Data acquisition method, device, medium and equipment for power system monitoring

Abstract

The application relates to the technical field of power system monitoring, and discloses a data acquisition method, device, medium and equipment for power system monitoring. The method comprises the following steps: analyzing monitoring data of a target power device to obtain analysis data, and performing fuzzy processing and multi-source data fusion; performing protocol adaptive conversion on the fused data to output standardized feature data; monitoring the real-time state of a communication link, switching to a local intelligent decision mode when the communication link is in an unusable state, predicting a future operation state based on historical standardized feature data, calculating a target adjustment amount through a multi-objective optimization algorithm, and performing local output; and when the communication link is in an available state, uploading the standardized feature data to a power monitoring master station. The above method improves the protocol compatibility of the device, has the ability of monitoring and optimizing the device in a network interruption scenario, and enhances the overall reliability and intelligent level of the power monitoring system in the face of device heterogeneity and environmental uncertainty.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring technology, and in particular to a data acquisition method, device, medium and equipment for power system monitoring. Background Technology

[0002] Power system monitoring technology is a key technology to ensure the safe, stable and economical operation of the power grid. With the large-scale access of distributed power sources and mobile emergency power sources, power monitoring systems need to collect data and monitor the status of various power equipment in real time. At present, field devices such as generator controllers, meters and protection devices have a variety of communication protocols, widely adopting standard industrial protocols such as Modbus-RTU, DLT645 and IEC104, as well as many proprietary protocols from various manufacturers, which leads to serious protocol heterogeneity problems in field data acquisition.

[0003] Traditional power monitoring data acquisition devices, such as remote terminal units (RTUs) or data acquisition terminals, typically use pre-built fixed protocol parsing libraries. This approach lacks the ability to identify and adaptively parse unknown or proprietary protocols. When new types of devices are connected or protocol versions are updated, manual intervention, program modification, and redeployment are required, resulting in poor system scalability, high deployment and maintenance costs, and difficulty in meeting the needs of rapid access and flexible configuration of power Internet of Things (IoT) devices. In addition, mobile emergency power supplies often operate in environments with poor network conditions, and their communication links with the power monitoring master station are prone to interruption or severe degradation. When communication is interrupted, local intelligent analysis and autonomous decision-making cannot be performed offline, meaning that the operating status of the power equipment cannot be optimized and controlled in real time when it is not under the monitoring of the master station, posing risks of safety hazards and low operating efficiency.

[0004] Therefore, a data acquisition method that is compatible with multiple protocols and has local intelligent decision-making capabilities under communication interruption conditions is urgently needed. Summary of the Invention

[0005] In view of this, this application provides a data acquisition method, device, medium and equipment for power system monitoring, the main purpose of which is to solve the technical problems in the prior art where power monitoring data acquisition devices are incompatible with multiple protocols and cannot make local intelligent decisions under the condition of communication interruption.

[0006] According to a first aspect of the present invention, a data acquisition method for power system monitoring is provided, the method being performed by an intelligent data acquisition device deployed on the power equipment side, comprising:

[0007] The monitoring data of the target power equipment is collected, the monitoring data is identified and parsed to obtain parsed data, and the parsed data is fuzzed and fused with multi-source data to generate fused data.

[0008] The fused data is subjected to protocol adaptive conversion to output standardized feature data;

[0009] The system monitors the real-time status of the communication link between the power monitoring master station and the power monitoring master station. When the communication link is unavailable, it switches to the local intelligent decision-making mode and predicts the future operating status of the target power equipment based on historical standardized feature data. It calculates the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and outputs the target adjustment amount locally.

[0010] When the communication link is available, the standardized feature data is uploaded to the power monitoring master station.

[0011] According to a second aspect of the present invention, a data acquisition device for power system monitoring is provided, comprising:

[0012] The data fusion module is used to collect monitoring data of the target power equipment, perform protocol identification and parsing on the monitoring data to obtain parsed data, and perform fuzzing processing on the parsed data and fusion with multi-source data to generate fused data;

[0013] The protocol conversion module is used to perform adaptive protocol conversion on the fused data and output standardized feature data;

[0014] The edge autonomous module is used to monitor the real-time status of the communication link with the power monitoring master station. When the communication link is unavailable, it switches to the local intelligent decision-making mode and predicts the future operating status of the target power equipment based on historical standardized feature data. It calculates the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and outputs the target adjustment amount locally.

[0015] The cloud collaboration module is used to upload the standardized feature data to the power monitoring master station when the communication link is available.

[0016] According to a third aspect of the present invention, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described data acquisition method for power system monitoring.

[0017] According to a fourth aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described data acquisition method for power system monitoring.

[0018] This invention provides a data acquisition method, device, medium, and equipment for power system monitoring. Through two processing stages—protocol identification and parsing, and protocol adaptive conversion—it achieves rapid matching of known protocols and dynamic parsing of unknown protocols, as well as automatic conversion of heterogeneous fused data into standardized feature data. This enables the device to adapt to the operation of multi-source heterogeneous protocols, allowing for plug-and-play functionality and online protocol learning for new devices without manual intervention. By continuously monitoring the communication link status, the device can smoothly switch to a local intelligent decision-making mode when the communication link is determined to be unavailable, and immediately initiate state prediction and multi-objective optimization calculations based on historical standardized feature data. Finally, the generated target adjustment amount is directly output through a local interface or control interface, ensuring that power equipment can maintain a safe and stable operating state even in extreme environments. Compared to existing technologies, which rely on fixed protocol libraries and require manual intervention to modify programs when connecting new devices, resulting in high expansion and maintenance costs, this application constructs a dual-mode protocol parsing mechanism. For unknown protocol data frames where feature code matching fails, it can autonomously infer the protocol type based on a dual-modal deep learning model, dynamically generate parsing rules, and cache them for subsequent rapid parsing. This enables plug-and-play functionality for new devices and online protocol learning without manual intervention, improving the device's runtime adaptability to multi-source heterogeneous protocols. Regarding network outage autonomy, existing technologies only have data caching capabilities after communication interruptions and cannot perform local intelligent analysis and autonomous decision-making in offline states. In contrast, this application continuously monitors... The system measures the communication link status and can smoothly switch to a local intelligent decision-making mode when communication is interrupted. It predicts the future operating status of the equipment based on a lightweight machine learning model and outputs the target adjustment amount locally through multi-objective optimization, enabling power equipment to maintain a safe and stable operating state even in network outage scenarios. Regarding resource scheduling mechanisms, existing devices lack dynamic scheduling capabilities when multiple tasks are concurrent, easily leading to system overload or delays in critical tasks. This application introduces a lightweight scheduling model that dynamically allocates resources for tasks such as data acquisition, protocol parsing, status prediction, and optimization calculations when the local computing resource occupancy rate exceeds a threshold, effectively avoiding the impact of resource bottlenecks on system real-time performance and reliability. In summary, this application improves the device's protocol compatibility, provides the ability to monitor and optimize equipment in network outage scenarios, and enhances the overall reliability and intelligence level of the power monitoring system in the face of equipment heterogeneity and environmental uncertainty.

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0021] Figure 1 A flowchart illustrating a data acquisition method for power system monitoring provided by an embodiment of the present invention is shown.

[0022] Figure 2 A flowchart illustrating another data acquisition method for power system monitoring provided by an embodiment of the present invention is shown.

[0023] Figure 3 This invention provides a schematic diagram of the structure of a data acquisition device for power system monitoring according to an embodiment of the present invention.

[0024] Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0025] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0026] This application provides a data acquisition method for power system monitoring, the method being executed by an intelligent data acquisition device deployed on the power equipment side, such as... Figure 1 As shown, the method includes the following steps:

[0027] 101. Collect monitoring data of the target power equipment, perform protocol identification and parsing on the monitoring data to obtain parsed data, and perform fuzzing processing on the parsed data and multi-source data fusion to generate fused data.

