Man-machine interaction method for park energy storage system

Abstract

A man-machine interaction method for a park energy storage system relates to the technical field of big data processing, and specifically comprises the following steps: respectively acquiring multi-dimensional real-time parameters of multiple types of energy storage devices in a park, and preprocessing to obtain state vectors; performing iterative processing on the state vector by adopting a graph neural network to obtain an equipment semantic vector; mapping the equipment semantic vector into digital twin semantic data; based on the digital twinning semantic data, constructing a virtual three-dimensional twinning scene consistent with the real environment of the park, defining a mapping relation between a state vector of each energy storage device and a visual attribute in the three-dimensional twinning scene, and performing dynamic rendering; generating a recommendation strategy by adopting a strategy planning algorithm according to the equipment semantic vector; executing a corresponding recommendation strategy according to the equipment nodes in the three-dimensional twinborn scene selected by the user, and generating an operation digital footprint; the method is suitable for the park energy storage system, and has the advantages of high visualization level and high interaction efficiency.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and more specifically to a human-computer interaction method for a park energy storage system. Background Technology

[0002] Currently, park energy storage systems typically include various energy storage devices (such as lithium batteries, supercapacitors, thermal energy storage, etc.), and their operation status and control strategies are complex, involving multiple aspects such as energy dispatch, system monitoring, and fault early warning.

[0003] However, traditional Energy Management Systems (EMS) are mainly based on two-dimensional interfaces for data display and interaction, which makes it difficult to intuitively reflect the spatial layout, operational status, and multi-device collaborative status of the energy storage system. In addition, EMS lacks real-time virtual mapping and intelligent linkage mechanisms for physical devices, resulting in low system management efficiency, delayed operation and maintenance response, and hindering operation and maintenance personnel from quickly grasping the overall situation and tracing faults.

[0004] Existing monitoring and management platforms for energy storage systems in industrial parks are mostly based on traditional EMS or SCADA architectures. Their core processes include: firstly, collecting operating parameters such as voltage, current, temperature, and SOC through BMS, PCS, and various sensors, and transmitting them to the monitoring platform via communication protocols such as Modbus and IEC 61850; then, processing the data by denoising, formatting, and archiving before storing it in a database; at the interface level, displaying the equipment's operating status and energy curves primarily through two-dimensional line charts, pie charts, and reports, while triggering alarm information when parameters exceed limits, prompting maintenance personnel with lists or dialog boxes; finally, issuing commands manually to start / stop equipment or switch operating strategies. However, such systems only operate at a two-dimensional visualization level, suffering from a single visualization dimension and insufficient spatial representation capabilities. They cannot intuitively reflect the actual layout and operating status of energy storage devices in the park, resulting in a lack of spatial awareness and global understanding for maintenance personnel in fault diagnosis and scheduling management. In addition, existing systems have limited data access and fusion capabilities for different types of energy storage devices, resulting in serious data silos and making it difficult to achieve unified management and intelligent collaboration of multi-source heterogeneous devices. Meanwhile, most systems lack digital twin capabilities, are unable to establish high-fidelity virtual models that synchronize the virtual and real worlds, and struggle to support state evolution prediction and interactive simulation in complex environments, severely restricting the system's intelligent operation and maintenance level and sustainable management capabilities.

[0005] While it is possible to achieve full-dimensional visibility, perception, and control of energy storage systems by constructing virtual mappings of real energy storage systems and synchronizing various operational data in real time, combined with 3D visualization rendering and human-computer interaction mechanisms, existing visualization platforms are mostly general-purpose and have not been customized and optimized for the complex equipment structures and interaction needs in park energy storage scenarios. As a result, there are problems such as insufficient data fusion, limited 3D scene expressiveness, and single interaction methods.

[0006] Therefore, we propose an interactive method that can improve the visualization level, interaction efficiency, and intelligent operation and management of energy storage systems. Summary of the Invention

[0007] The purpose of this invention is to provide a human-computer interaction method for park energy storage systems, which is applicable to park energy storage systems and has the advantages of high visualization level and strong interaction efficiency.

