Lithium battery cloud platform monitoring system for energy storage of power grid

Abstract

The invention belongs to the technical field of energy storage system monitoring, and particularly relates to an energy storage lithium battery cloud platform monitoring system for a power grid, which comprises an edge computing unit, a data transmission module, a cloud platform core processing unit, a data storage unit and an intelligent analysis module. The edge calculation unit performs data acquisition, preprocessing and event detection; the data transmission module supports low-delay dual-channel transmission; the cloud platform core processing unit adopts a micro-service and multi-agent architecture; the data storage unit is a multi-level architecture containing a high-speed cache and a time sequence database; the intelligent analysis module provides data snapshot, prediction and early warning, health management and energy optimization functions. According to the system, ultralow-delay transmission, high-concurrency processing, efficient storage and intelligent decision making are realized, and safe and efficient operation of the energy storage system is ensured.

Description

Technical Field

[0001] This invention belongs to the field of energy storage system monitoring technology, specifically a cloud platform monitoring system for lithium batteries used in power grid energy storage. Background Technology

[0002] As the global energy structure transitions towards renewable energy, lithium-ion battery energy storage systems, with their high energy density and high conversion efficiency, have become a key support for smoothing new energy fluctuations, participating in grid peak shaving and frequency regulation, and ensuring power system security.

[0003] To enable the monitoring of energy storage systems, existing technologies generally adopt cloud platform monitoring systems. These systems acquire parameters such as battery voltage, current, temperature, grid interaction power, and frequency through terminal acquisition devices, transmit them to the cloud for storage and analysis via the network, and provide users with services such as real-time status, historical queries, and health management through a visual interface. This approach has to some extent solved the information silo problem of traditional local monitoring.

[0004] Existing cloud platform monitoring systems have multiple deep-seated limitations when dealing with high real-time, high-concurrency, and high-data-volume scenarios in power grid energy storage: Transmission delay and insufficient determinism: Relying on the public network TCP / IP protocol, its retransmission and congestion control mechanisms prioritize data integrity and are not designed for real-time performance. Public network link latency, node queuing, etc., result in large fluctuations in transmission delay (200ms-800ms), which cannot meet the millisecond-level response requirements of overcurrent protection, thermal runaway early warning, etc. In emergency events, data is lost at the best processing time, and the monitoring system is reduced to a post-event recording tool. Storage performance bottlenecks and data loss: The traditional relational database “write-index-query” model is adopted. The ACID characteristics bring additional overhead such as table verification and index maintenance. A single write takes 10-20ms. High-concurrency writes are prone to queue congestion. High-frequency data is lost due to timeouts, resulting in a lack of complete data support for battery health analysis and fault tracing. Lack of dedicated acceleration technology: Without data snapshot technology, it is impossible to capture all operating data instantly during a fault, making it difficult to restore the state before the fault; Without high-speed caching and pre-reading mechanisms, high-frequency data access requires repeated reading of the database, with query time exceeding 1 second, and the delay is aggravated during peak periods, affecting the efficiency of operation and maintenance response and decision-making. To address this, the present invention provides a cloud platform monitoring system for lithium batteries used in power grid energy storage. Summary of the Invention

[0005] In order to overcome the shortcomings of existing technologies and solve the technical problems raised in the background art.

[0006] The technical solution adopted by the present invention to solve its technical problem is: the cloud platform monitoring system for lithium batteries for power grid energy storage described in the present invention includes an edge computing unit, a data transmission module, a cloud platform core processing unit, a data storage unit, an intelligent analysis module, and a user interaction interface module.

[0007] The edge computing unit is deployed at the energy storage power station site and is responsible for high-frequency, multi-dimensional data acquisition, local preprocessing, and real-time event detection of the operating parameters of the energy storage lithium battery system (including but not limited to battery clusters, battery modules, and individual cells). The edge computing unit includes a data acquisition interface module, a local data preprocessing module, a real-time event detection module, a local data caching module, and a data encapsulation and transmission module.

[0008] The data acquisition interface module is wired to the Battery Management System (BMS), Power Conversion System (PCS), and other auxiliary equipment (such as fire protection systems and temperature control systems) via industrial communication protocols (e.g., Modbus / TCP, CAN bus, IEC61850 protocol, or custom Ethernet protocol). It acquires operational data such as individual battery voltage, current, temperature, state of charge (SOC), state of health (SOH), and cycle count, as well as battery cluster power, capacity, charge / discharge status, and fault alarm information, in a periodic or event-driven manner. Simultaneously, it acquires grid interaction parameters such as frequency, voltage, active power, and reactive power from the grid side. The data acquisition interface module is configured with at least four independent physical communication ports, each supporting a communication rate of at least 10 Mbps to ensure high-speed concurrent acquisition of multi-source heterogeneous data. The local data preprocessing module receives the raw data transmitted from the data acquisition interface module and performs data denoising, format unification, timestamp calibration, and data compression operations. Data denoising is achieved through median filtering or moving average filtering algorithms to eliminate sensor noise. Format unification converts data from different sources into a unified JSON or Protocol Buffers format. The timestamp calibration is performed at the nanosecond level using a local clock server synchronized with the BeiDou or GPS timing module, ensuring the timing consistency of all data. The data compression employs lossless compression algorithms (such as Snappy or ZSTD) to reduce data transmission bandwidth requirements.

[0009] The real-time event detection module continuously monitors the pre-processed real-time data stream. When any monitored parameter (e.g., single-cell battery voltage, temperature) exceeds a preset safety threshold, a transient overcurrent occurs, or the rate of change exceeds a preset gradient, the real-time event detection module immediately triggers an alarm event. When a preset specific critical event occurs (e.g., initial signs of thermal runaway, severe overvoltage / undervoltage, short-circuit current), the real-time event detection module initiates a data snapshot mechanism. Within milliseconds, the data snapshot mechanism instructs the data acquisition interface module to synchronously collect and freeze all current full-scale operating data from all connected devices, forming a complete point-in-time snapshot dataset, which is then marked as the highest priority. The local data cache module uses an industrial-grade solid-state drive (e.g., NVM Express SSD) with high read / write bandwidth and low latency, with a capacity of at least 1TB, to temporarily store the pre-processed real-time data and triggered data snapshots, serving as a local buffer in case of cloud network interruptions or transmission congestion. The data encapsulation and transmission module encapsulates the data in the local data cache module into specific data packets according to the data priority and type, and sends them to the cloud platform core processing unit through the data transmission module. Data snapshots and emergency alarm data marked as highest priority are encapsulated using specific transport protocols and QoS levels.

[0010] The data transmission module is responsible for data communication between the edge computing unit and the cloud platform's core processing unit. The data transmission module is configured to support a dual-channel transmission mechanism: a critical channel for high-real-time, high-priority data, and a general-purpose channel for regular operational data.

[0011] The critical channel employs a customized application layer protocol based on User Datagram Protocol (UDP) or MQTToverTLS protocol with QoS2 level. The customized application layer protocol includes application layer acknowledgment mechanisms, error checking, and timeout retransmission logic to ensure reliable transmission of critical data while maintaining low latency. This critical channel provides a dedicated transmission link for data snapshots and emergency alarm data, with end-to-end transmission latency controlled within 100 milliseconds. The critical channel utilizes WAN acceleration technologies (such as SD-WAN) or dedicated network links to further optimize the transmission path and reduce physical latency. The general channel uses Transmission Control Protocol (TCP / IP) encrypted with Secure Sockets Layer / Transport Layer Security (TLS) for data transmission, suitable for periodically uploaded routine operational data. This general channel provides reliability and order guarantees for data transmission.