[0028] Specifically, monitoring data refers to the raw electrical signals or data packets collected in real time from target power equipment through various physical interfaces of intelligent data acquisition devices. Its format depends on the communication protocol used. Target power equipment typically includes generator controllers, protection devices, smart meters, etc. Protocol identification and parsing involves determining the protocol type and disassembling the monitoring data based on pre-set logical attribute tags within the device, extracting numerical points with physical meaning, and adding timestamps and quality stamps to generate structured parsed data. Logical attribute tags include protocol signatures, data parsing rule templates, etc. Data parsing rules refer to a set of logical instructions guiding how to extract and convert data point values ​​with physical meaning from raw data frames. Essentially, a set of standardized parsing logic defines the complete conversion process from binary messages to structured data. Specifically, data parsing rules include the following core logical elements: positioning logic, which specifies the starting position for data extraction within the data frame, typically expressed as a byte offset or register address, indicating the starting position of the target data in the message; and length logic, which specifies the number of bytes or bits occupied by the extracted data, determining the original data... The data parsing rules are as follows: physical length; format logic, which specifies the encoding format of the raw data and its interpretation method, including data type, byte order, and whether a sign bit is included; conversion logic, which specifies how to convert the raw numerical values ​​into engineering values ​​with physical meaning, usually manifested as linear transformation, table lookup mapping, or specific function calculation; and verification logic, which specifies how to verify the validity of the extracted data. These logical elements together constitute a complete data parsing rule. When the device executes protocol parsing, it completes the operations of positioning, extraction, format conversion, and validity judgment in sequence according to the logical order defined by the rule, and finally outputs parsed data with timestamps and quality stamps. Fusion processing is used to address uncertainties such as sensor noise and measurement errors that may exist in the parsed data. It maps precise numerical values ​​to membership vectors in fuzzy states such as normal, warning, and abnormal through a preset membership function. Multi-source data fusion takes multiple parsed data points from the same monitoring object and their corresponding membership vectors as a set of evidence, combines them with the confidence weight of the sensor, and uses an improved evidence theory synthesis rule to fuse them, finally generating a comprehensive confidence distribution and clarifying it into a representative numerical output, i.e., fused data.

[0029] In this embodiment, the original, messy field monitoring data is transformed into high-quality and highly consistent fused data, solving the data access barrier caused by the heterogeneity of protocols from multiple manufacturers' equipment and achieving compatibility with various standards and proprietary protocols. Through fuzzification processing and multi-source evidence fusion, the measurement noise and occasional failures of a single sensor are effectively suppressed, significantly improving the accuracy and reliability of the data. Furthermore, it provides reliable data input for subsequent protocol adaptive conversion and local intelligent decision-making, which is a prerequisite for achieving precise control and reliable autonomy in the data acquisition method of this application.

[0030] 102. Perform protocol adaptive conversion on the fused data and output standardized feature data.

[0031] Specifically, protocol adaptive conversion is a secondary processing step performed by the intelligent data acquisition device on the fused data. First, key features of the fused data, such as data point identifiers and source protocol types, are extracted and matched against feature signatures in the device's pre-built protocol rule base. If a match is successful, the fused data is directly converted into standard data points with standard physical semantics, standard units, and unified identifiers based on the corresponding data parsing rule template in the rule base. If a match fails, a deep learning model based on bimodal learning is activated to jointly analyze the contextual features and numerical statistical features of the fused data, inferring its physical semantics and mapping it to the closest standard data point definition in the pre-built power system monitoring data model. Simultaneously, a scaling conversion relationship is generated, ultimately completing the conversion to standard data points. After conversion, the device encapsulates the standard data points along with their metadata, such as original protocol information, conversion timestamps, and conversion confidence levels, into a message conforming to the target communication protocol, obtaining standardized feature data, specifically manifested as standard telemetry or teleindication data.

[0032] In this embodiment, the fused data processed internally by the device is uniformly converted into standardized feature data that can be identified and processed by the power monitoring master station. This achieves accurate conversion of known protocols and semantic mapping of unknown private protocols, significantly improving the device's compatibility with access to devices from different manufacturers and models. The output standardized feature data conforms to the power system monitoring data model specification, eliminating parsing obstacles on the master station side caused by data format differences, and achieving plug-and-play data access. Ultimately, it provides a standardized data payload for data upload when communication is normal, and also provides a unified format of historical data foundation for local intelligent decision-making after communication interruption, ensuring data consistency and reusability under both cloud-edge collaboration and edge autonomy modes.

[0033] 103. Monitor the real-time status of the communication link between the monitoring station and the power monitoring master station. When the communication link is unavailable, switch to the local intelligent decision-making mode and predict the future operating status of the target power equipment based on historical standardized feature data. Calculate the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and output the target adjustment amount locally.

[0034] Among them, the communication link refers to the communication connection established between the intelligent data acquisition device and the power monitoring master station for data transmission, usually based on 4G / 5G, fiber optic, or wireless private network; the unavailability state is a comprehensive assessment of communication quality, including two situations: communication quality degradation and communication interruption. Communication quality degradation means that the link still exists but its performance is severely degraded, while communication interruption means that the link is completely unavailable; the local intelligent decision-making mode is an autonomous operating state that the device actively switches to when it determines that the communication link is unavailable. In this mode, the device stops sending data to the master station and instead uses local computing power and built-in algorithm modules to achieve independent monitoring and control of the equipment; historical standardized feature data refers to the historical operating data sequence that conforms to the standard data model and has been adaptively converted by the protocol and stored in the local circular queue; the target adjustment amount is a specific numerical instruction generated by the device through prediction and optimization calculation to guide the adjustment of equipment operating parameters.

[0035] Specifically, this step involves a three-level progressive control logic. First, the device continuously monitors the heartbeat response latency and data packet delivery rate with the master station to assess the communication link quality in real time. When the heartbeat latency consistently exceeds a threshold and the delivery rate falls below a threshold, communication quality is deemed degraded. The device then activates a local caching mode, temporarily storing newly generated standardized feature data in a local circular queue while continuing reconnection attempts. When no heartbeat response is received and the number of failed reconnections reaches a preset value, communication is deemed interrupted, and the device immediately stops all uplink communication attempts and formally switches to local caching. In the intelligent decision-making mode, the device extracts a sequence of historical standardized feature data with configurable time windows from a local circular queue, and calls a built-in deep separable convolutional neural network to predict the key state parameters of the device at the next moment. Subsequently, with the optimization objective of approximating the predicted value to the grid target parameters pre-issued by the master station and minimizing the fluctuation of the control adjustment amount, a multi-objective cost function is constructed and the analytical solution of the target adjustment amount is obtained by differentiation. Finally, the target adjustment amount is output in the form of natural language commands or digital control commands through the local human-machine interface for reference by on-site operators or actuators.

[0036] In this embodiment, this step constructs a complete closed-loop process including communication sensing, state determination, mode switching, state prediction, and optimization decision-making. Through a two-level determination mechanism, it realizes the transition from cloud-edge collaboration to edge autonomy, avoiding frequent mode switching caused by network jitter, improving system stability. Furthermore, even in a completely offline state, the device can still predict the future state of the equipment based on local data and built-in models. Through multi-objective optimization, it generates scientific control instructions, enabling the power equipment to maintain a safe and stable operating state even when it is not under the monitoring of the main station. Finally, the locally output human-machine interaction instructions provide real-time and executable operation guidance for on-site maintenance personnel, which has important engineering practical value in scenarios such as disaster emergency response and remote area operations.

[0037] 104. When the communication link is available, upload the standardized feature data to the power monitoring master station.

[0038] Specifically, during continuous monitoring of the communication link status, when the device determines that the communication link meets the availability conditions, it enters the cloud-edge collaborative working mode. In this mode, the device uploads the generated standardized feature data to the power monitoring master station according to the preset transmission strategy. For critical telemetry data, the device typically adopts a real-time reporting strategy to ensure that the master station can promptly perceive the equipment's operating status. For non-critical data or parameters that change slowly, a periodic or change-based reporting strategy can be adopted to save communication bandwidth and device energy consumption. During the upload process, the device frames, encrypts, and verifies the data according to the requirements of the communication protocol to ensure the integrity and security of data transmission. At the same time, the device maintains real-time monitoring of the communication link during the data upload process. Once a link quality degradation or interruption is detected, a local caching or edge autonomous mechanism can be immediately triggered to form a complete closed-loop control logic.