[0008] This invention is achieved through the following technical solution: A human-computer interaction method for an energy storage system in a park, specifically including: Multidimensional real-time parameters of various types of energy storage devices in the park are collected and preprocessed to obtain state vectors; A graph neural network is used to iteratively process the state vector to obtain the device semantic vector; Map device semantic vectors to digital twin semantic data; Based on digital twin semantic data, a virtual three-dimensional twin scene consistent with the real environment of the park is constructed, and the mapping relationship between the state vector of each energy storage device and the visualization attributes in the three-dimensional twin scene is defined for dynamic rendering. Based on the device semantic vector, a strategy planning algorithm is used to generate a recommendation strategy; Based on the device nodes in the 3D twin scene selected by the user, the corresponding recommendation strategy is executed, and a running digital footprint is generated.

[0009] Furthermore, the multidimensional real-time parameters include device type, geographical location, state of charge, temperature, voltage, current, and power.

[0010] Furthermore, the graph neural network is used to process the state vector. Processing is performed to obtain the device semantic vector. The calculation formula is as follows:

[0011] in, For the first The node The device semantic vector of the layer, For a trainable weight matrix, It is a non-linear activation function. For neighboring nodes For nodes Attention weights For nodes The set of neighbors.

[0012] Furthermore, the mapping of device semantic vectors to digital twin semantic data specifically includes: Device semantic vector As input to the variational autoencoder, the state vector of the device in the topological context is obtained under normal system operating conditions. Anomaly detection is performed on the state vector of the device in the topological context, and the resulting device anomaly information is the digital twin semantic data.

[0013] Furthermore, the step of using the device semantic vector as input to the variational autoencoder to obtain the device's state vector in the topological context under normal system operating conditions specifically includes: Variational autoencoders convert device semantic vectors The mapping is to the latent distribution, and the calculation formula is:

[0014] In the formula, As latent variables, and These are the mean prediction function and the variance prediction function of the latent variables, respectively. It is a normal distribution function; The device's state vector in the topological context is obtained through the decoder. The calculation formula is:

[0015] In the formula, It is a non-linear mapping function.

[0016] Furthermore, the anomaly determination of the device's state vector within the topological context yields device anomaly data, which is the digital twin semantic data, specifically including: Calculate the reconstruction error of the state distribution The calculation formula is:

[0017] Set error threshold If the reconstruction error exceeds the error threshold, it is determined that there is an abnormality in the operation of the equipment. If the error threshold is not exceeded, the equipment is considered to be operating normally.

[0018] Furthermore, the step of generating a recommendation strategy based on the device semantic vector using a strategy planning algorithm specifically includes: Based on device semantic vector Search for similar device statuses in the historical device status database. ; The similarity distribution of device semantic vectors is calculated using cosine similarity metric. ; Calculate the normalized weights of the similarity distribution ; Select the one with the highest similarity Historical device status And extract its corresponding historical strategies. ; Based on normalized weights and historical strategy We perform weighted fusion to obtain the recommendation strategy.

[0019] Furthermore, the similarity distribution of the device semantic vectors is calculated using cosine similarity metric. The calculation formula is: .

[0020] Furthermore, the calculation of similarity distribution Normalized weights The calculation formula is: .

[0021] Furthermore, the normalized weights and historical strategy The recommendation strategy is obtained by weighted fusion, and the calculation formula is as follows: .

[0022] The technical solution of the present invention has at least the following advantages and beneficial effects: This invention discloses a human-computer interaction method for park energy storage systems. By constructing a virtual three-dimensional twin scene and combining it with dynamic rendering of the three-dimensional twin scene, abstract energy storage equipment data is transformed into a spatial and intuitive three-dimensional visual presentation. This solves the shortcomings of traditional two-dimensional interfaces in spatial expression, enabling operation and maintenance personnel to quickly grasp the equipment layout, operating status and energy flow path, significantly improving the system's perceptibility and global cognitive efficiency.

[0023] In addition, recommendation strategies are automatically generated based on high-level device semantic vectors, and users can directly select device nodes in the 3D scene to execute them. This realizes the transformation from passive monitoring to active intervention. Furthermore, it can provide a running digital footprint during user interaction, allowing users to obtain sufficient information and effectively improving the intelligence level and interaction efficiency of the system.

