Its elements include flow direction, power magnitude, time scale, and priority: the flow direction is determined by the cross-chain transaction topology (see Tangle structure); the power magnitude is optimized through real-time electricity price and SOC state; the time scale is calculated using the time window slicing algorithm of the Tangle structure, with a slicing period τ=2s; and the priority is determined by the dynamic reward function weight ω. i Output: The adaptive DRL model is an improvement on the widely accepted PPO (Proximal Policy Optimization) framework. The calculation logic of the reward function weight ωi references the dynamic weight allocation method of multi-objective reinforcement learning. Specifically, it achieves a dynamic trade-off between different objectives (safety, economy, grid support, and lifetime) by establishing a multi-objective reward function. The weight coefficients ensure that the priority output meets the real-time and security requirements of energy storage scheduling. The Tangle structure specifically refers to the DAG (Directed Acyclic Graph) architecture of the IOTA blockchain, and the time window slicing algorithm references the IOTA milestone timestamp mechanism.

[0042] The decentralized scheduling instruction is a structured data packet generated by a smart contract and confirmed by the consensus of the blockchain network. It includes the target node ID, interaction strategy parameters, digital signature and execution trigger conditions. After cross-chain security verification through the chain relay, the instruction is parsed and automatically executed by the target node's local smart contract. The execution result is recorded on the chain as an immutable transaction record.

[0043] Step 2: Based on the decentralized scheduling instructions, connect to the edge computing unit of the target node, generate a local monitoring network topology through dynamic sharding technology, and output network configuration parameters;

[0044] The edge computing unit for connecting to the target energy storage node is: an edge computing unit deployed at the target energy storage node establishes a secure communication connection with the cloud IoT platform and blockchain network. The edge computing unit has data caching, local computing and protocol conversion functions.

[0045] The dynamic generation of the local monitoring network includes: generating a local monitoring network topology based on the power transmission parameters and time scale in the node interaction strategy using dynamic sharding technology; the specific mapping rules are as follows: when the power transmission parameter > 100kW, the sharding granularity is set to 5 nodes / shard (to ensure real-time processing of high-frequency power data); when the power transmission parameter ≤ 100kW, the sharding granularity is set to 10 nodes / shard (to reduce the computing power consumption of edge computing units); the time scale τ = 2s is used as the update cycle for the sharded topology, and the sharding range is adjusted every 2s according to the real-time power value;

[0046] Specifically, this includes: generating a set of monitoring tasks for each interaction time node based on the power value, scheduling duration, and power direction in the node interaction strategy;

[0047] Among them, the monitoring task set is a collection of monitoring tasks generated for each interaction stage by referring to the node interaction strategy, and is set based on the knowledge base of energy storage equipment monitoring experience.

[0048] For example: Based on experience, it is determined that during the discharge process from energy storage node A to grid node B, it is necessary to verify whether the State of Charge (SOC) is higher than the lower limit threshold for discharge and whether the temperature is within a safe range during the discharge preparation phase; whether the discharge power ramp-up rate exceeds the maximum allowable value of the equipment and whether the actual discharge quantity deviates from the command by more than the tolerance during the real-time discharge phase; and whether the deviation between the final SOC state and the expected state is less than the allowable deviation during the discharge termination phase. Therefore, the monitoring task set is for the comparison of the discharge strategy from energy storage node A to grid node B. Before discharge, a monitoring task is generated to monitor whether the SOC state and temperature are within a safe range; during discharge, a monitoring task is generated to monitor the deviation of the power ramp-up rate and the discharge quantity; and after discharge, a monitoring task is generated to monitor the deviation of the final SOC state.

[0049] Based on the interaction time nodes and monitoring task set, a digital twin verification model is constructed. The specific connection logic is as follows: "Verify SOC > 20% during discharge preparation stage" in the monitoring task set is transformed into the initial simulation condition (initial SOC value ≥ 20%), "Verify power ramp rate ≤ 5MW / s during real-time discharge stage" is transformed into the power constraint condition of the thermodynamic model (simulation power change rate ≤ 5MW / s), and "Verify SOC deviation < 5% during discharge termination stage" is transformed into the deviation judgment standard of the simulation result (the deviation between the final simulated SOC value and the real-time value must be < 5%).

[0050] Among them, the digital twin verification model is a virtual simulation model built based on real-time device status and interaction strategies. It is used to simulate the expected operating state of energy storage devices under scheduling instructions and serves as a benchmark for data verification.