[0039] In this embodiment, efficient data collaboration between edge devices and the cloud master station is achieved. By uploading standardized feature data to the master station, the master station can monitor the operating status of field equipment in real time, providing accurate data support for global scheduling, situational awareness, and remote control. The uploaded data adopts a unified standard model, eliminating the parsing obstacles of data from different manufacturers' equipment on the master station side, and realizing unified access and centralized management of multi-source heterogeneous devices. This step, through the linkage with the communication status monitoring mechanism, forms a dual-mode operation architecture of uploading during normal operation and autonomous operation during abnormal operation. This ensures centralized monitoring capabilities under normal conditions and enables devices to have autonomous survival capabilities under abnormal conditions, improving the adaptability and robustness of the entire power monitoring system in the face of complex network environments.

[0040] This invention provides a data acquisition method for power system monitoring. Through two processing stages—protocol identification and parsing, and protocol adaptive conversion—it achieves rapid matching of known protocols and dynamic parsing of unknown protocols. It also automatically converts heterogeneous fused data into standardized feature data, enabling the device to adapt to multi-source heterogeneous protocols during operation. This allows for plug-and-play functionality for new devices and online protocol learning without manual intervention. By continuously monitoring the communication link status, the device can smoothly switch to a local intelligent decision-making mode when the communication link is determined to be unavailable. It immediately initiates state prediction and multi-objective optimization calculations based on historical standardized feature data, ultimately outputting the generated target adjustment amount directly through a local interface or control interface. This ensures that power equipment maintains a safe and stable operating state even in extreme environments. In summary, this application improves the device's protocol compatibility, provides the ability to monitor and optimize equipment in network outage scenarios, and enhances the overall reliability and intelligence level of the power monitoring system in the face of equipment heterogeneity and environmental uncertainty.

[0041] This application provides another data acquisition method for power system monitoring, which is executed by an intelligent data acquisition device deployed on the power equipment side. The basic functions of the intelligent data acquisition device provided in this application will be described first:

[0042] On the cloud-based power IoT platform, i.e., the power monitoring master station side, a hierarchical device description library can be built and distributed to intelligent data acquisition devices deployed on the power equipment side. This is a preferred implementation method to realize that the devices have pre-installed logical attribute tags, physical attribute tags, and protocol rule libraries. The hierarchical device description library defines physical attribute tags and logical attribute tags for each type of power monitoring equipment, such as generator controllers and battery management systems. Specifically, the cloud platform constructs a structured hierarchical device description library based on the technical specifications of various monitoring devices, defining physical attribute tags for each type of equipment, including rated electrical parameters and communication interface types, as well as logical attribute tags, including supported communication protocol types, etc. The protocol description library, consisting of protocol signatures, data parsing rule templates, and business function classifications, is distributed to the intelligent data acquisition devices in the field. This provides the devices with a unified and scalable equipment knowledge base, enabling them to dynamically adapt to devices from different manufacturers and using different protocols without modifying their internal code. This achieves flexible access based on configuration-driven principles. The protocol rule library, as a core component of logical attribute tags, is also distributed or pre-installed in the devices. To enable the intelligent data acquisition devices to pre-configure logical attribute tags, physical attribute tags, and the protocol rule library, a hierarchical equipment description library can be built on the power monitoring master station side and distributed to the devices. The specific construction of the hierarchical equipment description library may include the following steps:

[0043] First, a three-level description structure is constructed. For example, a three-level structured description system containing equipment manufacturer information, equipment model information, and equipment instance information can be built in the cloud management platform. The manufacturer information layer records the equipment manufacturer, such as ABB and Siemens; the model information layer is refined to the specific product series, such as the UC-2000 series generator controller; and the instance information layer corresponds to each unique device installed on a specific power equipment and is associated with its asset code and installation location. This hierarchical structure is conducive to realizing refined modeling and batch operation and maintenance management of equipment.

[0044] Secondly, define attribute tags. At the device model level, define physical attribute tags and logical attribute tags for each type of power monitoring device. The physical attribute tags include rated voltage (e.g., 400V AC), rated current (e.g., 630A), communication baud rate (e.g., 9600bps), and interface physical type (e.g., RS485 two-wire system), which are used to describe the inherent electrical and communication characteristics of the device. The logical attribute tags include the supported communication protocol type (e.g., Modbus-RTU), protocol signature for quick message identification, data parsing rule templates specifying how to extract data points from the original message, business function classification (e.g., power generation control or battery monitoring), and synchronization strategy with the master station (e.g., transmission frequency), which are used to define the data and behavioral logic of the device.

[0045] Finally, the protocol rule base is integrated. As a core component of logical attribute tags, the protocol rule base exists in a configurable form. It contains parsing rules and data point mapping relationships for standard power protocols such as Modbus-RTU, DLT645, and IEC 104. For example, it defines function codes and register address mappings for Modbus-RTU. Simultaneously, the protocol rule base supports extensions to proprietary power protocols. Users can customize message structures, verification methods, and data point mappings using graphical tools based on the equipment manual, and add new rules to the base. Based on these steps, by distributing the constructed hierarchical equipment description base to the intelligent data acquisition device—which contains logical attribute tags, physical attribute tags, and the integrated protocol rule base—the initial configuration or knowledge base update of the device is completed, thus equipping it with the prior knowledge required for subsequent adaptive data acquisition and intelligent decision-making.

[0046] Specifically, the logical attribute tags define the communication protocol types, protocol signatures, data parsing rule templates, and business function classifications supported by the power monitoring equipment. The physical attribute tags define the rated voltage, rated current, communication baud rate, and interface physical type of the power monitoring equipment. The protocol rule base contains standard parsing rules and data point mapping relationships for power protocols such as Modbus-RTU, DLT645, and IEC 104, as well as proprietary power protocols.

[0047] To facilitate understanding, a specific example can be used to illustrate the composition of logical attribute tags, physical attribute tags, and protocol rule base. Assume a PowerGen-8000 diesel generator controller is connected. Its logical attribute tag defines its support for the Modbus-RTU protocol, with the protocol feature code being function code 0x03. The data parsing rule template specifies that register address 40001 corresponds to phase A voltage and needs to be divided by 10 to convert it to a volt value. The business function is classified as power generation monitoring. Its physical attribute tag defines its rated voltage as 400V, rated current as 630A, communication baud rate as 9600bps, and interface type as RS-485. Simultaneously, the protocol rule base has pre-set parsing rules for standard protocols such as Modbus-RTU and DLT645, and can be expanded to support proprietary protocols. For example, when connecting a manufacturer's proprietary protocol temperature transmitter, its message structure and data point mapping relationship can be customized through a configuration tool and added to the library, enabling the device to have adaptive data parsing capabilities.

[0048] This application provides another data acquisition method for power system monitoring, such as... Figure 2 As shown, it specifically includes:

[0049] 201. Obtain the parsed data by performing protocol identification and parsing on the monitoring data of the target power equipment.

[0050] Specifically, the original data frames of the monitoring data are matched with a preset set of protocol feature codes. When a target feature code in the protocol feature code set matches the original data frame, data point values ​​are extracted from the original data frame based on the target data parsing rules associated with the target feature code, and these data point values ​​are used as parsed data. When no feature code in the protocol feature code set matches the original data frame, the structural and statistical features of the original data frame are extracted and input into a preset bimodal deep learning model to infer the protocol type and obtain the inference result. A temporary parsing rule is generated based on the inference result, and data point values ​​are extracted from the original data frame based on the temporary parsing rule, and these data point values ​​are used as parsed data.

