A real-time data integration method and platform based on a CMOS-CS hybrid architecture for monitoring power supply links and energy storage systems in power plants.

Abstract

This invention discloses a real-time data integration method and platform based on a CMOS-CS hybrid architecture for monitoring power supply links and energy storage systems in power plants. It belongs to the field of power supply or distribution circuit devices and energy storage technology. The method includes: acquiring raw data from power supply circuit devices, energy storage devices, and related components through edge acquisition units; aggregating this data via a unified acquisition bus; and performing structured encapsulation of state and continuous quantities through a channel type mapping table. Data splitting is achieved using transition judgment functions and residual deviation functions. Continuous data for energy storage and distribution is filtered using a weighted slope-constrained fitting residual function. Multi-source data fusion is completed through time synchronization and sliding time window clustering to form a unified event record. Event priority scoring functions are used to map the data to power supply regulation tasks. This invention improves the data integration efficiency of power supply circuit devices and energy storage systems, ensuring the real-time performance and accuracy of power supply control and energy storage scheduling.

Description

Technical Field

[0001] This invention relates to the field of power supply or distribution system technology, and more particularly to a real-time data integration method and platform for a COMS-CS hybrid architecture for monitoring power supply links and energy storage systems in power plants. Background Technology

[0002] With the integration and application of intelligent power supply systems and energy storage technology in power plants, the COMS platform, as the core hub for power supply circuit device scheduling and energy storage system monitoring, undertakes key responsibilities in power supply stability monitoring, power distribution link load allocation, energy storage charging and discharging coordination, and emergency power supply response. However, in current power plant power supply systems, the deployment of various data acquisition terminals (such as distribution switch quantity acquisition modules, energy storage unit monitoring terminals, bus voltage detection devices, and transmission line status sensors) is characterized by distribution and heterogeneity. The acquired data covers various types of data, including switch status quantities of power supply circuits, load continuity quantities, energy storage charging and discharging parameters, and transmission line loss data. These data structures differ significantly, update frequencies vary, and communication protocols (such as IEC61850 and Modbus) and reporting paths are heterogeneous. This leads to a core bottleneck in the application of the traditional COMS-CS hybrid architecture (where COMS represents a specific business function module, namely the Communication, Operation and Monitoring System, corresponding to the core business functions of communication interaction, operational status monitoring, and data encapsulation mapping for processing power equipment data; CS stands for Client / Server, typically established between local area networks, providing connection and data exchange services through dedicated servers) in power supply and energy storage systems.

[0003] On the one hand, the mixed data from power supply circuit devices and energy storage systems are reported indiscriminately, making it impossible for the system to accurately identify semantic attributes such as power distribution link status quantities and continuous energy storage quantities. This results in low-value data occupying bandwidth, and a lack of priority distinction between power supply circuit control commands and energy storage scheduling commands. On the other hand, issues such as edge terminal clock errors and network latency cause out-of-order timing of multi-channel data, affecting the accuracy of power supply fault tracing, power distribution load prediction, and energy storage charging and discharging coordination. In addition, the existing system relies on centralized computing and does not fully utilize edge resources for on-site preprocessing of power supply data and energy storage data, resulting in delayed control response of circuit devices and failing to meet the requirements of millisecond-level power supply adjustment and dynamic optimization of energy storage.

[0004] Therefore, establishing data semantic analysis, edge value judgment, central timing reconstruction, and closed-loop control mechanisms for power supply circuit devices and energy storage systems has become the core breakthrough point for intelligent power supply and efficient energy storage collaboration in the power supply or distribution system field. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time data integration method and platform for a COMS-CS hybrid architecture, particularly for monitoring power plant power supply links and energy storage systems, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A real-time data integration method for a CMOS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems includes:

[0008] The raw data from the power supply circuit device, energy storage device and related power equipment are acquired and aggregated to the data acquisition master station. The data type tags of the raw data in the data acquisition master station are obtained through the channel type mapping table. Based on the data type tags, structured encapsulation is completed to form a data set. The raw data includes boiler operation data, turbine protection cabinet status data, bus voltage power supply data, flue gas online detection data and energy storage device charging and discharging status data. The data type tags include status quantities and continuous quantities.

[0009] The original data of the state variables are subjected to transition judgment by a transition judgment function, and the original data filtered by the transition judgment function is used as the state variable data set. The original data of the continuous variables are subjected to residual offset judgment by a residual deviation function, and the original data filtered by the residual deviation function is used as the continuous variable data set. The state variable data set includes power distribution switch state change data, and the continuous variable data set includes abnormal data of energy storage charging and discharging parameters.

[0010] The continuous data set is filtered and inferred by the fitting residual function with weighted slope constraint, and a standard continuous data set is output. The parameters of the fitting residual function with weighted slope constraint include the sliding window prediction value, the slope of the fitted line and the channel characteristic adjustment factor.

[0011] Time alignment is performed on the state variable data set and the standard continuous variable data set. The state variable data set and the standard continuous variable data set are clustered using a sliding time window clustering function, and fusion calculation is performed to generate a unified event record.

