Energy storage on-site monitoring method and system of energy storage power station

Abstract

The invention relates to the technical field of energy storage systems, in particular to an energy storage on-site monitoring method and system for an energy storage power station, and the method is based on a pre-constructed device semantic meta-model of various devices of the energy storage power station, and a unified protocol analysis table and a semantic mapping table of the various devices of the energy storage power station. Comprising the steps of realizing multi-source heterogeneous monitoring data acquisition and unified formatting: acquiring data streams generated by equipment through an equipment communication interface aiming at each piece of equipment of an energy storage power station in real time; analyzing an original field in each data stream based on the protocol analysis table in real time; mapping each analyzed original field into a unified standard field in real time based on a semantic mapping table according to an equipment semantic meta-model; and performing space-time alignment resampling on the mapped standard field to obtain standard structured data of each device of the energy storage power station, and outputting the standard structured data. The method is used for solving the problems that information islands exist due to weak multi-source data interaction capability and a judgment mechanism is generally based on a static threshold and is difficult to deal with a complex operation trend.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage systems, in particular to an energy storage on-site monitoring method and system for energy storage power stations. BACKGROUND

[0002] With large-scale access of renewable energy, energy storage power stations have been widely used in grid peak shaving, peak load shifting, micro-grid regulation and other scenarios. In order to ensure the operation efficiency and safety of energy storage power stations, multiple monitoring devices are usually deployed inside the power station, such as battery management system (BMS), data acquisition module, environmental sensor, fire fighting system, etc.

[0003] Although the existing energy storage power station is equipped with a local station-level monitoring system for data collection and monitoring, it only bears the function of data aggregation and other functions, and the function is still weak. Currently, there are the following problems: 1) The data interaction capability of various monitoring devices in the system is weak, and there is an information island problem; 2) The judgment mechanism is generally based on static threshold, which is difficult to cope with complex operation trend; 3) The AI diagnosis is deployed in the cloud, the communication dependency is strong, the real-time performance and stability are insufficient, and it cannot respond on site. SUMMARY

[0004] In view of the above deficiencies, the present application provides an energy storage on-site monitoring method and system for energy storage power stations to solve at least one of the above problems.

[0005] In a first aspect, the present application provides an energy storage on-site monitoring method for energy storage power stations. The method is based on a pre-constructed device semantic meta-model of various devices of the energy storage power station, and a unified protocol analysis table and semantic mapping table of various devices of the energy storage power station. The device semantic meta-model includes a standard field dictionary; The method comprises: S1, multi-source heterogeneous monitoring data collection and unified formatting: Real-time data flow generated by each device of the energy storage power station is collected through the device communication interface; Real-time structured analysis of the original fields in each collected data flow based on the protocol analysis table; Real-time mapping of each original field parsed according to the standard field dictionary of the device semantic meta-model to a unified standard field based on the semantic mapping table, to obtain the standard field of each device of the energy storage power station; Using a master clock anchor point and a sliding time window strategy, the standard fields of each device of the energy storage power station are time and space aligned and resampled, to obtain the standard structured data of each device of the energy storage power station and output; S2, data fusion and feature extraction are performed on the standard structured data of each device output in S1 to obtain the structured feature vector of each device of the energy storage power station; S3, edge side intelligent diagnosis: input the obtained structured feature vector of each device of the energy storage power station into the pre-trained multi-task diagnosis model deployed on the local diagnosis node, and obtain the predicted current health status, current fault trend and current fault risk level of the energy storage power station through model operation.

[0006] Further, S2 comprises: For each device of the energy storage power station, the statistical features of continuous quantities and the behavior features of discrete state quantities in the standard structured data output in S1 are counted by using a unified sliding time window, and then the key derived quantities are derived by combining physical formulas and device logic rules, and the derived key derived quantities are collected as multi-dimensional aggregated features; For the standard structured data of each device of the energy storage power station in the sliding time window, multi-dimensional fusion of the standard structured data is performed through field association strategy according to the cooperative relationship between the energy storage power station devices, and then based on the multi-dimensional fusion, the cross-device coupling behavior between the devices of the energy storage power station is analyzed, and then the key joint features are extracted; The multi-dimensional aggregated features and the key joint features obtained under the sliding time window are uniformly encoded and fixed-length structured by the constructed feature configuration template, and the structured feature vector of each device of the energy storage power station is obtained. S3, edge side intelligent diagnosis: The obtained structured feature vector of each device of the energy storage power station is input into the pre-trained multi-task diagnosis model deployed on the local diagnosis node, and the predicted current health status, current fault trend and current fault risk level of the energy storage power station are obtained through model operation.

[0007] Further, the method further comprises: S4, according to the predicted current health status, current fault trend and current fault risk level, triggering the corresponding local response strategy.