[0051] Among them, the bimodal deep learning model is a pre-trained neural network model used to infer protocol types and generate parsing rules for unknown or proprietary protocol data frames that fail to match feature codes. The model employs a bimodal input architecture, enabling joint learning of the structural and statistical features of the data frame, thus achieving effective identification of unknown protocol patterns. Specifically, the input of the bimodal deep learning model contains feature vectors from two modalities: a structural feature modality, reflecting the format structure and field layout information of the data frame. Extracted structural features include the total frame length, the position and value of the frame header start flag, the position and value of the function code or type code, the position and value of the address field, the position and value of the control field, the position and type of the check field, the interval length between fixed fields, and the occurrence position of variable fields. Structural features are typically represented as numerical vectors or one-hot encoded forms, used to characterize the fixed features of the data frame. The format template is one type of modality; the other is the statistical feature modality, which reflects the numerical distribution and statistical regularity information of the data frame. The extracted statistical features include the mean and variance of the values ​​at each byte position, the distribution of the difference between adjacent bytes, the byte value entropy, the frequency of bytes appearing in a specific interval, the checksum pattern, the run length of consecutive identical bytes, and the numerical fluctuation characteristics. Statistical features are used to capture the implicit regular patterns in the data frame, and are especially suitable for analyzing private protocols without fixed field definitions. At the model architecture level, the bimodal deep learning model adopts a two-stream network structure. The input of the structural feature modality is connected to several fully connected layers or one-dimensional convolutional layers to extract the high-order representation of the structural features. The input of the statistical feature modality is connected to several fully connected layers to extract the high-order representation of the statistical features. The representation vectors of the two modalities are concatenated or weighted and fused in the fusion layer, and then connected to several fully connected layers and the softmax output layer to finally output the probability distribution of the protocol type.

[0052] In this embodiment, the acquisition of parsed data includes the following three steps: Step 1: Protocol feature code matching. After the intelligent data acquisition device receives the raw data frame, it first extracts the set of protocol feature codes preset in the logical attribute label, such as the function code 0x03 of the Modbus-RTU protocol, the frame start symbol 0x68 of the DLT645 protocol, etc., and quickly compares them with the header of the data frame. If the match is successful, the protocol type is immediately determined and the process proceeds to Step 2. If the match fails with all preset feature codes, it is determined to be an unknown or private protocol, and the process proceeds to Step 3. Step 2: Rule template parsing. Based on the data parsing rule template corresponding to the successfully matched protocol feature codes, the data point values ​​are accurately extracted from the data frame. This template specifies the extraction position, byte order, data type, and engineering value conversion coefficient of the data points. At the same time, a local timestamp accurate to milliseconds is automatically added to each data point, and a quality stamp is generated based on the frame verification result and data rationality to complete the parsing. Step 3: Auxiliary inference and parsing. When it is determined to be an unknown or private protocol, the auxiliary parsing process is started. The device first extracts multi-dimensional features from the data frame, including structural features such as frame length and byte values ​​at specific positions, as well as statistical features such as byte value distribution and checksum patterns. These features are fed into a pre-trained deep learning model based on bimodal learning. By jointly learning the features of structural and statistical modalities, the model infers the most likely protocol type. The model not only outputs a type label but also dynamically generates a set of temporary parsing rules based on the patterns it learns internally. For example, it infers that the first two bytes from the 5th byte represent temperature, and the data type is a signed short integer, which needs to be divided by 10 to get the actual value. The device applies this rule to complete the data parsing and caches this new protocol. When receiving similar messages later, it can directly call the cached rules for fast parsing without going through the model inference again, thus achieving online learning and adaptation of new protocols. It can be seen that this step achieves efficient and accurate parsing of known protocols, while also possessing adaptive parsing and online learning capabilities for unknown or proprietary protocols, significantly improving protocol compatibility and deployment flexibility in complex industrial environments.

[0053] 202. Perform fuzzification processing on the parsed data and fuse it with multi-source data to generate fused data.

[0054] Specifically, a membership function is determined based on the sensor type corresponding to the analytical data, and the numerical values ​​of the analytical data are mapped to membership vectors at at least two fuzzy state levels. At least one set of analytical data from the target power equipment is integrated into an evidence body, and the basic reliability assignment of the evidence body is determined based on the membership vector and the data quality stamp corresponding to the membership vector. The basic reliability assignment is discounted and corrected based on the preset sensor reliability weights to obtain the corrected evidence body. Multiple corrected evidence bodies are fused according to the preset evidence theory synthesis rules to obtain a comprehensive reliability distribution. The fuzzy state level corresponding to the maximum reliability in the comprehensive reliability distribution is selected as the fused state of the target power equipment, and the clarified value of the fused state is output as the fused data.

[0055] The evidence theory synthesis rule is based on Dempster-Shafer evidence theory and has been improved and optimized for power monitoring scenarios. It is used to fuse multi-source sensor data from the same monitoring object to generate a highly reliable comprehensive state assessment. Within the evidence theory framework, the identification framework {normal, warning, abnormal} is considered a complete set of possible system states. Each sensor data point is treated as a piece of evidence, and its basic reliability allocation is determined based on the membership vector and quality stamp of that data point. The core improvement of this application lies in introducing sensor reliability weights for discount correction: Let the preset sensor reliability weight be w (0≤w≤1), then the corrected basic reliability allocation is m'(A)=w×m(A), where A is a non-empty proper subset. The remaining reliability (1-w) is allocated to the entire set representing uncertainty. This correction mechanism integrates the long-term statistical reliability prior information of the sensors into the fusion process, effectively suppressing... To mitigate the excessive influence of low-quality sensors on the fusion results, multiple evidence bodies, after discounting correction, are fused using the Dempster synthesis rule. The resulting comprehensive reliability distribution is then used to determine the fusion state, employing the maximum reliability criterion—selecting the proposition with the highest reliability as the final state of the current monitored object. To meet subsequent quantitative calculation requirements, the fusion state is clearly output: if the state is normal, the median or weighted average of the membership interval is output; if the state is warning or abnormal, a preset warning threshold is output. Through this improved evidence theory synthesis rule, this application achieves efficient fusion of multi-source heterogeneous data, significantly improving the accuracy and reliability of state monitoring.

[0056] In this embodiment, firstly, based on the fuzzification processing of the membership function, for each parsed data point, such as the winding temperature value from a PT100 temperature sensor, the device selects a triangular or trapezoidal membership function according to the historical error distribution characteristics of the sensor type, mapping the precise value into a membership vector at three fuzzy state levels: normal, warning, and abnormal. Specifically, the normal temperature range is set to 0-120℃, the warning range to 115-135℃, and the abnormal range to >130℃. When the measured temperature is 118℃, by calculating its membership degree in each fuzzy interval, a three-dimensional vector can be obtained, such as [0.8 (normal), 0.2 (warning), 0.0 (abnormal)]. This vector quantifies the probability that the measured value belongs to each fuzzy state. Secondly, an evidence body is constructed and a basic confidence assignment is calculated, involving different sensors from the same monitoring object. Data points, such as direct contact temperature sensors, infrared thermometers, and vibration-based indirect temperature estimates, are treated as a set of evidence. The basic confidence assignment of each evidence is determined by the previously obtained membership vector and the quality stamp generated during its analysis. The higher the quality stamp, the higher the initial confidence assigned to its membership vector. Then, improved evidence fusion based on preset confidence weights is performed. Specifically, the improved Dempster-Shafer (DS) evidence synthesis rule is used to fuse the evidence. Before classic DS synthesis, the basic confidence assignment of each evidence is discounted based on the confidence weights of various sensors preset in the logical attribute labels. For example, the weight of a direct contact temperature sensor is α=0.9, the weight of an infrared thermometer is β=0.7, and the weight of a vibration inference sensor is γ=0.5. The discounting formula can be expressed as: For all nonempty propositions A, the remaining confidence (1-w) is assigned to the entire set representing uncertainty, where, Assigning the initial basic reliability of the evidence to proposition A. The new basic reliability distribution after discount correction is defined as follows: w is the discount coefficient or reliability weight. For all non-complete set propositions A, the weight w represents the portion of the evidence source considered reliable, which is assigned to each specific proposition. The remaining reliability (1-w) is assigned to the complete set representing uncertainty. Essentially, this introduces prior information on sensor reliability obtained from domain expert knowledge or long-term statistics into the fusion process, avoiding excessive influence of low-quality or easily drifting sensor evidence on the fusion result. Finally, a fusion decision and clarification output are performed. The multiple evidence bodies after discount correction are fused using the DS synthesis formula to obtain a comprehensive reliability distribution, for example: {Normal: 0.82, Warning: 0.15, Abnormal: 0.01, Uncertain: 0.02}. The fusion result output is a comprehensive reliability distribution. The fuzzy level corresponding to the maximum reliability is taken as the fusion state of the monitored object, and the clarified value of the state is taken as the fusion data output. Finally, the fuzzy level corresponding to the maximum reliability is selected as the current fusion state of the monitored object. In order to realize subsequent quantitative calculation, the fuzzy state is clarified. If the state is normal, the median of its membership interval or the weighted average based on membership can be used as the representative value. If the state is warning or abnormal, the preset warning threshold (such as 130℃ or 135℃) is output. The final representative value or threshold is the output fusion data.

