Intelligent decision-making method and device based on bus state

Abstract

The invention provides an intelligent decision-making method and device based on a bus state, and relates to the technical field of intelligent power grids. The method comprises the following steps: acquiring a power supply bus state decision result; searching in a local index database by taking the power supply bus state decision result as a query condition; according to a search result, extracting a corresponding decision basis interpretation text and maintenance operation guidance content from a local content library; and formatting and packaging the extracted decision basis interpretation text and the maintenance operation guidance content, and sending the result to terminal equipment of field maintenance personnel through a local communication interface. According to the method disclosed by the invention, a complex interpretation and guidance content generation process is moved forwards to a center side to be finished offline, and high-speed searching and pushing are carried out on an edge side by utilizing pre-generated indexes and contents; the technical problem that the resource-limited edge computing node is difficult to provide detailed decision basis interpretation and targeted operation guidance in real time is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and in particular to a method and device for intelligent decision-making based on bus status. Background Art

[0002] The power busbar in a large-scale power system is a critical hub connecting power generation, transmission, and distribution. Its operating status is crucial to the safe and stable operation of the entire power grid. To ensure reliable busbar operation, continuous and comprehensive monitoring is essential. Numerous sensors are typically deployed at substations and other sites to collect real-time data on various busbar operating parameters, including voltage, current, temperature, partial discharge, and vibration. These sensors continuously generate massive amounts of real-time monitoring data. Traditionally, all collected data is centrally transmitted via communication networks to a remote control center or data center for centralized processing, analysis, and storage. Professional maintenance personnel in the central system, or in the central system, use this data and historical experience to assess the busbar's operating status and make maintenance decisions. However, as the number of sensors and sampling frequency increase, the amount of raw data generated has exploded, placing significant pressure on communication network bandwidth. This is particularly true in remote areas with relatively weak communication infrastructure. Data transmission can suffer from high latency and poor reliability, directly impacting the immediacy and effectiveness of central decision-making.

[0003] To alleviate the processing pressure and communication bandwidth bottlenecks of central systems and improve the immediacy of decision-making, edge computing architectures have been introduced in recent years. In this architecture, edge computing nodes close to the data source (typically deployed within substations or regional data aggregation points) undertake some data processing and analysis tasks. Edge computing nodes receive real-time data collected by local sensors and use pre-deployed condition assessment models (typically trained based on extensive historical data) to process and analyze the data immediately, identifying abnormal bus patterns or signs of faults. Based on the assessment results, they quickly generate a condition assessment conclusion and maintenance decision recommendations for the power supply bus. For example, if the condition is normal, continue operation; if a minor anomaly is detected, further observation is recommended; if a significant anomaly is detected, scheduled maintenance is recommended; if a serious anomaly is detected, immediate shutdown and inspection is recommended. These edge-generated decision recommendations are transmitted via the communication network to the terminal devices of field maintenance personnel or displayed on local terminals.

[0004] After receiving the maintenance decision recommendations from the edge computing node, on-site maintenance personnel need to conduct on-site inspections of the power supply bus or perform corresponding maintenance operations based on the recommendations. However, the decision recommendations output by current edge computing nodes are usually highly abstract results, such as a status level, an exception code, or a simple instruction (such as "planned maintenance"). This output lacks the necessary transparency and detailed descriptions. On-site personnel often do not know the specific basis for the edge node to make this specific decision. For example, is it due to an abnormal increase in temperature at the busbar connection point, an increase in the partial discharge signal of the insulating component, or an abnormality in multiple monitoring parameters at the same time? Which sensor collected which data changes at which time point to trigger the early warning? What are the specific characteristics of these abnormal data? How does the edge node's status assessment model draw this conclusion based on this data?

[0005] Because edge nodes lack clear, specific explanations for their decisions, field maintenance personnel struggle to fully understand and trust their logic. This information asymmetry and lack of transparency can cause field personnel to question the accuracy of their decisions. To mitigate the risk of misjudgment, they may need to spend additional time and effort manually reviewing data (for example, logging into the central system to view raw data) or conducting preliminary on-site inspections to verify the edge node's judgment before executing. This undoubtedly increases the workload of field personnel, consumes valuable response time, and undermines the immediacy that edge computing is designed to provide.