[0012] The data transmission module establishes multiple encrypted tunnels (e.g., based on IPSecVPN or TLS) between the edge computing unit and the cloud platform to ensure the confidentiality, integrity, and authentication of data transmission.

[0013] The cloud platform core processing unit serves as the central brain of the entire monitoring system, built using a microservice architecture and a multi-agent system (MAS) paradigm. This core processing unit is deployed on a container orchestration platform with elastic scaling capabilities (such as a Kubernetes cluster) to ensure high availability and scalability.

[0014] The microservice architecture includes, but is not limited to: data access service, data preprocessing service, alarm management service, status estimation service, fault diagnosis service, predictive analytics service, energy management optimization service, user authentication and authorization service, and API gateway service. Each microservice communicates asynchronously via a distributed message queue (e.g., based on an Apache Kafka cluster), using Protocol Buffers or JSON as the message format.

[0015] The Multi-Agent System (MAS) paradigm operates on top of a microservice architecture, where each agent is a software entity with independent goals, behavioral capabilities, and collaborative abilities, corresponding to a specific microservice instance or functional module. The MAS includes, but is not limited to: The DataIngestionAgent is responsible for receiving raw data from the data transmission module, performing preliminary data verification and routing, and writing the data to the cache layer and the real-time time-series database layer. This agent is configured to have dynamic scaling capabilities to handle instantaneous peaks in data flow.

[0016] The DataPreprocessingAgent reads data from a real-time time-series database and performs advanced data cleaning (e.g., outlier detection, data imputation), feature extraction, and data normalization to provide high-quality input for subsequent analysis. Outlier detection employs statistical methods (e.g., the 3σ criterion) or machine learning models (e.g., IsolationForest). The StateEstimationAgent uses Kalman filtering, extended Kalman filtering, or unscented Kalman filtering algorithms, combined with a battery equivalent circuit model, to perform high-precision real-time estimation of the battery's state of charge (SOC), state of health (SOH), and remaining useful life (RUL).

[0017] The FaultDiagnosisAgent is based on a rule engine, decision tree, support vector machine, or deep learning classifier to detect and locate battery system faults (such as internal short circuits, abnormal connections, and abnormal temperature rise) in real time and generate detailed fault reports.

[0018] The Predictive Analytics Agent uses deep learning models such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), or Transformer to predict future trends of key battery parameters (such as voltage, temperature, and internal resistance), providing a basis for early warning.

[0019] The DispatchOptimizationAgent combines grid load forecasting, real-time electricity prices, weather forecasting, energy storage system constraints (such as maximum charge / discharge power, maximum / minimum SOC), and operational objectives, and uses reinforcement learning, dynamic programming, or linear programming algorithms to generate the optimal charge / discharge strategy.

[0020] The ResourceManagementAgent monitors the running status and resource consumption of each microservice and agent in the core processing unit of the cloud platform, performs dynamic load balancing, fault recovery and resource allocation, and ensures that the system runs stably in a high-concurrency environment.

[0021] The agents exchange information and collaborate by sharing a distributed high-speed cache and message queue. The communication protocol between the agents adopts a service-oriented message passing pattern to ensure the reliability and real-time performance of message passing, thereby enabling distributed decision-making and task decomposition, effectively distributing the upper-layer load, and reducing the decision-making complexity of a single agent.

[0022] The data storage unit adopts a multi-level storage architecture, including a high-speed cache layer, a real-time time-series database layer, and a long-term historical database layer.

[0023] The high-speed cache layer employs a distributed in-memory database cluster (e.g., based on a Redis cluster) to store triggered data snapshots transmitted by the edge computing unit, frequently accessed real-time operational data (e.g., power curves for the most recent hour, minute-level average temperature), and data pre-read and pre-fetched by the intelligent analysis module. The high-speed cache layer provides millisecond-level read / write response times.

[0024] The real-time time-series database layer employs a distributed, high-concurrency write-optimized time-series database cluster (e.g., based on Apache IoTDB or InfluxDB clusters) optimized for time-series data storage and querying. It is used to persistently store raw sensor data and computation results processed by a data preprocessing agent. The time-series database supports processing tens of thousands of write requests per second and provides efficient time-range query functionality.

[0025] The long-term historical database layer employs a distributed columnar storage database cluster (e.g., based on Apache Cassandra or ClickHouse clusters) or a large-scale object storage system (e.g., S3-compatible storage) to store massive amounts of low-frequency, long-term historical data (e.g., historical operational data over several years). This layer supports large-scale batch processing analysis and data archiving for offline modeling and trend analysis by the intelligent analysis module.

[0026] The intelligent analysis module provides advanced data analysis and decision support functions, including a data snapshot processing mechanism, real-time alarm and prediction early warning algorithms, health status management and lifespan prediction algorithms, energy management optimization algorithms, and a big data analysis platform.

[0027] The data snapshot processing mechanism is responsible for receiving and parsing the data snapshots marked as the highest priority sent by the edge computing unit, and quickly writing them into the cache layer and the real-time time-series database layer for accurate state restoration before and after the fault occurs and fault root cause analysis.

[0028] The real-time alarm and predictive warning algorithm continuously monitors real-time data streams, combining machine learning models (such as decision trees and random forests) and statistical methods to identify and issue fault alarms and abnormal event warnings in real time. The predictive warning algorithm utilizes battery parameter prediction results output by the aforementioned predictive analysis agent to identify potential thermal runaway risks, overcharge and over-discharge risks, and performance degradation trends in advance. The health status management and lifespan prediction algorithm, based on the SOH and RUL estimation results provided by the aforementioned state estimation agent, combines historical battery cycle data, ambient temperature data, and operating condition data to construct a more accurate battery lifespan prediction model and provide a battery pack health assessment report.

[0029] The energy management optimization algorithm utilizes the charging and discharging strategies generated by the aforementioned scheduling optimization agent and dynamically adjusts them according to the real-time demand of the power grid and the operating status of the power station, thereby achieving the optimal economic benefits and power grid support functions of the grid-side energy storage system.

[0030] The big data analytics platform employs a distributed computing framework (such as an Apache Spark cluster or an Apache Flink cluster) to perform offline or near real-time batch and stream processing analysis on massive amounts of historical data stored in a long-term historical database layer. This is used to discover deep operational patterns, optimize model parameters, generate operational reports, and support management decisions.

[0031] The user interaction interface module provides access interfaces for multiple users, implements a web management portal based on a B / S (browser / server) architecture, and provides a mobile application interface based on an M / S (mobile / server) architecture.

[0032] The user interaction interface module provides a real-time visual interface for the energy storage system's operating status, including individual battery voltage, current, temperature curves, state of charge, health status, power curves, and power statistics.

[0033] The user interaction interface module provides a historical data query function, supporting users to efficiently search and export data based on conditions such as time range, battery cell, and data type.

[0034] The user interaction interface module provides alarm information display and hierarchical response functions. The system divides alarms into multiple levels such as early warning, ordinary alarm, and emergency alarm, and notifies operation and maintenance personnel through SMS, email, APP push and other means according to the level. It supports alarm confirmation and processing flow management.

[0035] The user interaction interface module supports the configuration and distribution of charging and discharging strategies, enabling remote control.

[0036] The user interaction interface module enables multi-user, multi-role permission management, providing customized data access permissions and functional operation interfaces for different roles such as power plant operators, end users, and regulatory agencies.

[0037] The present invention also includes a hardware system supporting the core processing unit and data storage unit of the cloud platform. The hardware system adopts a loosely coupled architecture design and consists of standardized server nodes, network devices, and storage devices.

