Intelligent parameter scheme recommendation method, device and equipment for energy storage system

Abstract

The invention discloses an intelligent parameter scheme recommendation method, device and equipment for an energy storage system, and belongs to the technical field of energy storage systems. The method comprises the following steps: acquiring an original parameter scheme of the energy storage system and associated running context information, and receiving parameter adjustment information input by a user; mapping the parameter adjustment information, the original parameter scheme and the operation context information to the same potential feature space through a joint embedding model, and searching in the potential feature space to obtain multiple groups of candidate parameter combinations; based on the simulation substitution model, predicting the operation process of the energy storage system under the combination of the multiple groups of candidate parameters to obtain multiple groups of prediction results; performing multi-dimensional evaluation on the multiple groups of prediction results to determine a target parameter combination meeting user preferences and system constraint conditions; and feeding back the recommended parameter scheme to the user, and outputting explanation information for representing the difference among the original parameter scheme, the scheme assumed by the user and the recommended parameter scheme. The parameter adjustment decision cost can be reduced, and the interaction efficiency can be improved.

Description

Technical Field

[0001] This application belongs to the field of energy storage system technology, and in particular relates to a method, apparatus and equipment for recommending intelligent parameter schemes for energy storage systems. Background Technology

[0002] With the continuous development of new energy power generation and energy storage technologies, battery energy storage systems are widely used in scenarios such as photovoltaic-wind power microgrids, charging stations, and user-side energy management. These energy storage systems require comprehensive consideration of multiple factors during operation, including electricity pricing mechanisms, load forecasting, environmental factors, equipment operating status, and battery life constraints, to achieve safe, economical, and efficient operation. In practical applications, users often need to adjust the parameter configuration of the energy storage system according to operational requirements, such as changing the charging and discharging power, modifying the state-of-charge threshold, or resetting the operating period, to assess the impact of different parameter configurations on system revenue and equipment lifespan.

[0003] In related technologies, energy storage systems typically rely on physical mechanism-based numerical simulations or digital twin models for parameter adjustment and evaluation, performing complete operational calculations for given parameter schemes. While these methods can accurately reflect system behavior, the calculation process is complex and time-consuming, making it difficult to meet the need for immediate response in scenarios where users frequently adjust parameters or require rapid evaluation results. Furthermore, simulation results are mostly presented in curve or numerical form, requiring users to have a certain level of expertise to understand the differences between various schemes, thus limiting their ability to support parameter adjustment decisions.

[0004] Some related technologies attempt to provide parameter recommendations or operational strategy selection for energy storage systems through rule-driven or static optimization. However, these solutions are typically based on predefined objectives or fixed rules, making it difficult to flexibly reflect users' personalized needs or comprehensively weigh different parameter options under multi-objective constraints. Furthermore, the comparison and recommendation process between different solutions lacks an intuitive explanation mechanism, making it difficult for users to understand the rationale behind the recommendations. This limits the system's practicality in application scenarios such as high-frequency adjustments, rapid decision-making, and human-computer interaction. Therefore, these technologies still have shortcomings in terms of interactivity, interpretability, and personalized support, and require further improvement. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method, apparatus, and device for intelligent parameter scheme recommendation for energy storage systems, which supports rapid evaluation and scheme recommendation for user parameter adjustments in complex operating scenarios, reduces the cost of parameter adjustment decisions, and improves interaction efficiency.

[0006] Firstly, this application provides a method for recommending intelligent parameter schemes for energy storage systems, the method comprising: The system acquires the original parameter scheme and associated operating context information of the energy storage system, and receives parameter tuning information input by the user; the parameter tuning information is used to characterize the user's proposed scheme. The parameter tuning information, the original parameter scheme, and the runtime context information are mapped to the same latent feature space through a joint embedding model, and multiple sets of candidate parameter combinations are obtained by searching in the latent feature space. Based on the simulation substitution model, the operation process of the energy storage system under the multiple sets of candidate parameter combinations is predicted, and multiple sets of prediction results are obtained. The multiple sets of prediction results are evaluated in a multi-dimensional manner to determine the target parameter combination that satisfies user preferences and system constraints; the target parameter combination is used to characterize the recommendation parameter scheme. The system provides feedback to the user on the recommended parameter scheme and outputs explanatory information to characterize the differences between the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme.

[0007] Secondly, this application provides an intelligent parameter scheme recommendation device for energy storage systems, the device comprising: The acquisition module is used to acquire the original parameter scheme of the energy storage system and the associated operating context information, and to receive the parameter adjustment information input by the user; the parameter adjustment information is used to characterize the user's proposed scheme. The mapping module is used to map the parameter tuning information, the original parameter scheme, and the runtime context information to the same latent feature space through a joint embedding model, and to search for multiple sets of candidate parameter combinations in the latent feature space. The prediction module is used to predict the operation of the energy storage system under the multiple sets of candidate parameter combinations based on the simulation substitution model, and obtain multiple sets of prediction results. The evaluation module is used to perform multi-dimensional evaluation on the multiple sets of prediction results to determine the target parameter combination that satisfies user preferences and system constraints; the target parameter combination is used to characterize the recommendation parameter scheme. The recommendation module is used to provide feedback to the user on the recommended parameter scheme and output explanatory information to characterize the differences between the original parameter scheme, the user-initiated scheme and the recommended parameter scheme.

[0008] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent parameter scheme recommendation method for energy storage systems as described in the first aspect above.

[0009] Fourthly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent parameter scheme recommendation method for energy storage systems as described in the first aspect above.

[0010] Fifthly, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the intelligent parameter scheme recommendation method for energy storage systems as described in the first aspect.

[0011] Sixthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent parameter scheme recommendation method for energy storage systems as described in the first aspect above.

[0012] The intelligent parameter scheme recommendation method, device, electronic equipment, non-transitory computer-readable storage medium, chip, and computer program product provided in this application generate candidate parameter combinations by mapping the user's envisioned scheme's natural language parameter tuning information with the original parameter scheme and operating context information of the energy storage system to the same latent feature space. Based on a simulation substitution model with cross-scenario adaptability, the candidate parameter combinations are rapidly predicted for operation. Furthermore, the recommended parameter scheme is determined by combining multi-dimensional evaluation and interpretable feedback mechanisms. Thus, under the premise of ensuring system constraints, the method effectively connects the user's high-level parameter tuning intentions with the operating parameters of the energy storage system and supports rapid evaluation and understandable recommendation of parameter schemes.

[0013] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0014] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram illustrating an application scenario of the intelligent parameter scheme recommendation method for energy storage systems provided in this application embodiment; Figure 2 This is one of the flowcharts illustrating the intelligent parameter scheme recommendation method for energy storage systems provided in this application embodiment; Figure 3 This is the second flowchart illustrating the intelligent parameter scheme recommendation method for energy storage systems provided in this application embodiment; Figure 4 This is a schematic diagram of the intelligent parameter scheme recommendation device for energy storage systems provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0016] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0017] In practical applications, when users want to adjust the operating strategy of an energy storage system, such as changing the upper limit of charging or discharging power, modifying the constraint range of battery state of charge, or resetting the charging and discharging period, it is usually necessary to use numerical simulation to evaluate the adjusted parameter scheme in order to determine the impact of the parameter scheme on system revenue, battery life and operational safety under the current or predicted operating scenario.

[0018] With the development of artificial intelligence (AI) technology, general-purpose intelligent models with natural language understanding and generation capabilities have gradually attracted attention and are being explored and applied in power system-related fields. For example, some studies have attempted to use AI models to analyze power operation data to assist in applications such as fault identification, operational status monitoring, and dispatch decision support. This type of research has, to some extent, expanded the application scope of AI technology in power systems.

[0019] However, from the perspective of related technologies, current research largely focuses on single aspects such as system operation status analysis or scheduling optimization, failing to cover interactive application scenarios where users adjust parameters, conduct real-time evaluations, and select solutions during actual use. Furthermore, related technical solutions typically emphasize outputting analysis results or optimization decisions, lacking intuitive explanations of the differences between different parameter schemes and their trade-offs. Users find it difficult to understand the trade-offs in terms of benefits, lifespan, or constraints based on system outputs. Therefore, existing AI technologies remain insufficiently applicable in user-oriented parameter tuning and decision support applications.

[0020] In view of this, embodiments of this application provide an intelligent parameter scheme recommendation method for energy storage systems, which aims to support users in making parameter adjustments and scheme decisions in complex operating scenarios. By associating user intentions with system operating parameters and quickly evaluating and comprehensively analyzing candidate schemes, the method improves the interactive efficiency and decision rationality of the energy storage system parameter adjustment process.

[0021] The following description, in conjunction with the accompanying drawings, details the intelligent parameter scheme recommendation method for energy storage systems provided in this application, through specific embodiments and application scenarios.

[0022] The intelligent parameter scheme recommendation method for energy storage systems provided in this application embodiment can be applied to integrated energy systems containing energy storage units, such as residential energy storage systems, industrial and commercial energy storage systems, photovoltaic-energy storage microgrid systems, charging station energy storage systems, grid-side or independent energy storage power stations, and is used to support users in adjusting, evaluating and recommending schemes for energy storage system operating parameters under different operating modes and scheduling conditions.

[0023] Figure 1 This is a schematic diagram illustrating an application scenario of the intelligent parameter scheme recommendation method for energy storage systems provided in some embodiments of this application. For example... Figure 1 As shown, the application environment may include, for example, an energy storage system 102, a network, and electronic devices 104.

[0024] The energy storage system 102 includes, but is not limited to, energy storage batteries, battery management-related equipment, and energy conversion and control equipment connected to new energy power generation units and electrical loads. For example... Figure 1 In the residential energy storage scenario shown, the energy storage system can be connected to distributed photovoltaic power generation units and user-side loads to realize energy storage and release, and execute corresponding charging and discharging strategies according to the configured operating parameters.

[0025] The operational data generated by the energy storage system during operation can be transmitted to the electronic device 104 via a network. The operational data includes, but is not limited to, the operational status data of the energy storage battery, power and energy data, and scheduling and control information related to the operation of the energy storage system, which is used to support the analysis and evaluation of the operation under different parameter schemes.