[0012] The event priority scoring function identifies the operability of the unified event record, maps operable events to actual power supply circuit adjustment tasks, energy storage charging and discharging control commands, alarm push commands, and historical data recording operations, and binds them to generate a task structure set.

[0013] Preferably, the raw data in the data acquisition master station is a triplet structure, which includes channel number, device sampling time, and raw data.

[0014] Preferably, the parameters of the transition judgment function include the state value of the channel's previous upload, the value of the channel's previous non-upload but locally cached value, an indicator function, and a strategy coefficient, wherein the strategy coefficient is used to characterize the false alarm sensitivity.

[0015] Preferably, the parameters of the residual deviation function include the mean of the channel in the most recent sampling, the fitted predicted value of the channel at the current time, and the parameter coefficients, wherein the fitted predicted value includes the predicted value of the energy storage voltage.

[0016] Preferably, the parameters of the sliding time window clustering function include data units of the state quantity data set, data units of the standard continuous quantity data set, time after delay correction, time difference threshold, adjacency set of the channel, and adjacency weight coefficient, wherein the adjacency weight coefficient is used to adjust the influence of semantic similarity in clustering determination.

[0017] Preferably, the unified event record includes a unified event time, a structured data set, and events related to power supply status and energy storage parameters. The structured data set includes a state quantity data set and a standard continuous quantity data set.

[0018] Preferably, the parameters of the event priority scoring function include channel weight parameters, index functions, data density, and density weight parameters.

[0019] Preferably, the channel type mapping table is stored in text form in a configuration file or database table, and table lookup is achieved at runtime using a hash structure.

[0020] Preferably, the data set is formed by structural encapsulation based on the data type label, and the data set includes channel number, device sampling time, raw data and data type label.

[0021] A real-time data integration platform with a CMOS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems includes:

[0022] The data encapsulation module is used to acquire raw data from power equipment through edge acquisition units and aggregate the raw data to the data acquisition master station via a unified data acquisition bus. It obtains the data type tags of the raw data in the data acquisition master station through a channel type mapping table, completes structured encapsulation based on the data type tags, and generates an encapsulated data set. The raw data includes boiler operation data, turbine protection cabinet status data, bus voltage power supply data, and flue gas online detection data. The data type tags include status quantities and continuous quantities.

[0023] The data splitting module is used to perform jump judgment on the original data of state variables through the jump judgment function, and take the original data that meets the jump threshold as the state variable data set. It also performs residual offset judgment on the original data of continuous variables through the residual deviation function, and takes the original data that meets the deviation threshold as the continuous variable data set.

[0024] The data inference module is used to filter and infer continuous data sets through a fitting residual function with weighted slope constraints, and finally outputs a standard continuous data set after filtering. The fitting residual function with weighted slope constraints includes sliding window predicted values, the slope of the fitted line, and channel characteristic adjustment factors.

[0025] The data fusion module is used to align the state quantity data set and the standard continuous quantity data set in time through the time synchronization component, perform clustering processing on the state quantity data set and the standard continuous quantity data set through the sliding time window clustering function, and perform fusion calculation to generate a unified event record.

[0026] The task execution module is used to identify the operable status of unified event records through an event priority scoring function, map operable events to actual scheduling and control tasks, alarm push instructions, and historical data recording operations, and bind scheduling and control tasks, alarm push instructions, and historical data recording operations to generate a task structure set.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] This invention first constructs a channel-type mapping table during the channel access phase to achieve structured encapsulation and semantic annotation of raw data, providing a type foundation for subsequent differentiated scheduling. Then, a dual-channel data splitting strategy is executed in the acquisition agent, configuring upload rules for state variables and continuous variables separately. State variables are uploaded based on transition recognition, while continuous variables are judged based on sliding window bias and fitting residuals, achieving dual-objective control of transmission traffic compression and semantic preservation. Furthermore, a lightweight inference model is embedded in the continuous variable path to perform weighted scoring of fluctuation trends using fitting residuals and fluctuation factors, accurately identifying key turning points and ensuring that limited reporting bandwidth is utilized effectively. The platform transmits the most valuable scheduling information. After receiving the data, it uses a time correction mechanism and a two-factor similarity function based on channel adjacency to reconstruct the temporal consistency of multi-source heterogeneous data and cluster events within a sliding window, forming an event structure with unified timestamps and multi-channel semantic content. Finally, the platform uses an event scoring function to comprehensively analyze the matching degree of indicators and the internal signal density of events to generate a set of scheduling behaviors, including control commands, alarm pushes, and record archiving. A task binding structure keeps events and operations logically consistent, achieving a closed loop from perception, identification, reasoning, fusion, to control. This invention not only introduces targeted strategies suitable for power plant scenarios at each step but also constructs a complete technical path that is deployable, reproducible, and controllable. It significantly improves the COMS platform's processing capability and response efficiency for complex multi-source monitoring data, providing structured and systematic support for the intelligent scheduling of power plant operation systems. Attached Figure Description

[0029] Figure 1 This is a flowchart of a real-time data integration method for a COMS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems, as described in a specific embodiment of the present invention.