[0038] The hardware system dynamically distributes request traffic among multiple microservice instances by deploying a load balancer cluster (e.g., an L7 load balancer based on Nginx or HAProxy), supporting high-concurrency access. The load balancer cluster is configured to support session-based persistence, health checks, and automatic failover to ensure continuous service availability.

[0039] The hardware system supports third-party platform access and provides data interfaces and data subscription services compliant with open standards (such as RESTful API and GraphQL) through the API gateway service layer, enabling secure and efficient data sharing and collaboration with external systems such as power grid dispatch centers, meteorological service providers, and power trading market platforms. The API gateway is responsible for API authentication, request routing, traffic shaping, and protocol conversion.

[0040] The hardware system supports access for multiple energy types. Through standardized data models and protocol adaptation layers, it can integrate data from various energy types, such as photovoltaic power generation, wind power generation, hydropower, and traditional thermal power, into a unified platform for management and scheduling.

[0041] The hardware system integrates real-time retrieval of logs and fault reports throughout the entire lifecycle. It uses a distributed log collection system (such as ELKStack or Loki) to centrally collect, store, index, and visualize the operation logs of all microservices, agents, and hardware devices, enabling rapid fault location, diagnosis, and tracing.

[0042] The hardware system ensures high reliability and data persistence by deploying redundant servers, storage, and network devices, and by combining distributed file systems (such as Ceph or GlusterFS) and data backup and recovery mechanisms.

[0043] The beneficial effects of this invention are as follows: 1. The cloud platform monitoring system for lithium batteries used in power grid energy storage described in this invention reduces the end-to-end transmission latency of key data in the energy storage system to less than 100 milliseconds by introducing edge computing units and optimized data transmission modules, especially customized low-latency transmission protocols and multi-encryption tunnels for key channels. This solves the technical contradiction in the prior art where high data transmission latency prevents timely triggering of safety commands, and transforms the monitoring system from a post-event recording tool into an active prevention and control tool.

[0044] 2. The cloud platform monitoring system for lithium batteries used in power grids and energy storage, as described in this invention, effectively improves the concurrent processing capability of data writing and the query response speed by adopting a distributed time-series database cluster and a multi-level storage architecture, combined with a high-speed caching layer and a predictive prefetching mechanism. The time-series database can support continuous high-frequency writing of tens of thousands of data entries per second, while the high-speed caching layer can reduce the query response time for hot data to the millisecond level. This solves the problems of low efficiency and excessively long query times in traditional relational databases during high-frequency data writing, ensuring data integrity and real-time accessibility.

[0045] 3. The cloud platform monitoring system for lithium batteries used in power grids and energy storage, as described in this invention, decomposes complex monitoring and management tasks into multiple collaborative agents by introducing a multi-agent system (MAS) paradigm, and uses a distributed message queue for asynchronous communication and task scheduling. This architecture achieves effective load balancing of the system, significantly improves data processing quality and management efficiency in high-concurrency environments, and avoids data congestion and response delays that may result from traditional centralized processing.

[0046] 4. The cloud platform monitoring system for lithium batteries used in power grid energy storage described in this invention provides accurate and complete data support for fault root cause analysis by synchronously and instantaneously capturing all the full operating data of all battery cells within milliseconds of detecting abnormal events, which greatly improves the depth and efficiency of fault tracing.

[0047] 5. The cloud platform monitoring system for lithium batteries used in power grid energy storage described in this invention can identify potential risks in advance by performing high-precision prediction and trend analysis of battery status, and automatically generate and optimize charging and discharging strategies by combining external information such as power grid load and electricity price, thereby realizing the intelligent operation of the energy storage system and maximizing economic benefits.

[0048] 6. The cloud platform monitoring system for lithium batteries used in power grids and energy storage, as described in this invention, supports elastic scaling and modular upgrades of system resources through a loosely coupled microservice architecture and containerized deployment, exhibiting strong scalability. By providing standardized third-party platform access interfaces, it achieves seamless data sharing and collaboration with external systems, enhancing the interconnectivity and intelligence level of the entire energy ecosystem. Attached Figure Description

[0049] The invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a schematic diagram of the system architecture of the cloud platform monitoring system for lithium batteries used in power grid energy storage according to the present invention; Figure 2 This is a structural framework diagram of the edge computing unit of the present invention. Detailed Implementation

[0051] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0052] like Figure 1 As shown in the figure, the cloud platform monitoring system for lithium batteries used in power grid energy storage according to an embodiment of the present invention includes at least one edge computing unit, a data transmission module, a cloud platform core processing unit, a data storage unit, an intelligent analysis module, and a user interaction interface module. The overall system design adheres to the principles of modularity, scalability, and high reliability to adapt to the deployment needs of energy storage power stations of different scales and application scenarios.

[0053] First, edge computing units are strategically deployed at energy storage power station sites. Their core responsibility is to perform high-frequency, multi-dimensional data acquisition of energy storage lithium battery systems, specifically including battery clusters, battery modules, and the smallest controllable units such as individual cells, and to perform local preprocessing and real-time event detection. For example... Figure 2 As shown, the edge computing unit is not a single hardware entity, but a local intelligent processing node that integrates multiple functions. Its internal structure includes a data acquisition interface module, a local data preprocessing module, a real-time event detection module, a local data caching module, and a data encapsulation and transmission module. These modules work closely together to complete the intelligent management of on-site data.

[0054] Specifically, the data acquisition interface module is responsible for data interaction with various devices on-site. It establishes wired connections with the Battery Management System (BMS), Power Conversion System (PCS), and auxiliary equipment, including fire suppression systems and temperature control systems, through multiple industrial communication protocols such as Modbus / TCP, CAN bus, IEC61850, and Ethernet protocols customized for specific vendors' equipment. This multi-protocol support ensures broad compatibility with heterogeneous devices. The module can accurately acquire key parameters of the energy storage system's operation, such as voltage, current, temperature, state of charge (SOC), state of health (SOH), and cycle count for each individual battery cell, either through periodic polling or event-driven methods. Simultaneously, it can also collect power, charge, charge / discharge status, and fault alarm information at the battery cluster level. Furthermore, to achieve deep collaboration with the power grid, the module also synchronously collects grid interaction parameters such as frequency, voltage, active power, and reactive power. To ensure bandwidth and concurrency capabilities for data acquisition, the data acquisition interface module is configured with at least four independent physical communication ports, each supporting a communication rate of no less than 10Mbps. This ensures that multi-source heterogeneous data can be acquired at high speed and in a high-concurrency mode, avoiding data bottlenecks.

[0055] The local data preprocessing module receives and processes the transmitted raw data, following the output of the data acquisition interface module. Its primary task is to denoise the raw data, using methods such as median filtering or moving average filtering to effectively eliminate inherent sensor noise and transient fluctuations caused by environmental interference, ensuring data purity. Subsequently, it converts data from different sources and formats into standardized JSON or Protocol Buffers formats, providing a unified foundation for subsequent data processing and transmission. Timestamp calibration is another key function of this module. Through nanosecond-level calibration using a local clock server synchronized with the deployed BeiDou or GPS timing module, it ensures that all acquired data has strictly synchronized time sequence information, crucial for accurate event correlation and fault tracing. Finally, to optimize data transmission efficiency and reduce network bandwidth consumption, the local data preprocessing module employs lossless compression algorithms, such as Snappy or ZSTD, to compress the processed data in real time.