[0026] Electronic device 104 can be a server or an edge computing node, used to process received operational data and user-inputted parameter tuning information to generate candidate parameter schemes, evaluation results, and recommended parameter schemes. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, security services, and big data and artificial intelligence platforms.

[0027] It should be noted that, Figure 1 The application scenarios shown are merely examples, and the specific structural form, deployment scale, and deployment method of the computing equipment of the energy storage system do not constitute a limitation on this application.

[0028] The intelligent parameter scheme recommendation method for energy storage systems provided in this application can be executed by an electronic device or a functional module or entity within an electronic device capable of implementing the method. The electronic devices mentioned in this application include, but are not limited to, terminals or servers.

[0029] The following uses an electronic device as the execution subject to illustrate the method for recommending intelligent parameter schemes for energy storage systems provided in the embodiments of this application.

[0030] Figure 2 This is a flowchart illustrating a method for recommending intelligent parameter schemes for energy storage systems, provided in some embodiments of this application. For example... Figure 2 As shown, the recommended method for intelligent parameter schemes for energy storage systems includes steps 210 to 250.

[0031] Step 210: Obtain the original parameter scheme of the energy storage system and the associated operating context information, and receive the parameter tuning information input by the user; the parameter tuning information is used to characterize the user's proposed scheme.

[0032] The original parameter scheme refers to the set of parameter configurations used by the energy storage system before the current parameter adjustment time, which is used to constrain or describe the existing operating strategy of the energy storage system. The parameter configuration includes, but is not limited to, one or more of the following: upper limit of charging and discharging power, lower limit of charging and discharging power, upper and lower limits of battery state of charge (SOC), allowed charging and discharging period, and operating mode identifier.

[0033] Operating context information refers to the set of information corresponding to the original parameter scheme and used to characterize the current operating environment and scheduling conditions of the energy storage system. Operating context information includes, but is not limited to, one or more of the following: electricity price information, load forecast information, environmental condition information, equipment status information, and current scheduling strategy type, which are used to reflect the actual operating background of the energy storage system under the original parameter scheme.

[0034] In some embodiments, the original parameter scheme and operating context information can be obtained from the controller, energy management system, or historical configuration records of the energy storage system and used as the basic operating status input for the current parameter adjustment scenario.

[0035] Meanwhile, the electronic device receives parameter adjustment information input by the user through the application (APP) interface. Parameter adjustment information refers to the input information provided by the user to express their parameter adjustment intentions, which is used to characterize the user's proposed scheme for the operation strategy of the energy storage system.

[0036] The parameter tuning information can include direct user commands to modify one or more operating parameters, or it can include high-level parameter tuning descriptions entered in natural language, such as descriptions of operating objectives, strategy focus, or constraint preferences. For example, a user could enter in natural language, "I want to reduce battery wear during periods of large peak-valley price differences."

[0037] Step 220: Map the parameter tuning information, the original parameter scheme, and the running context information to the same latent feature space through the joint embedding model, and search for multiple sets of candidate parameter combinations in the latent feature space.

[0038] The joint embedding model refers to a model structure used to uniformly represent input information of different modes or types. The joint embedding model is configured to encode parameter tuning information representing the user's proposed scheme, as well as original parameter schemes and operating context information representing the existing operating state of the energy storage system, into vector representations with the same dimension and semantic comparability.

[0039] In this embodiment, the joint embedding model is used to model the correlation between the user's high-level parameter tuning intentions and the low-level operating parameters of the energy storage system. Specifically, the joint embedding model encodes the parameter tuning information that represents the user's proposed scheme (e.g., the operational focus goals described in natural language, such as "reducing battery wear and saving costs", "being more conservative in high-temperature seasons", "wanting to suppress evening peak electricity costs", etc.) into an intention vector. At the same time, it encodes the original parameter scheme (e.g., adjusting the charging power limit from 50kW to 40kW, a specific charging / discharging power limit / SOC limit / time window, etc.) and its corresponding operating context information into a context vector, and maps the intention vector and the context vector to the same latent feature space.

[0040] The latent feature space refers to the vector representation space constructed by the joint embedding model. In this space, the relative positions of different vectors reflect the degree of correlation between the corresponding parameter tuning intention and parameter configuration at the semantic and operational feature levels. By mapping parameter tuning information, as well as the original parameter scheme and operational context information, to the latent feature space, the relationship between the user's envisioned scheme and the current operating state of the energy storage system can be characterized at a unified representation level.

[0041] Electronic devices perform similarity searches on parameter configurations related to the current parameter tuning scenario based on a latent feature space. This identifies parameter configurations that are highly semantically relevant to the tuning information and compatible with the original parameter scheme and runtime context information at the runtime condition level. These parameter configurations are then output as candidate parameter combinations. Through inverse mapping, multiple sets of parameter configurations can be generated or selected from the latent feature space as candidate parameter combinations for the current parameter tuning scenario.

[0042] Among them, candidate parameter combination refers to one or more sets of parameter configurations generated in the potential feature space based on the correlation between the user's proposed scheme and the system's operating background under the current parameter adjustment scenario. Each candidate parameter combination is used to describe the operating parameter scheme that the energy storage system may adopt under the corresponding assumptions.

[0043] The above approach breaks away from the traditional method of adjusting configurations based solely on independent parameter inputs. It enables the understanding of the relationship between high-level user intent and low-level operating parameters at a unified representation level, and provides multiple candidate inputs for subsequent assessment of the operational impact of different parameter schemes.

[0044] Step 230: Based on the simulation substitution model, predict the operation process of the energy storage system under multiple combinations of candidate parameters and obtain multiple prediction results.

[0045] Among them, the simulation alternative model refers to the model used to approximate the operating behavior of the energy storage system. It is used to simulate and predict the operation of the energy storage system under given parameter configuration conditions without performing a complete physical simulation or digital twin calculation.

[0046] In this embodiment, the simulation substitution model is a meta-learning-based model configured to rapidly adapt to different operating scenarios and parameter distributions. Specifically, this simulation substitution model utilizes simulation rules formed by a digital twin model under various parameter scenarios for modeling. This enables the model to characterize the relationship between parameter changes and operating states during the operation of the energy storage system. When faced with new candidate parameter combinations, the model state is adjusted using a small number of samples or existing operating information to adapt to the current parameter tuning scenario. Compared to static proxy models, this simulation substitution model can be migrated and used across different operating scenarios, thereby meeting the needs of real-time prediction.

[0047] For any candidate parameter combination, the electronic device inputs this candidate parameter combination along with the corresponding operating context information into a meta-learning-based simulation substitution model to predict the operation of the energy storage system under that parameter configuration. The operation process refers to the dynamic operation of the energy storage system within a preset time range, reflecting the state evolution characteristics of the energy storage system under the constraints of the candidate parameter combination.

[0048] The prediction information output by the simulation alternative model constitutes the corresponding prediction results. The prediction results can include time series information used to characterize changes in the operating state of the energy storage system, reflecting changes in power, battery state of charge, and changes in economic indicators related to operation.

[0049] In some embodiments, to ensure the reliability of the prediction results, when the prediction results of the simulation alternative model for a certain candidate parameter combination have large uncertainties or are close to the system constraints, the electronic device can also combine a small number of high-precision simulation results to calibrate the prediction results, so as to ensure prediction efficiency while taking into account prediction accuracy.

[0050] Step 240: Evaluate multiple sets of prediction results from multiple dimensions to determine the target parameter combination that satisfies user preferences and system constraints; the target parameter combination is used to characterize the recommended parameter scheme.

[0051] The prediction results refer to the predictive information output by the simulation substitution model under different combinations of candidate parameters, used to characterize the operational performance of the energy storage system. Multi-dimensional evaluation refers to the process of comprehensively analyzing the prediction results from multiple evaluation dimensions.

[0052] Based on the prediction results corresponding to each candidate parameter combination, an evaluation model is constructed to characterize the relationship between parameter adjustments and operational results. This evaluation model is configured to analyze the impact of different parameter configurations on multiple evaluation indicators, thereby depicting the direction and degree of influence of parameter changes on the operational behavior of the energy storage system. Through this method, the different contributions of different parameter combinations to the operational results can be distinguished during the evaluation process.

[0053] In some embodiments, multi-dimensional evaluation includes at least a comprehensive analysis of multiple evaluation indicators related to the operation of the energy storage system. These evaluation indicators characterize the operating features of the energy storage system under different parameter schemes. The electronic device processes the prediction results of each candidate parameter combination across different evaluation indicator dimensions in a unified manner, and, under the premise of satisfying system constraints, comprehensively sorts or filters each candidate parameter combination.

[0054] System constraints refer to the restrictions that limit the safety, stability, or operational compliance of the energy storage system, and are used to define the feasible range of candidate parameter combinations. Candidate parameter combinations that do not meet the system constraints will not be considered as target parameter combinations.

[0055] After completing the multi-dimensional evaluation, the electronic device determines at least one set of parameter combinations whose comprehensive evaluation results meet the preset evaluation criteria from the candidate parameter combinations that satisfy the system constraints. This set is then used as the target parameter combination. The target parameter combination serves as the parameter scheme recommended to the user in the current parameter tuning scenario and as the basis for subsequent recommendation feedback and explanation output.

[0056] Step 250: Provide the user with the recommended parameter scheme and output explanatory information to characterize the differences between the original parameter scheme, the user's proposed scheme and the recommended parameter scheme.

[0057] Among them, explanatory information refers to natural language information used to explain the differences between different parameter schemes in terms of operational characteristics and evaluation results. It is used to help users understand the basis for the formation of recommended parameter schemes and their differences from other schemes.

[0058] Electronic devices compare and analyze the recommended parameter scheme corresponding to the target parameter combination with the original parameter scheme and the user-designed scheme to identify the differences in key parameter configurations and operational performance among different parameter schemes. The comparative analysis is used to characterize the relationship between parameter changes and benefit-related indicators, battery life-related indicators, and system constraint satisfaction, thereby reflecting the trade-offs of different parameter schemes across multiple evaluation dimensions.