[0051] The construction and operation of the digital twin verification model includes:

[0052] (1) Real-time data input: The real-time status of the device is obtained through the edge computing unit as the initial condition for simulation;

[0053] (2) Interactive strategy integration: The scheduling instruction parameters are converted into model boundary conditions, wherein the scheduling instruction parameters include the power magnitude P determined in step 1. cmd Priority weight ω i And the time scale (τ=2s), the specific conversion rule is: P cmd The target current density mapped to the electrochemical model (current density = P) cmd / (battery rated voltage × effective electrode area)), ω i Used to adjust the priority verification weights (e.g., ω) of simulation results i (When the safety weight is greater than 0.5, temperature deviation is checked first), and τ=2s is used as the simulation step size;

[0054] (3) Multi-domain coupled simulation: Solving the electrochemical-thermodynamic equations to generate the expected data stream; the electrochemical equations adopt the Butler-Volmer electrode reaction equations recognized in the field: i = i0 [exp(α n Fη / RT) - exp(-α p [Fη / RT)]; where i is the current density, i0 is the exchange current density, and α n、 α p η is the electron transfer coefficient, F is the Faraday constant (96485 C / mol), η is the overpotential, and R is the gas constant (8.314 J / (mol)). K), T is the absolute temperature of the battery; the thermodynamic equation adopts the lumped-parameter thermal balance equation: Q 生 =Q 散 +mcΔT; where Q 生 The heat of electrochemical reaction, Q 散 (where m is the heat dissipation power of the cooling system, c is the battery mass, c is the battery specific heat capacity, and ΔT is the battery temperature change). The equations are solved using the Euler numerical method, and the simulation step size is consistent with the time scale τ=2s.

[0055] (4) Dynamic Verification: Synchronize expected data and real-time monitoring data through the Kafka interface and perform deviation detection; the core configuration of the Kafka interface is: Topic name "ESS_DT_Verify", number of partitions = number of shards (to ensure independent synchronization of shard data), synchronization frequency = simulation step size = 2s; the deviation detection rule is: when the single-step deviation of key parameters (SOC / temperature / power) is >5% or the deviation of 3 consecutive steps is >3%, dual processing is triggered: ① Digital twin model parameter correction: adjust the exchange current density i of the electrochemical model (correction formula: i new = i old × (1 - 0.05 × deviation rate)); ② Alarm feedback: Send a "model deviation alarm" to the edge computing unit, along with details of the deviation parameters; When Kafka disconnects, enable the local caching mechanism (caching capacity of 100 data entries), and prioritize synchronizing cached data after the connection is restored.

[0056] Based on the monitoring tasks at each interaction time point, the edge computing unit obtains real-time data collected by the local monitoring agent of the energy interaction target node.

[0057] Among them, the monitoring agent is a lightweight data acquisition and preprocessing module deployed locally on the energy storage device, which undertakes real-time data acquisition, filtering and feature extraction tasks;

[0058] Align the digital twin verification model and monitoring data with time series, establish communication links between verification data streams and monitoring data streams with the same time series benchmark, and construct a local monitoring network;

[0059] The local monitoring network is an execution network that uses a digital twin verification model to compare expected operational data with real-time monitoring data input data in real time, and to perform deviation detection and alarms.

[0060] Step 3: Based on the network configuration parameters, perform global status monitoring through the end-edge-cloud collaborative computing architecture to obtain device anomaly data, including physical layer charging and discharging anomalies and service layer communication anomalies;

[0061] The global status monitoring includes: using the data comparison engine in the local monitoring network to perform real-time comparison and analysis of the expected operating data generated by the digital twin verification model and the real-time monitoring data collected by the monitoring agent; and reporting the anomaly detection results of each local monitoring network to the cloud to collaboratively form a global status monitoring view.

[0062] Specifically, this includes: acquiring, based on the communication links established in the local monitoring network, the expected operational data stream generated by the digital twin verification model and having the same time-series benchmark, and the real-time monitoring data stream collected by the monitoring agent;

[0063] The two data streams are input into the data comparison engine for point-by-point comparison and analysis, and the deviation values ​​of key parameters are calculated.

[0064] The comparison analysis employs a dynamic threshold comparison algorithm, which adaptively adjusts the deviation tolerance based on the current operating status of the equipment and environmental conditions.

[0065] The acquisition of abnormal device data includes: generating an abnormal event record when the data comparison engine detects that the deviation of key parameters exceeds the dynamic threshold;

[0066] Specifically, this includes recording the timestamp of the anomaly, the anomaly parameter identifier, the deviation value, and the current operating environment data for each anomaly event;

[0067] It also performs preliminary classification of abnormal events based on a pre-set abnormal classification rule base;

[0068] The physical layer charging and discharging anomalies refer to anomalies related to the physical state of the energy storage device and the energy conversion process.

[0069] This includes: abnormal voltage fluctuations, excessive current, abnormal changes in SOC status, abnormal temperature rise, decreased insulation resistance, and deviations between actual charge / discharge and expected values.