[0056] The real-time event detection module continuously monitors the pre-processed real-time data stream. This module is internally configured with multi-level detection logic and preset safety thresholds. When any key operating parameter, such as individual cell voltage or temperature, exceeds its preset safe operating range, or when a transient overcurrent occurs, or when the rate of change of any parameter exceeds a preset gradient, the real-time event detection module immediately triggers the corresponding alarm event. More importantly, when the system detects specific preset key events, such as early signs of thermal runaway, severe overvoltage or undervoltage conditions, or a short-circuit current event, the real-time event detection module immediately initiates a data snapshot mechanism. This mechanism can issue instructions to the data acquisition interface module within a millisecond-level time window, requesting the synchronous acquisition and freezing of all current full operating data from all connected devices. This includes, but is not limited to, the voltage, current, temperature, SOC, SOH of all individual cells, as well as the real-time status of all battery clusters and PCS, thus forming a complete and time-accurate data snapshot dataset. This data snapshot will be marked as the highest priority to ensure that it is transmitted and processed first, providing a valuable and irreproducible instantaneous data view for later fault diagnosis and root cause analysis.

[0057] To address unforeseen circumstances such as network outages or transmission congestion, the local data caching module provides reliable data buffering capabilities. This module utilizes industrial-grade solid-state drives (SSDs) with high read / write bandwidth and low latency, specifically NVMExpress (NVMe) SSDs, ensuring fast data writing and reading with their superior I / O performance. Its capacity is designed to be at least 1TB, sufficient for temporarily storing large amounts of pre-processed real-time data as well as triggered data snapshots. This local caching mechanism acts as a security barrier, ensuring continuous data collection and storage even when cloud network connections are unstable. Data is retransmitted after network recovery, preventing data loss and ensuring data integrity.

[0058] Ultimately, the data encapsulation and transmission module performs refined data encapsulation operations based on the priority and type of data in the local data cache module. For data snapshots and emergency alarm data marked as highest priority by the real-time event detection module, this module uses a specific transmission protocol and the highest QoS (Quality of Service) level for encapsulation. This differentiated encapsulation strategy ensures that critical data receives the highest priority and lowest latency during transmission, enabling it to reach the cloud platform's core processing unit as quickly as possible. Regular operational data uses standard encapsulation methods and is transmitted through a general channel.

[0059] Furthermore, the data transmission module constitutes a reliable and efficient communication bridge between the edge computing unit and the cloud platform's core processing unit. Its design fully considers the real-time, reliability, and security requirements of different types of data. This module innovatively features a dual-channel transmission mechanism: a critical channel dedicated to high-real-time, high-priority data, and a general-purpose channel for routine operational data.

[0060] The critical channel is designed to meet extreme real-time requirements, primarily carrying data snapshots and emergency alarm information. This channel can employ a customized application-layer protocol based on User Datagram Protocol (UDP). This protocol ensures low latency while guaranteeing reliable transmission of critical data through embedded application-layer acknowledgment mechanisms, error check codes, and timeout retransmission logic. Alternatively, the MQTToverTLS protocol with QoS2 level can be selected, providing an "ExactlyOnce" message transmission guarantee, further enhancing data reliability. The critical channel employs wide-area network acceleration technologies, such as SD-WAN (Software-Defined Wide Area Network), or utilizes dedicated network links provided by operators to intelligently optimize the transmission path, minimizing physical latency and precisely controlling end-to-end data transmission latency to within 100 milliseconds. This extremely low latency characteristic is crucial for the issuance of safety alarms and control commands in the grid energy storage system.

[0061] As a supplement, the general channel is primarily used for periodically uploaded routine operational data, such as non-urgent data like minute-level average battery voltage, current, and temperature. This channel uses Transmission Control Protocol (TCP / IP) encrypted with Secure Sockets Layer / Transport Layer Security (TLS) for data transmission. While TCP / IP itself provides reliability and order guarantees for data transmission, TLS encryption ensures confidentiality, integrity, and authentication of data during transmission, effectively preventing data eavesdropping and tampering.

[0062] To further enhance data transmission security, the data transmission module establishes multiple encrypted tunnels between the edge computing unit and the cloud platform, such as tunnels based on IPSec VPN or TLS protocols. These encrypted tunnels provide end-to-end confidentiality protection for all transmitted data, ensuring that data cannot be accessed by unauthorized third parties. Simultaneously, Message Authentication Codes (MACs) or digital signatures ensure that data is not tampered with during transmission, maintaining data integrity. An authentication mechanism ensures the legitimacy of both communicating parties, preventing unauthorized entities from accessing the system.

[0063] As the central brain of the entire monitoring system, the cloud platform's core processing unit is designed according to the microservice architecture and multi-agent system (MAS) paradigm to achieve unprecedented scalability, resilience, and intelligent decision-making capabilities. This core processing unit is deployed on a container orchestration platform with elastic scaling capabilities, such as a Kubernetes cluster, which ensures high service availability and dynamic resource allocation under high load.

[0064] Microservice architecture breaks down complex business logic into a series of independently deployed and running service units. These include, but are not limited to: a data access service, responsible for processing all data streams transmitted from edge computing units; a data preprocessing service, performing deeper data cleaning, transformation, and feature engineering; an alarm management service, responsible for alarm generation, classification, notification, and processing; a state estimation service, used to assess the internal state of the battery in real time; a fault diagnosis service, identifying and locating system faults; a predictive analytics service, predicting future trends; an energy management optimization service, generating optimal charging and discharging strategies; a user authentication and authorization service, managing system access permissions; and an API gateway service, serving as a unified entry point for external system access. These microservices are not directly coupled but communicate asynchronously through distributed message queues, such as message queues based on an Apache Kafka cluster, with message formats uniformly using efficient Protocol Buffers or the universally applicable JSON format. This asynchronous communication mechanism greatly improves the system's concurrent processing capabilities and fault isolation.

[0065] Furthermore, a Multi-Agent System (MAS) paradigm runs on top of the microservice architecture. In this paradigm, each agent is designed as a software entity with independent goals, behavioral capabilities, and collaborative abilities. They either correspond to a specific microservice instance or represent a functional module. This MAS architecture includes, but is not limited to: the DataIngestionAgent, which is the first point of entry into the cloud platform. It is responsible for receiving all raw data streams sent by the data transmission module, performing preliminary data format validation and integrity checks, and intelligently routing data based on data type and priority. High-frequency real-time data is directed to a high-speed caching layer and a real-time time-series database layer. To cope with instantaneous peaks in data flow, this agent is configured with dynamic scaling capabilities, automatically increasing or decreasing the number of instances based on the real-time data volume through Kubernetes' auto-scaling mechanism.

[0066] The DataPreprocessing Agent reads data from real-time time-series databases and performs more advanced and refined data cleaning operations than those at the edge. This includes outlier detection, which can employ statistical methods such as the 3σ (Three Sigma) criterion or utilize machine learning models such as Isolation Forest to identify and remove outliers from the data. Furthermore, it performs data imputation (e.g., based on linear interpolation or regression models) to address missing data; performs feature extraction to generate derived features more valuable for analysis; and normalizes the data to make it suitable for subsequent machine learning model training and inference, thus providing high-quality input for the intelligent analysis module.

[0067] The StateEstimation Agent is the core of refined battery management. It combines sophisticated battery equivalent circuit models (such as the Randle model or more complex N-order RC models) with advanced Kalman filtering, Extended Kalman Filter (EKF), or Unscented Kalman Filter (UKF) algorithms to perform high-precision real-time estimations of the battery's State of Charge (SOC), State of Health (SOH), and Remaining Life (RUL). These estimation results are crucial for battery operation and maintenance and lifespan management. The FaultDiagnosis Agent uses multiple intelligent algorithms to detect and locate battery system faults in real time. It integrates a rule engine for handling known fault modes; traditional machine learning models such as decision trees and support vector machines (SVM) for classifying fault types; and a deep learning classifier to identify more complex and hidden fault modes. This agent can identify and locate typical faults such as internal short circuits, abnormal connections, and abnormal temperature rises, and can generate detailed fault reports containing fault type, occurrence time, location information, and relevant parameters.