[0059] In some embodiments, based on the above comparative analysis results, the electronic device calls a language model to generate explanatory information in natural language form, which is used to explain the performance of different parameter schemes.

[0060] For example, the explanatory information is used to characterize at least one or more of the following differences: the main trends and key differences in the operation of different parameter schemes; the differences in revenue, lifespan-related indicators and system constraints of different parameter schemes and the reasons for these differences; potential operational risks or precautions related to the parameter schemes; and so on.

[0061] In some embodiments, when the parameter tuning information includes high-level operational objectives for characterizing the user's envisioned scheme, the explanatory information is also used to explain how the high-level operational objectives are reflected in the recommended parameter scheme, so as to help the user understand the correspondence between the recommended parameter scheme and their envisioned scheme.

[0062] The above explanations provide a way of explaining why the recommended parameter scheme differs from the original parameter scheme and the user's intended scheme, in a manner that aligns with the user's understanding.

[0063] Furthermore, electronic devices can present users with an explanation of the applicability of the recommended parameter scheme based on explanatory information, supporting users' understanding and confirmation of the recommendation results. By simultaneously providing users with the recommended parameter scheme and its corresponding explanatory information, interpretable feedback on the parameter tuning results is achieved, thereby providing decision support for users on whether to adopt or further adjust the parameter scheme.

[0064] According to the intelligent parameter scheme recommendation method for energy storage systems provided in the embodiments of this application, candidate parameter combinations are generated by mapping the natural language parameter tuning information of the user's envisioned scheme with the original parameter scheme and operating context information of the energy storage system to the same potential feature space. The candidate parameter combinations are then rapidly predicted based on a simulation substitution model with cross-scenario adaptability. Furthermore, the recommended parameter scheme is determined by combining multi-dimensional evaluation and interpretable feedback mechanisms. This achieves effective connection between the user's high-level parameter tuning intention and the operating parameters of the energy storage system while ensuring system constraints, and supports rapid evaluation and understandable recommendation of parameter schemes.

[0065] In the process of parameter adjustment and scheme recommendation for energy storage systems, users typically express their parameter adjustment needs or operational goals using natural language, while the scheduling strategies and operating parameters of the energy storage system exist in the form of numerical configurations and structured contextual information. In related technologies, the correspondence between user parameter adjustment needs and system parameter configurations is usually handled through rule matching, keyword mapping, or manual configuration, which makes it difficult to accurately understand user intent.

[0066] However, the above methods often treat natural language parameter tuning information and parameter configuration as independent input objects, lacking a unified modeling of the relationship between the semantic features of user intent and the features of system operating parameters. On the one hand, natural language expressions are abstract and diverse, making it difficult to accurately map them to specific parameter adjustments through fixed rules; on the other hand, the system operating state reflected by parameter schemes and operating context information cannot be directly compared with the user's parameter tuning intent at the semantic level, easily leading to inconsistencies between the user's envisioned scheme and the actual operating conditions of the system.

[0067] Therefore, it is difficult to simultaneously characterize the user's parameter tuning intention and the energy storage system's operating status within the same representation system in related technologies. This restricts the effectiveness of unified retrieval, combination, and evaluation of candidate parameter schemes in parameter tuning scenarios. It is necessary to introduce a technical means that can uniformly represent different types of information.

[0068] Therefore, in some embodiments, a joint embedding model is used to map the parameter tuning information, the original parameter scheme, and the runtime context information to the same latent feature space. This includes: mapping the natural language information in the parameter tuning information to the latent feature space through a text intent encoding network to obtain an intent vector representing the user's parameter tuning intent; and mapping the original parameter scheme and associated runtime context information to the latent feature space through a parameter-context encoding network to obtain a current context vector representing the current runtime state. The text intent encoding network and the parameter-context encoding network are obtained through joint training.

[0069] The joint embedding model includes a text intent encoding network and a parameter-context encoding network, which are used to represent different types of information in a unified manner.

[0070] The text intent encoding network is used to encode the natural language parameter tuning instructions input by the user, mapping the natural language information representing the parameter tuning intention into a fixed-dimensional vector representation. The resulting vector is used to characterize the user's intent vector in the parameter tuning scenario. Natural language parameter tuning instructions may include descriptions of operational goals or strategic focuses, such as preferences for battery life, operational benefits, or electricity price periods.

[0071] The parameter-context coding network is used to encode the parameter scheme of an energy storage system and its corresponding operational context information, mapping the numerical parameter configuration and the associated operational environment information into a vector representation consistent with the intent vector dimension. Parameter configuration may include upper limits for charging and discharging power, state of charge thresholds, and available charging and discharging periods, while operational context information may include site type, electricity price information, load forecasting results, and the current operational strategy mode.

[0072] In some embodiments, the text intent encoding network and the parameter-context encoding network can be implemented using neural network models. The text intent encoding network can include encoding structures for feature extraction and representation of natural language sequences, such as network structures built based on recurrent neural networks, convolutional neural networks, or attention mechanisms; the parameter-context encoding network can include multilayer perceptron networks or other feature encoding networks for joint representation of numerical parameters and contextual features.

[0073] It should be noted that the above network structure is only an example. Those skilled in the art can choose appropriate network forms or combinations thereof to implement the encoding function according to actual application scenarios. This application does not limit this.

[0074] On the one hand, electronic devices process natural language information (such as "pay more attention to battery life" and "control peak-hour electricity costs") in parameter tuning information through text intent coding networks. Text intent coding networks are used to semantically represent the natural language parameter tuning descriptions input by users, mapping the natural language information to vector representations in the latent feature space, thereby obtaining an intent vector that represents the user's parameter tuning intention.

[0075] For example, in the current hyperparameter tuning scenario, electronic devices use a Large Language Model (LLM) to perform semantic analysis on natural language hyperparameter tuning information. This identifies the target focus and potentially implicit operational constraints in the user's hyperparameter tuning intent, and combines this with the numerical modification information explicitly provided by the user to form an intent vector. By utilizing a language model to perform semantic analysis on complex natural language instructions, it helps to accurately identify the target and implicit constraints in the user's hyperparameter tuning intent. Combined with a joint embedding model, it achieves a collaborative mapping between natural language intent and numerical parameters, improving the matching degree between candidate parameter combinations and the user's actual needs.

[0076] On the other hand, the electronic device processes the original parameter scheme and its associated operating context information through a parameter-context coding network. It receives the parameter configuration of the energy storage system (such as the upper and lower limits of charging and discharging power, SOC upper and lower limits, and charging and discharging time periods) and the corresponding operating context information (such as the corresponding site type, electricity price and load forecast, and current strategy mode) through the parameter-context coding network, and maps them into the latent feature space to generate a current context vector to represent the current operating state as a reference characterization.

[0077] In some embodiments, the joint embedding model is constructed offline before being applied to the parameter tuning scheme recommendation. During the training phase, the text intent encoding network and the parameter-context encoding network are constructed through joint training, making the intent vector mapped to the latent feature space comparable to the current context vector under the same representation system. This supports the correlation analysis of user parameter tuning intent and system operating status within a unified latent feature space.

[0078] Specifically, the electronic device system jointly trains the text intent encoding network and the parameter-context encoding network based on historical parameter tuning samples. These historical samples represent the relationship between the user's tuning intent, the actual parameter scheme used, and the corresponding operating scenario during the historical tuning process. Through joint training, the correlation between intent vectors and parameter-context vectors matching the same tuning intent and operating scenario is enhanced in the latent feature space, presenting a "user intent - actual parameters used - current scenario" relationship. This supports similarity measurement of natural language intent and numerical parameter schemes within a unified latent feature space.

[0079] Through the training methods described above, the joint embedding model can provide a basic representation capability for the search and generation of candidate parameter combinations based on the latent feature space in subsequent parameter tuning scenarios, while maintaining the comparability of different types of input information.

[0080] In the above embodiments, by introducing a text intent encoding network and a parameter-context encoding network, the parameter tuning information in natural language form and the parameter scheme and operating context information in numerical form are mapped to the same latent feature space. This allows the user's high-level parameter tuning intent to be associated with the underlying operating parameters of the energy storage system within a unified representation space, thereby avoiding the inconsistency problem caused by direct matching between natural language information and numerical parameters. Furthermore, by jointly embedding different types of information, a foundation is provided for generating candidate parameter combinations based on semantic relevance and operating state similarity within the latent feature space. This enables the subsequent parameter scheme generation process to simultaneously consider both user expectations and actual system operating conditions, improving the rationality and consistency of parameter scheme recommendations.

[0081] During the parameter tuning process of relevant energy storage systems, even when a preliminary mapping between users' natural language tuning requirements and system parameters has been established, the problem of generating unreasonable candidate parameter schemes persists. On the one hand, complex natural language instructions often contain implicit goals or constraints, and relying solely on keyword or rule parsing makes it difficult to accurately understand the user's true intent. On the other hand, when generating parameter schemes, if matching is based solely on the user's current intent, it is easy to overlook the constraints of the existing scheduling strategies and operating environment of the energy storage system, resulting in parameter schemes that clearly conflict with existing operating conditions and have high implementation costs.

[0082] Furthermore, related technologies typically involve direct searching or enumeration within the parameter space, resulting in an excessively large set of candidate parameters or insufficient relevance, which is detrimental to subsequent simulation evaluation and scheme selection.

[0083] To this end, in some embodiments, multiple sets of candidate parameter combinations are searched in the latent feature space, including: extracting the running context information associated with each parameter configuration in the candidate parameter library, and mapping each parameter configuration and the associated running context information to the latent feature space to obtain a candidate context vector for representing each candidate running state; calculating the intent similarity between the intent vector and each candidate context vector, and the environment similarity between each candidate context vector and the current context vector in the latent feature space; determining multiple target parameter configurations in the candidate parameter library based on the intent similarity and environment similarity, and generating multiple sets of candidate parameter combinations based on the multiple target parameter configurations.

[0084] During the candidate parameter generation process, the electronic device encodes each parameter configuration in the candidate parameter library in conjunction with its historical operating context information to obtain the corresponding candidate context vector.