[0024] Furthermore, by introducing a variational autoencoder to determine anomalies in the device semantic vector, the system can accurately learn the state distribution under normal operating conditions and provide early warnings for subtle and potential operational anomalies, thereby preventing problems before they occur and enhancing the safety assurance capability of the energy storage system. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the human-computer interaction method for the park energy storage system of the present invention; Figure 2 This is a schematic diagram of the three-dimensional digital twin system structure used in the park energy storage system of the present invention; Figure 3 This is a schematic diagram of the electronic device in this invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0027] Example 1 like Figure 1 As shown, a human-computer interaction method for an energy storage system in a park specifically includes: Multidimensional real-time parameters of various types of energy storage devices in the park are collected and preprocessed to obtain state vectors; In addition, the multidimensional real-time parameters include device type, geographical location, state of charge, temperature, voltage, current, and power; The equipment type can be obtained from the system configuration database, equipment nameplate information, or asset management system. This parameter is used for subsequent classification management and differentiated processing. Since different types of energy storage equipment (such as lithium batteries, supercapacitors, and flywheel energy storage) have different characteristics, working modes, and monitoring priorities, this method can subsequently load the corresponding three-dimensional model according to the type, set different normal operation parameter thresholds, and apply different analysis models and control strategies. Geographic location can be obtained from the Global Positioning System, Geographic Information System, or equipment layout diagrams in the park construction drawings. This parameter is used to achieve spatial positioning and topology construction, which is the geographical basis for building a three-dimensional digital twin scene consistent with the real park environment. At the same time, location information is a key input for building the equipment connection topology in the graph neural network, used to determine the electrical and physical adjacency relationships between devices. The State of Charge (SOC) can be provided in real-time by the energy storage device's own Battery Management System (BMS). This parameter is a core indicator of the operating status, directly determining the device's charge / discharge capability, and is used to achieve: 1) Visualization, which refers to mapping the fill height or color of the battery bar in the virtual twin 3D scene; 2) Alarms: When the SOC is too low or too high, an alert will be triggered in the virtual twin 3D scene; 3) Health assessment: The trend of SOC changes can help assess the health status of the energy storage device's own battery; Temperature can be obtained by temperature sensors installed inside or in critical parts of the energy storage device. This parameter is a core safety and status monitoring indicator, and temperature anomalies are usually an early signal of equipment failure. Therefore, by analyzing the data changes of this parameter, it can be mapped into a heat map chromatogram in a virtual twin 3D scene to intuitively display the heat distribution of the device; and a high-level alarm can be triggered immediately when the temperature exceeds the limit to prevent thermal runaway. Voltage / current can be obtained from the sensors of the device's own battery management system (BMS) and power conversion system (PCS), while power can be directly provided by the power conversion system (PCS) or calculated in real time through voltage and current values. Power parameters can be used to represent the flow speed and density of "energy flow" particle effects in the 3D scene of the virtual twin, intuitively showing the real-time flow direction and magnitude of energy. Furthermore, the corresponding hardware processor carrying this method supports multiple industrial communication protocols (such as Modbus, IEC 61850, MQTT) when receiving output signals from other sensors or detection devices to complete data acquisition, and is compatible with edge acquisition devices. In addition, when preprocessing multidimensional real-time parameters, time alignment is first performed to ensure that data from different sampling frequencies have a unified timestamp; secondly, noise reduction is performed, using moving average or Kalman filtering to suppress high-frequency noise; then normalization and unit conversion are performed. Taking voltage normalization as an example, the calculation formula is as follows:

[0028] In the formula, This is the original voltage measurement value. and These are the lower and upper limits of the normal operating voltage range, respectively. This is the normalized value. Ultimately, this module transforms multi-source heterogeneous data into a standardized input stream with a unified structure and consistent timing, providing high-quality input for subsequent state modeling and visualization.

[0029] A graph neural network is used to iteratively process the state vector to obtain the device semantic vector. The calculation formula is as follows:

[0030] in, For the first The node The device semantic vector of the layer, where the first layer is the device semantic vector. The device semantic vector of each node at layer 0 is... It is a state vector. , For a trainable weight matrix, It is a non-linear activation function. For neighboring nodes For nodes Attention weights For nodes The set of neighbors is obtained from the graph structure; When constructing a graph neural network, the energy storage system in the park is first abstracted into a graph structure. , where the set of nodes Represents energy storage devices, edge collection This represents electrical connections; by formally describing the energy flow and connection topology between devices using a graph structure, information propagation and fusion of different device states can be achieved within the graph, thereby enabling global modeling and intelligent reasoning of the park's energy storage system. This provides a structured foundation for subsequent anomaly detection and scheduling optimization. Since the node state update process is performed on a set of nodes... Each node in This process is carried out sequentially, meaning that each node dynamically updates its own state based on the feature information of its neighboring nodes through a weighted aggregation method. Then, feature propagation and aggregation operations are performed on the graph structure to obtain a graph neural network. The graph neural network can realize the propagation of state information and feature fusion between devices, so that the representation of each energy storage device not only includes its own operating parameters (such as voltage, power, SOC, etc.), but also integrates the influence of the devices electrically connected to it, thereby obtaining a high-level device semantic vector with topology awareness, providing an intelligent modeling foundation for subsequent anomaly detection and 3D visualization.