[0030] Figure 2 This is a data transmission diagram in the real-time data integration method of the COMS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems in a specific embodiment of the present invention. Detailed Implementation

[0031] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] refer to Figure 1 As shown, this invention proposes a real-time data integration method based on a CMOS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems, including:

[0033] Step 1: Acquire raw data from the power supply circuit device, energy storage device, and associated power equipment and aggregate it to the data acquisition master station. Obtain the data type tags of the raw data from the data acquisition master station through the channel type mapping table. Based on these data type tags, perform structured encapsulation to form a data set. The raw data includes boiler operation data, turbine protection cabinet status data, bus voltage power supply data, flue gas online detection data, and energy storage device charging and discharging status data. The data type tags include status quantities and continuous quantities, specifically including:

[0034] This step primarily involves the structured encapsulation and type labeling of data collected from the power plant site, serving as the starting point for the entire real-time data integration process. Power plant operation generates a large amount of equipment monitoring data, originating from various industrial field devices such as distributed control systems, programmable logic controllers, smart meters, and protection devices. This data is aggregated by channel number and includes equipment operating status (e.g., pump start-up status, valve on / off status), analog signals (e.g., steam flow, main steam temperature, bus voltage), and alarm quantities (e.g., high-pressure over-limit, low oil level alarm). The signal types of different devices vary significantly, and the data formats and semantics are completely different. However, during the acquisition and transmission phase, most acquisition systems only distinguish channels by number or register address, without including signal type information. This prevents subsequent systems from performing differentiated scheduling based on semantics when processing data. Therefore, the first step in data access must be to complete the type identification and structure labeling of data channels, ensuring that the raw collected data has a unified data structure and clear semantic tags.

[0035] Raw data acquisition is accomplished through edge acquisition units. For example, data acquisition modules installed in boiler control systems connect to temperature and pressure transmitters via the Modbus RTU protocol; PLCs in turbine protection cabinets acquire status bits and upload them to the central control room via the PROFINET protocol; intelligent circuit breakers in high-voltage electrical systems upload remote signaling and telemetry signals via the IEC61850 interface; and flue gas online monitoring devices in environmental protection systems periodically send data packets via Ethernet. All the data from these devices is ultimately aggregated to the data acquisition master station via a unified data acquisition bus (such as industrial Ethernet, CAN bus, or serial bus). When each data point arrives at the master station, its original structure is generally represented as a triplet. ,in It is the channel number. This refers to the device-side sampling time of the data. This is the raw data. At this point, the data does not carry any information about "data semantics".

[0036] To implement type labeling, during the deployment phase, the system's operations and maintenance personnel configure a channel type mapping table. This table maps the data type of the data acquisition master station to the corresponding channel type and sends the corresponding raw data to the channel corresponding to that channel type. The channel type is denoted as... The format is as follows:

[0037] ;

[0038] in, Channel number, This is the data type label corresponding to this channel. The value is limited to either a "state quantity" or a "continuous quantity." The former represents discrete device states, such as start / stop signals and switch positions, while the latter represents continuously changing physical quantities, such as temperature, voltage, and pressure. This mapping table is stored in text form (such as a configuration file or database table) and is loaded into the memory of the data acquisition service module at startup. During runtime, a hash structure is used to achieve fast table lookup.

[0039] When the acquisition module receives a piece of raw data At that time, the program automatically looks up its type identifier in a table. And complete the structured encapsulation, the output is:

[0040] ;

[0041] This structured data unit This refers to data record items with a unified semantic format. The entire process is independent of protocol content and can be completed solely based on the channel number. The advantage of this mechanism is that it can shield against differences in underlying protocols, and only structured data with a unified format needs to be processed when handling data.

[0042] For example, during one operation, the steam flow meter (numbered as...) A data point was received: the device reported a value of Sampling time is The original structure is The system retrieves the result from a table lookup. If the corresponding type is "continuous quantity", then the output encapsulation result is:

[0043] ;

[0044] For example, data collected from the start button of the turbine oil pump is numbered... The signal, value is (Indicates that it has been started), sampling time is If its channel is configured as a "state variable", then the encapsulation result is:

[0045] ;

[0046] All of this encapsulated structured data will be written into the collection. This serves as the basis for the data splitting and uploading strategy in the next step, and the final output is a packaged data set. Its elements are in a four-tuple structure, each item including channel number, sampling time, type identifier and value, with the following uniform format: .

[0047] In summary, this step does not rely on the embedded semantic fields of each acquisition protocol. Instead, it enables the system to directly complete data structure standardization and semantic recognition at runtime by constructing a channel-type binding relationship determined during the deployment phase.

[0048] It is particularly efficient in scenarios such as power plants where multiple devices, interfaces, and signal types coexist. It can significantly improve the efficiency and accuracy of data channel management, model judgment, and fusion calculation in subsequent steps, avoid structural uncertainty caused by delayed type identification, and has good implementability and portability when deployed on any device with access capabilities, such as the main station acquisition server, protocol conversion gateway, and edge computing unit.