[0008] Further, the method for obtaining the trained multi-task diagnosis model in S3 comprises: S31, constructing a training set; S32, constructing a network architecture of the multi-task diagnosis model; S33, training the network architecture by using the training set to obtain the trained multi-task diagnosis model; In S31, the method for constructing the training set comprises: S311, collecting full life cycle historical operation data of each device of the energy storage power station; S312, using the same method as S1 and S2 to process the collected full life cycle historical operation data of each device of the energy storage power station to obtain a plurality of historical structured feature vectors of the energy storage power station; S313, labeling each historical structured feature vector, the label includes a health state label of the energy storage power station, a failure trend label and a failure risk score label; the health state label includes "101", "102" and "103", representing normal, slight abnormality and serious abnormality in turn; the failure trend label includes "201", "202" and "203", representing failure rising, failure falling and trend stability in turn; the risk score label includes "301", "302", "303", "304", "305" and "306", corresponding to score intervals [0, 20), [20, 40), [40, 60), [60, 80), [80, 95), [95, 100] in turn, representing extremely low risk, low risk, medium risk, medium-high risk, high risk and extremely high risk in turn; The labeled historical structured feature vectors are collected and sorted in ascending order of data collection time to form a training set.

[0009] Further, the network architecture of the multi-task diagnosis model comprises: An input layer for inputting the structured feature vector to provide input for subsequent feature extraction; A shared hidden layer connected to the input layer for extracting a global representation of the input structured feature vector; A task branch layer connected to the shared hidden layer, specifically comprising a state classification branch, a trend prediction time series branch and a risk score branch; wherein the state classification branch is used to predict the health state category probability distribution of the energy storage power station; the trend prediction time series branch is used to predict the state change amount of the energy storage power station; and the risk score branch is used to predict the continuous distribution of the risk score of the energy storage power station; An output layer comprising a first softmax function layer, a second softmax function layer and a Linear layer; the first softmax function layer is connected to the output end of the state classification branch, used to obtain the corresponding health state classification of the energy storage power station based on the output of the state classification branch and output; the second softmax function layer is connected to the output end of the trend prediction time series branch, used to obtain the corresponding failure trend classification of the energy storage power station based on the output of the trend prediction time series branch and output; and the Linear layer is connected to the output end of the risk score branch, used to map the output of the risk score branch to the corresponding risk score classification and output; S33, training the network architecture using the training set to obtain a trained multi-task diagnosis model.

[0010] Further, S33 comprises: S331, initializing the network architecture parameters to obtain an initial model; S332, input the training set into the initial model for forward propagation iterative training, in each iteration training, the loss of each task branch is calculated, and then the total loss is calculated by dynamically fusing the loss of each branch; wherein the state classification branch adopts a cross entropy loss function, the trend prediction time series branch adopts an MSE loss function, and the risk score branch adopts an MSE loss function; S333, update the model parameters by back propagation until the total loss converges or the preset training number is reached, to obtain the trained multi-task diagnosis model.

[0011] Further, the device semantic meta-model further comprises a device semantic processing module; The device semantic processing module is configured to implement standard field derivation calculation, field addition, device registration and semantic conflict processing in the standard field dictionary.

[0012] Further, when the number of local diagnosis nodes in S3 is multiple, the trained multi-task diagnosis model is deployed on each local diagnosis node, and S3 comprises: The obtained structured feature vector of each device of the energy storage power station is input into the trained multi-task diagnosis model deployed on each local diagnosis node; The trained multi-task diagnosis model deployed on each local diagnosis node respectively operates on the input structured feature vector, obtains the predicted current health state, current fault trend and current fault risk level of the energy storage power station, and sends them in a structured form to the collaborative judgment module; The collaborative judgment module forms a candidate result set from the prediction results sent by each local diagnosis node, and determines the final prediction result through majority voting or weighted voting mechanism for the same type of abnormal event.

[0013] Further, the method further comprises: After the local response strategy is triggered to execute, the change of the preset target field of each device of the energy storage power station is monitored for a preset length of time, the improvement degree after the strategy response is calculated, and a feedback score is generated based on this; The control strength of the local response strategy is adjusted based on the feedback score to form a self-closing loop response optimization mechanism.

[0014] In a second aspect, the application provides an energy storage on-site monitoring system for an energy storage power station, which comprises a pre-constructed device semantic meta-model of each type of device of the energy storage power station, and a unified protocol analysis table and semantic mapping table of each type of device of the energy storage power station; the device semantic meta-model comprises a standard field dictionary; The system further comprises: A monitoring data acquisition module for collecting and uniformly formatting multi-source heterogeneous monitoring data of the energy storage power station; The monitoring data acquisition module is configured to perform the following steps: Real-time data streams generated by each device of the energy storage power station are collected through a device communication interface; Real-time structured analysis of original fields in each collected data stream based on a protocol analysis table; Real-time mapping of each parsed original field to a unified standard field according to a standard field dictionary of a device semantic meta-model based on a semantic mapping table, to obtain standard fields of each device of the energy storage power station; The standard fields of each device of the energy storage power station are spatio-temporally aligned and resampled using a master clock anchor point and a sliding time window strategy, to obtain and output standard structured data of each device of the energy storage power station; The system further comprises: A feature extraction module for data fusion and feature extraction on the standard structured data of each device output by the monitoring data collection module, to obtain structured feature vectors of each device of the energy storage power station; An edge-side intelligent diagnosis module for inputting the obtained structured feature vectors of each device of the energy storage power station into a pre-trained multi-task diagnosis model deployed on a local diagnosis node, and obtaining a predicted current health status, current fault trend and current fault risk level of the energy storage power station through model operation.

[0015] As can be seen from the above technical solutions, the present application has the following advantages: The present application constructs a device semantic meta-model, a protocol analysis table and a semantic mapping table, can adapt to device data of different manufacturers and different protocols, eliminates the problem of weak interaction ability between multi-source data and information island, and has stronger scalability and adaptability compared with the traditional mode relying on static mapping relationship, and is suitable for data integration requirements of complex heterogeneous systems.