[0068] Predictive Analytics Agents utilize advanced deep learning models, such as Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs), or Transformer networks, to make high-precision predictions of future trends in key battery parameters, such as changes in voltage, temperature, and internal resistance over the next few hours or days. These predictions provide a scientific basis for early warning of potential thermal runaway risks, overcharge and over-discharge risks, and performance degradation trends.

[0069] The Dispatch Optimization Agent is responsible for generating the optimal charging and discharging strategy for the energy storage system. It comprehensively considers grid load forecast data, real-time electricity price information, weather forecast data, the energy storage system's own operational constraints (such as maximum charging and discharging power, maximum / minimum SOC range), and operational objectives (such as revenue maximization, peak-valley arbitrage, and ancillary service support). This agent employs reinforcement learning, dynamic programming, or linear programming algorithms to solve complex mathematical models, generating optimal charging and discharging curves that adapt to grid demand and market conditions.

[0070] The ResourceManagementAgent continuously monitors the operational status and resource consumption of each microservice and agent within the core processing unit of the cloud platform. It performs dynamic load balancing to ensure that request traffic is evenly distributed across service instances, preventing overload of any single instance. When a service instance failure is detected, it automatically triggers a fault recovery mechanism. Furthermore, it dynamically allocates computing resources such as CPU and memory based on system load, ensuring stable and efficient system operation under high concurrency.

[0071] Agents exchange information and collaborate efficiently through a shared distributed cache and message queue. This inter-agent communication protocol employs a service-oriented message passing pattern, ensuring the reliability and real-time nature of message delivery. This enables distributed decision-making and task decomposition, effectively distributing the upper-layer load that traditional centralized systems might face and significantly reducing the decision-making complexity of a single agent.

[0072] As the cornerstone for carrying massive amounts of energy storage system data, the data storage unit adopts a hierarchical storage architecture, designed to balance real-time data access performance, high-concurrency write capabilities, and the economical storage and analysis needs of long-term historical data. This architecture includes a high-speed cache layer, a real-time time-series database layer, and a long-term historical database layer, with each layer coordinating through data flow and management strategies.

[0073] Specifically, the caching layer employs a distributed in-memory database cluster, such as a Redis cluster, designed to provide extremely fast read and write response times. This layer stores data snapshots transmitted by the edge computing unit and triggered by the real-time event detection module; these snapshots are crucial for instantaneous fault location. Simultaneously, frequently accessed real-time operational data, such as the power curve of the most recent hour and minute-level average temperature—"hot data"—is also stored here for millisecond-level fast queries by the user interface and intelligent analysis module. Furthermore, the intelligent analysis module uses pre-read and pre-fetch mechanisms to load data to be analyzed into the cache in advance, further optimizing analysis efficiency. This caching layer can typically provide read and write response times of less than 5 milliseconds.

[0074] Above this, the real-time time-series database layer employs a distributed, high-concurrency write-optimized time-series database cluster, with options including Apache IoTDB or InfluxDB clusters. This layer is deeply optimized for the storage and querying of time-series data, serving as the persistent storage location for all raw sensor data and computational results processed by the data preprocessing agent. The time-series database is designed to support processing tens of thousands or even hundreds of thousands of write requests per second, efficiently handling high-frequency data reporting from large-scale energy storage power plants. Its unique index structure and storage mechanism provide efficient time-range query capabilities, making it fast and convenient to retrieve data for specific time periods from massive amounts of time-series data.

[0075] To address the massive amounts of long-term historical data with low access frequency, this system employs a long-term historical database layer. This layer utilizes a distributed columnar storage database cluster, such as one based on Apache Cassandra or ClickHouse, or leverages a large-scale object storage system (e.g., S3-compatible storage). This layer primarily stores historical operational data spanning several years or even decades, including all raw data, processing results, and alarm records. Designed to support large-scale batch processing analysis and data archiving, this layer enables the intelligent analysis module to perform complex offline modeling, trend analysis, capacity decay studies, and support management decisions and report generation, while simultaneously optimizing storage costs.

[0076] Furthermore, the intelligent analysis module is the core of this system for achieving intelligent decision-making and early warning. It provides advanced data analysis and decision support functions, and integrates multiple complex algorithms and platform support to extract value from massive amounts of data and achieve preventive maintenance and optimized operation.

[0077] The data snapshot processing mechanism is a crucial component of rapid fault diagnosis. This mechanism is specifically designed to receive and parse high-priority data snapshots sent by edge computing units. Once received, the system immediately writes them to the cache layer and the real-time time-series database layer. This instant storage ensures that critical data before and after a fault is completely and accurately preserved, providing operations and maintenance personnel with a complete and high-quality dataset for fault root cause analysis and event chain reconstruction, significantly reducing fault location time.

[0078] Real-time alarm and predictive early warning algorithms are implemented through continuous and uninterrupted monitoring of the incoming real-time data stream. This module combines various machine learning models, such as decision trees and random forests, with statistical methods (such as moving averages and exponential smoothing) to identify abnormal patterns in the data. When any parameter deviates from the normal range or exhibits an abnormal trend, the system can issue a fault alarm in real time. More importantly, the predictive early warning algorithm fully utilizes the prediction results of key battery parameters (such as voltage, temperature, and internal resistance) output by the aforementioned predictive analysis agent to identify potential risks of thermal runaway, overcharging and over-discharging, and performance degradation trends hours or days in advance. This allows maintenance personnel to take preventative measures and avoid serious accidents.

[0079] The health status management and lifespan prediction algorithm is based on the estimated state of charge (SOC), state of health (SOH), and remaining useful life (RUL) provided by the aforementioned state estimation agent, combined with historical battery cycle data, ambient temperature data, and actual operating condition data (such as depth of charge / discharge, rate, etc.). By constructing a more accurate battery lifespan prediction model, this module can provide detailed health assessment reports and longer-term lifespan predictions for each battery cluster or the entire energy storage system, guiding maintenance personnel to conduct targeted maintenance and replacement plans and optimizing asset management.

[0080] The energy management optimization algorithm utilizes the charging and discharging strategies generated by the aforementioned scheduling optimization agent and dynamically adjusts them based on factors such as real-time grid demand, market electricity prices, the actual operating status of energy storage power stations, and renewable energy output forecasts. The algorithm aims to achieve optimal economic benefits and maximum grid support for grid-side energy storage systems. This is achieved through precise peak-valley arbitrage, participation in frequency regulation and peak shaving ancillary services, and support for grid emergency response, thereby enhancing the commercial value of energy storage systems and grid stability.

[0081] To support the aforementioned advanced analytics and model training, the big data analytics platform employs a distributed computing framework, such as an Apache Spark cluster or an Apache Flink cluster. This platform can perform offline or near real-time batch and stream processing analysis on massive amounts of historical data stored in a long-term historical database layer. Through its powerful computing capabilities, the platform can uncover deep-seated patterns in the operation of energy storage systems, optimize the parameters of various intelligent models (such as adjusting the weights of predictive models or the thresholds of fault diagnosis models), automatically generate various operational reports, and provide data-driven decision support for management, thus promoting the intelligent operation of energy storage power plants.