[0070] The service layer communication anomalies are those related to data transmission, instruction execution, and network connectivity.

[0071] These include: edge computing unit communication latency timeout, excessively high monitoring data packet loss rate, command response timeout, interruption of connection with the blockchain network, and data verification failure.

[0072] Step 4: Based on the equipment anomaly data, generate anomaly handling strategies through an adaptive deep reinforcement learning model. The model dynamically adjusts the weight coefficients of the reward function based on real-time environmental data.

[0073] The decision-making process of the adaptive deep reinforcement learning model includes:

[0074] Based on the received device anomaly data and combined with real-time environmental data obtained from the IoT platform, a handling strategy for specific anomaly scenarios is generated.

[0075] The real-time environmental data includes: real-time electricity price signals from the power grid, load demand change rate, renewable energy generation forecast deviation rate, and system frequency deviation value;

[0076] The dynamic adjustment of the reward function weight coefficients includes:

[0077] Establish a multi-objective reward function R t= ω1(t)·R safety +ω2(t)·R economy + ω3(t)·R grid + ω4(t)·R lifespan The weight generation network under the meta-learning framework dynamically calculates the weight coefficients ω based on the real-time environmental state and device health status. i (t); The meta-learning framework adopts the MAML (Model Independent Meta-Learning) structure recognized in the field, and the MAML framework is the specific implementation of the "dynamic weight allocation method for multi-objective reinforcement learning" in step 1. The weight generation network outputs ωi(t) that conforms to the real-time environment and equipment health status through the inner loop (quickly adapting weights for the current abnormal scenario) and outer loop (optimizing the general weight calculation capability based on historical abnormal data) of MAML. The specific parameters and topology are as follows: MAML training parameters: 500 meta-training iterations, inner loop learning rate 0.01, outer loop learning rate 0.001; Weight generation network topology: Input layer (dimension 12, including 12 environmental equipment status parameters such as power grid frequency deviation, SOC, temperature, etc.) - Hidden layer (2 layers, 64 neurons per layer, activation function is ReLU) - Output layer (dimension 4, corresponding to ω1~ω4); Equipment health status adaptation: When SOH < 80%, the bias value of ω4 (lifetime weight) is increased by 0.2 in the network output layer to prioritize the protection of equipment life.

[0078] Specifically, the full list of 12 parameters includes: grid frequency deviation (Hz), energy storage device SOC (%), average temperature of battery cells (°C), real-time charging and discharging power (kW), battery insulation resistance (MΩ), local electricity price signal (yuan / kWh), grid load demand change rate (% / min), energy storage device SOH (%), cooling system inlet air temperature (°C), edge computing unit communication latency (ms), frequency of similar anomalies in the past 7 days (times / week), and renewable energy generation prediction deviation rate (%).

[0079] Specifically, when a grid frequency deviation exceeding a threshold is detected, R is automatically increased. grid The weight ω3(t) is used to ensure the stability of the power grid;

[0080] When the electricity price signal indicates a widening peak-to-valley difference, R is automatically increased. economy The weight ω2(t);

[0081] When the equipment temperature rises abnormally, the current or voltage exceeds the limit, or the insulation resistance decreases (i.e., an abnormal physical layer charging or discharging occurs), the R value will automatically increase. safety The weight ω1(t) is used to ensure equipment safety and prevent accidents;

[0082] When the energy storage device's State of Health (SOH) is detected to be declining rapidly, the number of charge / discharge cycles is approaching the threshold, or there are frequent historical anomalies, the R value is automatically increased. lifespan The weight ω4(t) is used to protect the equipment life and reduce destructive operations.

[0083] The generation of the exception handling strategy includes:

[0084] In response to physical layer charging and discharging anomalies, the generated handling strategies include: adjusting charging and discharging power commands, activating the backup cooling system, and performing isolation operations.

[0085] In response to service layer communication anomalies, the generated handling strategies include: switching communication channels, enabling redundant links, and resynchronizing blockchain data.

[0086] Step 5: Based on the aforementioned anomaly handling strategy, recalculate the energy scheduling target node through a smart contract, form a closed-loop control command, and feed it back to the blockchain network.

[0087] The recalculation of the energy dispatch target node includes:

[0088] Based on the specific handling measures in the exception handling strategy, the rescheduling smart contract deployed on the transaction request chain is triggered;

[0089] The rescheduling smart contract accesses the latest device status data on the current energy storage state chain and obtains real-time power grid operating parameters;

[0090] The optimal scheduling scheme is recalculated using a multi-objective optimization algorithm to generate a new list of energy scheduling target nodes. The multi-objective optimization algorithm adopts the NSGA-II (non-dominated sorting genetic algorithm) recognized in the field, with core parameters set as follows: population size 50, number of iterations 100, crossover probability 0.8, and mutation probability 0.01. The weight allocation rule for the optimization objectives is: system operation safety (weight 0.3) > equipment life loss minimization (weight 0.25) > grid ancillary service demand satisfaction (weight 0.25) > economic benefit maximization (weight 0.2), ensuring a safety-first scheduling logic.