[0049] Step Two: The raw data of the state variables are subjected to transition judgment function to determine transitions. The raw data filtered by the transition judgment function is used as the state variable data set. The raw data of the continuous variables are subjected to residual deviation function to determine residual offset. The raw data filtered by the residual deviation function is used as the continuous variable data set. The state variable data set includes data on sudden changes in the state of distribution switches, and the continuous variable data set includes abnormal data on charging and discharging parameters of energy storage. Specifically, it includes:

[0050] This step is a critical node in data processing. Because the monitoring data from power plant power supply links and energy storage systems exhibit significant heterogeneity and varying update characteristics—including both equipment start-up / shutdown status change data and high-frequency, continuously changing process monitoring data—processing this data uniformly can lead to decreased scheduling efficiency and increased system response latency. This problem is particularly pronounced in client / server architectures with a large number of terminals and limited bandwidth. Therefore, it is essential to perform semantic-driven routing and uploading scheduling before the structured data enters the platform.

[0051] The input is the data set output from step one. ,in Indicates that the channel number is Data unit, Sampling time, These are the original collected values. For type labels, values ​​include "state quantity" and "continuous quantity," which were already configured in the previous step. Complete static annotation. The system performs annotation during runtime. Type identification is performed, and the process proceeds to the triage stage.

[0052] against For state data, this step adopts a dual judgment strategy of event jump judgment and confirmation of steady state write-back. Since state data usually does not have the characteristic of continuous change, but only changes when actual operation occurs, it is only uploaded when the value of the data changes.

[0053] To ensure the system correctly identifies "continuous state" instead of misidentifying it as "jump rollback," this step designs a jump function with an acknowledgment mechanism, where the jump function is:

[0054] ;

[0055] in, This is the status value of the last upload to this channel. This is the value that was not uploaded to the channel last time but is cached locally. This is an indicator function that returns either 1 or 0. This is the strategy coefficient, used to control false alarm sensitivity, and is usually set to 1. .

[0056] Specifically, when When the system determines that the channel status has indeed changed, it enters the status data upload buffer.

[0057] The dual-judgment strategy not only increases the ability to identify repeated state changes, but also avoids misidentifying transient fluctuations as valid transitions. At the same time, the introduction of a local state confirmation mechanism improves the effectiveness of platform data. This mechanism is adapted to the short-term fluctuation signals that may be generated during the operation of switchgear in power plants. For example, the opening and closing commands of high-voltage circuit breakers may generate multiple 0 / 1 changes due to contact jitter. The system can use this strategy to identify the valid changes and upload a confirmation result.

[0058] for For data that is "continuous," a difference upload mechanism driven by both sliding window deviation and trend fitting residuals is adopted. Considering that signals such as steam flow and main steam pressure in power plant operation data change at varying rates, relying solely on a fixed offset would lead to some low-variable signals being mistakenly considered invalid. Therefore, the system designs the following residual deviation term as the upload criterion, where the residual deviation function is:

[0059] ;

[0060] in, Indicates channel In recent The mean of the samples. This represents the current time-time predicted value obtained by fitting the channel using a single linear least squares method. This is the residual weighting coefficient (which can be set from 0.8 to 1.2, fine-tuned according to channel characteristics). The larger the current value deviates from the window mean and the fitted value, the greater the weighting coefficient. The larger the value, the more likely the system is to classify it as a "high-potential mutation critical point," which can trigger an upload.

[0061] Specifically, only when At that time, the data was included in the continuous upload set.

[0062] This joint judgment mechanism has three key advantages for power plant scenarios: First, it can identify "trend abrupt changes" signals, such as main steam pressure that may not significantly deviate from the average value under short-term load fluctuations but whose fitting residuals increase rapidly, enabling timely reporting; second, it avoids over-uploading of long-term stable signals such as drum water level due to slight fluctuations; and third, the combined design of sliding window and residuals can adapt to different signal change characteristics, and the window size can be adjusted according to the physical meaning of the signal during the deployment phase. With residual weights .

[0063] After the above-mentioned flow division and strategy determination, the system outputs two data sets: a state data set and a continuous quantity data set.

[0064] ;

[0065] ;

[0066] These two sets are sent to different upload channels for subsequent edge model processing (for...). Integration with the central platform (for) () can be directly input.

[0067] Step 3: Filter and infer the continuous data set using a weighted slope-constrained fitting residual function, and output a standard continuous data set. The parameters of the weighted slope-constrained fitting residual function include the sliding window predicted value, the slope of the fitted line, and the channel characteristic adjustment factor, specifically including:

[0068] This step is located in the peripheral reasoning model part of the entire invention process. Its main function is to process the continuous data set that has been determined in step two. Further screening and inference are performed, using a lightweight model to determine the upload value of each piece of data locally at the edge. This step is designed not merely for data compression, but to address typical problems in power plant CMOS-C / S architectures—high frequency of data collection from field terminals, significant data redundancy, limited upload channel bandwidth, and limited processing capacity of the central platform. Before platform processing, a local inference module performs an initial screening of high-information-density data, achieving edge offloading, data assignment, and platform deduplication. This step acts as a "value judgment" filter in the entire system, determining which data has global event relevance and should enter the unified fusion process.

