Energy storage system application interaction method and system based on voice voice control model

Abstract

This invention relates to an interactive method and system for energy storage systems based on a voice-controlled model. The method includes: receiving user voice commands, performing voice recognition and semantic parsing, and extracting control intentions and parameters; matching or generating system operation strategies based on the parsing results; and decomposing the strategies into device control commands and issuing them for execution. The system includes modules for voice input, processing, escape extraction, strategy escape, strategy decomposition, and output, supporting multilingual recognition, local offline operation, strategy customization, and security verification. This invention lowers the barrier to entry for new energy storage systems, improves interactive intelligence and security, and is suitable for residential, commercial, off-grid, and mobile energy storage scenarios.

Description

Technical Field

[0001] This invention belongs to the field of human-computer interaction technology for new energy and energy storage systems. Specifically, it relates to a method and system for application interaction (HMI) of energy storage systems based on a voice control model. It is applicable to the human-computer interaction of users in off-grid or grid-connected power consumption scenarios where new energy (including wind power generation, photovoltaic power generation, etc.) is coupled with energy storage (including electrochemical energy storage or hydrogen energy storage or other forms of energy storage). Background Technology

[0002] With the popularization of new energy and energy storage systems, users have placed higher demands on the ease of operation and intelligence of these systems. Traditional control methods mostly rely on professional interfaces, physical buttons, or mobile apps, which pose high barriers to entry and complex operation for non-professional users. Although there are patents related to voice interaction in existing technologies, they are mostly focused on general voice recognition or simple command execution, and fail to deeply integrate with the operating characteristics of new energy-energy storage systems, multilingual support, localized strategy generation, and safety constraint verification, lacking a complete voice control solution specifically for energy management scenarios.

[0003] A search revealed that most existing patents focus on energy storage system architecture, remote data transmission, or simple human-machine interaction processes, without addressing deep integration technologies such as natural language-based semantic understanding, multi-strategy matching, offline operation, and safety control. Therefore, it is necessary to propose a more intelligent, secure, and user-friendly power supply interaction solution. Summary of the Invention

[0004] This invention provides an application interaction method and system for energy storage systems based on a voice control model, enabling users to directly control system operation through natural language, reducing the barrier to entry and improving interaction efficiency and security.

[0005] To address the above technical problems, this invention provides an interactive system for energy storage systems based on a voice-controlled model, characterized by comprising: The voice input module is used to receive the user's voice commands; The voice processing module is used to convert voice signals into character information; The escape information extraction module is used to perform semantic parsing and intent recognition on the character information, and extract target object, action type and parameter information; The system policy escaping module is used to generate or match corresponding system operation policies; The strategy decomposition module is used to decompose the system operation strategy into device control instructions; The output module is used to send device control commands to the corresponding devices.

[0006] Furthermore, the system policy escaping module is configured as follows: First, the target object, action type, and parameter information are converted into semantic vectors; Then, the semantic vector is compared with the strategy vector in the preset strategy vector library to calculate the similarity and match the corresponding system operation strategy. When no preset strategy is matched, a corresponding system operation strategy is generated based on the intent recognition model.

[0007] Furthermore, the system policy escaping module has a built-in rule engine, which is configured with a preset rule library. The preset rule library contains the mapping relationship between keyword slot combinations of object-action-target and system operation strategies. When the information output by the escaping information extraction module matches the keyword slot combination, the system policy escaping module directly calls the corresponding system operation strategy. The system policy escaping module also has a built-in lightweight machine learning model, which includes an intent recognition module and a slot filling module. When the information output by the escaping information extraction module does not match the preset rules, the intent classification module is used to identify the user intent category, the slot filling module is used to extract specific parameters, and the system policy escaping module generates the corresponding system operation strategy based on the intent category and parameters.

[0008] Furthermore, the strategy decomposition module is configured as follows: 1) First, comprehensively organize and sort out at least one of the following information, and standardize the multi-source data and status parameters: Energy input forms include renewable energy generation, grid power supply, or a combination thereof; Energy storage forms include one or more of electrochemical energy storage, hydrogen energy storage, and compressed air energy storage; Energy usage patterns, including the time of energy use, power output, operating time, and energy-consuming objects; Environmental and external conditions information, including weather conditions, electricity market information, or electricity price information; Based on the above information, a corresponding customized operation strategy process is generated; 2) Perform strategy matching and task decomposition based on the current system operating status, input characteristics, and target description, and extract relevant information of the strategy object from the decomposition results to establish physical mapping relationships; 3) Conduct deviation analysis between system operation indicators and strategy objectives to obtain control parameters; 4) Encapsulate standardized control commands, clarify the direction of action of the control commands, and determine the device tag number, type, and parameter values.