[0016] The present application completes diagnosis model deployment at the edge node, and compared with the traditional scheme relying on cloud reasoning, the mechanism reduces the communication dependence on the cloud, enhances the real-time performance and stability, and facilitates on-site response. DETAILED DESCRIPTION

[0017] The present application will be described in detail below. Specific details such as specific system structures, techniques, etc. are presented to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details.

[0018] The key terms appearing in the present application are explained below.

[0019] BMS: battery management system.

[0020] PCS: energy storage converter.

[0021] SOC: System On Chip, system on chip or system on a chip.

[0022] PCCK: Pack Control Circuit Kit, battery pack control assembly.

[0023] It should be understood that the term "comprising" as used in the specification indicates the presence of the recited features, integers, steps, operations, elements, and / or components but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The terms "comprising", "including", "having" and their conjugates mean "including but not limited to", unless otherwise expressly stated.

[0024] The phrase "one embodiment" or "some embodiments" appearing in the specification is intended to denote, in one or more embodiments of the present application, the presence of a feature, structure, characteristic, or the like described in connection with that embodiment. Thus, when a phrase such as "in some embodiments", "in other embodiments", or the like appears in the specification, a claim following such a phrase is not necessarily dependent on the same embodiment as a previous claim, but is intended to denote "one or more but not all embodiments of the present application", unless otherwise expressly stated.

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0026] The energy storage in-situ monitoring method for an energy storage power station provided by the embodiment can be executed by an energy storage in-situ monitoring system for an energy storage power station. The energy storage in-situ monitoring method for an energy storage power station provided by the embodiment is executed by a computer device, and accordingly, the energy storage in-situ monitoring system for an energy storage power station is run in the computer device.

[0027] In one embodiment of the present application, the method is based on a pre-constructed device semantic meta-model of various types of devices in an energy storage power station, and a unified protocol analysis table and semantic mapping table of various types of devices in the energy storage power station.

[0028] Construction and registration of the device semantic meta-model: based on industry standards and related device protocol documents, the device semantic meta-model of various types of devices in the energy storage power station (such as BMS, PCS, liquid cooling, etc.) is abstractly constructed.

[0029] The device semantic meta-model includes a standard field dictionary.

[0030] The standard field dictionary defines the core parameters of various devices of the energy storage power station. Specifically, the standard field dictionary defines the core parameters of the device, including a standard field name (including a field name, a physical meaning, a standard unit, a data type, etc.), a field hierarchical structure (an electric core, a PCCK, a cluster), a field type (a state quantity, a control quantity, an environmental quantity), a source device identifier, etc., as a basis for subsequent device semantic mapping and data normalization.

[0031] Meanwhile, the field dictionary supports a scope constraint mechanism for specifying the validity of a field under a specific state or condition, and supports a dynamic semantic labeling mechanism, which can dynamically adjust the semantic interpretation of a field according to a device running state, a scene label or an environmental variable, so that the same field has differential semantic expression in different contexts.

[0032] For various devices (such as BMS, PCS, liquid cooling machine, fire fighting, dehumidifier, electric meter, etc.) in the energy storage power station, a unified "protocol analysis-semantic mapping table" is constructed to realize decoding and semantic binding of device data.

[0033] The "protocol analysis-semantic mapping table" is a protocol analysis table and a semantic mapping table.

[0034] The protocol analysis table is used for structured analysis of raw fields of data streams generated by various devices of the energy storage power station, and is specifically used for describing the physical position, data type, offset, unit conversion and other rules of each field in the raw message, so as to structure and analyze the raw data stream collected through the communication interface of various devices in the energy storage power station, such as Modbus TCP / RTU, CAN, etc., to extract field name, actual value, timestamp, source device ID and other raw fields.

[0035] The semantic mapping table is used for mapping the raw fields obtained by analysis into unified standard fields according to the standard field dictionary of the device semantic meta model, realizing field semantic alignment and standardized reconstruction, and specifically including unified processing of field name, unit, dimension, precision, data range, etc. This process supports retaining the original field name and context information for subsequent data fusion, traceability and consistency analysis.

[0036] The method of the embodiment includes the step S1 of multi-source heterogeneous monitoring data acquisition and unified formatting.

[0037] This step is used to realize data semantic unification and provide a high-quality, dimension-unified data basis for downstream fusion analysis.

[0038] Specifically, the step S1 includes: Step S11, real-time acquisition of data streams generated by devices of the energy storage power station through a device communication interface (such as Modbus TCP / RTU, CAN, etc.) for each device of the energy storage power station; Step S12, structurally parse the original fields in each data stream collected in real time based on a protocol parsing table; Step S13, map each parsed original field to a unified standard field according to a standard field dictionary of the device semantic meta model in real time based on a semantic mapping table, to obtain standard fields of each device of the energy storage power station. Step S14, perform spatio-temporal alignment resampling on the standard fields of each device of the energy storage power station by using a master clock anchor point and a sliding time window strategy, to obtain and output standard structured data of each device of the energy storage power station.

[0039] The step S1 outputs the reconstructed data that is semantically consistent, format-unified, and spatio-temporally aligned, in a JSON format, while supporting local synchronous caching.