[0057] 203. Perform protocol adaptive conversion on the fused data and output standardized feature data.

[0058] Specifically, key features of the fused data are extracted and matched with feature signatures in a pre-defined protocol rule base. When a target feature signature matches a key feature, the fused data is converted into standard data points according to the data parsing rule template corresponding to the target feature signature. When no feature signature matches a key feature, a pre-defined bimodal deep learning model is used to jointly analyze the fused data, outputting the physical semantics, scaling transformation relationship, and transformation confidence of the fused data. The physical semantics are then mapped to the corresponding standard data point definition in the pre-defined power system monitoring data model. Based on the standard data point definition and scaling transformation relationship, the fused data is converted into standard data points. The standard data points, their corresponding metadata, and the transformation confidence are encapsulated into standardized feature data, which includes standard telemetry or teleindication data conforming to the target communication protocol.

[0059] In this embodiment, through this step, data from multi-source heterogeneous sensors is transformed into highly reliable, highly consistent, and anti-interference-capable fused data, providing high-quality input for subsequent intelligent decision-making. Specific steps include: First, key feature matching and rule lookup: The intelligent data acquisition device extracts key features of the fused data to be transformed, such as data point identifiers, source communication protocol types, and location information in the original message. These features are then quickly compared with protocol feature signatures in a pre-set protocol rule base. Protocol feature signatures include function codes, register address ranges, message structure templates, etc.; Second, standardized transformation based on the rule base: If a corresponding feature signature is matched in the protocol rule base, the data parsing rule template associated with the feature signature is called. Based on the template, the original data value is converted into data with clear physical meaning, standard units, and uniformity. The first step involves standard data points with data point identifiers, which are then scaled according to coefficients defined in the template to obtain directly usable engineering values. The second step involves semantic inference and mapping based on a large model. If no matching feature is found in the protocol rule base, the data point is determined to belong to a non-standard definition part of an unknown or proprietary protocol. In this case, the device calls a deep learning model based on bimodal learning. The model infers the most likely physical semantics by jointly analyzing the contextual features and numerical statistical features of the data point, and automatically maps it to the closest standard data point definition in the power system monitoring data model, while recommending reasonable scaling relationships. The third step involves encapsulating the generated standard data point, its metadata, and the conversion confidence level into a standard telemetry or teleindication data message conforming to the target communication protocol, such as IEC 60870-5-104, for uploading or local use.

[0060] 204. Monitor the real-time status of the communication link between the monitoring station and the power monitoring master station. When the communication link is unavailable, switch to the local intelligent decision-making mode.

[0061] Specifically, the system monitors the heartbeat response delay and data packet delivery rate between the system and the power monitoring master station. When the heartbeat response delay exceeds the heartbeat response delay threshold and the data packet delivery rate is lower than the data packet delivery rate threshold, the system determines that the communication quality has deteriorated, activates the local caching mode, and stores the standardized feature data in the local circular queue. When no heartbeat response is received and the number of failed connection retries with the power monitoring master station reaches a preset number, the system determines that the communication is interrupted, stops sending standardized feature data to the power monitoring master station, and switches to the local intelligent decision-making mode.

[0062] In this embodiment, the device continuously monitors the heartbeat response delay and uplink data packet delivery rate with the master station. When the heartbeat delay continuously exceeds the heartbeat response delay threshold T1 and the delivery rate is lower than the data packet delivery rate threshold R1, it is determined that the communication quality is severely degraded, and the local caching mode is activated to temporarily store the data to be uploaded in the local circular queue. If no heartbeat response is received continuously and the active reconnection attempt fails N times, it is finally determined that the communication is interrupted. After the interruption is determined, the device officially switches to the local intelligent decision-making mode, stops data upload attempts, activates the embedded lightweight machine learning model and multi-objective optimization algorithm module, and switches the human-machine interface to the local decision-making interface. This ensures that the device can autonomously and smoothly switch from the data upload mode to the autonomous mode with local intelligent analysis capabilities when network conditions deteriorate, thereby maintaining the critical monitoring and operation guidance capabilities for power equipment during network interruption.

[0063] Furthermore, if communication is interrupted, a mechanism for triggering and switching to the local intelligent decision-making mode is implemented through a state machine, comprising three core stages: monitoring, early warning, and switching. The first stage involves real-time monitoring and early warning of communication quality. The communication management thread built into the intelligent data acquisition device sends a heartbeat message to the power monitoring master station system platform of the Dianhong device every second, and simultaneously sends a packet of standard telemetry and teleindication data. The device continuously calculates two key indicators: heartbeat response latency (the time difference between sending a heartbeat and receiving confirmation from the power monitoring master station system, typically <200ms in a 4G / 5G private network environment); and data packet delivery success rate (the number of packets successfully received from the power monitoring master station system TCP within the past minute). The device compares the proportion of ACK-confirmed data packets with preset thresholds. The heartbeat response delay threshold T1 is defined as a heartbeat response delay exceeding 500ms for five consecutive times, and the data packet success rate threshold R1 is defined as a data packet success rate below 90%. When both conditions are met simultaneously, the device determines that the network connection is in a warning state of impending interruption or severe quality degradation. The second stage involves the activation of local caching mode. Once the warning state is entered, the device immediately activates local caching mode. In this mode, the device suspends initiating new data transmission connections to the power monitoring master station system. All newly generated standard telemetry and teleindication data are stored in a fixed-capacity local circular queue. The queue follows a first-in, first-out principle; when the queue is full, it automatically overwrites the oldest data. The device still maintains heartbeat transmission and reconnection attempts, but data upload is suspended, effectively providing a buffer period for possible temporary network jitter and avoiding frequent mode switching. The third stage involves communication interruption determination and mode switching. The device continuously attempts to... The device attempts to re-establish a stable connection with the power monitoring master station system. If no response is received for 10 consecutive heartbeats, and the TCP layer reconnection attempt fails 3 times consecutively, the device ultimately determines that the communication is interrupted. At this time, the device performs the following actions to formally switch to the local intelligent decision-making mode: completely stop all attempts to send data to the power monitoring master station system, release communication thread resources, then start the lightweight machine learning model and multi-objective optimization algorithm module that was originally in standby mode, allocate computing resources to them, and finally control the local touch screen display unit to automatically switch from the original data upload status interface to the local decision support interface. The interface will prominently display key operating status, early warning information, and subsequent control quantity adjustment suggestions / operation instructions. Based on this, through the above three-level progressive triggering and switching mechanism, the device achieves a smooth and reliable transition from cloud-edge collaboration to edge autonomy, ensuring that continuous intelligent operation support can still be obtained on site during the period when the connection to the power monitoring master station system is unavailable.

[0064] 205. Under the local intelligent decision-making mode, the future operating status of the target power equipment is predicted based on historical standardized feature data, and the target adjustment amount for adjusting the operation of the target power equipment is calculated through a multi-objective optimization algorithm.

[0065] Specifically, a deep separable convolutional neural network is invoked, using historical standardized feature data with configurable time windows as input, to output the predicted values ​​of key state parameters of the target power equipment at the next moment; with the key state parameter prediction values ​​approximating the power grid target parameters issued by the power monitoring master station and minimizing the fluctuation of the target adjustment amount as the optimization objective, a multi-objective cost function is constructed; the target adjustment amount is obtained by differentiating the multi-objective cost function.