[0091] The optimization objectives of the multi-objective optimization algorithm include: system operation safety, maximizing economic benefits, satisfying the demand for power grid ancillary services, and minimizing equipment lifespan loss.

[0092] The formation of the closed-loop control command includes:

[0093] The rescheduling smart contract encapsulates the new list of energy scheduling target nodes and the updated node interaction strategy into a structured data packet.

[0094] Data packets are digitally signed and timestamped to form closed-loop control commands with complete authentication and tamper-proof features;

[0095] The feedback to the blockchain network includes:

[0096] Broadcast closed-loop control instructions as new transactions to the blockchain network;

[0097] Verification and consensus confirmation are performed using the Byzantine fault-tolerant consensus algorithm; after verification, the instruction is recorded on the transaction request chain and synchronized to the relevant energy storage state chain via the chain relay.

[0098] The execution of the closed-loop control command includes:

[0099] 5011. After the edge computing unit of the target energy storage node detects a new closed-loop control command on the chain, it verifies the validity of the digital signature.

[0100] 5012. After successful verification, the updated node interaction strategy contained in the instruction will be executed automatically, and anomaly reproduction monitoring will be initiated.

[0101] 5013. If an "abnormality is still triggered after execution" (i.e., a secondary abnormality) is detected, tiered processing is triggered: 5013A. First secondary abnormality: The "backup node pool" (a list of redundant nodes pre-deployed in the energy storage state chain, containing 5-10 backup nodes) is invoked, the multi-objective optimization algorithm in step 5 is re-executed, and the node is switched to a backup node; 5013B. Two consecutive secondary abnormalities (i.e., tertiary abnormalities): An "equipment maintenance instruction" is triggered, the energy interaction of the node is suspended, the operation and maintenance system is notified through a smart contract, and the regional backup energy storage system is activated to replenish energy; The node selection criteria for the backup node pool are: SOH≥90%, no abnormal records in the past 7 days, and distance from the original target node ≤5km (ensuring communication latency ≤50ms).

[0102] 5014. If there are no secondary anomalies, the dynamic adjustment of energy dispatch is completed, and the execution result is recorded in the energy storage state chain.

[0103] The working principle and beneficial effects of the above technical solution are as follows: Firstly, high-frequency real-time data from energy storage devices is collected based on an IoT edge-cloud collaborative architecture. Then, a blockchain network consisting of an energy storage state chain, a transaction request chain, and a distributed chain relay is used. An adaptive DRL model is combined to determine the energy dispatch target node and node interaction strategy, generating decentralized dispatch instructions. Next, dynamic sharding technology is used to connect to the edge computing units of the target nodes, constructing a local monitoring network containing a digital twin verification model. Subsequently, global monitoring is conducted in an edge-cloud collaborative mode, distinguishing between physical layer and service layer anomalies. Finally, an anomaly handling strategy is generated based on the adaptive DRL model, forming closed-loop control instructions through smart contracts and feeding them back to the blockchain network. In the event of a secondary anomaly, a backup node pool or maintenance instructions are activated. The beneficial effects are that the decentralized blockchain architecture ensures tamper-proof and cross-chain security for dispatch instructions; dynamic sharding and digital twins improve the real-time performance and accuracy of monitoring; the adaptive DRL model achieves multi-objective optimization of anomaly handling (safety, economy, grid, and lifespan); and the closed-loop control mechanism and hierarchical anomaly handling strategy effectively reduce equipment failure risks, ensuring the safety, economy, and grid adaptability of energy storage dispatch.

[0104] In one embodiment, step 3: Based on the network configuration parameters, perform global status monitoring through an edge-cloud collaborative computing architecture to obtain device anomaly data, wherein the anomaly types include physical layer charging / discharging anomalies and service layer communication anomalies, including:

[0105] Based on the monitoring task set and time series configuration in the network configuration parameters, start the edge-cloud collaborative monitoring process.

[0106] The edge-cloud collaborative monitoring process includes a three-layer collaborative architecture: monitoring agent collects raw data, edge computing unit performs real-time preprocessing, and cloud platform performs in-depth analysis.

[0107] Real-time data collection of equipment operation is achieved through monitoring agents deployed locally on the energy storage devices.