[0069] Input data set Derived from step two, each structured data entry in the set is... ,in Indicates the channel number. For data sampling time, The current value, The data is considered a "continuous quantity." All data has already undergone preliminary buffering and scheduling strategies to ensure frequency stability and numerical validity; therefore, this step does not involve data cleaning but focuses on "identifying which values ​​are significant in their historical trends." To this end, the system deploys a lightweight inference module within each edge acquisition terminal for each channel. Maintain a size of Data window Save channel within window The recent data values The system performs fitting modeling and dynamic fluctuation analysis on the window to determine the current value. Whether it constitutes a "trend turning point" or "fluctuation anomaly" relative to the historical sequence.

[0070] To achieve more targeted judgment capabilities, this step designs a weighted slope-constrained fitting residual function to calculate the upward priority index of the current value. As shown below:

[0071] ;

[0072] in, Indicates a sliding window The predicted values ​​obtained using first-order linear least squares fitting are shown above. Let be the slope of the fitted line (i.e., the rate of trend growth). This represents the variance of the values ​​within the window. (Coefficient) and These are channel characteristic adjustment factors. The former controls the weight of the trend change term, while the latter controls the impact of historical volatility on the current judgment. The general range is... ~ This is configured by the system initialization settings.

[0073] The above design has the following advantages: (1) The trend slope is used as the regularization term of the residual rather than an independent index, which means that when a signal is in a continuous growth or decline process, even if the instantaneous deviation is not large, it will be identified as a key point; (2) Historical fluctuations are introduced as a dynamic adjustment factor, that is, for signals that are in the high jitter range, the system tends to report more cautiously, while when the signal is stable for a long time and suddenly changes, the system can respond quickly; (3) The fitting process is limited to the first-order linear model, and the whole process is completed in a single channel loop, which is suitable for use on terminals with low computing power such as PLC and edge gateway.

[0074] In practice, the system handles each channel. It maintains an independent data window and historical parameter cache, and updates the window and recalculates immediately upon arrival of a sampled value. , and Then substitute into the above formula to calculate. If the following conditions are met, then the data... Determined to have upload value, added to the uplink queue:

[0075] ;

[0076] in For channel The upload threshold is configured during platform initialization and set based on the characteristics of the signal's physical quantity. For example, it is set to the temperature channel of a steam turbine bearing. This indicates that the sum of indicators such as fitting error and trend change exceeds [a certain threshold]. It must be reported.

[0077] For example, suppose the number is The main steam flow channel has a relatively stable value within the most recent window, but the currently measured value... Significantly higher than the fitted predicted value slope Indicates an upward trend, with a low window variance. ,set up , Then the calculation yields:

[0078] ;

[0079] If a threshold is set ,but This data will be reported.

[0080] The final output is the filtered standard continuous data set, denoted as... Each data item in the set retains the same structure and remains the same. It possesses the attributes for determining the value of uploaded content.

[0081] Step 4: Time alignment is performed on the state variable data set and the standard continuous variable data set. A sliding time window clustering function is used to cluster the state variable data set and the standard continuous variable data set, and a fusion calculation is performed to generate a unified event record. Specifically, this includes:

[0082] This step is performed on the COMS central platform and is used for standard continuous data sets from the edge side. With state dataset Perform unified temporal reconstruction and multi-source fusion;

[0083] After the previous steps, the edge devices have completed data semantic labeling, channel splitting and value filtering. However, due to the fact that each edge node is distributed in different areas of the power plant, there are differences in network latency, sampling frequency and cache reporting strategy, which causes the data received by the central CMOS to have time misalignment and semantic interleaving problems.

[0084] For example, the sampling frequency of the turbine main shaft vibration signal is 100Hz, while the upload period of the cooling valve opening signal is 1 second. Furthermore, due to different network paths, the system may receive valve change data first, and then receive its corresponding flow response value later. This out-of-order phenomenon, if not time-series reordering, will lead to incorrect event identification or delayed judgment. Therefore, the task of this step is to construct a reproducible time-series reconstruction and event fusion mechanism to ensure that all data from different devices and channels can be correctly integrated into a unified temporal semantic structure at the central side.

[0085] After the data enters the CMOS processing module, the time synchronization component first aligns the local time of each edge node with the central system clock. Time synchronization is achieved through the master station's time synchronization signal, combined with the local timestamp in each uploaded data packet. With data reception time Calculate network transmission delay The central system maintains a node latency statistics table and calculates the moving average latency for each node. During data fusion, time correction is performed to map all data to a unified reference time axis. In this way, even if the reporting intervals of each terminal are different, a unified time scale can be maintained at the central level.

[0086] To further achieve event-level fusion and rearrangement, this step designs a "sliding time window clustering mechanism" in the CMOS data bus, with each window having a length of... The system dynamically adjusts the window length based on the device sampling frequency, typically taking 2 to 3 times the upload cycle of the main control channel (e.g., if the main control signal has a 1-second cycle, the window length is set to 3 seconds). All data within the window undergoes time correction before being processed for clustering, using a composite decision formula based on time proximity and device correlation.