[0009] Furthermore, the system policy escaping module supports voice command parsing and policy generation through a local rule engine or local semantic model without an internet connection; the policy decomposition module verifies device operation safety constraints, power constraints, and state constraints before generating control commands.

[0010] Furthermore, the voice input module communicates with the new energy-storage system via at least one of Bluetooth, near-field communication, wireless direct connection, or wired connection.

[0011] The interaction method for energy storage systems based on voice control models is characterized by the following specific steps: Receive user voice input commands; Perform speech recognition processing on the speech signal and convert the speech signal into character information; The character information is semantically parsed and intent is identified to extract the target object, action type, and parameter information; Match or generate corresponding system operation strategies based on the parsing results; The system operation strategy is decomposed into specific equipment control commands and sent to the corresponding equipment in the new energy-storage system for execution.

[0012] Furthermore, the voice input supports multiple natural languages, and normalizes voice commands in different languages ​​through a unified semantic intermediate representation, thereby achieving consistent control of the new energy-storage system in a cross-language environment.

[0013] Furthermore, the system operation strategy is a customizable strategy applicable to different types of energy storage systems or power system equipment. The customizable strategy includes at least one of preset rule strategy customization, semantic and control strategy mapping customization, and equipment capability boundary constraints. The customization process includes strategy definition, strategy verification, and strategy binding, and the implementation steps include strategy generation, strategy decomposition, and strategy execution.

[0014] Furthermore, the semantic parsing and intent recognition are implemented through a rule engine and / or machine learning model, including a rule matching method based on keyword slot extraction and a semantic reasoning method based on intent recognition and slot filling.

[0015] Beneficial Effects: This invention primarily serves the needs of users without technical expertise (such as those in homes, outdoors, and mobile settings). It effectively addresses the optimization of overall energy reliability or cost minimization across various energy usage scenarios, environmental conditions, market conditions, new energy power generation conditions, energy storage conditions, and electricity demand. Specifically, this invention effectively supports customized intelligent control needs for residential, commercial, and non-large-scale professional power systems, such as off-grid and isolated power systems (or energy sources). Users can freely issue commands to the configured new energy-storage power supply system using voice commands—without relying on other methods—achieving rapid, accurate, and flexible control of the system's charging and discharging (energy release) for stable, safe, and precise supply, storage, and release, without requiring specialized knowledge or technical training. Attached Figure Description

[0016] Figure 1 The organizational structure and flowcharts for each module of the system; Figure 2 This is a schematic diagram of the system assembly of the present invention. Detailed Implementation

[0017] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below.

[0018] The present invention proposes an interactive system for energy storage system based on a voice control model, comprising: a voice input module, a voice processing module for converting voice signals into character information, an escape information extraction module, a system strategy escape module, a strategy decomposition module, and an output module; The system's "input module" is for voice input. After receiving the user's wake-up prompt and entering a ready state, the system directly receives human voice commands (which can use different languages) and transmits them to the microphone or other type of sound receiving device on the "voice processing module," converting the sound wave (vibration) signal of the voice into an electrical signal.

[0019] The "speech processing module" is a digital transformation module that uses algorithms to map time-domain sound waves to character sequences. This module is in the form of an integrated circuit or a module board with chips, capable of converting speech sound wave electrical signals into digital character information through automatic speech recognition (ASR) parsing technology. The technical means required for this step are currently relatively mature and can be achieved through existing mature implementation methods or means on the market.

[0020] The "Escape Information Extraction Module" is used to capture contextual features of information that has been converted into characters (strings) after speech processing, based on the definition of device capability boundaries, preset dictionaries or pre-trained models, and whether the intent of the received information is explicit or implicit. It considers either word segmentation methods of natural language processing or large model methods similar to bidirectional long short-term memory (BiLSTM) to capture contextual features.