[0108] Includes raw sampled values ​​of voltage, current, SOC, temperature, and charge / discharge power;

[0109] After filtering and extracting features from the raw data, the monitoring agent uploads it to the cloud IoT platform through the protocol conversion function of the edge computing unit.

[0110] The edge computing unit receives the preprocessed data and performs the first level of anomaly detection:

[0111] Real-time data comparison is performed based on the expected data range generated by the digital twin verification model;

[0112] When a data deviation is detected to exceed a preset threshold, a preliminary anomaly alarm is generated and the anomaly type is marked.

[0113] The cloud platform receives anomaly alerts and complete data streams uploaded by the edge computing unit and performs a second-level in-depth analysis:

[0114] Machine learning algorithms are applied to perform pattern recognition and root cause analysis on abnormal data. The machine learning algorithm uses the C4.5 decision tree algorithm, and the input features include: abnormal parameter type (voltage / current / SOC, etc.), abnormal duration (seconds), ambient temperature at the time of the abnormality (°C), equipment health status (SOH), and historical frequency of similar abnormalities (times / month). The model training data comes from a dataset of energy storage equipment fault records from the past 3 years (containing 10,000+ labeled samples, with labeling dimensions including "root cause of the abnormality (e.g., battery cell short circuit / communication module failure)" and "handling result").

[0115] Based on the anomaly classification rule base, anomalies are accurately classified to distinguish between physical layer charging and discharging anomalies and service layer communication anomalies.

[0116] The specific identification of physical layer charging and discharging anomalies includes:

[0117] When voltage fluctuations exceed ±10% of the rated value, current exceeds the limit, SOC abnormal jump exceeds 5%, temperature exceeds the safety threshold, or the actual charge / discharge quantity deviates from the expected value for a continuous period of more than 3%, it is determined to be an abnormal physical layer charge / discharge.

[0118] The specific identification of service layer communication anomalies includes:

[0119] When an edge computing unit communication latency exceeding 100ms, a packet loss rate exceeding 2%, a command response timeout exceeding 5 seconds, or a connection interruption with the blockchain network exceeding 10 seconds is detected, it is determined to be a service layer communication anomaly. The above thresholds are the basic reference values ​​for the dynamic threshold algorithm. In actual detection, the threshold will be adaptively adjusted according to the device's current SOC (e.g., when SOC < 20%, the voltage fluctuation threshold is relaxed to ±15%) and the ambient temperature (e.g., when the temperature > 40℃, the temperature anomaly threshold is lowered by 5℃). Specifically, the intermediate scenario adjustment formula is: Voltage fluctuation dynamic threshold = ±10% × (1 + (20% - SOC) / 20% × 0.5), where ±10% is used when SOC ∈ [20%, 100%], and the formula is used when SOC ∈ [0%, 20%) (e.g., when SOC = 10%, the threshold = ±12.5%); Temperature anomaly dynamic threshold = Safety threshold (default 45℃) - 5℃ × (temperature - 40℃) / 20℃, where temperature ∈ [0℃, 40℃]. When the temperature is 45℃, the threshold is calculated according to the formula when the temperature is ∈ (40℃, 60℃) (e.g., when the temperature is 50℃, the threshold is 42.5℃).

[0120] The final output contains complete device anomaly data, including:

[0121] The anomaly includes the timestamp of the anomaly occurrence, the anomaly type identifier, detailed data of the anomaly parameters, an anomaly level assessment, and preliminary handling suggestions.

[0122] The beneficial effects of the above technical solution are as follows: the three-layer collaborative architecture realizes the efficient division of labor of "local rapid response + cloud deep analysis", reducing data transmission redundancy and latency; the dynamic threshold algorithm adapts to different operating conditions of the equipment (SOC, ambient temperature) to avoid false alarms and missed alarms; the C4.5 decision tree algorithm relies on massive historical data to improve the accuracy of anomaly root cause identification, providing accurate basis for subsequent anomaly handling and further ensuring the operational stability of energy storage equipment.

[0123] In one embodiment, determining emergency response capability based on the device anomaly data includes:

[0124] Based on the abnormal content of physical layer charging and discharging anomalies, the corresponding emergency handling strategies are matched through the anomaly-emergency strategy mapping table;

[0125] The exception-emergency strategy mapping table is a pre-configured table that maps exception types to handling strategies.

[0126] For example, when the abnormal content is "frequency drops suddenly for 1 second", the matching emergency strategy is to call the rotating reserve response strategy (rapid response of energy storage and hydropower units).

[0127] Non-rotating reserve response strategy (gas turbine cold start);

[0128] When the abnormal message is "generator tripped", the matching emergency strategy is to invoke the spinning reserve response strategy (gas turbine rapid ramp-up, pumped storage unit increased generation).