[0031] The device semantic vector is mapped to digital twin semantic data, where the digital twin semantic data is a label or rating with clear business semantics corresponding to the device semantic vector.

[0032] Based on digital twin semantic data, a virtual three-dimensional twin scene consistent with the real environment of the park is constructed, and the mapping relationship between the state vector of each energy storage device and the visualization attributes in the three-dimensional twin scene is defined for dynamic rendering. This step dynamically displays the operating status and energy flow of energy storage devices. It utilizes the Cesium or Three.js engine to build the scene and combines it with park GIS data for geographic reconstruction. Furthermore, by defining the mapping relationship between the state vectors of each energy storage device and the visualization attributes in the 3D twin scene, dynamic rendering ensures that each visualization attribute of the 3D twin scene has corresponding data support. This allows users to intuitively perceive the dynamic operation of the devices in 3D space when interacting with the 3D twin scene through the display screen. The interactive operations also support multi-scale zooming, free viewpoint switching, and scene roaming. For example, temperature in the state vector of an energy storage device. The mapping between the visualization property and the thermochromatogram is as follows:

[0033] In the formula, For mapping functions, and These are the maximum and minimum temperature values, respectively. For color output in thermal chromatography, similarly, the state of charge is mapped to an animation of changes in charge tank height, and power... This is then mapped to the particle flow velocity of energy streamlines; In summary, the 3D twin scene directly displays a virtual scene on the screen that is consistent with the real environment of the park, showing the distribution of various energy storage devices. When the user selects one of the energy storage devices, the screen will further display the multi-dimensional real-time parameters of that energy storage device. Among them, temperature, state of charge and power parameters are represented by thermochromatography, animation of changes in the height of the charge tank and particle flow of energy flow lines, respectively.

[0034] Based on the device semantic vector, a strategy planning algorithm is used to generate a recommendation strategy; The device semantic vector is vector data processed by a graph neural network. It contains contextual information about the location and relationship of the energy storage device in the overall park energy storage system. Therefore, through a data-driven strategy planning algorithm based on historical experience, a recommended strategy that can meet the current scenario can be obtained.

[0035] Based on the device nodes in the 3D twin scene selected by the user, the corresponding recommendation strategy is executed, and a running digital footprint is generated.

[0036] The digital footprint includes reports such as operation logs, policy records, abnormal equipment lists, and 3D interactive heatmaps, and supports export in multiple formats, such as PDF, Excel, and JSON. It can also achieve data sharing and linkage with energy management systems (EMS), supervisory control systems, or operation and maintenance platforms through standardized API interfaces.

[0037] Furthermore, when outputting the digital footprint of operation, a user preference modeling method based on interaction behavior sequences can be introduced, by analyzing user interaction sequences. Clustering or pattern recognition can be used to analyze user habits and provide intelligent assisted recommendations. User preference modeling methods can include multivariate time series prediction mechanisms based on Transformers, which use historical state sequences as a basis for prediction. For input, Output the future based on the length of the historical time step. The prediction result of the step is calculated using the following formula:

[0038] in This represents a time-series modeling function based on an attention mechanism. Then, from the predicted future semantic vector, the corresponding operating parameters are decoded, allowing for early prediction of potential SOC drops, overheating, or power fluctuations in the equipment. This provides early warnings to maintenance personnel and, combined with the strategy recommendation mechanism of the interactive module, enables proactive scheduling optimization.