[0085] The candidate parameter library stores or provides searchable parameter configurations and their corresponding operating context information. Parameter configurations can be derived from one or more of the following: historical operating parameter schemes of the energy storage system, pre-configured parameter schemes, or extended parameter schemes generated based on existing parameter schemes.

[0086] Within the latent feature space, the electronic device simultaneously evaluates the similarity of candidate parameters based on the intent vector and the original policy context vector.

[0087] On the one hand, electronic devices measure the degree of fit between the candidate parameter scheme and the user's current parameter tuning intention by comparing the correlation between the parameter-context vector and the intent vector corresponding to the candidate parameter scheme in the latent feature space.

[0088] On the other hand, electronic devices measure the consistency between candidate parameter schemes and existing operating strategies and operating environments by comparing the correlation between the parameter-context vectors corresponding to candidate parameter schemes and the current context vectors in the latent feature space.

[0089] By simultaneously considering the relationship between the two aspects in the potential feature space, the system can take into account both the user's proposed scheme and the existing operating conditions of the energy storage system when generating candidate parameter combinations, thereby reducing the possibility of parameter schemes that are significantly inconsistent with the current operating environment or are difficult to implement entering the candidate set.

[0090] Based on the above similarity evaluation results, the electronic device can prioritize the selection of several parameter configurations in the potential feature space that are highly relevant to the user's parameter tuning intention and highly compatible with the original strategy's operating conditions. On this basis, it can locally expand through one or more of the following methods: parameter interpolation, parameter fine-tuning, or parameter generation, thereby forming multiple sets of candidate parameter combinations.

[0091] In this way, the candidate parameter combinations obtained by the search can reflect the user's current natural language parameter tuning intention while retaining the original parameter scheme's constraints in terms of safety constraints and station operation characteristics. Furthermore, it can control the candidate size while ensuring the relevance of the candidate set. Thus, under the premise of meeting the requirements of safety constraints and station characteristics, it provides a structurally reasonable and highly feasible parameter input basis for subsequent rapid simulation evaluation and recommendation ranking.

[0092] Some technical solutions, when generating candidate parameter combinations, only interpolate or enumerate around the target parameter configuration itself, without fully considering the compatibility between the parameter configuration and the original operating strategy and operating context of the energy storage system. This can easily lead to parameter schemes that are theoretically feasible but difficult to implement under actual operating conditions.

[0093] Therefore, in some embodiments, multiple sets of candidate parameter combinations are generated based on multiple target parameter configurations, including: evaluating the effectiveness of each target parameter configuration based on the compatibility between each target parameter configuration and the runtime context information corresponding to the original parameter scheme; and expanding the parameters of the target parameter configurations that pass the effectiveness evaluation to generate multiple sets of candidate parameter combinations.

[0094] First, the electronic device evaluates the compatibility between each target parameter configuration and the corresponding operating context information of the original parameter scheme. The compatibility level reflects the feasibility of the target parameter configuration under the current operating conditions.

[0095] During this process, the electronic device performs correlation analysis between each target parameter configuration and the corresponding operating context information of the original parameter scheme to determine whether the target parameter configuration meets the basic constraints related to the current operating environment. Target parameter configurations that fail the compatibility assessment will not be used for subsequent parameter expansion.

[0096] For the target parameter configuration that passes the effectiveness evaluation, the electronic device then performs a parameter expansion operation based on the target parameter configuration to generate multiple sets of candidate parameter combinations. Parameter expansion may include interpolation, fine-tuning, or combination adjustment of some parameters in the target parameter configuration, thereby forming a set of parameter schemes with certain differences in parameter values.

[0097] The candidate parameter combinations generated in the above manner inherit the user's parameter tuning intention reflected in the target parameter configuration, while maintaining compatibility with the original parameter scheme and its runtime context, and can be used as input for subsequent simulation evaluation.

[0098] In the above embodiments, by introducing a runtime context-based validity assessment before parameter expansion, the compatibility between the target parameter configuration and the current runtime environment is screened. This reduces the number of parameter schemes that do not match existing runtime conditions from entering the candidate set, improving the feasibility of candidate parameter combinations. Simultaneously, expanding parameters only for target parameter configurations that pass the validity assessment helps control the candidate size while ensuring candidate parameter diversity, thereby reducing the computational burden in subsequent simulation and evaluation stages and improving the overall efficiency of the parameter tuning process. Furthermore, by generating candidate parameter combinations based on the target parameter configuration, the candidate parameter schemes inherit the tuning direction reflected in the target parameter configuration while maintaining consistency with the runtime conditions of the original parameter schemes, providing a more stable and reliable input basis for subsequent scheme selection and recommendation based on simulation results.

[0099] In the evaluation of parameter schemes for energy storage systems, it is usually necessary to conduct operational simulations on multiple sets of candidate parameter combinations to analyze their impact on system power changes, energy state, and related operational indicators. Related technologies commonly employ high-precision simulation models or digital twin systems based on physical mechanisms to evaluate parameter schemes. While these methods can accurately reflect system operating behavior, the calculation process is complex and time-consuming, making it difficult to meet the need for rapid response in scenarios with a large number of candidate parameter combinations or requiring frequent parameter tuning.

[0100] Furthermore, directly using simplified models for prediction can easily lead to unstable or significantly biased prediction results due to a lack of characterization of complex operational constraints and historical operational characteristics. Especially when parameter schemes change, ensuring both prediction efficiency and predictive rationality becomes a pressing issue in related technologies. Therefore, it is necessary to introduce a simulation method that combines a basic system operation model with prior information for auxiliary prediction, in order to support rapid evaluation of multiple candidate parameter combinations.

[0101] Therefore, in some embodiments, based on a simulation substitution model, the operation process of the energy storage system under multiple combinations of candidate parameters is predicted to obtain multiple prediction results, including: for any combination of candidate parameters, based on the candidate parameter combination and the associated operating context information, calling a language model to generate prior information to characterize the distribution characteristics of the simulation output; based on the prior information, calling a simulation substitution model to predict the operation process of the energy storage system under the combination of candidate parameters to obtain the corresponding prediction results.

[0102] For any given set of candidate parameter combinations, the electronic device takes the candidate parameter combinations and their associated operating context information as input, and calls the language model to generate prior information to characterize the distribution characteristics of the simulation output. The prior information is used to reflect the prior characteristics of the energy storage system's operating results in terms of time variation trends, value ranges, or constraint satisfaction under similar operating conditions and parameter configurations.

[0103] Subsequently, when the electronic device invokes the simulation substitution model to predict the operation of the energy storage system, it uses prior information to guide or constrain the prediction process of the simulation substitution model. Specifically, the prior information is used to limit the reasonable distribution range of the output results of the simulation substitution model, or to verify the consistency of the changes in operating state generated during the prediction process, thereby reducing the possibility that the prediction results deviate from existing operating rules or system constraints.

[0104] Based on this, the simulation substitution model predicts the operation of the energy storage system under candidate parameter combinations, operating context information, and prior information, obtaining prediction results corresponding to those combinations. By executing the above prediction process for different candidate parameter combinations, the system obtains multiple sets of prediction results for subsequent scheme evaluation and selection.

[0105] Among them, the simulation substitution model is used to make approximate predictions of changes in the operating state of the energy storage system without performing a full high-precision simulation.

[0106] In some embodiments, to support the execution of the aforementioned simulation prediction process, the electronic device can construct a basic model characterizing the operational behavior of the energy storage system in the early stages. Specifically, the electronic device deploys sensors within the energy storage system to collect time-series data related to operation, including power, voltage, current, state of charge, temperature, electricity price information, and load forecast results. Furthermore, based on the time-series data and the physical model and equipment parameters of the energy storage device, the electronic device constructs a high-precision digital twin model, which serves as a benchmark reference for simulation evaluation. Simultaneously, the system parses and processes textual information related to the operation of the energy storage system, such as scheduling strategies and lifetime constraints, to form calculable constraints, thereby providing a constraint basis for the prediction process of the simulation alternative model.

[0107] In the above embodiments, by introducing a simulation substitution model in the application stage and combining it with prior information generated by the language model, the operation process of multiple sets of candidate parameter combinations can be quickly predicted. This provides usable prediction results for parameter scheme evaluation while ensuring prediction efficiency. At the same time, by forming a digital twin model and computable operating constraints in the system construction stage, a basic support is provided for the prediction process of the simulation substitution model, enabling the prediction results to reflect the actual operating characteristics of the energy storage system to a certain extent, thus balancing prediction speed and prediction rationality. In addition, by reducing the number of direct calls to the high-precision simulation model, the overall computational cost of parameter scheme evaluation can be effectively reduced, providing an efficient input basis for subsequent multi-dimensional evaluation and recommendation ranking.

[0108] Related technologies often lack a systematic evaluation mechanism for the reliability of prediction results, making it difficult to identify low-reliability predictions or high-risk solutions in a timely manner. This can easily lead to the introduction of unstable or unfeasible parameter schemes into subsequent evaluation and recommendation processes. On the one hand, simulation-based alternative models may lead to prediction biases due to model uncertainties when faced with insufficiently covered parameter combinations or operating scenarios. On the other hand, even if the prediction results are numerically feasible, they may approach the system constraint boundaries in terms of power, state of charge, or operating sequence, thus posing a high risk in actual operation.

[0109] Therefore, it is necessary to introduce a processing mechanism based on multi-dimensional confidence assessment after the prediction results are generated, and to combine a high-precision digital twin model to calibrate the prediction results when necessary, so as to improve the reliability of the overall recommendation process.

[0110] Therefore, in some embodiments, after obtaining multiple sets of prediction results, the method further includes: estimating the confidence level corresponding to the prediction results from multiple dimensions; for any set of prediction results, if the corresponding confidence level is lower than a preset confidence threshold, and / or the margin between at least one key operating indicator and the corresponding system constraint boundary is less than a preset margin threshold, calling a digital twin model to locally calibrate the corresponding prediction results.