[0087] ;

[0088] In the formula, and They represent from the set and Data unit; , This is the time after the delay correction; The time difference threshold (for example, signals within 1 second are considered to occur simultaneously); Indicates channel The adjacency set is used to characterize physical relationships. For example, the adjacency set of the main steam valve opening channel can include the main steam flow and main steam pressure channels. This refers to the adjacency weight coefficient, which typically ranges from [value missing]. This is used to adjust the influence of semantic similarity in clustering decisions. When When two pieces of data belong to the same time sequence event, the system considers them to belong to the same event cluster and adds them to the same event cluster.

[0089] After clustering, the system performs a fusion calculation on each event cluster to generate a unified event record. Event Time The weighted median of all samples within the cluster is taken. The weighting coefficients are assigned according to the data type. In one embodiment, state variables are assigned a weight of 1.2, and continuous variables are assigned a weight of 1.0, to ensure that the event occurs closer to the control action timing. (Set) This stores all data units within the event cluster, preserving their original structure to support subsequent analysis. For example, in a valve opening operation... of The data may include multiple figures: valve opening increased from 30% to 50%, main steam flow increased by 3.2 tons / hour, and main steam pressure increased by 0.05 MPa.

[0090] To address situations where data reporting delays or interruptions occur on certain channels, this step also incorporates a timing completion strategy. If an adjacent channel is found within the same window... The data is missing, but its previous moment With the next moment If all data exist, then linear interpolation is used to estimate the intermediate missing values. ;

[0091] The event structure is marked with a "complete flag". This complete data will not participate in alarm or scheduling decisions, but is only used to maintain the integrity of the event data structure so that it can be used for trend display or historical tracking at the CMOS interface layer.

[0092] For example: Suppose the CMOS system receives the following data segments within the same time window: status variables from the main steam valve channel (number 1201). This indicates that the valve is open; continuous flow from the main steam flow channel (number 2301) This indicates an increase in flow rate; the continuous flow from the main steam pressure channel (number 2401) This indicates an increase in pressure. Because their time differences are all less than the set threshold... And they have an adjacency relationship, the system according to Clustering them into the same event cluster based on the conditions generates a unified event. The three data entries are then merged into a single logical event record for use by the platform's execution module in the next step. The final output is a fused event set. Each of them Each represents a temporally and semantically independent runtime event. Each event structure contains a unified time frame. and related data sets This provides basic data input for subsequent dispatch instruction generation, alarm triggering, and operational status analysis.

[0093] Step 5: Identify the operability of the unified event record using the event priority scoring function, map operable events to actual power supply circuit adjustment tasks, energy storage charging and discharging control commands, alarm push commands, and historical data recording operations, and bind them to generate a task structure set, specifically including:

[0094] This step serves as the concluding stage of the entire invention system, and its core function is to integrate the event set that has been merged in the previous step. This is mapped to actual scheduling and control tasks, alarm push commands, or historical data recording operations, achieving a closed-loop process from data integration to platform behavior. As the power plant's COMS platform serves as the central scheduling core of the monitoring system, its final output must meet the basic requirements of being "executable, verifiable, and recordable," ensuring the system can automatically respond after event identification while avoiding false triggers or inconsistent behavior. Therefore, the key challenge of the platform's response mechanism is: how to determine "which events need to be responded to, what to respond to, and how to respond" after multi-signal fusion, and transform the results into operational-level task entities. The mechanism designed in this step is innovatively constructed from four aspects: structural logic, judgment method, response action, and execution binding, ensuring a logical closed loop from event identification to task issuance.

[0095] The input for this step is the set of fusion events. Each event , Indicates the moment of unity of events. The aggregated structured data set consists of continuous data from step three. Compared with the state data in step two These data are collectively composed of several elements. During actual platform deployment, these data are collected via edge terminals and connected to the platform data bus through the scheduling and control host. Each data item is a structured quadruple. The format and naming have been standardized in the preceding steps. Event Collection The event is written into the platform's event handling module, where it is parsed, evaluated, and processed one by one.

[0096] To identify which events require a response, this step introduces an event priority scoring function for industrial scheduling scenarios. This scoring function consists of two parts: an indicator matching score and a structure density score. The formula for the event priority scoring function is as follows:

[0097] ;

[0098] in Indicates channel The weighting parameters are set according to the importance of the equipment in the entire process chain, such as setting the boiler main steam temperature channel to 1.5 and the auxiliary cooling water system to 0.8. It is an indicator function, when the data value Falling into the "action trigger zone" set by this channel It returns 1 if the motor current exceeds the rated current by 10%, otherwise it returns 0. This range can be set according to the safety threshold or operating procedures. This indicates the event data density, which is the amount of data contained in the event per unit of time. It is a density weight parameter, usually set to 1.0. The higher the score, the more likely the event has the characteristics of "indicator overshooting" and "multiple signals converging", and is more likely to be a key event that the platform should focus on responding to.