[0006] In addition, even if the on-site personnel accept the decision recommendations of the edge node and are ready to implement them, they still need specific and targeted operational guidance to complete the subsequent maintenance tasks. Different bus types (such as different voltage levels and different structures), different abnormal modes (such as overheating, partial discharge, and abnormal vibration), and different on-site environmental conditions (such as high temperature, high humidity, rain and snow) may require different inspection steps, safety protection measures, and the preparation of different tools, equipment, or spare parts. The current edge nodes only provide high-level decision recommendations (such as "recommended planned maintenance"), but do not provide detailed operational guidelines that are closely related to the decision and targeted at the current specific situation.

[0007] The lack of targeted operating guidelines forces on-site personnel to rely on their personal experience and memory when performing maintenance tasks, or to spend time reviewing thick manuals and technical documentation. This not only reduces the efficiency of on-site operations, especially for relatively inexperienced new operators, but also increases the risk of incorrect operations due to incomplete information or misunderstandings. In serious cases, this can cause secondary damage to equipment and even endanger personnel safety.

[0008] Therefore, to fully leverage the effectiveness of power bus condition maintenance decision-making systems based on edge computing nodes, it's imperative to address the issue of insufficient edge node output information. Edge nodes need to enhance their output capabilities, enabling them to not only make decisions quickly but also generate and provide field maintenance personnel with clear, easy-to-understand explanations of the decision-making basis. Furthermore, they need to be able to automatically generate and push targeted, actionable maintenance operation instructions tailored to the decision, based on the current decision recommendation, identified anomaly patterns, the bus type involved, and potential field environmental information. This enhanced output capability plays a key role in increasing field personnel's trust in edge-side decisions, reducing unnecessary on-site review work, and improving the efficiency and safety of field maintenance operations. It represents a core technical challenge in achieving a closed-loop decision-making process at the edge and fully unleashing the potential of edge computing. Summary of the Invention

[0009] The purpose of the present invention is to provide an intelligent decision-making method and device based on the bus status. By moving the complex explanation and guidance content generation process to the center side for offline completion, and using pre-generated indexes and content on the edge side for high-speed search and push, it effectively solves the technical problem that resource-constrained edge computing nodes have difficulty in providing detailed decision-making basis explanations and targeted operation guidance in real time.

[0010] In a first aspect, the present invention provides an intelligent decision-making method based on bus status, comprising the following steps:

[0011] Obtain the power supply bus status decision result;

[0012] The power supply bus status decision result is used as a query condition to search in a local index library; the local index library stores a multi-dimensional index of decision-making explanation texts and maintenance operation guidance content pre-generated by the central side system;

[0013] Based on the search results, the corresponding decision-making basis explanation text and maintenance operation guidance content are extracted from the local content library; the local content library stores the decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system;

[0014] The extracted decision-making basis explanation text and maintenance operation guidance content are formatted and packaged, and sent to the terminal equipment of the on-site maintenance personnel through the local communication interface.

[0015] The intelligent decision-making method based on bus status provided by the present invention, at the site of a large-scale power system substation, faces edge computing nodes with limited computing and storage resources. After the decision results are output by the bus status assessment model, based on the results and specific site conditions (such as bus model, abnormal mode, environmental factors, etc.), clear decision-making basis explanations and targeted operational maintenance guidance are provided to on-site maintenance personnel at high speed and in real time, so as to overcome the problems of information opacity and insufficient guidance in existing solutions and effectively utilize the immediacy advantage of edge computing.

[0016] Furthermore, the step of using the power supply bus state decision result as a query condition and searching in the local index library includes:

[0017] A1. Construct a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and on-site environment information code; each layer of the multidimensional hash index corresponds to the decision type, abnormal mode, bus model, key monitoring parameter, and on-site environment;

[0018] A2. Use the constructed multi-dimensional hash index to search layer by layer to obtain the final index result.