[0040] In the implementation of the step S14, a BMS or another high-frequency device of the energy storage power station is taken as a time anchor point, a sliding time window is constructed, the parsed data of each device is projected into the target window according to the time stamp, and a full-station state snapshot at a unified sampling time is generated. For fields not in the time window, interpolation, previous value retention, default value filling, or other methods are used for completion according to the field characteristics.

[0041] In addition, in the implementation, a time integrity label can be added to each frame of data and a data quality score (such as whether there is an interpolated field or a missing field ratio) can be generated according to actual conditions, to ensure the consistency of the "unified time state" in downstream semantic calculation and state evaluation modules.

[0042] In the embodiment, the method further includes: Step S2, data fusion and feature extraction; Step S3, edge-side intelligent diagnosis.

[0043] The standard structured data of each device of the energy storage power station output in the step S1 is input to the step S2 for data fusion and feature extraction, to obtain a structured feature vector of each device of the energy storage power station.

[0044] In the method of the embodiment, the step S3 includes: The structured feature vector of each device of the energy storage power station obtained by the step S2 is input to a pre-trained multi-task diagnosis model deployed on a local diagnosis node, and after model operation, a predicted current health status, a current fault trend, and a current fault risk level of the energy storage power station are obtained.

[0045] The step S2 collects the multi-source data output in the step S1, and performs information extraction, fusion processing, and potential risk signal mining on the collected multi-source data, to generate a high-quality feature vector that can be used for model calculation.

[0046] Specifically, step S2 comprises: Step S21, for each device of the energy storage power station, a unified sliding time window is adopted to count the statistical characteristics of continuous quantities and the behavior characteristics of discrete state quantities in the standard structured data of each device output in step S1, and then the key derived quantities are derived in combination with physical formulas and device logic rules, and the derived key derived quantities are collected as multi-dimensional aggregated features; Step S22, for the standard structured data of each device of the energy storage power station in the sliding time window, the multi-dimensional fusion of the standard structured data is performed through a field association strategy according to the cooperative relationship between the devices of the energy storage power station, and then based on the multi-dimensional fusion, the cross-device coupling behavior between the devices of the energy storage power station is analyzed, and then the key joint features are extracted; Step S23, the multi-dimensional aggregated features and the key joint features obtained under the sliding time window are uniformly coded and fixed-length structured by the constructed feature configuration template, and the structured feature vectors of each device of the energy storage power station are obtained.

[0047] Understandably, step S21 adopts field aggregation and derived feature calculation to derive more valuable statistical quantities and behavior characteristics from a large number of basic fields, thereby obtaining a high-dimensional but sparse statistical feature vector. Specifically: on the basis of setting a sliding time window, the statistical characteristics of continuous quantities (including cell voltage, current, temperature, SOC, PCS power, and environmental temperature and humidity) in the standard structured data of each device output in step S1 are counted to characterize the stability and trend of device operation; at the same time, the behavior characteristics (including switching frequency, duration, and proportion) of discrete state quantities (including dehumidification, cooling, discharge, and fire protection) in the standard structured data of each device output in step S1 are identified to reflect the working state and response mode; then based on physical formulas (such as P=U×I and energy integration) and device logic rules, key derived quantities are derived to lay a semantic foundation for multi-source data fusion.

[0048] In this embodiment, the statistical characteristics of the counted continuous quantities include mean, maximum, slope, standard deviation, skewness, and range. The derived key derived quantities include power, energy, pressure difference, temperature difference, consistency index, and cooling response delay.

[0049] Understandably, step S22 cross-device semantic fusion and abnormal signal identification, excavates the cooperative relationship between devices in the same time window. Specifically, it includes: The "full station state snapshot" under the unified time reference constructed by the above steps is subjected to multi-dimensional fusion of key fields (including temperature, current, power, SOC, operating state, alarm flag) through a field association strategy, semantic association between devices in operating behavior, response relationship and abnormal performance is mined, cross-device coupling behavior (such as whether high temperature triggers the cooling system, whether rapid change of SOC is accompanied by PCS start-stop, whether alarms occur synchronously, etc.) is analyzed, and these cross-device semantic behaviors are modeled and matched based on rule driving, frequency statistics and graph structure modeling, from which abnormal linkage modes (such as temperature rise + cooling failure + power drop) are identified, and key joint features are extracted.

[0050] Optionally, the cross-device semantic behaviors are modeled and matched based on rule driving, frequency statistics and graph structure modeling, from which abnormal linkage modes are identified, and key joint features are extracted, including: Step 1: Rule engine matching.

[0051] First, the key fields (including cell voltage, temperature, SOC, PCS power, liquid cooling state, dehumidification state, fire alarm) in the panoramic state are input into the rule engine according to the time window. The rule engine has a pre-defined trigger-response rule library built-in, for example, the liquid cooling system responds at full speed when the cell temperature exceeds a threshold, and for example, the PCS load is reduced when the SOC rapidly decreases. Whenever the data in the time window meets the condition of a rule in the trigger-response rule library, the corresponding trigger event (such as the above-mentioned cell temperature exceeding the threshold, and the above-mentioned SOC rapidly decreasing) is recorded, along with the field, device ID, timestamp and trigger type, providing basic data for subsequent cross-device association analysis.

[0052] Step 2: Frequency modeling.