[0039] Example 2 As one embodiment, mapping the device semantic vector to digital twin semantic data specifically includes: Device semantic vector As input to the variational autoencoder, the state vector of the device in the topological context under normal system operating conditions is obtained: By introducing a variational autoencoder (VAE), deep learning and probabilistic assessment of device health status can be achieved, thereby discovering potential risks hidden in complex data relationships that traditional thresholding methods cannot capture. A variational autoencoder is a generative model consisting of an encoder and a decoder. Its core idea is to learn the probability distribution of input data in a low-dimensional latent space. The encoder consists of multiple fully connected layers, and its function is to process the input device semantic vector. Mapping to the latent space, unlike ordinary autoencoders, VAE encoders do not output a fixed code, but rather output parameters of a probability distribution. Specifically, variational autoencoders map the device semantic vector... Mapping to latent distribution The calculation formula is:

[0040] In the formula, As latent variables, and These are the mean prediction function and the variance prediction function of the latent variables, respectively. It is a normal distribution function; The decoder also consists of multiple fully connected layers, and its function is to extract latent variables. It reconstructs the original input data and attempts to... Mapping back to the device semantic vector space, the output is a reconstructed vector. Specifically, the decoder obtains the device's state vector in the topological context. That is, the reconstructed vector, calculated using the following formula:

[0041] In the formula, It is a nonlinear mapping function; However, since the training objective of VAE is not only to minimize the reconstruction error, but also to require the latent distribution of the latent variables... Approximately standard normal distribution Therefore, its loss function is constructed as follows: Anomaly detection is performed on the state vector of the device within the topological context. The resulting device anomaly information constitutes the digital twin semantic data, which specifically includes: Calculate the reconstruction error of the state distribution The calculation formula is:

[0042] Set error threshold If the reconstruction error exceeds the error threshold, it is determined that there is an abnormality in the operation of the equipment. If the error threshold is not exceeded, the equipment is considered to be operating normally.

[0043] Based on this judgment result—"normal operation" or "potential anomaly"—the final generated digital twin semantic data is a high-level status label with clear business meaning. Furthermore, the semantic data of "potential anomaly" will directly trigger a highlight alarm in the 3D twin scene and can be used as input for the interactive control module to automatically start the diagnosis and strategy recommendation process, forming a complete intelligent closed loop from perception to cognition to action.

[0044] Example 3 As one embodiment, the step of generating a recommendation strategy based on the device semantic vector using a strategy planning algorithm specifically includes: Based on device semantic vector Search for similar device statuses in the historical device status database. ; It should be noted that the historical equipment status database is a database built in advance based on the historical operating data of the park's energy storage system; The similarity distribution of device semantic vectors is calculated using cosine similarity metric. The calculation formula is: .

[0045] Cosine similarity focuses on the difference in direction between two vectors and is not sensitive to absolute values, making it very suitable for measuring state similarity in high-dimensional semantic spaces. Calculate the normalized weights of the similarity distribution The calculation formula is: .

[0046] This can further amplify the weight of highly similar cases, making recommendations more dependent on the most similar historical scenarios; Select the one with the highest similarity Historical device status ,in The value is a hyperparameter, typically set based on the size of the historical database and the desired diversity of recommendations, and its corresponding historical strategies are extracted. ; Based on normalized weights and historical strategy The recommendation strategy is obtained by weighted fusion, and the calculation formula is as follows:

[0047] Specifically, for each control command, such as "charging power of device A", the recommended value is the weighted average of that power in various historical strategies. This fusion method generates a smooth and robust strategy that integrates multiple successful experiences and avoids the randomness that may exist in a single historical case.

[0048] Example 4 As one example, such as Figure 2 As shown, a three-dimensional digital twin system for a park energy storage system can be used to realize the human-computer interaction method for the park energy storage system as described in Examples 1-3. This three-dimensional digital twin system includes: The acquisition layer collects multi-dimensional real-time parameters of various types of energy storage devices in the park and preprocesses them to obtain state vectors. The perception layer uses a graph neural network to iteratively process the state vector to obtain the device semantic vector; Map device semantic vectors to digital twin semantic data; The processing layer, based on digital twin semantic data, constructs a virtual 3D twin scene consistent with the real environment of the park, defines the mapping relationship between the state vectors of each energy storage device and the visual attributes in the 3D twin scene, and performs dynamic rendering; Based on the device semantic vector, a strategy planning algorithm is used to generate a recommendation strategy; The output layer executes the corresponding recommendation strategy based on the device nodes in the user-selected 3D twin scene and generates a running digital footprint.