[0099] Once an event Greater than the preset response threshold The system treats this event as an "operable event" and initiates the task generation process. Task generation includes three types of response actions, each corresponding to a different platform module. The first is control task generation; for example, if the event indicates that the main pump current exceeds the upper limit and the cooling system flow is insufficient, a control command is triggered. ,in It is the target device number. It refers to the type of action (e.g., "pump shutdown", "load reduction"). The execution time is typically set to the event identification completion time plus a fixed response delay. Control commands are written to the control link interface and issued by the platform via communication protocols (such as MODBUS, IEC104).

[0100] Secondly, there is the alarm push mechanism. If an event, although not meeting the control conditions, is determined to be a potential risk or a sign of equipment malfunction, the platform generates an alarm message. The risk level is set according to the equipment's risk level, and the description field is automatically generated by the platform based on the event content, such as "Abnormal pressure in the auxiliary machine's cooling circuit; manual inspection recommended." This information is sent via SMS, desktop pop-ups, dispatch interfaces, etc.

[0101] Finally, there is the historical archive record. All events, regardless of whether they trigger controls or alarms, are recorded in the operation log, with the following structure: It is used for system review, scheduling verification, or trend analysis.

[0102] To ensure task traceability and execution consistency, this step binds all events and their derived response tasks into structured tuples:

[0103] ;

[0104] The meaning and data source of each element in this structure are recorded in the platform system's logs and interfaces. For example, From the instruction generation module, From the alarm generation module, It is directly archived from the event structure. The platform outputs a record after processing each event. The structure is routed to the operation log queue. For example, during a certain operation, the main steam valve opening jumps from 30% to 60%, while the main steam pressure rises from 12.5 MPa to 13.9 MPa, and the turbine main shaft speed increases beyond the scheduled prediction rate. The platform identifies this as a "rapid load increase event" based on the event structure, matches it with multiple out-of-bounds thresholds, and ultimately... The triggering actions include: "issuing a dispatch warning, controlling the auxiliary sprinkler system to reduce pressure, and recording the event structure," and the task structure. The task is fully written to the task queue and executed. The final output is a collection of task structures. It not only preserves the original data structure and response behavior of events, but also establishes a logical binding between events and tasks, which is the key structure for the COMS platform to realize "data-driven scheduling behavior".

[0105] refer to Figure 2 As shown, the second invention of this application also proposes a real-time data integration platform with a CMOS-CS hybrid architecture for monitoring power plant power supply links and energy storage systems, including:

[0106] The data encapsulation module is used to acquire raw data from power supply circuit devices, energy storage devices, and associated power equipment and aggregate it to the data acquisition master station. It obtains the data type tags of the raw data in the data acquisition master station through the channel type mapping table, and performs structured encapsulation based on the data type tags to form a data set. The raw data includes boiler operation data, turbine protection cabinet status data, bus voltage power supply data, flue gas online detection data, and energy storage device charging and discharging status data. The data type tags include status quantities and continuous quantities.

[0107] The data splitting module is used to perform transition judgment on the raw data of state quantities through a transition judgment function, and to use the raw data filtered by the transition judgment function as a state quantity data set. The module also performs residual offset judgment on the raw data of continuous quantities through a residual deviation function, and to use the raw data filtered by the residual deviation function as a continuous quantity data set. The state quantity data set includes power distribution switch state change data, and the continuous quantity data set includes abnormal data of energy storage charging and discharging parameters.

[0108] The data inference module is used to filter and infer continuous data sets through a weighted slope-constrained fitting residual function, and finally output the filtered standard continuous data set. The parameters of the weighted slope-constrained fitting residual function include the sliding window prediction value, the slope of the fitted line, and the channel characteristic adjustment factor.

[0109] The data fusion module is used to perform time alignment on the state variable data set and the standard continuous variable data set. It uses a sliding time window clustering function to cluster the state variable data set and the standard continuous variable data set, and performs fusion calculations to generate a unified event record.

[0110] The task execution module is used to identify the operable status of unified event records through an event priority scoring function, map operable events to actual power supply circuit adjustment tasks, energy storage charging and discharging control commands, alarm push commands, and historical data recording operations, and bind power supply circuit adjustment tasks, energy storage charging and discharging control commands, alarm push commands, and historical data recording operations to generate a task structure set.