[0066] In this implementation, the lightweight machine learning model employs a deep separable convolutional neural network (DSN) architecture. By decomposing standard convolution operations, it significantly reduces the number of parameters and computational complexity while maintaining feature extraction capabilities, making it suitable for deployment on resource-constrained edge devices. Specifically, the model's input is a configurable time window sequence of historical standard telemetry and teleindication data, such as data from the most recent dozens of sampling periods. The sequence is extracted from a local circular queue in chronological order. The model's output is the predicted value of key operating state parameters of the power equipment at the next moment. These key parameters include at least the grid connection point voltage, system frequency, and total active power. Furthermore, the model adopts a cloud-edge collaborative update mechanism: cloud-edge collaborative updates... The new feature is that, when communication is normal, the device receives global model incremental parameters from the power monitoring master station and securely integrates them with local model parameters, enabling the local model to absorb global knowledge trained based on multi-device data; and edge autonomous fine-tuning, that is, when communication is interrupted, the device uses continuously added operating data in the local cache to fine-tune the model through online learning algorithms, such as mini-batch gradient descent, so as to quickly adapt to the current operating characteristics and changes of the local power equipment. Finally, the predicted value output by the model comes with a confidence interval, which is calculated based on the statistical distribution of the model's historical prediction errors and is used to quantify the uncertainty of this prediction.

[0067] In practice, the time window length is set to 30 sampling points via a configuration file, corresponding to approximately 5 minutes of data. Assuming a sampling period of 10 seconds, after the local intelligent decision-making mode is triggered, the model extracts standard data from the local circular cache queue in chronological order for the most recent 30 moments, including voltage, current, frequency, and power, constructing a two-dimensional tensor as input. The model's output layer directly outputs the predicted grid-connected voltage, system frequency, and total output power for the next sampling moment. This allows the model to not only output predicted values ​​but also calculate and attach confidence intervals based on the statistics of the prediction errors of the past 100 times, quantifying the uncertainty of the prediction.

[0068] Furthermore, to address situations where local computing resources are limited or prediction confidence is too low, the device is designed with an adaptive prediction mode switching mechanism. When the switching conditions are met, the device switches from the aforementioned deep separable convolutional neural network mode to a lightweight linear prediction mode based on recursive least squares (RLS) with lower computational load. In this mode, the device uses the RLS algorithm to dynamically identify a short-term linear prediction model online based on pre-stored grid target parameters and recent historical data extracted from the local circular queue. Then, with the optimization objectives of approximating the grid target voltage and frequency while minimizing control quantity fluctuations, a multi-objective cost function is constructed, and the optimal adjustment amount of the control quantity is obtained analytically by differentiation.

[0069] Specifically, the prediction model embedded in the device employs a deepwise separable convolutional neural network. The network architecture decomposes standard convolution into depthwise convolution and pointwise convolution, significantly reducing model parameters and computational cost while maintaining feature extraction capabilities. This makes it ideal for deployment on resource-constrained edge devices. The model's input is a configurable time window sequence of historical standard telemetry and teleindication data. In practice, the time window length is set to 30 sampling points (approximately 5 minutes of data) via a configuration file. Assuming a sampling period of 10 seconds, after the local intelligent decision-making mode is triggered, the model extracts the most recent 30 moments of standard data from the local circular cache queue in chronological order, constructing a two-dimensional tensor as input. The model's output layer corresponds to three key regression tasks, directly outputting the predicted values ​​of three key state parameters of the monitored power equipment at the next sampling moment: grid connection point voltage, device frequency, and total output power. These three parameters are core indicators for evaluating the grid connection stability and load capacity of the monitored power equipment. The model adopts a dual-mode collaborative update strategy, led by the power monitoring master station system and supplemented by the edge. The cloud-edge collaborative update refers to updating the data while maintaining normal communication. The device periodically receives global model incremental parameters from the power monitoring master station system. It securely integrates these incremental parameters with local model weights, enabling the edge model to absorb general knowledge gained from training data from multiple monitored power devices across the network, maintaining synchronization with the global model. Edge autonomous fine-tuning, in a local intelligent decision-making mode during communication interruptions, utilizes newly generated operational data continuously accumulated in the local cache queue for online fine-tuning using a mini-batch gradient descent method. For example, every 10 new samples accumulated, the model performs forward and backward propagation once using this as the training set. Fine-tuning the weights allows the model to quickly adapt to specific changes in the current operating conditions of the monitored power device. Furthermore, the model not only outputs predicted values ​​but also calculates and adds a confidence interval to each predicted value based on historical statistics of the model's prediction errors. The device records the errors between the past 100 predicted values ​​and subsequent actual measurements, calculating their mean and standard deviation. The confidence interval is output as part of the prediction result, quantitatively reflecting the uncertainty of the current prediction, providing a basis for risk perception decisions by the subsequent optimization decision-making module.

[0070] Furthermore, under the local intelligent decision-making mode, the lightweight machine learning model performs parameter identification and state prediction based on the online least squares method, and calls a multi-objective optimization algorithm to generate quantitative control quantity adjustment suggestions or operation instructions. Specifically, this includes: state prediction and model identification, based on the grid target parameters obtained before the communication interruption and the historical operating data of the most recent m cycles extracted from the local circular queue, using the recursive least squares method to dynamically identify the short-term linear prediction model of the operating parameters of the monitored power equipment; multi-objective optimization parameter tuning, with the goal of approximating the grid target voltage and frequency and avoiding large fluctuations in control quantity, constructing a cost function, and obtaining the analytical solution of the optimal adjustment quantity by differentiation; mapping and output of control quantity adjustment suggestions or operation instructions, mapping the optimal adjustment quantity to natural language parameter tuning instructions, the mapping rules include fuzzy classification based on the positive and negative signs and magnitude of the adjustment quantity and direction and action mapping, and determining the parameter tuning object as excitation or throttle based on the contribution analysis of voltage and frequency deviation; output and update mechanism, after each decision, the data used this time is added to the local training set, triggering the model to fine-tune the model based on the small batch gradient descent based on the new data.

[0071] The specific implementation method is as follows:

[0072] First, state prediction and model identification: based on power grid target parameters obtained before the communication interruption. Based on the most recent m-cycle historical operating data extracted from the local circular queue, a short-term linear prediction model for the monitored power equipment / power equipment operating parameters is dynamically identified using the recursive least squares method. The specific expression is as follows:

[0073] ,

[0074] ,

[0075] In the formula, , These represent the current grid connection point voltage and the device frequency, respectively. For the previous control quantity, and These are all model coefficients to be solved, and the coefficient vector is... and The historical data matrix X and the observation vector V are updated online using the least squares formula, which is expressed as follows:

[0076]

[0077] Second, multi-objective optimization parameter tuning: to approximate the target voltage of the power grid. and frequency To avoid large fluctuations in the control quantity, a multi-objective cost function is constructed:

[0078] ,

[0079] Where α, β, and γ are configurable weight coefficients, specifically determined by differentiation.

[0080] make

[0081]

[0082] The analytical solution for the optimal adjustment amount is obtained:

[0083]

[0084] In the formula, A and B are the prediction base values ​​calculated based on the current state and the previous control quantity, respectively.

[0085] Third, the control quantity adjustment suggestions or operation command mapping and output will The mapping is translated into natural language-style parameter tuning commands, and the mapping rules include: based on The sign and magnitude of the voltage and frequency are fuzzy classified and mapped to direction and action. Based on the contribution analysis of voltage and frequency deviation, the parameter adjustment object is determined to be excitation or throttle. Finally, the expected effect information is output on the local interface.

[0086] Fourth, the output and update mechanism adds the data used in each decision to the local training set, triggering the model to fine-tune the model based on the new data in mini-batch gradient descent, continuously improving the accuracy of prediction and parameter tuning.

[0087] 206. When the communication link is available, upload the standardized feature data to the power monitoring master station.

[0088] In this embodiment, when the communication link is in a normal and available state, the standardized feature data can be directly uploaded to the power monitoring master station. See step 104 for details, which will not be elaborated here.

[0089] 207. If local computing resources are detected to be scarce, processing resources will be dynamically allocated through a preset lightweight scheduling model.

[0090] Specifically, the system detects the occupancy status of local computing resources; when the occupancy rate of local computing resources exceeds a preset threshold, it uses preset physical attribute labels and predicted values ​​of key status parameters as inputs to call a preset lightweight scheduling model to generate resource allocation decisions for data processing tasks.