[0111] The electronic device analyzes each prediction result from multiple preset dimensions to obtain the confidence level of each prediction result. The confidence level estimation includes at least one or more of the following: model uncertainty assessment based on the stability or consistency of the prediction results, and constraint consistency assessment based on the degree of conformity between the prediction results and system constraints.

[0112] Model uncertainty assessment is used to reflect the predictive stability or consistency of a simulation alternative model under the current candidate parameter combination and corresponding scenario characteristics. For example, electronic devices estimate the uncertainty of prediction results based on the similarity between candidate parameter combinations and their scenario characteristics and the historical applicable scenarios of the simulation alternative model, and / or based on the prediction differences of the simulation alternative model under multiple sampling or multi-model output conditions.

[0113] Constraint consistency assessment is used to reflect the degree of conformity between the prediction results and the system operating constraints. For example, electronic devices compare the predicted power change curves, state of charge change curves, or other key operating indicators with the pre-analyzed system constraints to determine whether the prediction results violate constraints for a long period of time or deviate significantly from typical operating modes.

[0114] After completing the above confidence level estimation, the electronic device, for any set of prediction results, if the corresponding confidence level is lower than the preset confidence level threshold, and / or if the margin between at least one key operating indicator and the corresponding system constraint boundary is less than the preset margin threshold, calls the digital twin model to perform local calibration of the prediction results.

[0115] Local calibration is used to perform high-precision simulation calculations on the parts of the operating range with high uncertainty or close to the constraint boundary in the prediction results, so as to obtain a more reliable description of the operating results.

[0116] For prediction results that pass the confidence assessment and do not require triggering local calibration, the electronic device retains its corresponding prediction curve and key operating indicators, and records its margin information in each system constraint dimension; for prediction results after local calibration, the system uses the calibrated results as input for subsequent evaluation and recommendation.

[0117] In the above embodiments, by introducing a multi-dimensional confidence estimation mechanism after the prediction results are generated, the reliability of the prediction results of the simulation alternative model can be quantitatively evaluated, thereby timely identifying parameter schemes that are unstable or high-risk. Furthermore, by triggering local calibration of the digital twin model when the confidence of the prediction results is low or the key operating indicators are close to the system constraint boundaries, high-precision simulation is only invoked in necessary scenarios, thereby ensuring the reliability of the prediction while avoiding the computational overhead caused by full-scale high-precision simulation.

[0118] For example, the electronic device combines candidate parameter combinations with contextual information such as the current site's ambient temperature, photovoltaic or load forecasts, electricity price curves, equipment rated parameters, and current operating mode to form the scenario features corresponding to the candidate solution, which are used as input to the simulation alternative model. Candidate combinations that clearly violate hard constraints such as equipment rated power, SOC limits, safety policies, or grid connection specifications are directly determined to be infeasible and eliminated.

[0119] In the offline phase, the simulation replacement model uses the digital twin model as a "teacher" and is pre-trained in multiple sites and under multiple parameter distributions through meta-learning, enabling it to quickly adapt to different operating conditions and cross-parameter adjustment scenarios. In the online phase, the electronic equipment uses a small amount of historical operating data from the current site or the results of digital twin simulation to quickly fine-tune the replacement model, making it closer to the characteristics of the current equipment and environment.

[0120] Meanwhile, electronic devices use a large language model to analyze textual knowledge such as operating constraints, scheduling rules, and typical operating modes, extracting prior rules and physical constraints such as power curves, SOC curves, and charging and discharging timing, and using these priors as constraints or rationality check rules in the training and inference stages of alternative models.

[0121] Based on this, a simulation substitution model is invoked for each set of candidate parameter combinations to quickly generate corresponding predicted outputs such as power curves, SOC curves, and revenue curves, achieving simulation evaluation that is orders of magnitude faster than high-precision digital twin models.

[0122] Furthermore, the electronic device evaluates the confidence of the prediction results of the alternative model from multiple dimensions: on the one hand, it estimates the uncertainty of the model itself based on the similarity between the candidate parameters and scene features and the historical training distribution, as well as the differences in multi-model / multi-sampling outputs within the alternative model; on the other hand, it compares the predicted power, SOC, demand, and other curves with the operating constraints and typical patterns extracted by the large language model. If it finds that the constraints are violated for a long time or that the pattern is obviously inconsistent with engineering experience, it lowers the confidence of the candidate solution or triggers the high-precision digital twin model to perform local calibration.

[0123] For candidate combinations that pass the above checks and have high confidence, the electronic device retains its prediction curve and key economic indicators (such as total revenue, peak-to-valley difference, battery life loss, etc.) and marks its margin information under various constraints as input for subsequent multi-objective evaluation and recommendation ranking; for candidate combinations with low confidence or that violate the constraint boundaries multiple times, they are marked as low-confidence or infeasible solutions and are downweighted or eliminated in subsequent steps.

[0124] When the prediction confidence of a candidate parameter combination is low, or when the predicted key indicators such as SOC, power, and demand are close to the preset safety constraints, equipment constraints, or grid connection constraints, the electronic equipment will mark the candidate parameter combination as a scheme that requires local calibration.

[0125] For the above scheme, the electronic device first inputs the candidate parameters and their corresponding information such as the site environment, electricity price and load forecast into the high-precision digital twin model while keeping the set of parameters unchanged. This yields more accurate power curves, SOC curves, revenue curves and constraint triggering conditions. The digital twin results are then used to replace the prediction output of the original simulation replacement model. At the same time, the sample is used as incremental data to locally fine-tune the simulation replacement model to improve its prediction accuracy in the nearby parameter region.

[0126] If the digital twin simulation results show that the above candidate parameter combinations still have obvious constraint violations or insufficient constraint margins under actual physical constraints, the electronic device, under the premise of limiting the adjustable range of a few key parameters (such as the upper limit of charging and discharging power, the upper and lower limits of SOC, peak power limit, etc.), performs a small-range search within its neighborhood for the parameter combination, calls the digital twin model to evaluate the searched perturbation parameter combinations, and prioritizes the parameter combination that satisfies all constraints and maintains the original benefits and lifetime performance as much as possible, as a local calibration version of the candidate parameter to participate in the subsequent effect evaluation and recommendation ranking.

[0127] In energy storage system parameter recommendation scenarios, for multiple candidate parameter combinations, related technologies typically employ simple multi-index weighting or rule-based ranking methods for evaluation. For example, they might calculate benefits, lifetime depreciation, or constraint risks separately and then directly sum them using weighted methods. However, these methods often rely solely on static comparisons of numerical results, failing to clearly distinguish the impact between "parameter adjustments" and "changes in evaluation indicators," and thus failing to reflect the source and extent of different parameters' influence on each indicator. In multi-objective trade-off scenarios, the coupling relationships between parameters are easily overlooked, leading to evaluation results that are insensitive to changes in user preferences or lacking a reasonable penalty mechanism when constraint conflicts exist.

[0128] Based on this, in some embodiments, multiple sets of prediction results are evaluated in multiple dimensions to determine the target parameter combination that satisfies user preferences and system constraints. This includes: constructing a causal influence model to characterize the influence of parameter adjustments on multiple evaluation indicators; quantifying the prediction results of each candidate parameter combination on multiple evaluation indicator dimensions based on the causal influence model to obtain the individual standardized scores of each candidate parameter combination on each evaluation indicator dimension; obtaining user preference information and determining the weight parameters corresponding to each evaluation indicator dimension based on the user preference information; weighting the individual standardized scores of each candidate parameter combination on each evaluation indicator dimension based on the weight parameters to obtain the comprehensive evaluation score corresponding to each candidate parameter combination; wherein, the weighting process further includes: imposing constraint penalty terms on selected or candidate parameter combinations that violate system constraints; and selecting the target parameter combination with the highest comprehensive evaluation score from among the candidate parameter combinations.

[0129] First, an electronic device constructs a causal relationship model to characterize the influence between parameter adjustments and multiple evaluation indicators. Evaluation indicators may include, but are not limited to, one or more of the following: revenue indicators, battery life degradation indicators, constraint risk indicators, and self-consumption rate indicators. This causal relationship model describes the direction and extent of the impact of changes in various parameter dimensions (such as upper and lower limits of charge / discharge power, SOC limits, and time window settings) on each evaluation indicator.

[0130] Furthermore, based on a causal relationship model, the electronic device quantifies the prediction results of each candidate parameter combination across multiple evaluation index dimensions. Specifically, the electronic device maps the prediction results under different evaluation indexes to comparable numerical ranges and performs standardization processing to obtain the individual standardized scores of each candidate parameter combination across each evaluation index dimension.

[0131] In some embodiments, for each candidate parameter combination, the electronic device constructs a causal influence matrix between parameters and evaluation indicators through single-parameter intervention. Specifically, the electronic device first runs a simulation substitution model (calibrated in conjunction with a high-precision digital twin model if necessary) under the current candidate parameter combination and the context of the corresponding site environment, electricity price and load forecast, to obtain the baseline power curve, SOC curve, and evaluation indicators such as revenue, lifetime, number of constraint defaults, and self-consumption rate of the candidate scheme.

[0132] Based on this, the system sequentially applies small-amplitude perturbations to each adjustable parameter, generating "slightly increased" and "slightly decreased" parameter versions while keeping other parameters constant. The system then calls the simulation replacement model to recalculate the corresponding evaluation indicators. By comparing the changes in indicators before and after the perturbation, the system obtains the direction and intensity of the parameter's influence on each evaluation indicator in the current scenario, and fills this information into a two-dimensional table of "parameter * indicator" to form a causal influence matrix. The rows of this matrix correspond to each adjustable parameter, and the columns correspond to each evaluation indicator. The matrix elements characterize the positive or negative impact and degree of influence of slightly adjusting a parameter on the corresponding evaluation indicator while keeping other conditions constant, thus reflecting the local causal relationship between the parameter and the indicator. Therefore, a causal influence model is constructed.

[0133] During this period, electronic devices can synchronously monitor the constraint triggering status and constraint margin changes in the evaluation results before and after the disturbance. For cases where the equipment's rated limits, safety constraints, or grid connection constraints are significantly triggered due to parameter adjustments, the significant adverse effects are recorded on the corresponding constraint-related indicators or marked as infeasible directions. This is so that in subsequent multi-objective trade-offs and strategy recommendations, the parameter adjustment direction of "increased benefits but significantly increased constraint risks" can be identified, thus avoiding the recommendation of parameter adjustment schemes that do not conform to engineering constraints.