[0053] Based on the recorded trigger events, frequency statistics and quantitative modeling are performed on each type of trigger event. The frequency statistics include event trigger frequency, response delay or amplitude change range, which are used to judge the stability and repeatability of the linkage event. Through quantitative analysis of this step, high-probability and high-impact abnormal linkage modes can be selected based on threshold conditions, serving as candidate modes for subsequent coding.

[0054] Step 3: Graph structure construction and reasoning.

[0055] The trigger events and their response relationships are constructed into a directed graph, where the nodes of the directed graph represent device fields, the edges represent trigger or response relationships, and the weights of the edges are the above-mentioned frequency statistics. Through directed graph traversal and path matching algorithms, chain linkage abnormalities across devices and multiple fields are identified, such as "cell temperature rise → liquid cooling non-response → PCS power drop". This structured representation can reveal complex systematic abnormal paths.

[0056] Step 4: high-dimensional joint feature vector coding.

[0057] The rule engine trigger record, frequency statistics result and directed graph structure path feature are integrated, and are coded into a high-dimensional joint feature vector (i.e., a key joint feature) according to a preset field template. When coding into a high-dimensional joint feature vector according to the preset field template, continuous quantities are normalized, discrete quantities are mapped by coding, and missing fields are filled with default values and marked with a missing bitmap.

[0058] It can be understood that step S23 compresses multi-dimensional data into a fixed-structure, unified-format model input vector: a fixed-length, structured, model-directly-usable feature vector array.

[0059] In a specific implementation, step S23 can include: By constructing a feature configuration template, feature values are extracted in a preset field order, continuous values are normalized, discrete fields are mapped by coding, and missing fields are filled with default values and marked with a missing bitmap, so as to ensure that the vector length is fixed and consistent, and finally a structured fixed-length feature array is output.

[0060] The output of step S23 can support JSON, Protobuf and other format serialization.

[0061] Optionally, the method for obtaining the trained multi-task diagnosis model includes: Step S31, constructing a training set; Step S32, constructing a network architecture of the multi-task diagnosis model; Step S33, training the network architecture using the training set to obtain the trained multi-task diagnosis model.

[0062] In S31, the method for constructing the training set includes: Step S311, collecting full-life-cycle historical operation data of each device of the energy storage power station; Step S312, using the same method as steps S1 and S2 to process the collected full-life-cycle historical operation data of each device of the energy storage power station to obtain a plurality of historical structured feature vectors of the energy storage power station; Step S313, each historical structured feature vector is labeled, and the label includes a health state label, a failure trend label and a failure risk score label of the energy storage power station; the health state label includes "101", "102" and "103", representing normal, slight abnormality and serious abnormality in sequence; the failure trend label includes "201", "202" and "203", representing failure rising, failure falling and trend stability in sequence; the risk score label includes "301", "302", "303", "304", "305" and "306", corresponding to score intervals [0, 20), [20, 40), [40, 60), [60, 80), [80, 95) and [95, 100] in sequence, representing extremely low risk, low risk, medium risk, medium-high risk, high risk and extremely high risk in sequence. The labeled historical structured feature vectors are collected and sorted in ascending order of data collection time to form a training set.

[0063] In a specific implementation, the label set can be constructed based on expert experience or field records, and the label set records the health state label, the failure trend label and the failure risk score label of the energy storage power station. During labeling, the sample data is labeled using the labels recorded in the label set.

[0064] Exemplarily, the network architecture of the multi-task diagnosis model constructed above includes: An input layer is configured to input the structured feature vector, providing input for subsequent feature extraction; A shared hidden layer is connected to the input layer and configured to extract a global representation of the input structured feature vector; A task branch layer is connected to the shared hidden layer and specifically includes a state classification branch, a trend prediction time series branch and a risk score branch; the state classification branch is configured to predict a health state category probability distribution of the energy storage power station; the trend prediction time series branch is configured to predict a state change amount of the energy storage power station; and the risk score branch is configured to predict a continuous distribution of a risk score of the energy storage power station; An output layer includes a first softmax function layer, a second softmax function layer and a Linear layer; the first softmax function layer is connected to an output end of the state classification branch and configured to obtain and output a corresponding health state classification of the energy storage power station based on the output of the state classification branch; the second softmax function layer is connected to an output end of the trend prediction time series branch and configured to obtain and output a corresponding failure trend classification of the energy storage power station based on the output of the trend prediction time series branch; and the Linear layer is connected to an output end of the risk score branch and configured to map and output a corresponding risk score classification of the energy storage power station based on the output of the risk score branch.

[0065] Optionally, the network architecture is trained using the training set in step S33 to obtain a trained multi-task diagnosis model, including: S331, initialize the network architecture parameters to obtain an initial model; S332, input the training set into the initial model for forward propagation iterative training, in each iteration training, calculate the loss of each task branch, and then calculate the total loss by dynamically fusing the branch losses; S333, update the model parameters by back propagation until the total loss converges or a preset training number is reached, to obtain a trained multi-task diagnosis model.

[0066] Among them, the state classification branch adopts a cross-entropy loss function, the trend prediction time series branch adopts an MSE loss function, and the risk score branch adopts an MSE loss function.

[0067] In the present application, the trained multi-task diagnosis model is deployed on a local diagnosis node. When deploying, the model is first processed into a lightweight model file that can be deployed on an edge computing device, and then deployed on the local diagnosis node.