[0091] In this embodiment, the model's input state is a multi-dimensional vector comprehensively representing the system's real-time operating status, mainly including three types of information: static resource endowment, derived from the device's computing power level defined in the preset physical attribute labels; dynamic task load, including the queue length and estimated computation time of each of the four core processing tasks: real-time monitoring data acquisition, protocol parsing, state prediction, and optimization calculation; and business urgency, calculated based on the urgency score derived from the key state parameters predicted by the lightweight machine learning model. The model's output action is a decision vector that directly guides resource allocation, containing two types of control commands: one is computing resource scheduling. The model dynamically allocates processor time slices to the four core tasks mentioned above, ensuring that their total is 100%. Another approach is to optimize storage resources by making binary decisions to determine whether to convert some historical data from high-precision storage to low-precision storage in order to free up memory space. The model continuously collects states, actions, rewards, and new state experiences and stores them in a local replay pool. It is trained online using random sampling. Its reward function is designed as a weighted combination of three indicators: task completion timeliness, prediction accuracy maintenance level, and overall device energy consumption. This guides the model to learn intelligent scheduling strategies that can balance system real-time performance, accuracy, and energy efficiency under resource-constrained conditions.

[0092] In a preferred embodiment, to reduce the runtime overhead of the scheduling model itself, the system introduces an energy-saving observation mode. After the model completes a round of dynamic scheduling, if the system detects that the lengths of each task queue tend to stabilize and the prediction accuracy of key state parameters recovers to an acceptable level, the scheduling model will automatically switch to a low-power observation mode. In this mode, the model suspends periodic rescheduling calculations and only maintains light monitoring of the system state until the system load or prediction accuracy deteriorates again to the threshold requiring re-triggering of scheduling. This mechanism effectively reduces the resource consumption of the resource scheduling and management function itself while ensuring the overall service quality.

[0093] 208. Map the optimal adjustment amount to executable operation instructions and output them through the local interface.

[0094] In this embodiment, fuzzy classification and direction determination are performed based on the sign and magnitude of the optimal adjustment amount. The specific control object, such as the excitation system or speed regulation system, is determined based on the contribution analysis of voltage and frequency deviations. Finally, the analysis results are converted into natural language parameter adjustment instructions that can be understood by on-site operators, or digital control instructions that can be directly issued to the equipment control unit. These instructions are displayed through a local human-machine interface. After each state prediction, optimization calculation, and instruction generation is completed, the data used in this operation will be added to the local training set to trigger the next round of online fine-tuning of the prediction model, thereby achieving continuous self-optimization of the model.

[0095] Furthermore, as Figure 1 In terms of specific implementation, this application provides a data acquisition device for power system monitoring, such as... Figure 3 As shown, the device includes: a data fusion module 301, a protocol conversion module 302, an edge autonomy module 303, and a cloud collaboration module 304.

[0096] The data fusion module 301 is used to collect monitoring data of the target power equipment, perform protocol identification and parsing on the monitoring data to obtain parsed data, and perform fuzzing processing on the parsed data and fusion with multi-source data to generate fused data;

[0097] Protocol conversion module 302 is used to perform adaptive protocol conversion on the fused data and output standardized feature data;

[0098] The edge autonomous module 303 is used to monitor the real-time status of the communication link with the power monitoring master station. When the communication link is unavailable, it switches to the local intelligent decision-making mode and predicts the future operating status of the target power equipment based on historical standardized feature data. It calculates the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and outputs the target adjustment amount locally.

[0099] The cloud collaboration module 304 is used to upload standardized feature data to the power monitoring master station when the communication link is available.

[0100] In specific application scenarios, the data fusion module 301 is specifically used to match the original data frame of the monitoring data with a preset set of protocol feature codes. When a target feature code in the protocol feature code set matches the original data frame, data point values ​​are extracted from the original data frame based on the target data parsing rules associated with the target feature code, and the data point values ​​are used as parsed data. When no feature code in the protocol feature code set matches the original data frame, the structural and statistical features of the original data frame are extracted and input into a preset bimodal deep learning model to infer the protocol type and obtain the inference result. Based on the inference result, a temporary parsing rule is generated, and data point values ​​are extracted from the original data frame based on the temporary parsing rule, and the data point values ​​are used as parsed data.

[0101] In specific application scenarios, the data fusion module 301 is also used to determine the membership function based on the sensor type corresponding to the parsed data, and map the numerical values ​​of the parsed data to membership vectors at at least two fuzzy state levels; integrate at least one set of parsed data of the target power equipment into an evidence body, and determine the basic reliability allocation of the evidence body based on the membership vector and the data quality stamp corresponding to the membership vector; discount and correct the basic reliability allocation based on the preset sensor reliability weight to obtain the corrected evidence body; fuse multiple corrected evidence bodies according to the preset evidence theory synthesis rules to obtain a comprehensive reliability distribution; select the fuzzy state level corresponding to the maximum reliability in the comprehensive reliability distribution as the fusion state of the target power equipment, and output the clarified value of the fusion state as the fusion data.

[0102] In specific application scenarios, the protocol conversion module 302 is specifically used to extract key features of the fused data and match the key features with feature signatures in the preset protocol rule base. When a target feature signature matches a key feature, the fused data is converted into standard data points according to the data parsing rule template corresponding to the target feature signature. When no feature signature matches a key feature, the fused data is jointly analyzed using a preset bimodal deep learning model, outputting the physical semantics, scaling transformation relationship, and transformation confidence of the fused data. The physical semantics are then mapped to the corresponding standard data point definition in the preset power system monitoring data model. Based on the standard data point definition and scaling transformation relationship, the fused data is converted into standard data points. The standard data points, their corresponding metadata, and the transformation confidence are encapsulated into standardized feature data, which includes standard telemetry or teleindication data conforming to the target communication protocol.

[0103] In specific application scenarios, unavailable states include communication quality degradation and communication interruption. The edge autonomous module 303 is specifically used to monitor the heartbeat response latency and data packet success rate between the module and the power monitoring master station. When the heartbeat response latency exceeds the heartbeat response latency threshold and the data packet success rate is lower than the data packet success rate threshold, communication quality degradation is determined, the local caching mode is activated, and standardized feature data is stored in the local circular queue. When no heartbeat response is received and the number of failed connection retries with the power monitoring master station reaches the preset number, communication interruption is determined, the sending of standardized feature data to the power monitoring master station is stopped, and the module switches to the local intelligent decision-making mode.

[0104] In specific application scenarios, the edge autonomous module 303 is used to call a deep separable convolutional neural network, which takes historical standardized feature data with configurable time windows as input and outputs the predicted value of the key state parameters of the target power equipment at the next moment; it constructs a multi-objective cost function with the goal of approximating the power grid target parameters issued by the power monitoring master station with the predicted value of the key state parameters and minimizing the fluctuation of the target adjustment amount; and it obtains the target adjustment amount by taking the derivative of the multi-objective cost function.

[0105] In specific application scenarios, the edge autonomous module 303 is also used to detect the occupancy status of local computing resources; when the occupancy rate of local computing resources exceeds a preset threshold, it uses preset physical attribute labels and key status parameter prediction values ​​as inputs to call a preset lightweight scheduling model to generate resource allocation decisions for data processing tasks.

[0106] It should be noted that other corresponding descriptions of the functional units involved in the data acquisition device for power system monitoring provided in this embodiment can be found in [reference needed]. Figure 1 and Figure 2 The corresponding description in [the document] will not be repeated here.

[0107] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described data acquisition method for power system monitoring.

[0108] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), including several instructions to enable a computer device (such as a personal computer, server, or network device) to execute the data acquisition method for power system monitoring in various implementation scenarios of this application.

[0109] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 The illustrated embodiment of a data acquisition device for power system monitoring, in order to achieve the above objectives, such as... Figure 4 As shown, this embodiment also provides a physical device for data acquisition for power system monitoring. This device includes a communication bus, a processor, a memory, and a communication interface. It may also include input / output interfaces and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the data acquisition method for power system monitoring described in the above embodiment.