[0134] Based on this, electronic devices acquire user preference information. This user preference information is used to characterize the degree of emphasis a user places on different evaluation indicator dimensions (e.g., valuing battery life more, or valuing benefits more), and accordingly determines the weight parameters corresponding to each evaluation indicator dimension.

[0135] Then, based on the weight parameters, the electronic device performs a weighted calculation of the standardized individual scores of each candidate parameter combination across each evaluation index dimension to obtain the comprehensive evaluation score for each candidate parameter combination. During the weighted calculation process, for candidate parameter combinations that violate system constraints, a constraint penalty term can be introduced into their comprehensive evaluation score to reduce their comprehensive evaluation score.

[0136] For example, electronic devices can normalize multiple evaluation indicators such as benefits, battery life depletion, constraint risks, and self-consumption rate to obtain a 0-1 score for each candidate parameter combination on each evaluation dimension. Then, a large language model can analyze the user's natural language preferences, mapping preference information such as "more value lifespan or more value benefits" into weight vectors for each evaluation dimension. A comprehensive scoring model can be constructed using a weighted summation method, and constraint violation penalties can be superimposed when necessary. This allows for a comprehensive ranking and selection of multiple candidate solutions based on user preferences.

[0137] Finally, based on the comprehensive evaluation scores corresponding to each candidate parameter combination, the electronic device selects the target parameter combination with the highest comprehensive evaluation score from multiple candidate parameter combinations as the recommended parameter scheme that satisfies user preferences and system constraints.

[0138] In the above embodiments, by introducing a causal influence model, the influence relationship between parameter adjustment and evaluation indicators is structurally characterized. This allows the evaluation process of multiple candidate parameter combinations to no longer rely solely on numerical comparisons of results, but to reflect the sources of influence of different parameters on each evaluation indicator. By standardizing different evaluation indicators and determining weight parameters in conjunction with user preferences, the comprehensive evaluation results can be adjusted according to different dimensions emphasized by users, improving the consistency between recommendation results and user needs. By introducing constraint penalty terms in the comprehensive evaluation process, candidate parameter combinations that violate system constraints can be effectively distinguished during the evaluation stage, which is beneficial for selecting parameter schemes that balance performance and feasibility during the recommendation stage.

[0139] Ordinary users often find it difficult to understand the differences between different parameter schemes and the reasons for their formation. On the one hand, simulation or prediction results are usually presented in the form of curves or indicators, lacking intuitive explanations of key operational characteristics; on the other hand, the source of the difference between the user's proposed scheme and the system's recommended scheme is unclear, making it difficult for users to judge whether the recommended scheme truly meets their own goals or constraints.

[0140] Furthermore, when recommended solutions involve multi-dimensional trade-offs such as benefits, lifespan, or constraint risks, the relevant technologies generally lack a unified explanatory mechanism, making it impossible to compare and explain different solutions by combining prediction results and evaluation conclusions, and also making it difficult to provide clear warnings when potential risks exist.

[0141] To clearly present the differences between the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme, and to improve the understandability and interactivity of the recommendation process, in some embodiments, the recommended parameter scheme is fed back to the user, and explanatory information is output to characterize the differences between the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme. This includes: based on the prediction results corresponding to the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme, calling a language model to generate explanatory information to characterize the operating characteristics of each scheme.

[0142] The electronic device first uses the prediction results of the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme under the corresponding operating scenario to call the language model to generate a comparative description to characterize the differences of each scheme in terms of benefit indicators, battery life related indicators, and system constraint risk indicators, and provides the corresponding reason analysis.

[0143] The explanatory information should include at least a description of the differences between the various schemes in terms of power change trends, energy status change trends, and key operating characteristics.

[0144] In some embodiments, the language model is invoked to generate explanatory information to characterize the operational characteristics of each scheme, including: invoking the language model based on multi-dimensional evaluation results to generate comparative descriptions to characterize the differences of each scheme in terms of revenue indicators, battery life-related indicators, and system constraint risk indicators, and providing corresponding causal analysis.

[0145] For example, the language model can generate explanatory information based on the differences in comprehensive evaluation scores, benefit indicators, and lifespan-related indicators of different parameter schemes. This information indicates the change in total benefit of the recommended parameter scheme compared to the original parameter scheme, and the correlation between this change and the allocation of charging and discharging power or the adjustment of operating time. At the same time, it explains why the user's proposed scheme has higher battery life loss indicators than the recommended parameter scheme due to changes in the SOC fluctuation range or the depth of charge and discharge. This provides a comparative explanation of the differences between the schemes in terms of benefit and lifespan, helping users understand the trade-off logic of the recommended parameter scheme in terms of benefit, lifespan, and constraint risks.

[0146] In some embodiments, the process of calling a language model to generate explanatory information to characterize the operational characteristics of each scheme further includes: generating prompt information to characterize the potential operational risks of the user-initiated scheme or recommended parameter scheme based on the prediction results and system constraints, and providing corresponding optimization suggestions.

[0147] For example, when prediction results show that the user's proposed solution may experience power close to the equipment's rated upper limit or SOC close to the safety boundary during certain periods, the language model can generate corresponding risk warnings to explain the potential constraint risks in long-term operation. It can also provide further suggestions, such as reducing the charging and discharging power during the corresponding periods or adjusting the upper and lower limits of SOC, to decrease the probability of triggering system constraints. By highlighting potential operational risks and providing optimization suggestions, the understandability and operability of parameter recommendations in actual operation are improved.

[0148] In some embodiments, calling a language model to generate explanatory information for characterizing the operational characteristics of each scheme further includes: when the parameter tuning information includes the user's high-level goal, generating explanatory information to explain the correspondence between the user's high-level goal and the recommended parameter scheme.

[0149] For example, when a user explicitly states in the parameter tuning information that the main goal is to reduce battery life loss or smooth peak power, the language model can generate explanatory information to explain how the corresponding parameter configuration in the recommended parameter scheme corresponds to the higher-level goal, and the impact of related parameter adjustments on life indicators or constraint risk indicators during multi-dimensional evaluation, thereby making the recommendation results easier for users to accept and verify.

[0150] In the above embodiments, by combining the prediction results to generate operating characteristic descriptions for different parameter schemes, users can intuitively understand the differences in power, energy status, and other aspects of each scheme.

[0151] In some embodiments, the method further includes: receiving updated parameter tuning information input by a user based on explanatory information; and triggering an update of the candidate parameter combination and / or recommended parameter scheme based on the updated parameter tuning information.

[0152] After providing the user with recommended parameter schemes and explanatory information, the system further receives updated parameter tuning information input by the user based on the explanatory information. Updated parameter tuning information can include modifications to existing parameter tuning information, such as resetting the charge / discharge power range, SOC upper and lower limits, operating period, or target emphasis.

[0153] Upon receiving updated parameter tuning information, the electronic device can trigger an update of the candidate parameter combinations and / or recommended parameter schemes based on the updated parameter tuning information. Specifically, the updated parameter tuning information can be input into the joint embedding model as new parameter tuning information, and combined with the original parameter scheme and runtime context information, candidate parameter combinations can be re-retrieved or generated in the latent feature space, or the original candidate parameter combinations can be filtered or adjusted to form updated candidate parameter combinations.

[0154] In some embodiments, when updating parameter tuning information involves only local parameter range or preference adjustment, the electronic device can re-evaluate and sort parameter combinations that meet the updated parameter tuning information based on existing candidate parameter combinations, thereby obtaining an updated recommended parameter scheme without having to re-execute the complete parameter tuning process.

[0155] By establishing a parameter tuning mechanism based on explanatory information, this application embodiment enables users to interactively adjust parameters based on their understanding of the differences in operation between different parameter schemes. This transforms the parameter tuning process from a one-time recommendation into an iterative dynamic optimization process, thereby improving the flexibility and efficiency of parameter scheme adjustments. Simultaneously, this mechanism helps reduce repetitive parameter tuning operations, enhancing the system's applicability in high-frequency parameter tuning and real-time decision-making scenarios.

[0156] In the application of energy storage system dispatch and recommendation, different users, different sites, and different operating stages often have stable but implicit preference characteristics, such as different ways of making trade-offs between benefits, lifespan, or constraint risks.

[0157] Therefore, in some embodiments, the electronic device may also record the target parameter scheme actually adopted by the user and the corresponding operational feedback information.

[0158] The operational feedback information includes at least one or more of the following: candidate parameter combinations and prediction results, target parameter scheme, actual operational status corresponding to the target parameter scheme, and user feedback information regarding the scheme.

[0159] In some embodiments, the electronic device updates the weight parameters corresponding to each evaluation metric dimension based on operational feedback information, so that the target parameter combination determined based on the weight parameters converges in a direction consistent with the user's long-term selection habits under the same or similar parameter tuning scenarios. Specifically, the electronic device can use the user's multiple adoption results under the same or similar parameter tuning scenarios as observation samples to adjust the weight parameters used for multi-dimensional evaluation, so that subsequent evaluations based on the weight parameters in the same or similar parameter tuning scenarios are more likely to produce a target parameter combination consistent with the user's previous adoption results.

[0160] In other embodiments, the electronic device also uses runtime feedback information to ensure that the candidate parameter combinations retrieved by the joint embedding model converge in a direction that aligns with the user's long-term selection habits, under conditions of similar user tuning intentions and similar runtime contexts. The joint embedding model is incrementally updated. Specifically, the electronic device can use the comparison between the target parameter scheme actually adopted by the user and the candidate parameter combinations that were not adopted to update the mapping in the joint embedding model that represents the relationship between tuning intentions and parameter configurations, thereby improving the consistency between the retrieved candidate parameter combinations and the user's actual adoption results under conditions of similar user tuning intentions and similar runtime contexts.