[0068] Optionally, when processing the model into a lightweight model file that can be deployed on an edge computing device, the model can be generated into a lightweight model file suitable for deployment on an edge computing device by compression means such as channel pruning, parameter quantization and knowledge distillation, while outputting a model input-output definition structure and a standardized configuration template.

[0069] Among them, the model input-output definition structure and the standardized configuration template provide standardized guidance for model deployment and use: Model input-output definition structure: including the input data format received by the model, the meaning and format of the model output result.

[0070] Standardized configuration template: standardize hardware parameters (such as computing power allocation, storage occupancy setting) and running environment configuration (such as dependent library version, inference engine parameter) when deploying the model, to ensure the standardization and reusability of the deployment process.

[0071] Optionally, the present application supports an abnormal sample backflow mechanism during model training.

[0072] The abnormal sample backflow mechanism includes: In each iteration training, calculate the model confidence and determine whether the model confidence is too low or there is an unknown state sample according to a preset threshold condition; If the model confidence is too low or there is an unknown state sample, the related features and their context are packaged into a micro sample set, which is returned to the cloud for model retraining or parameter fine-tuning.

[0073] As another embodiment of the present application, the method further includes: Step S4, triggering a corresponding local response strategy according to the predicted current health state, the current failure trend and the current failure risk level.

[0074] In a specific implementation, a prediction result-response rule mapping table can be preset according to actual conditions. In use, after the predicted current health state, the current failure trend and the current failure risk level are obtained, a corresponding local response strategy is triggered according to the mapping table.

[0075] The prediction result-response rule mapping table can be formulated by experts in the field according to experience and actual conditions.

[0076] The local response strategy includes a cooling system strengthening strategy, a PCS load shedding strategy, an isolation discharge strategy and an alarm prompt strategy.

[0077] As an exemplary embodiment of the application, the device semantic meta-model further includes a device semantic processing module. The device semantic processing module is used to realize standard field derivation calculation, field addition, device registration and semantic conflict processing in the standard field dictionary.

[0078] The field derivation calculation is for some parameters that cannot be directly measured, such as efficiency and electric quantity (for example, the dischargeable quantity is calculated by the nominal electric quantity x SOC). When a new device model or a new field is registered through the device semantic meta-model, the application automatically checks whether the field name, unit and derivation logic of the new device model or the new field have semantic overlap or conflict with the existing device model or the new field in the device semantic meta-model. If there is, a prompt or merging suggestion is provided to ensure the accuracy and consistency of the overall semantic system.

[0079] Optionally, when the number of local diagnosis nodes is multiple, the trained multi-task diagnosis model is deployed on each local diagnosis node.

[0080] Optionally, the application supports semantic drift detection for the multi-task diagnosis model deployed on each local diagnosis node.

[0081] The semantic drift detection includes: monitoring the degree of deviation between the model input distribution and the sample data distribution of the training set; determining whether to trigger model hot update or parameter adjustment based on the degree of deviation, to ensure the robustness of the model in long-term operation.

[0082] Optionally, step S3 specifically includes: inputting the obtained structured feature vector of each device of the energy storage power station into the trained multi-task diagnosis model deployed on each local diagnosis node; The trained multi-task diagnosis model deployed on each local diagnosis node respectively operates on the input structured feature vector to obtain the predicted current health status, current fault trend and current fault risk level of the energy storage power station, and sends them to the collaborative judgment module in a structured form. The collaborative judgment module forms a candidate result set from the prediction results sent by each local diagnosis node, and determines the final prediction result for the same type of abnormal event through majority voting or weighted voting mechanism.

[0083] In specific implementation, the initial weights of each local diagnosis node are the same.

[0084] In use, the accuracy of the historical prediction results (i.e. historical diagnosis results) of each local diagnosis node can be adjusted. For example, the judgment accuracy of each local diagnosis node can be periodically counted (compared with the actual fault verification result), and the diagnosis node whose accuracy is 10% higher than that of other local diagnosis nodes has its weight increased by 0.1, while the local diagnosis node whose accuracy is 10% lower than that of other local diagnosis nodes has its weight decreased by 0.1. This adjustment continues until only one local diagnosis node has a weight of 0 and the weights of the remaining local diagnosis nodes are all 0.

[0085] In specific implementation, when only one local diagnosis node has a weight of 0 and the weights of the remaining local diagnosis nodes are all 0, an alarm can be sent to remind the relevant staff to update the model deployed by the diagnosis node whose weight is 0.

[0086] As some other embodiments of the present application, the method further comprises: After the local response strategy is triggered to execute, the change of the preset target field of each device of the energy storage power station is monitored for a preset length of time, the improvement degree after the strategy response is calculated, and a feedback score is generated based on this; The control strength of the local response strategy is adjusted based on the feedback score to form a self-closing loop response optimization mechanism.

[0087] The preset target field includes temperature, current and SOC.

[0088] The feedback score includes response effect score, delay time and residual anomaly.

[0089] The present application provides an energy storage power station energy storage on-site monitoring system, specifically, the system comprises: A pre-constructed device semantic meta-model of various devices of the energy storage power station; A pre-constructed protocol analysis table and semantic mapping table of various devices of the energy storage power station; A monitoring data acquisition module; A feature extraction module; An edge side intelligent diagnosis module.

[0090] The device semantic meta-model includes a standard field dictionary, which defines core parameters of various types of devices in the energy storage power station.