[0021] The "System Policy Escaping Module" works by vectorizing the escaped character (string) information and quantifying the matching degree by calculating the similarity between this information and the vectorized system policies in a pre-defined dictionary list. Specifically: When the received information is a system-built-in preset command (users can also define these commands by customizing semantic information), such as "System power on / off," "Highest performance mode," "Longest battery life mode," "Optimal economy mode," "System self-check," "Troubleshooting," "Contact customer service," or "Emergency stop," the system will complete the corresponding system action according to these predefined execution strategy commands. The specific method is as follows: This invention constructs a mechanism based on a "rule engine + keyword slot extraction" in a local system. This mechanism can run offline with fast response and good data privacy. Furthermore, this method does not rely on a large online model that requires an internet connection. The main implementation steps are: 1. Tokenization: Segmenting the string (sentence) processed by the speech processing module.

[0022] 2. Regular Expression / Keyword Matching: Extracts core parameters based on the word segmentation results according to three categories: "object," "action," and "target" (including quantity or magnitude). The core parameters within the rule set defined for this system are shown in the table below:

[0023] 3. Rule Engine: Based on the rule set, a corresponding set of operation strategies (referred to as "strategy set") is established. The strategy set is a collection of operation strategies related to system devices, and each operation strategy has a corresponding system device control instruction to match. The following is a summary of the actual control objectives transformed from the rule set and mapped to specific device operation instructions (strategy set), that is, translating natural language into specific device control logic: Serial Number Rule set combination Control Target Strategy set (device object and point table operation logic) 1 System / Inverter + On / Start System startup 1. Check if the BMS status is normal. 2. Send command 1 (Start) to the PCS / inverter "Start / Stop Control Word" address. 2 System / Inverter + Off / Stop System shutdown 1. Execute power descent logic. 2. Send command 0 (Stop) to the PCS / inverter "Start / Stop Control Word" address. 3. Disconnect the AC / DC side contactor. 3 Mode + Settings + Peak Performance Switch to high performance mode 1. Set the "Operating Mode" register to High_Perf. 2. Remove power limiting and set the "Maximum Charge / Discharge Power" to 100% of the rated value. 3. Adjust the temperature control system threshold. 4 Mode + Settings + Optimal Economy Switch to energy-saving / economy mode 1. Set the "Operating Mode" register to Eco_Mode. 2. Enable the peak shaving and valley filling algorithm. 3. Automatically adjust the charging and discharging logic based on the current electricity price period. 5 Mode + Settings + Maximum Battery Life Switch to long battery life mode 1. Set the "Running Mode" register to Long_Range. 2. Limit the maximum output power (e.g., set it to 80%). 3. Optimize the power consumption of auxiliary equipment (fans / air conditioners). 6 Battery / PCS + Settings + Percentage <number> (e.g., Battery setting 80%) Setpoint setting (SOC / Power) If the context is charging limitation: Write a value <number> (e.g., 80) to the "SOC Upper Limit Setting" register. If the context is power output: Write a percentage of the rated power <number> to the "Active Power Setpoint" register. 7 Inverter / PCS + Increase / Enlarge + Some / A Little Step-by-step fine-tuning (upward adjustment) Read the current "active power" setting value, calculate (with a set step size, such as 5%), and write it to the power setting register. 8 Inverter / PCS + Reduce / Lower + Some / A Little Step-by-step fine-tuning (downward adjustment) Read the current "active power" setting value, calculate (set step size), and write it to the power setting register. 9 System / PCS + Self-test Trigger device self-diagnosis 1. Send the Self_Check flag to the "System Instructions" register. 2. Poll the fault status register and return the result to the interface. 10 Customer service / manufacturer + Contact Get human support 1. No hardware commands are issued. 2. A pre-set manufacturer's 400 number or customer service QR code pops up on the human-machine interface. 11 Boost converter + Increase + Percentage (number) Boost / Voltage Regulation Write a percentage of the rated voltage to the "Output Voltage Setpoint" register of the boost converter (or increase that percentage from the current value). … …… …… …… This invention mainly focuses on the human-computer interaction process; the system's own strategy execution process and the specific instructions and actions of each device are omitted here.