[0082] To ensure system usability and multi-party collaboration capabilities, the user interaction interface module provides access interfaces for multiple users, and its design fully considers the needs and usage scenarios of different user roles. This module implements a fully functional web management portal based on a B / S (Browser / Server) architecture, while also providing a mobile application interface based on an M / S (Mobile / Server) architecture, allowing users to access the system anytime, anywhere via smartphones or tablets.

[0083] One of the core functions of the user interaction interface module is to provide a real-time visual interface for the operation status of the energy storage system. This interface can display key operating parameters, including individual battery voltage, current, temperature curves, state of charge, health status, power curves, and power statistics, in the form of intuitive charts and dashboards, enabling operation and maintenance personnel to have a clear understanding of the overall and local status of the system.

[0084] In addition, this module provides an efficient historical data query function. Users can perform efficient searches based on time range, specific battery units (such as a battery cluster or a battery module), data type, and other conditions, and support exporting the query results to various formats (such as CSV and Excel) for subsequent offline analysis.

[0085] In terms of alarm management, the user interface module provides detailed alarm information display and tiered response capabilities. The system categorizes alarms into multiple levels based on their urgency, such as early warning, general alarms, and emergency alarms. Once an alarm is triggered, the system will promptly notify relevant maintenance personnel via SMS, email, app push notifications, and other methods, according to the alarm level and preset notification policies. This module also supports alarm confirmation and processing workflow management, ensuring that every alarm receives a timely response and is effectively handled.

[0086] To enable remote and precise management of energy storage systems, the user interface module supports the configuration and distribution of charging and discharging strategies. Maintenance personnel or authorized users can set parameters such as charging and discharging plans and power limits through the interface, and then distribute these commands to the edge computing units or PCS systems in the field in real time, thereby achieving remote control of the energy storage system.

[0087] To address the varying permission requirements of different user roles, the user interface module implements a robust multi-user, multi-role permission management mechanism. This mechanism provides customized data access permissions and functional interfaces for different roles, including power plant operators, end users, and regulatory agencies, ensuring data security and business process compliance, and achieving granular permission control.

[0088] To support the stable and efficient operation of the cloud platform's core processing and data storage units, this invention also includes a meticulously designed hardware system. This hardware system employs a loosely coupled architecture, consisting of standardized server nodes, high-performance network equipment, and highly reliable storage devices, ensuring the system's scalability and resilience.

[0089] The hardware system dynamically distributes request traffic among multiple microservice instances by deploying a load balancer cluster, such as an L7 (application layer) load balancer based on Nginx or HAProxy, thereby supporting high-concurrency access and avoiding single points of failure. The load balancer cluster is configured to support session-based persistence to ensure that requests from the same user are always routed to the same service instance, maintaining session state; it also features health checks to automatically remove failed service instances; and it supports automatic failover, quickly switching to a standby instance when the primary service instance fails, ensuring continuous service availability.

[0090] To achieve deep integration with the external ecosystem, the hardware system supports access from third-party platforms. This is achieved through an API gateway service layer, which provides data interfaces and data subscription services compliant with open standards (such as RESTful API and GraphQL). This enables the system to securely and efficiently share and collaborate with external systems such as power grid dispatch centers, meteorological service providers, and power trading market platforms. The API gateway is not only responsible for API authentication to verify the legitimacy of the request source, but also performs request routing, traffic shaping to control the request rate, and protocol conversion to ensure compatibility with the communication standards of different systems.

[0091] The hardware system also supports multi-energy type access. Through standardized data models and flexible protocol adaptation layers, it can integrate data from various energy types, such as photovoltaic power generation, wind power generation, hydropower, and traditional thermal power, into a unified platform for management and scheduling. This capability enables the system to serve as the core platform for multi-energy microgrids or regional energy management.

[0092] To enable rapid fault location and tracing within the system, the hardware system integrates real-time retrieval of logs and fault reports throughout the entire lifecycle. This functionality employs a distributed log collection system, such as ELKStack (Elasticsearch, Logstash, Kibana) or Loki, to centrally collect, store, index, and visualize operational logs from all microservices, agents, and underlying hardware devices. This allows operations personnel to quickly retrieve, analyze, and visualize all log information through a unified interface, thereby achieving rapid fault location, diagnosis, and root cause analysis.

[0093] Ultimately, to ensure high reliability and data persistence under various extreme conditions, the hardware system deploys redundant servers, storage, and network devices. Simultaneously, combined with distributed file systems (such as Ceph or GlusterFS) and robust data backup and recovery mechanisms, it ensures that even in the event of partial hardware failure, the system can continue to operate without data loss.

[0094] Example 1 In a specific application scenario, the cloud platform monitoring system for lithium batteries used in power grid energy storage according to the present invention was deployed in a 100MW / 200MWh lithium-ion battery energy storage power station under the jurisdiction of a provincial power grid. The power station contains 50 independent battery clusters, each battery cluster consists of 20 battery modules, and each battery module integrates 200 individual lithium-ion cells, that is, a total of 200,000 individual cells in a single station.

[0095] In this embodiment, edge computing units are deployed in the area where each battery cluster is located, with one industrial-grade edge server in each area, totaling 50 units. Each edge server's data acquisition interface module communicates directly with the BMS and PCS of its respective battery cluster via CAN bus and Modbus / TCP protocol, and simultaneously connects to auxiliary equipment such as temperature control and fire suppression systems via Ethernet interface. Each port is configured with a communication rate of no less than 10Mbps to handle the periodic acquisition of up to 2000 parameters per second (including individual cell voltage, current, temperature, SOC, SOH, etc.). The local data preprocessing module uses the Snappy algorithm to compress the data, achieving an average compression ratio of 3:1. The real-time event detection module sets over- and under-voltage thresholds of 2.5V and 4.2V for individual battery voltages, and an over-temperature threshold of 55°C for temperature. When the voltage of any individual cell changes by more than 0.1V within 100ms or the temperature changes by more than 5°C within 1 second, the data snapshot mechanism is immediately triggered, and instantaneous data of 50,000 parameters from the entire cluster are acquired within milliseconds. The local data caching module uses a 2TB NVMeSSD, which can store at least 72 hours of full high-frequency data offline.

[0096] The critical channel of the data transmission module employs a custom UDP-based protocol, providing an independent transmission link for data snapshots and emergency alarms. This link is optimized using SD-WAN technology to ensure that the end-to-end transmission latency from the edge computing unit to the cloud platform's core processing unit is controlled within 80 milliseconds. The general channel, on the other hand, uses a TLS-encrypted TCP / IP protocol to upload regular operational data at a frequency of minutes, with an average data transmission rate of 5MB / s.

[0097] The core processing unit of the cloud platform is deployed in a Kubernetes-based hybrid cloud environment, leveraging the elastic resources of the public cloud and the data security advantages of the private cloud. The data access service processes over 100,000 data records per second and distributes the data to subsequent microservices via an Apache Kafka cluster. The state estimation agent uses the unscented Kalman filter (UKF) algorithm, combined with measured data and a battery equivalent circuit model, to estimate the SOC and SOH of each individual cell in real time, with an estimation error controlled within 1%. The predictive analytics agent uses an LSTM neural network to predict battery temperature and voltage trends 48 hours in advance based on historical data and weather forecasts, achieving a prediction accuracy of over 95%. The scheduling optimization agent uses a reinforcement learning algorithm, combined with real-time electricity prices (using second-level electricity trading market data), grid dispatch instructions, and battery constraints, to generate the optimal charge / discharge curve for each 24-hour period, maximizing peak-valley arbitrage and ancillary service revenue.