[0129] Non-rotating reserve response strategy (cross-provincial standby capacity call);

[0130] The matching emergency response strategies are expanded on the timeline according to response time requirements:

[0131] Based on the response time characteristics of various reserve resources in the emergency strategy, the emergency sub-strategies are unfolded according to time sequence.

[0132] The rotating reserve sub-strategy is deployed on a timescale ranging from seconds to minutes.

[0133] Non-spinning reserve sub-strategies are deployed on timescales ranging from minutes to hours;

[0134] Based on the actual operating status of energy storage devices, construct a storage capacity description vector:

[0135] The reserve capacity description vector is represented in four-dimensional vector form [response time, duration, adjustment rate, rated capacity].

[0136] Specifically, for battery energy storage rotational storage, the vector is [200ms, 2 hours, 20MW / s, 100MW];

[0137] For the non-rotating reserve of the gas turbine, the vector is [5min, 8 hours, 10MW / min, 200MW];

[0138] Based on the time characteristics of the reserve capacity description vector, available reserves are marked on the time axis:

[0139] The appropriate intervention time point for reserve resources is determined based on the response time component of the vector.

[0140] The duration range of reserve resources is determined based on the duration component.

[0141] Mark the available time window for each reserve resource on the timeline;

[0142] Starting from the point when the anomaly occurred, traverse the timeline to match reserve resources:

[0143] At each point in time, check whether the currently available reserve resources meet the needs of the emergency sub-strategy.

[0144] If a combination of reserve resources that meets the criteria exists, then the time point can be marked as covered.

[0145] If all time points can be covered, then it is determined that there is an emergency response capability.

[0146] When an emergency response capability is determined to exist, an emergency response instruction sequence is generated:

[0147] Based on the timeline matching results, emergency response instructions are generated and executed in chronological order.

[0148] This includes the type of reserve resources to be called at each point in time, the power command, and the duration parameters;

[0149] The emergency response instruction sequence is submitted to the blockchain network via a smart contract.

[0150] After consensus is reached, the data will be distributed to each target energy storage node for execution.

[0151] The beneficial effects of the above technical solutions are as follows: the anomaly-strategy mapping table enables rapid matching of emergency solutions and avoids decision-making delays; time axis expansion and reserve capacity vector quantization ensure the timeliness and accuracy of emergency resource scheduling and avoid resource mismatch; the resource matching mechanism based on time windows can effectively verify the feasibility of emergency handling, ensure that suitable resources can be quickly mobilized when anomalies occur, and minimize the impact of anomalies on the operation of energy storage equipment and grid stability.

[0152] In one embodiment, the method for dynamically determining the importance threshold includes:

[0153] Based on the content of the power supply anomaly, the malicious value is quantified through a multi-dimensional anomaly detection index system.

[0154] The multidimensional anomaly detection index system includes data integrity dimension, service availability dimension, and resource security dimension;

[0155] Among them, the detection index for the data integrity dimension is the power / status data deviation rate, which is calculated as: |actual value - expected value| / expected value × 100%;

[0156] The service availability metric is the command non-response rate, calculated as: (Number of non-response commands / Total number of commands) × 100%.

[0157] The detection metric for resource security is the unauthorized API call rate, calculated as: (Number of unauthorized calls / Total number of API calls) × 100%.

[0158] The various test indicators were normalized:

[0159] The min-max normalization method is used to map each index value to the [0,1] interval.

[0160] The normalization formula is: X norm = (XX min ) / (X max -X min ),

[0161] Where X min and X max Determined based on historical data statistics;

[0162] Malicious values ​​are calculated using a weighted fusion method based on preset weights.

[0163] Malicious value = w1 × data integrity index + w2 × service availability index + w3 × resource security index;

[0164] The weighting coefficients w1, w2, and w3 were determined through historical optimization experiments.

[0165] Specifically, a grid search method is used to find the weight combination that minimizes the overall loss on the historical dataset;

[0166] The importance threshold is set to be equal to the malicious value:

[0167] When the importance threshold equals the malicious value, the severity of the anomaly is matched with the strictness of the handling.

[0168] The important value threshold dynamic adjustment mechanism is automatically updated based on the malicious value calculated in real time;

[0169] The optimization process with preset weights includes:

[0170] Collect historical records of power supply anomaly handling and construct a dataset containing anomaly indicators and handling results;

[0171] The loss function is defined as the weighted sum of processing delay time and resource waste.

[0172] The weight coefficients are iteratively optimized using the gradient descent method to minimize the overall loss.

[0173] The optimal weight combination is ultimately obtained as the preset weight.