[0111] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A COMS-CS hybrid architecture real-time data integration method for power plant power supply chain and energy storage system monitoring, characterized in that, The application relates to a data collection method for power supply circuit and energy storage equipment, which comprises the following steps: acquiring original data of power supply circuit devices, energy storage equipment and associated power equipment and converging to a data collection master station, acquiring data type tags of the original data in the data collection master station through a channel type mapping table, completing structured packaging based on the data type tags to form a data set, the original data comprising boiler operation data, steam turbine protection cabinet state data, bus voltage power supply data, flue gas online detection data and energy storage equipment charging and discharging state data, and the data type tags comprising state quantity and continuous quantity; performing jump judgment on the original data of the state quantity through a jump judgment function, taking the original data filtered through the jump judgment function as a state quantity data set, performing residual deviation judgment on the original data of the continuous quantity through a residual deviation function, taking the original data filtered through the residual deviation function as a continuous quantity data set, the state quantity data set comprising power distribution switch state mutation data, and the continuous quantity data set comprising energy storage charging and discharging parameter abnormal data; filtering and reasoning the continuous quantity data set through a fitting residual function with a weighted slope constraint, and outputting a standard continuous quantity data set, parameters of the fitting residual function with the weighted slope constraint comprising a sliding window prediction value, a slope of a fitting line and a channel characteristic adjustment factor; performing time alignment on the state quantity data set and the standard continuous quantity data set, performing clustering processing on the state quantity data set and the standard continuous quantity data set through a sliding time window clustering function, and performing fusion calculation to generate a unified event record; identifying the operable conditions of the unified event record through an event priority scoring function, mapping the operable events into actual power supply circuit adjustment tasks, energy storage charging and discharging control instructions, alarm pushing instructions and historical data record operations and binding to generate a task structure set. The original data in the data collection master station is in a triple structure, the triple structure comprising a channel number, a device end sampling time and original data. Parameters of the jump judgment function comprise a last uploaded state value of a channel corresponding to the channel number, a value not uploaded but locally cached, an indication function and a strategy coefficient, and the strategy coefficient is used for representing false alarm sensitivity. Parameters of the residual deviation function comprise a mean value of a channel in the latest sampling, a fitting prediction value of the channel at a current time and a parameter coefficient, and the fitting prediction value comprises an energy storage voltage prediction value. Parameters of the sliding time window clustering function comprise data units of the state quantity data set, data units of the standard continuous quantity data set, a time after delay correction, a time difference threshold, a neighbor set of a channel and a neighbor weight coefficient, and the neighbor weight coefficient is used for adjusting the influence of semantic similarity in clustering determination. The unified event record comprises an event unified time, a structured data set and an event associated with power supply state and energy storage parameter, and the structured data set comprises the state quantity data set and the standard continuous quantity data set.

2. The COMS-CS hybrid architecture real-time data integration method for power plant power supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, Parameters of the event priority scoring function comprise a channel weight parameter, an index function, a data density and a density weight parameter.

3. The COMS-CS hybrid architecture real-time data integration method for power plant power supply chain link and electrical energy storage system monitoring according to claim 2, characterized in that, ​ 4. The COMS-CS hybrid architecture real-time data integration method for power plant power supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, ​ 5. The COMS-CS hybrid architecture real-time data integration method for power plant supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, ​ 6. The COMS-CS hybrid architecture real-time data integration method for power plant supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, ​ 7. The COMS-CS hybrid architecture real-time data integration method for power plant supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, ​ 8. The COMS-CS hybrid architecture real-time data integration method for power plant supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, The channel type mapping table is stored in a configuration file or a database table in text form, and is implemented by a hash structure for table lookup at runtime.

9. The COMS-CS hybrid architecture real-time data integration method for power plant supply chain link and electrical energy storage system monitoring according to claim 1, characterized in that, The structured packaging is completed based on the data type label to form a data set, which includes a channel number, a device end sampling time, raw data, and a data type label.

10. A COMS-CS hybrid architecture real-time data integration platform oriented to power plant power supply chain and energy storage system monitoring, characterized in that, The application comprises: a data packaging module configured to acquire raw data of power supply circuit devices, power storage devices, and associated power equipment and aggregate the raw data to a data acquisition master station, acquire a data type label of the raw data in the data acquisition master station through a channel type mapping table, and complete structured packaging based on the data type label to form a data set, wherein the raw data includes boiler operation data, steam turbine protection cabinet state data, bus voltage power supply data, flue gas online detection data, and power storage device charging and discharging state data, and the data type label includes a state variable and a continuous variable; a data shunting module configured to perform jump judgment on raw data of the state variable through a jump judgment function, filter the raw data through the jump judgment function as a state variable data set, perform residual deviation judgment on raw data of the continuous variable through a residual deviation function, filter the raw data through the residual deviation function as a continuous variable data set, the state variable data set includes power distribution switch state mutation data, and the continuous variable data set includes power storage charging and discharging parameter abnormal data; a data reasoning module configured to filter and reason the continuous variable data set through a fitting residual function with a weighted slope constraint, and finally output a filtered standard continuous variable data set, wherein parameters of the fitting residual function with the weighted slope constraint include a sliding window prediction value, a slope of a fitting line, and a channel characteristic adjustment factor; a data fusion module configured to perform time alignment on the state variable data set and the standard continuous variable data set, perform clustering processing on the state variable data set and the standard continuous variable data set through a sliding time window clustering function, and perform fusion calculation to generate a unified event record; a task execution module configured to identify an operable condition of the unified event record through an event priority scoring function, map the operable event to an actual power supply circuit adjustment task, a power storage charging and discharging control instruction, an alarm pushing instruction, and a historical data record operation, and bind the power supply circuit adjustment task, the power storage charging and discharging control instruction, the alarm pushing instruction, and the historical data record operation to generate a task structure set.

Images