[0098] The data storage unit's high-speed caching layer uses a Redis cluster to cache the most recent two hours of real-time high-frequency data and prediction results, with a query response time of less than 10 milliseconds. The real-time time-series database layer uses an InfluxDB cluster, capable of writing over 150,000 timestamped data entries per second and supporting millisecond-level historical data queries. The long-term historical database layer uses a ClickHouse cluster to store all raw and processed data from the past five years, totaling over 500TB, for offline big data analysis and model training.

[0099] The intelligent analysis module's data snapshot processing mechanism can parse and write data to the high-speed cache within 50 milliseconds after the data snapshot arrives at the cloud platform. The real-time alarm and predictive warning algorithm, based on a random forest model, performs real-time analysis of over 200 features, providing thermal runaway risk warnings up to 15 minutes in advance with a false alarm rate of less than 0.5%. The health status management and lifespan prediction algorithm, based on the SOH decay curve, predicts that the overall lifespan of the power plant can be extended by 5%. The big data analysis platform, based on an Apache Spark cluster, performs offline analysis of historical data weekly, generating monthly maintenance reports and quarterly performance optimization suggestions.

[0100] The user interaction interface module provides web and mobile app access, allowing maintenance personnel to view the power plant's operational status in real time, receive alarm notifications, and remotely issue charging and discharging commands via the mobile app. Access control is divided into four roles: power plant manager, operations engineer, maintenance personnel, and grid dispatcher, ensuring data security and operational compliance.

[0101] Comparative Example 1 To demonstrate the technological advancements and advantages of the cloud platform monitoring system for lithium batteries used in power grid energy storage as described in this invention, we propose a comparative example of a traditional energy storage power station monitoring system that does not employ the core technical features of this invention. This comparative system is deployed in the same 100MW / 200MWh lithium-ion battery energy storage power station and includes a traditional SCADA (Supervisory Control and Data Acquisition) system, a centralized database, and a rule-based alarm system.

[0102] In this comparative system, field data acquisition primarily relies on Remote Terminal Units (RTUs) deployed at the battery cluster or PCS layer. These RTUs report aggregated data to the central SCADA server via Modbus / RTU or Modbus / TCP protocols at a frequency of seconds or tens of seconds. Detailed parameters of individual cells are typically not directly uploaded, or are only processed locally by the BMS at a low frequency (e.g., minutes) and then aggregated to report a small number of key statistical values ​​(e.g., maximum voltage, minimum voltage, average temperature), lacking real-time full data at the individual cell level. Data preprocessing capabilities are extremely limited, mainly performing format conversion and a small amount of data type checking, lacking real-time noise reduction and advanced compression capabilities. No local data caching is set up on-site; all data is uploaded in real-time, and data will be lost if the network is interrupted. Event detection is limited to simple threshold judgments, unable to trigger data snapshots within milliseconds, and lacks the ability to provide early warnings of thermal runaway.

[0103] Data transmission uses the standard TCP / IP protocol, without prioritizing data. All data is transmitted through a single link, lacking dedicated critical channels and wide area network acceleration mechanisms. Transmission latency is limited by the quality of the public network, typically fluctuating between hundreds of milliseconds and several seconds, and end-to-end transmission reliability is not guaranteed.

[0104] The core processing unit of the cloud platform adopts a traditional centralized monolithic application architecture or a small number of service stacks. The data access server is a single instance with limited high-concurrency write capabilities. The database typically uses relational databases (such as MySQL or PostgreSQL), which have significant performance bottlenecks in high-frequency time-series data writing and querying, making it difficult to handle the write requests of tens of thousands of data entries per second from large-scale energy storage power stations. The system lacks microservices and multi-agent system paradigms; all analysis and decision-making logic is concentrated in one or a few modules, resulting in poor scalability and susceptibility to response delays and blockages in high-concurrency environments. State estimation, fault diagnosis, and predictive analysis are mostly based on simple mathematical models and empirical rules, lacking the intelligent capabilities of machine learning and deep learning, resulting in low estimation accuracy, short prediction cycles, insufficient accuracy, and weak early warning capabilities. Energy management optimization is also limited to preset static strategies or manual intervention, unable to dynamically adjust in real time to maximize economic benefits.

[0105] For data storage, only a single-tier or two-tier database architecture is used. All real-time and historical data are written to a relational database, resulting in low data write throughput and query efficiency. Especially when querying historical data with a large span, the response time can reach tens of seconds or even minutes. Without a high-speed caching layer, the performance of querying hot data is limited by the database IOPS.

[0106] The intelligent analysis module is weak, lacking data snapshot processing capabilities. The alarm system primarily uses threshold-based alarms, with virtually no predictive or early warning capabilities; alarms are only issued after a fault occurs, preventing proactive detection. Health status management and lifespan prediction are based on simple statistical models, with accuracy and reliability far lower than deep learning-based models. Big data analytics capabilities are limited, typically relying on reporting tools for simple statistical analysis of limited historical data.

[0107] The user interface uses traditional PC-based SCADA client software, which has insufficient support for mobile devices. The system functions are fixed, and user permission management is coarse-grained.

[0108] Data-driven comparison Through the deployment and operation of the above-mentioned Example 1 and Comparative Example 1, we collected key performance indicators for quantitative comparison to highlight the significant advantages of the cloud platform monitoring system for power grid energy storage of the present invention in terms of real-time performance, processing power, intelligence level and system reliability.

[0109] Performance indicators Example 1 (This Invention) Comparative Example 1 (Traditional System) Performance improvement (relative to Comparative Example 1) End-to-end transmission delay of critical data 80ms (data snapshot, emergency alarm) 500ms~3000ms (depending on network) 84%-97% reduction Real-time data write throughput >150,000 messages / second (single site) <5,000 messages / second (single site) 2900% improvement Hot data query response time <10ms >500ms >98% reduction Battery SOC / SOH estimation error <1% 5%-10% >80% reduction Thermal runaway risk warning time 15 minutes in advance 0 minutes (alarm triggered after fault occurs) From nothing to something Predictive analysis accuracy (voltage / temperature) >95% (48 hours in advance) <70% (1-2 hours in advance) >35% improvement High-concurrency user request processing capability >10,000 QPS (via microservices, load balancing) <500 QPS (bottleneck for a single application) >1900% improvement Data integrity (after network outage) 100% (Data cached locally for 72 hours, to be re-uploaded after recovery) 0% (Data loss during network outage) From nothing to something System resource elastic scaling capability Automatic elastic scaling to handle peak loads Manual capacity expansion may require system shutdown for maintenance. Automation, second-level response Energy management optimization improves revenue 5%-15% (optimized in real-time via RL) <2% (static strategy) 150%-650% improvement The above data clearly demonstrates that the cloud platform monitoring system for lithium batteries used in power grids and energy storage exhibits overwhelming advantages over traditional systems in several key performance indicators. For example, the significant reduction in end-to-end transmission latency of critical data (at least an 84% reduction) enables the system to truly transform from a "post-event recording" tool into a "proactive prevention and control" tool, gaining valuable time for timely safety measures. Real-time data write throughput reaches up to 150,000 records per second, dozens of times that of traditional systems, effectively solving the database bottleneck problem caused by the massive influx of high-frequency data from large-scale energy storage power stations and ensuring real-time data integrity. Hot data query response time has been significantly optimized from seconds to milliseconds, greatly improving user experience and analysis efficiency.