[0174] IoT-based energy storage device management systems, such as Figure 2 As shown, it includes:

[0175] This invention provides an IoT-based energy storage device management system, such as... Figure 2 As shown, it includes:

[0176] The node determination module is used to determine the target nodes for energy dispatch and node interaction strategies through the blockchain network based on the real-time operating data of energy storage devices collected by the IoT platform, and to generate decentralized dispatch instructions.

[0177] The local monitoring network generation module is used to connect with the edge computing units of the target nodes according to decentralized scheduling instructions, generate the local monitoring network topology through dynamic sharding technology, and output network configuration parameters.

[0178] The monitoring module is used to perform global status monitoring based on network configuration parameters through an edge-cloud collaborative computing architecture, and to obtain abnormal device data, including physical layer charging and discharging abnormalities and service layer communication abnormalities.

[0179] The rescheduling module is used to generate anomaly handling strategies based on abnormal equipment data through an adaptive deep reinforcement learning model, and recalculate the energy scheduling target node through a smart contract based on the anomaly handling strategies to form a closed-loop control command and feed it back to the blockchain network.

[0180] The rescheduling module performs corresponding anomaly handling based on the anomaly type, and then reschedules the energy scheduling target node, including:

[0181] After performing physical power supply isolation in the first physical power supply isolation scenario, the first anomaly handling strategy is determined based on the physical layer charging and discharging anomaly content;

[0182] After processing based on the first anomaly handling strategy, the corresponding energy scheduling target nodes are rescheduled.

[0183] If the anomaly type is a service layer communication anomaly, obtain the second physical power supply isolation condition satisfied by the energy scheduling target node;

[0184] After implementing physical power isolation for the second physical power isolation scenario, the second anomaly handling strategy is determined based on the content of the service layer communication anomaly.

[0185] After processing based on the second anomaly handling strategy, the corresponding energy scheduling target nodes are rescheduled.

[0186] Among them, the first physical power supply isolation condition satisfied by the target node for energy dispatching includes:

[0187] Determine the reserve information of the target nodes for energy dispatch;

[0188] Extract reserve characteristics based on the reserve characteristic extraction template and reserve information;

[0189] Assess emergency response capabilities based on reserve characteristics;

[0190] If an emergency can be handled, physical power supply isolation of the corresponding energy dispatch target nodes should be carried out after the emergency is handled.

[0191] If emergency response is not possible, physical power supply isolation should be implemented directly for the corresponding energy dispatch target nodes.

[0192] Among them, the second physical power supply isolation condition satisfied by the target node for energy dispatching includes:

[0193] If the importance value of the energy interaction task is greater than or equal to the importance value threshold, then the physical power supply during the execution of the energy interaction task is guaranteed. After the energy interaction task is completed, the physical power supply isolation of the corresponding energy scheduling target node is then carried out.

[0194] If the importance value is less than the importance value threshold, then physical power supply isolation of the corresponding energy scheduling target node will be directly implemented.

[0195] The beneficial effects of the IoT-based energy storage device management system are as follows: Firstly, relying on a blockchain network (including the energy storage status chain and transaction request chain) to achieve decentralized scheduling command generation and cross-chain security verification, effectively ensuring the immutability of commands and the trust level of node interactions, avoiding the risks of data forgery or command tampering. Secondly, the local monitoring network generation module adopts dynamic sharding technology combined with a digital twin verification model, which can reduce the computing power consumption of edge computing units and improve the real-time performance and comparison accuracy of monitoring data, laying a reliable foundation for anomaly detection. Thirdly, the monitoring module adopts a three-layer collaborative architecture of "end-edge-cloud," coupled with dynamic threshold algorithms and C4.5. The decision tree algorithm can accurately distinguish between physical layer charging and discharging anomalies and service layer communication anomalies, significantly reducing false alarms and missed alarms, and enabling rapid anomaly location and root cause analysis. The rescheduling module generates multi-objective optimized anomaly handling strategies based on an adaptive DRL model, and combines hierarchical anomaly handling with differentiated physical power supply isolation. This prioritizes equipment safety and grid stability, maximizes economic benefits, and extends equipment lifespan, ultimately achieving a comprehensive improvement in the safety, economy, real-time performance, and stability of energy storage equipment operation.