[0024] When the received information is not a pre-defined command built into the system, semantic reasoning is used to infer the user's true purpose from the user's input, thereby determining the action strategy of the system and related devices. The main methods applied are "Intent Recognition" and "Slot Filling" (especially when specific numerical values ​​are involved). The relevant steps are as follows: 1. Establish semantic teaching materials 1.1 Establish a "tag" system for relevant actions based on system operation objectives: LABEL_0: Startup LABEL_1: Stop LABEL_2: Charging LABEL_3: Discharge LABEL_X: ... (other related strategies or actions) LABEL_END: ​​No action (corresponds to scenarios where user input is not related to system actions). 1.2 Preparing Training Data (Data Annotation) Collect and summarize a large amount of possible user input language chat, and complete the tagging, i.e., "corpus annotation". An example is as follows: Input information True Intent (Label) Remark "Using energy storage systems" start up direct instructions "I don't want to rely solely on wired electricity." start up Implicit instructions There will be a power outage tomorrow. Charge Implicit Needs Set SOC to 50% Set SOC to 50% Specific parameters "Turn it off." stop direct instructions "Hello" No action Non-system actions …… …… 2. Architecture and Training We employ the BERT text classification algorithm model, retaining the core of BERT (a 12-layer Transformer structure) and adding one output layer (a fully connected layer). Within this output layer, we utilize an "attention" mechanism to focus on the key information of the text being classified. This method yields the differential probabilities of relevant labels, ultimately determining which label to use as the basis for the specific action.

[0025] For actions involving setting specific data parameters, the BiLSTM-CRF algorithm can be used to obtain the specific values ​​and units by performing sequence labeling, thereby extracting the specific parameter values ​​and unit information.

[0026] Building upon the model described above, further training can be used to execute specific behavioral actions within the policy set. The steps are as follows: Input data: Input the completed and summarized label set into the model.

[0027] Computational error: The model is verified through the forward computation process of the neural network and the error between the result and the expected label is obtained.

[0028] Backpropagation: Based on the label results and the quantized deviation, the gradient of each layer is calculated in reverse.

[0029] Adjusting parameters: Fine-tune the model's internal weight values ​​through gradient calculations via backpropagation.

[0030] Convergence: After several rounds of training, the model is able to accurately (or with a high probability) obtain specific policy instructions by summarizing the possible user input information.

[0031] The "Strategy Decomposition Module" is designed to decompose control actions according to the system operation strategy determined in the previous module ("System Strategy Escape Module"). This is done in conjunction with information such as the actual online status of system equipment under input conditions, relevant parameters, historical data, and technical characteristic indicators, following a pre-defined strategy decomposition table. This module also performs action logic analysis and calculation to achieve the strategy objectives from the current system operation indicators. This includes extracting the equipment number or control point table number of the controlled object or device from the strategy, clarifying the specific control signal types and directions (e.g., "on," "off," "charging," "discharging," "alarm"), and specifying the values ​​of the control parameters. This module consists of the following specific components: 1. Multi-source data access and status standardization (1) Data aggregation: Real-time acquisition of system equipment online status (Online / Offline), key parameters (such as voltage, temperature, SOC, etc.), historical operating data and technical characteristic indicators.

[0032] (2) Policy loading: Read the currently effective system operation policy configuration and load the associated "policy decomposition table".

[0033] (3) Data cleaning: The input data is validated and standardized to remove abnormal noise and ensure that subsequent logical calculations are based on accurate state benchmarks.

[0034] 2. Strategy Matching and Task Decomposition (1) Scene recognition: Based on the current system running status and input characteristics, match the corresponding policy entries in the "Policy Decomposition Table".

[0035] (2) Action decomposition: Break down the macro-level strategic objectives (such as "peak shaving and valley filling" and "energy-saving operation") into specific execution sub-tasks.

[0036] (3) Object locking: Extract controlled object information from the decomposition results, identify the specific equipment number or control point ID, and establish the mapping relationship between the strategy and the physical equipment.

[0037] 3. Target Deviation Analysis and Logical Calculation (1) Current status assessment: Compare the current system operation indicators with the target values ​​set by the strategy, and calculate the deviation (Gap Analysis).

[0038] (2) Logical deduction: Based on the preset control algorithm or rule engine, analyze the action logic required to achieve the goal (e.g., whether to increase power or reduce load).

[0039] (3) Parameter calculation: Based on the deviation and the technical characteristics of the equipment, calculate the specific control parameter values ​​(such as: set power value, target temperature, charge / discharge rate, etc.) to ensure that the parameters are within the safe operating range of the equipment.