[0161] In some other embodiments, the electronic device further incrementally updates the simulation substitution model based on operational feedback information to improve the prediction accuracy of the simulation substitution model within the current operating scenario and high-frequency parameter region. Specifically, the electronic device can use the actual operating state data corresponding to the target parameter scheme as new samples to update the prediction capability of the simulation substitution model within the current operating scenario and high-frequency parameter region, thereby improving the matching degree between subsequent simulation prediction results and actual operating states.

[0162] For example, during each interaction involving parameter selection and recommendation, the electronic device continuously records the following information: the candidate parameter combinations shown to the user and their predicted scores, the solution actually adopted by the user, the actual benefits obtained from the subsequent operation of the solution, changes in lifespan, constraint defaults, and the user's subjective feedback (e.g., satisfaction level, whether to adjust again). This information is organized into a feedback sample of "scenario features - candidate solutions - system prediction - user selection - actual effect" to drive subsequent self-learning updates.

[0163] Electronic devices treat various weights (such as benefit weights, lifespan weights, and constraint risk weights) in a multi-objective comprehensive scoring model as user preference parameters to be estimated, and maintain a set of prior distributions for them. Whenever a user makes a choice among a set of candidate solutions, or provides positive / negative feedback on the recommendation result, the electronic device uses reinforcement learning or Bayesian updates on the distribution of these weights based on observations of which type of compromise the user prefers. This ensures that, in the same or similar scenarios, the comprehensive scoring model gradually converges towards a weight configuration that "better matches the user's long-term selection habits," achieving personalized adaptive adjustment of multi-objective weights. For different sites or different users, separate preference distributions can be maintained to achieve differentiated learning.

[0164] For the joint embedding model, the electronic device constructs preference ranking samples by comparing the "user's final adopted solution with the unadopted solution". It incrementally fine-tunes the matching relationship between text intent encoding and parameter-context encoding, so that the candidate parameters retrieved in similar intents and scenarios in the future are closer to the user's actual choices. For the simulation substitution model, the electronic device uses "candidate parameter combinations and their real curves and indicators obtained from real-time operation or digital twin simulation" as new samples, and periodically performs small-batch incremental training or parameter updates. This allows the substitution model to continuously improve prediction accuracy in the current site and the parameter area with the most recent use frequency, while remaining cautious in areas with high uncertainty.

[0165] Through the aforementioned self-learning process, the joint embedding model, simulation substitution model, and multi-objective weights can be dynamically adjusted, enabling the energy storage system to gradually align with actual scenarios and user preferences during long-term operation, thereby improving the effectiveness and credibility of subsequent adjustments and recommendations.

[0166] The following description, with reference to the accompanying drawings, illustrates the intelligent parameter scheme recommendation method for energy storage systems provided in this application through a specific example.

[0167] For example, such as Figure 4 As shown in the figure, this application provides an anomaly detection and interpretation method for multi-source time-series data of energy storage systems. This method constructs a complete processing chain from user parameter tuning intent acquisition, candidate parameter generation, rapid simulation prediction, multi-dimensional evaluation to recommendation feedback and interactive updates. It is used to achieve intelligent generation, rapid evaluation, and interpretable recommendation of parameter schemes under different operating scenarios, aiming to support users in making efficient and iterative parameter adjustment decisions based on natural language and operational feedback. The method includes: Step one involves data integration and digital twin construction, which includes collecting multi-source operational data such as power, current, voltage, SOC, temperature, electricity price, and load forecasting, performing unified feature processing and time alignment, and constructing a basic digital twin model that can be used for simulation.

[0168] Step two, parameter-intent joint embedding and candidate combination generation, includes, for example, parsing the user's natural language parameter tuning intent and jointly embedding it with the original parameter scheme and the runtime context, and generating candidate parameter combinations that satisfy multiple objectives and constraints through inverse mapping.

[0169] Step 3: Rapid prediction of meta-learning simulation alternative models, including: rapid prediction of power, SOC, revenue and lifetime under candidate parameter combinations using meta-learning-driven simulation alternative models, and calibration by combining a small number of high-precision simulations when necessary.

[0170] Step four involves causal-multi-objective fusion evaluation and recommendation, including multiple evaluation indicators such as comprehensive benefits, lifespan, and constraint risks. Candidate parameter combinations are quantitatively evaluated and ranked to select recommended parameter schemes that meet user preferences and system constraints.

[0171] Step 5, Natural Language Interpretation and Interactive Recommendation, includes, for example, generating explanations of the differences between the original parameter scheme, the user-initiated scheme, and the recommended scheme based on the prediction results and evaluation conclusions, and supporting users to make interactive adjustments and real-time reviews based on the interpretation results.

[0172] Step six involves self-learning and continuous optimization, including recording the actual solutions adopted by users and operational feedback information, and updating evaluation weights, parameter mapping relationships, and simulation models to adapt to changes in device status and improve the consistency and stability of subsequent recommendations.

[0173] This has led to the formation of a closed-loop intelligent dispatch and recommendation process for energy storage systems.

[0174] It should be noted that, Figure 4 Steps one to six shown are a functional summary description of the overall processing flow of the method in the embodiments of this application, used to illustrate the logical division of each processing stage; the specific implementation process corresponding to them has been described in detail above in conjunction with steps 210 to 260, and the two are consistent in technical content, only differing in the level of expression.

[0175] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0176] The intelligent parameter scheme recommendation method for energy storage systems provided in this application can be executed by an intelligent parameter scheme recommendation device for energy storage systems. This application uses the example of an intelligent parameter scheme recommendation device for energy storage systems executing the intelligent parameter scheme recommendation method for energy storage systems to illustrate the intelligent parameter scheme recommendation device for energy storage systems provided in this application.

[0177] like Figure 4As shown, the intelligent parameter scheme recommendation device for energy storage systems includes: an acquisition module 401, a mapping module 402, a prediction module 403, an evaluation module 404, and a recommendation module 405.

[0178] The acquisition module 401 is used to acquire the original parameter scheme of the energy storage system and the associated operating context information, and to receive the parameter adjustment information input by the user; the parameter adjustment information is used to characterize the user's proposed scheme.

[0179] The mapping module 402 is used to map the parameter tuning information, the original parameter scheme and the running context information to the same latent feature space through the joint embedding model, and to search for multiple sets of candidate parameter combinations in the latent feature space.

[0180] The prediction module 403 is used to predict the operation of the energy storage system under multiple combinations of candidate parameters based on the simulation substitution model, and obtain multiple prediction results.

[0181] Evaluation module 404 is used to evaluate multiple sets of prediction results in multiple dimensions to determine the combination of target parameters that meet user preferences and system constraints; the combination of target parameters is used to characterize the recommended parameter scheme.

[0182] The recommendation module 405 is used to provide users with recommended parameter schemes and output explanatory information to characterize the differences between the original parameter scheme, the user's proposed scheme, and the recommended parameter scheme.

[0183] According to the intelligent parameter scheme recommendation device for energy storage systems provided in the embodiments of this application, candidate parameter combinations are generated by mapping the natural language parameter tuning information of the user's proposed scheme with the original parameter scheme and operating context information of the energy storage system to the same potential feature space. The candidate parameter combinations are then rapidly predicted based on a simulation substitution model with cross-scenario adaptability. Furthermore, the recommended parameter scheme is determined by combining multi-dimensional evaluation and interpretable feedback mechanisms. Thus, under the premise of ensuring system constraints, the device effectively connects the user's high-level parameter tuning intentions with the operating parameters of the energy storage system, and supports rapid evaluation and understandable recommendation of parameter schemes.

[0184] In some embodiments, the mapping module is further configured to map the natural language information in the parameter tuning information to the latent feature space through a text intent coding network to obtain an intent vector representing the user's parameter tuning intent; and to map the original parameter scheme and associated runtime context information to the latent feature space through a parameter-context coding network to obtain a current context vector representing the current runtime state; wherein the text intent coding network and the parameter-context coding network are obtained through joint training.

[0185] In some embodiments, the mapping module is further configured to extract the running context information associated with each parameter configuration in the candidate parameter library, and map each parameter configuration and the associated running context information to the latent feature space to obtain a candidate context vector for representing each candidate running state; in the latent feature space, calculate the intent similarity between the intent vector and each candidate context vector, and the environment similarity between each candidate context vector and the current context vector; based on the intent similarity and environment similarity, determine multiple target parameter configurations in the candidate parameter library, and generate multiple sets of candidate parameter combinations based on the multiple target parameter configurations.

[0186] In some embodiments, the mapping module is further configured to evaluate the effectiveness of each target parameter configuration based on the compatibility between each target parameter configuration and the runtime context information corresponding to the original parameter scheme; and to expand the parameters of the target parameter configurations that pass the effectiveness evaluation to generate multiple sets of candidate parameter combinations.

[0187] In some embodiments, the prediction module is further configured to, for any set of candidate parameter combinations, based on the candidate parameter combinations and associated operating context information, call a language model to generate prior information to characterize the distribution characteristics of the simulation output; based on the prior information, call a simulation alternative model to predict the operation process of the energy storage system under the candidate parameter combinations, and obtain the corresponding prediction results.

[0188] In some embodiments, the prediction module is further configured to estimate the confidence level corresponding to the prediction results from multiple dimensions; for any set of prediction results, if the corresponding confidence level is lower than a preset confidence threshold, and / or the margin between at least one of the key operating indicators and the corresponding system constraint boundary is less than a preset margin threshold, the digital twin model is invoked to locally calibrate the corresponding prediction results.

[0189] In some embodiments, the evaluation module is further configured to construct a causal influence model to characterize the influence of parameter adjustment on multiple evaluation indicators; based on the causal influence model, quantify the prediction results of each candidate parameter combination on multiple evaluation indicator dimensions to obtain the individual standardized scores of each candidate parameter combination on each evaluation indicator dimension; obtain user preference information and determine the weight parameters corresponding to each evaluation indicator dimension based on the user preference information; perform weighted processing on the individual standardized scores of each candidate parameter combination on each evaluation indicator dimension based on the weight parameters to obtain the comprehensive evaluation score corresponding to each candidate parameter combination; wherein, the weighted processing further includes: imposing constraint penalty terms on selected candidate parameter combinations or those that violate system constraints; and selecting the target parameter combination with the highest comprehensive evaluation score among the candidate parameter combinations.