[0110] Of particular note is the system's ability to provide a 15-minute advance warning of thermal runaway risk through the introduction of an intelligent analysis module—a capability completely absent in traditional systems. This effectively mitigates potential major safety incidents. Simultaneously, the application of deep learning models significantly reduces battery SOC / SOH estimation errors and greatly improves prediction accuracy, providing a solid foundation for refined operation and maintenance and asset management. The system demonstrates exceptional resilience in handling high-concurrency user requests and, through energy management optimization algorithms, brings considerable economic benefits to energy storage power stations. Furthermore, the local caching mechanism ensures data integrity during network interruptions, eliminating the risk of data loss inherent in traditional systems. These quantitative data fully demonstrate the non-obvious innovation and practicality of this invention in addressing the pain points of existing technologies and improving the operational efficiency and safety of grid-connected energy storage systems.

[0111] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cloud platform monitoring system for lithium batteries used in power grid energy storage, characterized in that, include: Edge computing units are deployed at energy storage power stations to perform high-frequency, multi-dimensional data acquisition, local preprocessing, and real-time event detection for energy storage lithium battery systems. The data transmission module is configured with a dual-channel transmission mechanism to enable data communication between the edge computing unit and the cloud platform core processing unit; The core processing unit of the cloud platform is built using a microservice architecture and a multi-agent system paradigm. It is deployed on an elastically scalable container orchestration platform and is used to receive, process, and analyze data transmitted by the edge computing units through the data transmission module. The data storage unit adopts a multi-level storage architecture to store various types of data collected, processed, and analyzed by the monitoring system; The intelligent analysis module is used for advanced analysis, real-time alarms and predictive warnings, and decision optimization of the data stored in the data storage unit; The user interaction interface module is used to provide real-time operation status visualization, historical data query, alarm information display and hierarchical response, and charging and discharging strategy configuration and distribution functions for energy storage systems.

2. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The edge computing unit includes: The data acquisition interface module is wired to the battery management system, energy conversion system and other auxiliary equipment via industrial communication protocols to acquire single-cell battery operating data in a periodic or event-driven manner, and synchronously collect grid-side parameters. It is configured to have at least 4 independent physical communication ports. The local data preprocessing module receives raw data transmitted from the data acquisition interface module and performs data denoising, format unification, timestamp calibration, and lossless data compression operations. The real-time event detection module continuously monitors the real-time data stream processed by the local data preprocessing module. When any parameter exceeds the preset safety threshold, instantaneous overcurrent occurs, or the rate of change exceeds the preset gradient, an alarm event is triggered. When a preset specific key event occurs, a data snapshot mechanism is activated. The local data caching module uses an industrial-grade solid-state drive with high read / write bandwidth and low latency to temporarily store real-time data and data snapshots processed by the local data preprocessing module. The data encapsulation and transmission module encapsulates data from the local data cache module into specific data packets based on data priority and type, and sends them to the cloud platform core processing unit through the data transmission module. Data snapshots and emergency alarm data marked as the highest priority are encapsulated using specific transmission protocols and service quality levels.

3. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 2, characterized in that, When the real-time event detection module initiates the data snapshot mechanism, it instructs the data acquisition interface module to synchronously collect and freeze the current full running data of all connected devices within a millisecond time, forming a complete snapshot dataset at a specific time point, and marking it as the highest priority.

4. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The data transmission module includes: The critical channel is used for high real-time and high-priority data transmission. It adopts a customized application layer protocol based on the User Datagram Protocol or a message queue telemetry transmission protocol with quality of service level. The customized application layer protocol includes application layer acknowledgment mechanism, error checking and timeout retransmission logic, and utilizes wide area network acceleration technology or dedicated network links to optimize the transmission path. The general channel is used for routine data transmission. It uses a transmission control protocol encrypted with a secure socket layer / transport layer security protocol to transmit data, providing reliability and ordering guarantees for data transmission. The data transmission module establishes multiple encrypted tunnels between the edge computing unit and the cloud platform's core processing unit to ensure the confidentiality, integrity, and authentication of data transmission.

5. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The microservice architecture of the core processing unit of the cloud platform includes data access service, data preprocessing service, alarm management service, status estimation service, fault diagnosis service, predictive analysis service, energy management optimization service, user authentication and authorization service, and application programming interface gateway service. Each microservice communicates asynchronously through a distributed message queue.

6. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 5, characterized in that, The multi-agent system paradigm of the core processing unit of the cloud platform includes: The data acquisition agent is responsible for receiving raw data sent by the data transmission module, performing preliminary data verification and routing, writing the data into the high-speed cache layer and the real-time time series database layer, and is configured to have dynamic scaling capabilities to cope with instantaneous peaks in the data flow. The state estimation agent uses Kalman filtering, extended Kalman filtering, or unscented Kalman filtering algorithms, combined with the battery equivalent circuit model, to perform high-precision real-time estimation of the battery's state of charge, state of health, and remaining lifespan. The scheduling optimization agent combines grid load forecasting, real-time electricity prices, weather forecasting, energy storage system constraints, and operational objectives, and uses reinforcement learning, dynamic programming, or linear programming algorithms to generate the optimal charging and discharging strategy. Resource management agent monitors the running status and resource consumption of each microservice and agent in the core processing unit of the cloud platform, and performs dynamic load balancing, fault recovery and resource allocation.

7. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The data storage unit includes: The high-speed cache layer employs a distributed in-memory database cluster to store triggered data snapshots transmitted by edge computing units, frequently accessed real-time running data, and data pre-read and pre-fetched by the intelligent analysis module, providing millisecond-level read and write response times. The real-time time series database layer adopts a distributed, high-concurrency write-optimized time series database cluster for persistent storage of raw sensor data and calculation results processed by the data preprocessing agent. It supports processing tens of thousands of write requests per second and provides efficient time range query functions. The long-term historical database layer employs a distributed columnar storage database cluster or a large-scale object storage system to store massive amounts of long-term historical data with low access frequency, supporting large-scale batch processing analysis and data archiving.

8. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The intelligent analysis module includes: The data snapshot processing mechanism is responsible for receiving and parsing the data snapshots marked as the highest priority sent by the edge computing unit, and quickly writing them into the cache layer and the real-time time series database layer for accurate state restoration before and after the failure and root cause analysis of the failure. The real-time alarm and predictive warning algorithm continuously monitors the real-time data stream, combines machine learning models and statistical methods to identify and issue fault alarms and abnormal event warnings in real time, and uses the battery parameter prediction results output by the predictive analysis agent to identify potential thermal runaway risks, overcharge and over-discharge risks and performance degradation trends in advance. The health status management and life prediction algorithm, based on the state of charge, health status and remaining life estimation results provided by the state estimation agent, combined with battery historical cycle data, ambient temperature data and operating condition data, constructs a battery life prediction model and provides a battery pack health assessment report. The energy management optimization algorithm utilizes the charging and discharging strategies generated by the scheduling optimization agent to dynamically adjust the system based on real-time grid demand and power plant operating status, thereby achieving optimal economic benefits and grid support functions for the grid-side energy storage system.

9. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 1, characterized in that, The user interaction interface module implements a web management portal based on a browser / server architecture and provides a mobile application interface based on a mobile device / server architecture. It provides a real-time operation status visualization interface for the energy storage system, historical data query function, alarm information display and hierarchical response function, charging and discharging strategy configuration and distribution function, and supports multi-user and multi-role permission management.

10. The cloud platform monitoring system for lithium batteries used in power grid energy storage according to claim 9, characterized in that, It also includes a hardware system with a loosely coupled architecture, consisting of standardized server nodes, network devices and storage devices. By deploying a load balancer cluster, it dynamically distributes request traffic among multiple microservice instances, supports access from third-party platforms and multiple energy types, and integrates real-time retrieval of logs and fault reports throughout the entire lifecycle.

Images