[0091] The protocol parsing table is used to structurally parse raw fields in data streams generated by various types of devices in the energy storage power station.

[0092] The semantic mapping table is used to map the parsed raw fields to unified standard fields according to the standard field dictionary of the device semantic meta-model.

[0093] The monitoring data collection module is used to collect and uniformly format multi-source heterogeneous monitoring data of the energy storage power station.

[0094] Specifically, the monitoring data collection module is configured to perform the following steps: Real-time collection of data streams generated by devices in the energy storage power station through device communication interfaces; Real-time structural parsing of raw fields in the collected data streams based on the protocol parsing table; Real-time mapping of parsed raw fields to unified standard fields according to the standard field dictionary of the device semantic meta-model based on the semantic mapping table, to obtain standard fields of devices in the energy storage power station; Temporal and spatial alignment and resampling of the standard fields of devices in the energy storage power station using a master clock anchor and a sliding time window strategy, to obtain and output standard structured data of devices in the energy storage power station.

[0095] The feature extraction module is used to perform data fusion and feature extraction on the standard structured data of devices output by the monitoring data collection module, to obtain structured feature vectors of devices in the energy storage power station.

[0096] The edge-side intelligent diagnosis module is used to input the obtained structured feature vectors of devices in the energy storage power station into a pre-trained multi-task diagnosis model deployed on a local diagnosis node, to obtain a predicted current health status, current fault trend, and current fault risk level of the energy storage power station through model operation.

[0097] The embodiment of the energy storage in-situ monitoring system of the energy storage power station of the present embodiment belongs to the same inventive concept as the energy storage in-situ monitoring method of the energy storage power station of the above embodiments, and details not described in the embodiment of the energy storage in-situ monitoring system of the energy storage power station can be referred to the embodiment of the energy storage in-situ monitoring method of the energy storage power station.

[0098] The same and similar parts among the various embodiments in the present specification can be referred to each other. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the description in the method embodiment.

[0099] In the above embodiments of the present application, the description of each embodiment has a focus, and the part not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0100] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will accord with the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for in-situ monitoring of energy storage of an energy storage power plant, characterized in that, The method is based on a pre-constructed equipment semantic meta-model of various types of equipment of the energy storage power station, and a unified protocol analysis table and a semantic mapping table of the various types of equipment of the energy storage power station; the equipment semantic meta-model comprises a standard field dictionary; The method comprises: S1, multi-source heterogeneous monitoring data acquisition and unified formatting: Real-time acquisition of data streams generated by each device of the energy storage power station through a device communication interface; Real-time structured analysis of original fields in each acquired data stream based on a protocol analysis table; Real-time mapping of each analyzed original field to a unified standard field according to a standard field dictionary of the equipment semantic meta-model based on a semantic mapping table, to obtain standard fields of each device of the energy storage power station; Using a master clock anchor point and a sliding time window strategy, the standard fields of each device of the energy storage power station are spatio-temporally aligned and resampled, to obtain standard structured data of each device of the energy storage power station and output the same; S2, data fusion and feature extraction on the standard structured data of each device output in S1, to obtain structured feature vectors of each device of the energy storage power station; S3, edge-side intelligent diagnosis: inputting the obtained structured feature vectors of each device of the energy storage power station into a pre-trained multi-task diagnosis model deployed on a local diagnosis node, and obtaining a predicted current health status, a current fault trend and a current fault risk level of the energy storage power station through model operation.

2. The energy storage on-site monitoring method of the energy storage power station according to claim 1, characterized in that S2 comprises: For each device of the energy storage power station, using a unified sliding time window, statistical features of continuous quantities and behavior features of discrete state quantities in the standard structured data of each device output in S1 are counted, and then key derived quantities are deduced by combining physical formulas and device logic rules, and the deduced key derived quantities are collected as multi-dimensional aggregated features; For the standard structured data of each device of the energy storage power station in the sliding time window, multi-dimensional fusion of the standard structured data is performed through a field association strategy according to the cooperative relationship between the devices of the energy storage power station, and then based on the multi-dimensional fusion, cross-device coupling behaviors between the devices of the energy storage power station are analyzed, and then key joint features are extracted therefrom; The multi-dimensional aggregated features and the key joint features obtained under the sliding time window are uniformly encoded and structured through a constructed feature configuration template, to obtain the structured feature vectors of each device of the energy storage power station.

3. The energy storage plant energy storage in-situ monitoring method of claim 2, wherein, The method further comprises: S4, triggering a corresponding local response strategy according to the predicted current health status, the current fault trend and the current fault risk level.

4. The energy storage on-site monitoring method of the energy storage power station according to claim 2, characterized in that The method for obtaining the trained multi-task diagnosis model in S3 comprises: S31, constructing a training set; S32, constructing a network architecture of the multi-task diagnosis model; S33, training the network architecture using the training set to obtain the trained multi-task diagnosis model; In S31, the method for constructing the training set comprises: S311, collecting full-life-cycle historical operation data of each device of the energy storage power station; S312, using the same method as S1 and S2, processing the collected full life cycle historical operation data of each device of the energy storage power station to obtain a plurality of historical structured feature vectors of the energy storage power station; S313, labeling each historical structured feature vector, and the labeling labels include an energy storage power station health state label, a fault trend label, and a fault risk score label; the health state label includes "101", "102", and "103", representing normal, slight abnormality, and serious abnormality in turn; the fault trend label includes "201", "202", and "203", representing fault rising, fault falling, and trend stability in turn; the risk score label includes "301", "302", "303", "304", "305", and "306", corresponding to score intervals [0, 20), [20, 40), [40, 60), [60, 80), [80, 95), [95, 100] in turn, representing extremely low risk, low risk, medium risk, medium-high risk, high risk, and extremely high risk in turn; The labeled historical structured feature vectors are collected and sorted in ascending order of data collection time to form a training set.