[0196] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for managing energy storage devices based on Internet of Things, characterized in that, Comprise: Step 1: According to the real-time operation data of energy storage equipment collected by the Internet of Things platform, determine the energy scheduling target node and node interaction strategy through the blockchain network, and generate decentralized scheduling instructions; Step 2: According to the decentralized scheduling instructions, interface the edge computing unit of the target node, generate a local monitoring network topology through dynamic sharding technology, and output network configuration parameters; Step 3: According to the network configuration parameters, perform global state monitoring through an end-edge-cloud collaborative computing architecture, and obtain device exception data, wherein the exception types include physical layer charge and discharge exceptions and service layer communication exceptions; Step 4: According to the device exception data, generate an exception handling strategy through an adaptive deep reinforcement learning model, which dynamically adjusts the reward function weight coefficient according to real-time environmental data; Step 5: According to the exception handling strategy, recalculate the energy scheduling target node through the smart contract, form a closed-loop control instruction, and feed back to the blockchain network; The generation of the local monitoring network topology also includes the construction of a digital twin verification model for simulating the expected operation state of the energy storage equipment under the scheduling instructions, and the model generates expected data streams by solving electrochemical-thermodynamic equations; The electrochemical equation uses the Butler-Volmer electrode reaction equation, and the thermodynamic equation uses the lumped parameter heat balance equation. The equation solving uses the Euler numerical method, and the simulation step size and time scale are consistent; The electrochemical equation employs the Butler-Volmer electrode reaction equation: i = i0[exp(α n Fη / RT) - exp(-α p Fη / RT)]; where i is the current density, i0is the exchange current density, α n、 α p is the electron transfer coefficient, F is the Faraday constant, η is the overpotential, R is the gas constant, and T is the cell absolute temperature. 2.The Internet of Things based energy storage device management method of claim 1, wherein, The blockchain network is a decentralized architecture composed of multiple parallel chains with different functions, including an energy storage state chain for storing device state and a transaction request chain for processing transaction requests, and cross-chain communication between chains is performed through distributed chain relays. 3.The Internet of Things based energy storage device management method of claim 1, wherein, The node interaction strategy is an optimized scheduling scheme generated by a deep reinforcement learning (DRL) model based on an adaptive proximal policy optimization (PPO) framework, and its elements include flow direction, power size, time scale, and priority. The non-dominated sorting genetic algorithm II (NSGA-II) is used for multi-objective optimization when recalculating the energy scheduling target node through the smart contract, and the optimization objectives include system operation safety, maximum economic benefit, power grid auxiliary service demand satisfaction, and minimum device life loss. 4.The Internet of Things based energy storage device management method of claim 1, wherein, The generation of the local monitoring network topology through dynamic sharding technology includes: According to the power transmission parameters and time scale in the node interaction strategy, dynamically adjust the sharding granularity; When the power transmission parameter is greater than 100kW, the sharding granularity is set to 5 nodes per slice; When the power transmission parameter is less than or equal to 100kW, the sharding granularity is set to 10 nodes per slice; The time scale is the update period of the sharding topology. 5.The Internet of Things based energy storage device management method of claim 1, wherein, The global state monitoring through the end-edge-cloud collaborative computing architecture includes: Collecting raw data through the monitoring agent deployed locally on the energy storage equipment; Real-time preprocessing and first-level anomaly detection through the edge computing unit; Deep analysis and second-level anomaly detection through the cloud platform, applying machine learning algorithms for pattern recognition and root cause analysis of abnormal data. 6.The Internet of Things based energy storage device management method of claim 5, wherein, The anomaly detection uses a dynamic threshold comparison algorithm that adjusts the deviation tolerance adaptively according to the current operating state of the device and environmental conditions.

7. A management system for the method of managing energy storage devices based on the Internet of Things according to any one of claims 1 to 6, characterized in that, The system comprises: A node determination module is configured to determine an energy scheduling target node and a node interaction strategy through a blockchain network according to real-time operation data of the energy storage device collected by the Internet of Things platform, and generate a decentralized scheduling instruction. A local monitoring network generation module is configured to generate a local monitoring network topology through dynamic sharding technology for an edge computing unit of the target node according to the decentralized scheduling instruction, and output network configuration parameters. A monitoring module is configured to perform global state monitoring through an end-edge-cloud collaborative computing architecture according to the network configuration parameters, and obtain device abnormal data, wherein the abnormal types include physical layer charge and discharge abnormalities and service layer communication abnormalities. A rescheduling module is configured to generate an abnormality processing strategy through an adaptive deep reinforcement learning model according to the device abnormal data, and recalculate the energy scheduling target node through a smart contract according to the abnormality processing strategy, form a closed-loop control instruction, and feed back to the blockchain network. 8.The IoT-based energy storage device management system of claim 7, wherein, The rescheduling module includes a digital signature for verifying the closed-loop control instruction, and an updated node interaction strategy; if a first secondary abnormality occurs, a backup node pool is called to recalculate the scheduling target; if a continuous secondary abnormality occurs, a device repair instruction is triggered and a regional backup energy storage system is enabled.

9. An electronic device comprising a processor and a memory, characterized in that The memory stores a computer program, which, when executed by the processor, implements the method of any one of claims 1-6.

Images