[0040] The system's "output module" functions as follows: It handles control commands and output signals for the downstream controlled equipment, including adjustment parameters such as timing, duration, increase, and decrease within the control strategy, as well as control parameters for each device (e.g., inverters, boost converters, BMS), such as on / off states and timing logic. Specifically: 1. Standardized encapsulation of control instructions (1) Signal type definition: Based on the logic calculation results, clarify the specific control signal type (such as: digital quantity "on / off", analog quantity "set value", status quantity "charging / discharging", event quantity "alarm", etc.).

[0041] (2) Control direction confirmation: Clarify the direction of control action (e.g., positive regulation, reverse inhibition, emergency cut-off) to ensure that the intention of the command is unambiguous.

[0042] (3) Message assembly: Encapsulate the device tag number, signal type, control direction and control parameter values ​​into a standardized control command message.

[0043] 2. Security Verification and Command Issuance (1) Interlock check: Before the instruction is issued, a safety interlock logic check is performed (e.g., whether the equipment is under maintenance, whether it is in a fault state, whether the parameters are out of limit) to prevent misoperation.

[0044] (2) Command output: After the verification is successful, the control command is sent to the underlying execution unit (such as PLC, SCADA or equipment controller).

[0045] (3) Execution feedback: Listen for device execution feedback signals, record command sending and execution results, form a closed-loop log, and provide historical data support for subsequent strategy optimization.

[0046] An interactive method for energy storage systems based on a voice-controlled model is presented. The system receives user voice through a microphone and converts it into text via a voice processing module. An escape information extraction module uses a rule engine or BERT model for intent recognition and slot filling. A system policy escape module matches a preset policy or generates a new policy based on the recognition results. A policy decomposition module combines real-time device status and safety constraints to generate specific control commands, which are then sent to inverters, BMS, PCS, and other devices for execution via an output module.

[0047] Specifically, a data connection is established between the voice input device and the new energy-storage power generation system equipment (hereinafter referred to as "energy storage power supply"). This connection is completed using a mature technology chain, and the specific technical routes that can be adopted include: Bluetooth communication Near Field Communication (NFC) Wi-Fi Direct Wired connection (USB-OTG) After the connection is established, a complete intelligent human-machine interface (HMI) + energy storage power supply system assembly as described in this invention is formed, such as... Figure 2 As shown.

[0048] Through the above-mentioned system assembly, the present invention enables the transmission and execution of information from human voice input to device control and system adjustment.

[0049] Summarize: This invention enables users to directly control the operating status of an energy storage system using natural language. Users do not need prior experience in power systems, energy storage control, or specialized operations; they can perform operations such as system start / stop, operating mode switching, charging / discharging parameter setting, power adjustment, and system self-checks via voice commands. This invention transforms complex energy management and control logic into an intuitive and natural language interaction method, thereby significantly lowering the barrier to entry for using new energy storage systems.

[0050] The overall system architecture of this invention adopts a modular design, consisting of a voice input terminal, a voice processing and semantic parsing module, a system policy translation module, a policy decomposition module, and a device control output module. The system can be deployed independently in the energy storage power supply unit or edge controller, or integrated as a functional extension module of an existing new energy-energy storage system. The architecture supports local operation and offline control, and possesses good scalability, compatibility, and reliability.

[0051] Each functional module has clearly defined responsibilities and collaborates with others: The voice input module is responsible for collecting the user's voice and completing basic signal conversion; The speech processing module completes the conversion of speech to characters; The escape information extraction module performs semantic analysis on character information to identify user intent and key parameters; The system policy escaping module maps abstract semantics to concrete, executable system operation policies; The strategy decomposition module combines device status and safety constraints to further decompose the strategy into device-level control instructions; The output module sends commands to the corresponding inverter, energy storage unit or auxiliary equipment to complete the actual control action.

[0052] At the hardware level, the system establishes a connection with new energy storage devices through microphones, processing chips, and communication interfaces; at the software level, it achieves semantic understanding and strategy generation through speech recognition, natural language processing, rule engines, and machine learning models. The entire process begins with voice acquisition, proceeds through recognition, parsing, decision-making, and decomposition, and ultimately forms safe and controllable equipment operation commands, constituting a complete closed-loop control process.

[0053] This invention significantly improves the human-machine interaction efficiency and intelligence level of new energy-storage systems, making energy management more intuitive, efficient, and safe. It demonstrates excellent application results in scenarios such as residential energy storage, industrial and commercial energy storage, off-grid power supplies, mobile power supply, and microgrids. It can improve system availability, reduce operation and maintenance complexity, and promote the widespread application of new energy and energy storage systems among non-professional users, thus possessing broad industrial application prospects.