5. The energy storage power station energy storage on-site monitoring method according to claim 4, characterized in that, the network architecture of the multi-task diagnosis model comprises: an input layer for inputting the structured feature vector to provide input for subsequent feature extraction; a shared hidden layer connected to the input layer for extracting a global representation of the input structured feature vector; a task branch layer connected to the shared hidden layer and specifically comprising a state classification branch, a trend prediction time series branch, and a risk score branch; wherein the state classification branch is used to predict the health state category probability distribution of the energy storage power station; the trend prediction time series branch is used to predict the state change amount of the energy storage power station; and the risk score branch is used to predict the continuous distribution of the risk score of the energy storage power station; an output layer comprising a first softmax function layer, a second softmax function layer, and a Linear layer; the first softmax function layer is connected to the output end of the state classification branch and is used to obtain and output the corresponding health state classification of the energy storage power station based on the output of the state classification branch; the second softmax function layer is connected to the output end of the trend prediction time series branch and is used to obtain and output the corresponding fault trend classification of the energy storage power station based on the output of the trend prediction time series branch; and the Linear layer is connected to the output end of the risk score branch and is used to map and output the corresponding risk score classification of the energy storage power station based on the output of the risk score branch; S33, training the network architecture using the training set to obtain a trained multi-task diagnosis model.

6. The energy storage plant energy storage in-situ monitoring method of claim 5, wherein, S33 comprises: S331, initializing the network architecture parameters to obtain an initial model; S332, inputting the training set into the initial model for forward propagation iterative training; in each iteration, the loss of each task branch is calculated, and then the total loss is calculated by dynamically fusing the branch losses; wherein the state classification branch adopts a cross-entropy loss function, the trend prediction time series branch adopts an MSE loss function, and the risk score branch adopts an MSE loss function. S333, the model parameters are updated by back propagation until the total loss converges or a preset number of training times is reached, to obtain a trained multi-task diagnosis model.

7. The energy storage power station energy storage on-site monitoring method of claim 1, wherein, The device semantic meta-model further comprises a device semantic processing module; The device semantic processing module is configured to implement standard field derivation calculation, field addition, device registration, and semantic conflict processing in the standard field dictionary.

8. The energy storage power station energy storage on-site monitoring method of claim 1, wherein, When the number of local diagnosis nodes in S3 is multiple, the trained multi-task diagnosis model is deployed on each local diagnosis node, and S3 comprises: The obtained structured feature vectors of each device of the energy storage power station are input into the trained multi-task diagnosis model deployed on each local diagnosis node; The trained multi-task diagnosis model deployed on each local diagnosis node respectively operates on the input structured feature vectors to obtain predicted current health status, current fault trend, and current fault risk level of the energy storage power station, and sends the predicted results in a structured form to the collaborative judgment module; The collaborative judgment module forms a candidate result set from the predicted results sent by each local diagnosis node, and determines the final prediction result for the same type of abnormal event through a majority vote or a weighted vote mechanism.

9. The energy storage plant energy storage in-situ monitoring method of claim 3, wherein, The method further comprises: After the local response strategy is triggered to execute, the change of the preset target field of each device of the energy storage power station is monitored for a preset length of time, the improvement degree after the strategy response is calculated, and a feedback score is generated based on the improvement degree; The control strength of the local response strategy is adjusted based on the feedback score to form a self-closing loop response optimization mechanism.

10. An energy storage in-situ monitoring system for an energy storage power plant, characterized in that, The system comprises a pre-constructed device semantic meta-model of each type of device of the energy storage power station, and a unified protocol analysis table and semantic mapping table of each type of device of the energy storage power station; the device semantic meta-model comprises a standard field dictionary; The system further comprises: A monitoring data acquisition module for collecting and uniformly formatting multi-source heterogeneous monitoring data of the energy storage power station; The monitoring data acquisition module is configured to perform the following steps: Real-time collection of data streams generated by each device of the energy storage power station through a device communication interface; Real-time structured analysis of original fields in each collected data stream based on the protocol analysis table; Real-time mapping of each analyzed original field to a unified standard field according to the standard field dictionary of the device semantic meta-model based on the semantic mapping table, to obtain standard fields of each device of the energy storage power station; Real-time spatio-temporal alignment and resampling of the standard fields of each device of the energy storage power station using a master clock anchor point and a sliding time window strategy, to obtain and output standard structured data of each device of the energy storage power station; The system further comprises: A feature extraction module for data fusion and feature extraction of the standard structured data of each device output by the monitoring data acquisition module, to obtain structured feature vectors of each device of the energy storage power station; The edge side intelligent diagnosis module is used for inputting the obtained structured feature vector of each device of the energy storage power station into a pre-trained multi-task diagnosis model deployed on the local diagnosis node, and obtaining a predicted current health status, current fault trend and current fault risk level of the energy storage power station through model operation.