[0049] The acquisition layer consists of various sensors, metering devices, data acquisition units, and communication units; the perception layer consists of edge servers or cloud servers; the processing layer corresponds to graphics workstations or rendering servers and enterprise-level servers; and the output layer consists of monitoring screens made up of multiple LEDs and high-performance desktop computers.

[0050] Example 5 As attached Figure 3 An electronic device shown includes: Processor, memory, communication interface; The memory is used to store the executable instructions of the processor; The processor is configured to execute the human-machine interaction method for the above-mentioned park energy storage system by executing the executable instructions.

[0051] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the human-computer interaction method for the aforementioned park energy storage system.

[0052] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A human-computer interaction method for an energy storage system in a park, characterized in that, Specifically, it includes: Multidimensional real-time parameters of various types of energy storage devices in the park are collected and preprocessed to obtain state vectors; A graph neural network is used to iteratively process the state vector to obtain the device semantic vector; Map device semantic vectors to digital twin semantic data; Based on digital twin semantic data, a virtual three-dimensional twin scene consistent with the real environment of the park is constructed, and the mapping relationship between the state vector of each energy storage device and the visualization attributes in the three-dimensional twin scene is defined for dynamic rendering. Based on the device semantic vector, a strategy planning algorithm is used to generate a recommendation strategy; Based on the device nodes in the 3D twin scene selected by the user, the corresponding recommendation strategy is executed, and a running digital footprint is generated.

2. The human-computer interaction method for a park energy storage system according to claim 1, characterized in that: The multidimensional real-time parameters include device type, geographical location, state of charge, temperature, voltage, current, and power.

3. The human-computer interaction method for a park energy storage system according to claim 1, characterized in that: The graph neural network is used to process the state vector. Processing is performed to obtain the device semantic vector. The calculation formula is as follows: in, For the first The node The device semantic vector of the layer, For a trainable weight matrix, It is a non-linear activation function. Neighboring nodes For nodes Attention weights For nodes The set of neighbors.

4. The human-computer interaction method for the park energy storage system according to claim 3, characterized in that: The mapping of device semantic vectors to digital twin semantic data specifically includes: Device semantic vector As input to the variational autoencoder, the state vector of the device in the topological context is obtained under normal system operating conditions. Anomaly detection is performed on the state vector of the device in the topological context, and the resulting device anomaly information is the digital twin semantic data.

5. The human-computer interaction method for a park energy storage system according to claim 4, characterized in that: The step of using the device semantic vector as input to a variational autoencoder to obtain the device's state vector in the topological context under normal system operating conditions specifically includes: Variational autoencoders convert device semantic vectors The mapping is to the latent distribution, and the calculation formula is: In the formula, As latent variables, and These are the mean prediction function and the variance prediction function of the latent variables, respectively. It is a normal distribution function; The device's state vector in the topological context is obtained through the decoder. The calculation formula is: In the formula, It is a non-linear mapping function.

6. The human-computer interaction method for a park energy storage system according to claim 5, characterized in that: The anomaly detection of the device's state vector within the topological context yields digital twin semantic data, specifically including: Calculate the reconstruction error of the state distribution The calculation formula is: Set error threshold If the reconstruction error exceeds the error threshold, it is determined that there is an abnormality in the operation of the equipment. If the error threshold is not exceeded, the equipment is considered to be operating normally.

7. The human-computer interaction method for a park energy storage system according to claim 3, characterized in that: The step of generating a recommendation strategy based on the device semantic vector using a strategy planning algorithm specifically includes: Based on device semantic vector Search for similar device statuses in the historical device status database. ; The similarity distribution of device semantic vectors is calculated using cosine similarity metric. ; Calculate the normalized weights of the similarity distribution ; Select the one with the highest similarity Historical device status And extract its corresponding historical strategies. ; Based on normalized weights and historical strategy We perform weighted fusion to obtain the recommendation strategy.

8. The human-computer interaction method for a park energy storage system according to claim 7, characterized in that: The similarity distribution of device semantic vectors is calculated using cosine similarity metric. The calculation formula is: 。 9. The human-computer interaction method for a park energy storage system according to claim 8, characterized in that: The calculation of similarity distribution Normalized weights The calculation formula is: 。 10. The human-computer interaction method for a park energy storage system according to claim 9, characterized in that: The normalized weights and historical strategy The recommendation strategy is obtained by weighted fusion, and the calculation formula is as follows: 。

Images