[0054] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An interactive system for energy storage systems based on a voice-controlled model, characterized in that, include: The voice input module is used to receive the user's voice commands; The voice processing module is used to convert voice signals into character information; The escape information extraction module is used to perform semantic parsing and intent recognition on the character information, and extract target object, action type and parameter information; The system policy escaping module is used to generate or match corresponding system operation policies; The strategy decomposition module is used to decompose the system operation strategy into device control instructions; The output module is used to send device control commands to the corresponding devices.

2. The system according to claim 1, characterized in that: The system policy escaping module is configured as follows: First, the target object, action type, and parameter information are converted into semantic vectors; Then, the semantic vector is compared with the strategy vector in the preset strategy vector library to calculate the similarity and match the corresponding system operation strategy. When no preset strategy is matched, a corresponding system operation strategy is generated based on the intent recognition model.

3. The system according to claim 2, characterized in that: The system strategy escaping module has a built-in rule engine, which is configured with a preset rule library. The preset rule library contains the mapping relationship between keyword slot combinations of object-action-target and system operation strategies. When the information output by the escaping information extraction module matches the keyword slot combination, the system strategy escaping module directly calls the corresponding system operation strategy. The system policy escaping module also has a built-in lightweight machine learning model, which includes an intent recognition module and a slot filling module. When the information output by the escaping information extraction module does not match the preset rules, the intent classification module is used to identify the user intent category, the slot filling module is used to extract specific parameters, and the system policy escaping module generates the corresponding system operation strategy based on the intent category and parameters.

4. The system according to claim 1, characterized in that: The strategy decomposition module is configured as follows: 1) First, comprehensively organize and sort out at least one of the following information, and standardize the multi-source data and status parameters: Energy input forms include renewable energy generation, grid power supply, or a combination thereof; Energy storage forms include one or more of electrochemical energy storage, hydrogen energy storage, and compressed air energy storage; Energy usage patterns, including the time of energy use, power output, operating time, and energy-consuming objects; Environmental and external conditions information, including weather conditions, electricity market information, or electricity price information; Based on the above information, a corresponding customized operation strategy process is generated; 2) Perform strategy matching and task decomposition based on the current system operating status, input characteristics, and target description, and extract relevant information of the strategy object from the decomposition results to establish physical mapping relationships; 3) Conduct deviation analysis between system operation indicators and strategy objectives to obtain control parameters; 4) Encapsulate standardized control commands, clarify the direction of action of the control commands, and determine the device tag number, type, and parameter values.

5. The system according to claim 1, characterized in that: The system policy escaping module supports voice command parsing and policy generation through a local rule engine or local semantic model without an internet connection; the policy decomposition module verifies device operation safety constraints, power constraints, and state constraints before generating control commands.

6. The system according to claim 1, characterized in that: The voice input module communicates with the new energy-storage system via at least one of Bluetooth, near-field communication, wireless direct connection, or wired connection.

7. An interactive method for energy storage systems based on a voice-controlled model according to any one of claims 1-6, characterized in that, The specific steps are as follows: Receive user voice input commands; Perform speech recognition processing on the speech signal and convert the speech signal into character information; The character information is semantically parsed and intent is identified to extract the target object, action type, and parameter information; Match or generate corresponding system operation strategies based on the parsing results; The system operation strategy is decomposed into specific equipment control commands and sent to the corresponding equipment in the new energy-storage system for execution.

8. The method according to claim 7, characterized in that: The voice input supports multiple natural languages. It normalizes voice commands in different languages ​​through a unified semantic intermediate representation, thereby achieving consistent control of the new energy-storage system in a cross-language environment.

9. The method according to claim 7, characterized in that: The system operation strategy is a customizable strategy applicable to different types of energy storage systems or power system equipment. The customizable strategy includes at least one of the following: preset rule strategy customization, semantic and control strategy mapping customization, and equipment capability boundary constraints. The customization process includes strategy definition, strategy verification, and strategy binding. The implementation steps include strategy generation, strategy decomposition, and strategy execution.

10. The method according to claim 7, characterized in that: The semantic parsing and intent recognition are implemented through a rule engine and / or machine learning model, including a rule matching method based on keyword slot extraction and a semantic reasoning method based on intent recognition and slot filling.

Images