[0190] In some embodiments, the recommendation module is further configured to, based on the prediction results corresponding to the original parameter scheme, the user-conceptual scheme, and the recommended parameter scheme, invoke a language model to generate explanatory information characterizing the operational characteristics of each scheme; wherein, the explanatory information includes at least a description of the differences between the schemes in power change trends, energy state change trends, and key operational characteristics; invoking the language model to generate explanatory information characterizing the operational characteristics of each scheme includes: invoking the language model to generate a comparative description characterizing the differences between the schemes in terms of revenue indicators, battery life-related indicators, and system constraint risk indicators based on multi-dimensional evaluation results, and providing corresponding causal analysis; generating prompt information characterizing the potential operational risks of the user-conceptual scheme or recommended parameter scheme based on prediction results and system constraints, and providing corresponding optimization suggestions; and, if the parameter tuning information includes user high-level objectives, generating explanatory information explaining the correspondence between user high-level objectives and recommended parameter schemes.

[0191] In some embodiments, the above-described apparatus further includes an update module, configured to receive updated parameter tuning information input by the user based on the explanation information; and to trigger an update of the candidate parameter combination and / or recommended parameter scheme based on the updated parameter tuning information.

[0192] In some embodiments, the update module is further configured to record the target parameter scheme actually adopted by the user and the corresponding operational feedback information; wherein, the operational feedback information includes at least the candidate parameter combination and prediction results, the target parameter scheme, the actual operational status corresponding to the target parameter scheme, and the user feedback information; based on the operational feedback information, at least one of the weight parameters, joint embedding models, and simulation substitution models corresponding to each evaluation index dimension is updated.

[0193] The intelligent parameter scheme recommendation device for energy storage systems in the embodiments of this application can be an electronic device or a component in an electronic device, such as an integrated circuit or a chip.

[0194] The intelligent parameter scheme recommendation device for energy storage systems provided in this application embodiment can realize all the processes implemented in the above-described intelligent parameter scheme recommendation method embodiment for energy storage systems. To avoid repetition, it will not be described again here.

[0195] In some embodiments, such as Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, it implements the various processes of the above-described intelligent parameter scheme recommendation method embodiment for energy storage systems and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0196] This application provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described intelligent parameter scheme recommendation method embodiment for energy storage systems and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0197] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-storable media, such as computer read-only memory (ROM), random-access memory (RAM), magnetic disks, or optical disks.

[0198] The computer-readable storage medium may include: read-only memory (ROM), random-access memory (RAM), magnetic disk or optical disk, etc.

[0199] This application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for recommending intelligent parameter schemes for energy storage systems.

[0200] This application provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-described intelligent parameter scheme recommendation method embodiment for energy storage systems, and can achieve the same technical effect. To avoid repetition, it will not be described again.

[0201] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0202] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for recommending intelligent parameter schemes for energy storage systems, characterized in that, include: Obtain the original parameter scheme of the energy storage system and the associated operating context information, and receive the parameter adjustment information input by the user; The parameter tuning information is used to characterize the user's proposed solution; The parameter tuning information, the original parameter scheme, and the runtime context information are mapped to the same latent feature space through a joint embedding model, and multiple sets of candidate parameter combinations are obtained by searching in the latent feature space. Based on the simulation substitution model, the operation process of the energy storage system under the multiple sets of candidate parameter combinations is predicted, and multiple sets of prediction results are obtained. The multiple sets of prediction results are evaluated in a multi-dimensional manner to determine the combination of target parameters that meet user preferences and system constraints. The target parameter combination is used to characterize the recommended parameter scheme; The system provides feedback to the user on the recommended parameter scheme and outputs explanatory information to characterize the differences between the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme.

2. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1, characterized in that, The step of mapping the parameter tuning information, the original parameter scheme, and the runtime context information to the same latent feature space through a joint embedding model includes: The natural language information in the parameter tuning information is mapped to the latent feature space through a text intent coding network to obtain an intent vector that represents the user's parameter tuning intent; The original parameter scheme and associated runtime context information are mapped to the latent feature space through a parameter-context encoding network to obtain a current context vector that represents the current runtime state. The text intent encoding network and the parameter-context encoding network are obtained through joint training.

3. The intelligent parameter scheme recommendation method for energy storage systems according to claim 2, characterized in that, The search in the latent feature space yields multiple sets of candidate parameter combinations, including: For each parameter configuration in the candidate parameter library, extract the running context information associated with each parameter configuration, and map each parameter configuration and its associated running context information to the latent feature space to obtain a candidate context vector for characterizing each candidate running state. In the latent feature space, the intent similarity between the intent vector and each candidate context vector, and the environment similarity between each candidate context vector and the current context vector are calculated respectively. Based on the intent similarity and the environment similarity, multiple target parameter configurations are determined in the candidate parameter library, and multiple sets of candidate parameter combinations are generated based on the multiple target parameter configurations.

4. The intelligent parameter scheme recommendation method for energy storage systems according to claim 3, characterized in that, The process of generating multiple sets of candidate parameter combinations based on the configuration of the multiple target parameters includes: The effectiveness of each target parameter configuration is evaluated based on the compatibility between each target parameter configuration and the runtime context information corresponding to the original parameter scheme. The target parameter configuration that has passed the effectiveness evaluation is expanded to generate multiple sets of candidate parameter combinations.

5. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1, characterized in that, The simulation-based substitution model is used to predict the operation of the energy storage system under multiple combinations of candidate parameters, resulting in multiple prediction results, including: For any set of candidate parameter combinations, based on the candidate parameter combinations and associated runtime context information, a language model is invoked to generate prior information to characterize the distribution features of the simulation output. Based on the prior information, a simulation substitution model is invoked to predict the operation of the energy storage system under the candidate parameter combinations, and the corresponding prediction results are obtained.

6. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1 or 5, characterized in that, After obtaining multiple sets of prediction results, the method further includes: The confidence level corresponding to the prediction result is estimated from multiple dimensions; For any set of prediction results, if the corresponding confidence level is lower than the preset confidence threshold, and / or the margin between at least one of the key operating indicators and the corresponding system constraint boundary is less than the preset margin threshold, the digital twin model is invoked to perform local calibration on the corresponding prediction results.

7. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1, characterized in that, The multi-dimensional evaluation of the multiple sets of prediction results to determine the target parameter combination that satisfies user preferences and system constraints includes: Construct a causal relationship model to characterize the influence of parameter adjustment on multiple evaluation indicators; Based on the causal relationship model, the prediction results of each candidate parameter combination on multiple evaluation index dimensions are quantified to obtain the single standardized score of each candidate parameter combination on each evaluation index dimension. Obtain user preference information and determine the weight parameters corresponding to each evaluation indicator dimension based on the user preference information; Based on the weight parameters, the standardized scores of each candidate parameter combination on each evaluation index dimension are weighted to obtain the comprehensive evaluation score corresponding to each candidate parameter combination; wherein, the weighting process further includes: imposing constraint penalty terms on selected candidate parameter combinations or those that violate system constraints; Select the target parameter combination with the highest comprehensive evaluation score from among the candidate parameter combinations.

8. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1, characterized in that, The step of providing feedback to the user on the recommended parameter scheme and outputting explanatory information characterizing the differences between the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme includes: Based on the prediction results corresponding to the original parameter scheme, the user-initiated scheme, and the recommended parameter scheme, a language model is invoked to generate explanatory information to characterize the operational features of each scheme; wherein, the explanatory information includes at least a description of the differences between the schemes in terms of power change trends, power status change trends, and key operational features; The calling language model generates explanatory information to characterize the operational features of each scheme, including: Based on the multi-dimensional evaluation results, the language model is invoked to generate a comparative description that characterizes the differences of each scheme in terms of benefit indicators, battery life-related indicators, and system constraint risk indicators, and provides corresponding causal analysis. Based on the prediction results and system constraints, a prompt message is generated to characterize the potential operational risks of the user-initiated scheme or the recommended parameter scheme, and corresponding optimization suggestions are given. If the parameter tuning information includes the user's high-level goals, explanatory information is generated to illustrate the correspondence between the user's high-level goals and the recommended parameter scheme.

9. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1 or 8, characterized in that, The method further includes: Receive parameter tuning information input by the user based on the explanation information; Based on the updated parameter tuning information, the update of candidate parameter combinations and / or recommended parameter schemes is triggered.

10. The intelligent parameter scheme recommendation method for energy storage systems according to claim 1, characterized in that, The method further includes: Record the target parameter scheme actually adopted by the user and the corresponding operation feedback information; wherein, the operation feedback information includes at least the candidate parameter combination and prediction results, the target parameter scheme, the actual operation status corresponding to the target parameter scheme, and user feedback information; Based on the operational feedback information, at least one of the weight parameters, joint embedding models, and simulation substitution models corresponding to each evaluation index dimension is updated.

11. A smart parameter scheme recommendation device for energy storage systems, characterized in that, The device includes: The acquisition module is used to acquire the original parameter scheme of the energy storage system and the associated operating context information, and to receive the parameter adjustment information input by the user; the parameter adjustment information is used to characterize the user's proposed scheme. The mapping module is used to map the parameter tuning information, the original parameter scheme, and the runtime context information to the same latent feature space through a joint embedding model, and to search for multiple sets of candidate parameter combinations in the latent feature space. The prediction module is used to predict the operation of the energy storage system under the multiple sets of candidate parameter combinations based on the simulation substitution model, and obtain multiple sets of prediction results. The evaluation module is used to perform multi-dimensional evaluation on the multiple sets of prediction results to determine the target parameter combination that satisfies user preferences and system constraints; the target parameter combination is used to characterize the recommendation parameter scheme. The recommendation module is used to provide feedback to the user on the recommended parameter scheme and output explanatory information to characterize the differences between the original parameter scheme, the user-initiated scheme and the recommended parameter scheme.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent parameter scheme recommendation method for energy storage systems as described in any one of claims 1-10.

Images