[0110] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0111] Those skilled in the art will understand that the data acquisition physical device structure for power system monitoring provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0112] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs to be identified. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented through hardware. By applying the technical solution of this application, through the two processing stages of protocol identification and parsing and protocol adaptive conversion, it achieves rapid matching of known protocols and dynamic parsing of unknown protocols, as well as automatic conversion of heterogeneous fused data into standardized feature data. This enables the device to have adaptive capabilities for multi-source heterogeneous protocols during operation, and can complete plug-and-play of new devices and online protocol learning without manual intervention. By continuously monitoring the communication link status, the device can smoothly switch to local intelligent decision-making mode when the communication link is determined to be unavailable, and immediately initiate state prediction and multi-objective optimization calculation based on historical standardized feature data. Finally, the generated target adjustment amount is directly output through the local interface or control interface to ensure that the power equipment can maintain a safe and stable operating state even in extreme environments. In summary, this application improves the protocol compatibility of the device, has the ability to monitor and optimize the device in network outage scenarios, and enhances the overall reliability and intelligence level of the power monitoring system in the face of device heterogeneity and environmental uncertainty.

[0114] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0115] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A data acquisition method for power system monitoring, characterized in that, The method is executed by an intelligent data acquisition device deployed on the power equipment side, including: The monitoring data of the target power equipment is collected, the monitoring data is identified and parsed to obtain parsed data, and the parsed data is fuzzed and fused with multi-source data to generate fused data. The fused data is subjected to protocol adaptive conversion to output standardized feature data; The system monitors the real-time status of the communication link between the power monitoring master station and the power monitoring master station. When the communication link is unavailable, it switches to the local intelligent decision-making mode and predicts the future operating status of the target power equipment based on historical standardized feature data. It calculates the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and outputs the target adjustment amount locally. When the communication link is available, the standardized feature data is uploaded to the power monitoring master station; The process of obtaining parsed data by protocol identification and parsing of the monitoring data includes: The original data frames of the monitoring data are matched with a preset set of protocol feature codes; When a target feature code in the protocol feature code set matches the original data frame, data point values ​​are extracted from the original data frame based on the target data parsing rules associated with the target feature code, and the data point values ​​are used as the parsed data. When no feature code in the protocol feature code set matches the original data frame, the structural and statistical features of the original data frame are extracted and input into a preset bimodal deep learning model to infer the protocol type and obtain the inference result. A temporary parsing rule is generated based on the inference result. Data point values ​​are extracted from the original data frame based on the temporary parsing rule and the data point values ​​are used as the parsed data. The process of fuzzing the parsed data and fusing it with multi-source data to generate fused data includes: Based on the sensor type corresponding to the parsed data, a membership function is determined, and the numerical values ​​of the parsed data are mapped to membership vectors at at least two fuzzy state levels. The at least one set of the parsed data of the target power equipment is integrated into an evidence body, and the basic reliability assignment of the evidence body is determined based on the membership vector and the data quality stamp corresponding to the membership vector. The basic confidence distribution is discounted and corrected based on the preset sensor confidence weights to obtain the corrected evidence body. Multiple corrected evidence bodies are then fused according to the preset evidence theory synthesis rules to obtain the comprehensive confidence distribution. The fuzzy state level corresponding to the maximum confidence in the comprehensive confidence distribution is selected as the fusion state of the target power equipment, and the value of the fuzzy state after clarification is output as the fusion data.

2. The method according to claim 1, characterized in that, The step of performing protocol adaptive transformation on the fused data and outputting standardized feature data includes: Extract the key features of the fused data and match the key features with the feature signatures in the preset protocol rule base; When a target feature signature matches the key feature, the fused data is converted into standard data points according to the data parsing rule template corresponding to the target feature signature; When no feature signature matches the key feature, the fused data is jointly analyzed using a preset bimodal deep learning model, and the physical semantics, scaling transformation relationship and transformation confidence corresponding to the fused data are output. The physical semantics are then mapped to the corresponding standard data point definition in the preset power system monitoring data model. Based on the standard data point definition and the scaling transformation relationship, the fused data is converted into standard data points. The standard data points and corresponding metadata, along with the transformation confidence, are encapsulated into standardized feature data, wherein the standardized feature data includes standard telemetry or teleindication data that conforms to the target communication protocol.

3. The method according to claim 1, characterized in that, The unavailability state includes communication quality degradation and communication interruption; the real-time status of the communication link between the monitoring station and the power monitoring master station, when the communication link is unavailable, switches to the local intelligent decision-making mode, including: Heartbeat response latency and data packet success rate between the monitoring and power monitoring master station; When the heartbeat response delay exceeds the heartbeat response delay threshold and the data packet success rate is lower than the data packet success rate threshold, communication quality is determined to be degraded, local caching mode is activated, and the standardized feature data is stored in a local circular queue. When no heartbeat response is received and the number of failed connection retries with the power monitoring master station reaches a preset number, communication is determined to be interrupted, the transmission of the standardized feature data to the power monitoring master station is stopped, and the system switches to local intelligent decision-making mode.

4. The method according to claim 1, characterized in that, The prediction of the future operating state of the target power equipment based on historical standardized feature data, and the calculation of the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm, include: A deep separable convolutional neural network is invoked, taking the historical standardized feature data with a configurable time window as input, and outputting the predicted values ​​of the key state parameters of the target power equipment at the next moment; The multi-objective cost function is constructed with the predicted values ​​of the key state parameters approximating the target parameters of the power grid issued by the power monitoring master station, and with the goal of minimizing the fluctuation of the target adjustment amount. The objective adjustment amount is obtained by differentiating the multi-objective cost function.

5. The method according to claim 4, characterized in that, The method further includes: Detect the usage status of local computing resources; When the local computing resource utilization rate exceeds a preset threshold, the preset physical attribute label and the predicted value of the key state parameter are used as inputs to call the preset lightweight scheduling model to generate resource allocation decisions for the data processing task.

6. A data acquisition device for power system monitoring, characterized in that, include: The data fusion module is used to collect monitoring data of the target power equipment, perform protocol identification and parsing on the monitoring data to obtain parsed data, and perform fuzzing processing on the parsed data and fusion with multi-source data to generate fused data; The protocol conversion module is used to perform adaptive protocol conversion on the fused data and output standardized feature data; The edge autonomous module is used to monitor the real-time status of the communication link with the power monitoring master station. When the communication link is unavailable, it switches to the local intelligent decision-making mode and predicts the future operating status of the target power equipment based on historical standardized feature data. It calculates the target adjustment amount for adjusting the operation of the target power equipment through a multi-objective optimization algorithm and outputs the target adjustment amount locally. The cloud collaboration module is used to upload the standardized feature data to the power monitoring master station when the communication link is available. The data fusion module is also used to match the original data frames of the monitoring data with a preset set of protocol feature codes; When a target feature code in the protocol feature code set matches the original data frame, data point values ​​are extracted from the original data frame based on the target data parsing rules associated with the target feature code, and the data point values ​​are used as the parsed data. When no feature code in the protocol feature code set matches the original data frame, the structural and statistical features of the original data frame are extracted and input into a preset bimodal deep learning model to infer the protocol type and obtain the inference result. A temporary parsing rule is generated based on the inference result. Data point values ​​are extracted from the original data frame based on the temporary parsing rule and the data point values ​​are used as the parsed data. The data fusion module is also used to determine the membership function based on the sensor type corresponding to the parsed data, and to map the numerical value of the parsed data into a membership vector at at least two fuzzy state levels. The at least one set of the parsed data of the target power equipment is integrated into an evidence body, and the basic reliability assignment of the evidence body is determined based on the membership vector and the data quality stamp corresponding to the membership vector. The basic confidence distribution is discounted and corrected based on the preset sensor confidence weights to obtain the corrected evidence body. Multiple corrected evidence bodies are then fused according to the preset evidence theory synthesis rules to obtain the comprehensive confidence distribution. The fuzzy state level corresponding to the maximum confidence in the comprehensive confidence distribution is selected as the fusion state of the target power equipment, and the value of the fuzzy state after clarification is output as the fusion data.

7. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images