[0019] A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

[0020] Furthermore, the specific steps in step A1 include:

[0021] A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window.

[0022] A12. If the CPU occupancy rate exceeds a preset first threshold or the memory occupancy rate exceeds a preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity;

[0023] A13. Calculate a data quality score based on the missing rate and the noise level. If the data quality score is lower than a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index;

[0024] A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

[0025] Furthermore, the specific steps in step A13 include:

[0026] A131. When a preset data quality assessment model is used to identify suspected abnormal data in sensor data, relevant historical data is obtained and the similarity between the suspected abnormal data and the historical data is calculated. If the similarity is lower than a preset similarity threshold, the suspected abnormal data is determined to be true abnormal data and an abnormality alarm is generated. Otherwise, the suspected abnormal data is determined to be noise data.

[0027] A132. If the abnormal alarm information is generated, the abnormal data ratio is calculated based on the actual abnormal data, and the data quality score is calculated based on the missing rate, the noise level, and the abnormal data ratio. Otherwise, the data quality score is calculated based on the missing rate and the noise level.

[0028] A133. If the data quality score is lower than the third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index.

[0029] Furthermore, the steps of formatting and packaging the extracted decision-making basis explanation text and maintenance operation instruction content and sending them to the terminal device of the on-site maintenance personnel through the local communication interface include:

[0030] Acquiring on-site information sent by a terminal device; the on-site information includes on-site environment brightness and on-site environment noise;

[0031] If the brightness of the on-site environment is lower than a preset fourth threshold, the extracted decision-making basis explanation text and maintenance operation instruction content are formatted and encapsulated into a broadcastable voice message, and the voice message is sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can listen to the voice message in a broadcast manner;

[0032] If the on-site environmental noise is greater than the preset fifth threshold, the extracted decision-making basis explanatory text and maintenance operation guidance content will be formatted and encapsulated into readable text information, and the text information will be sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can read the text information by reference.

[0033] Furthermore, the specific steps in step A14 include:

[0034] A141. Monitor the remaining storage space of edge computing nodes;

[0035] A142. If the remaining storage space is less than the preset storage threshold, an incremental rebuild is performed. Specifically, a multidimensional hash index is constructed for the newly added data based on the adjusted hash function and dimension weights, and the newly added index is merged into the original multidimensional hash index.

[0036] A143. If the remaining storage space is greater than or equal to the storage threshold, a full reconstruction method is adopted, specifically: the multidimensional hash index is rebuilt according to the adjusted hash function and dimension weights.

[0037] Furthermore, the specific steps in step A142 include:

[0038] When using the incremental reconstruction method, if the newly added index conflicts with the original multidimensional hash index, the conflict resolution strategy is dynamically adjusted according to the number and type of conflicts; the conflict resolution strategy includes open addressing, chain addressing and rehashing; wherein, the number of conflicts is calculated by a hash conflict detection algorithm, and the conflict types include data conflicts and structural conflicts. The data conflict refers to that the storage location pointed to by the newly added index is already occupied, and the structural conflict refers to that the dimensional structure of the newly added index is inconsistent with that of the original index.

[0039] In a second aspect, the present invention provides an intelligent decision-making device based on bus status, comprising:

[0040] An acquisition module is used to obtain the power supply bus status decision result;

[0041] A search module is used to use the power supply bus status decision result as a query condition to search in a local index library; the local index library stores a multi-dimensional index of decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system;

[0042] An extraction module is used to extract the corresponding decision-making basis explanation text and maintenance operation guidance content from the local content library based on the search results; the local content library stores the decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system;

[0043] The sending module is used to format and encapsulate the extracted decision-making basis explanation text and maintenance operation guidance content, and send them to the terminal equipment of the on-site maintenance personnel through the local communication interface.

[0044] Furthermore, the search module is used to perform the following operations when searching in the local index library using the power supply bus status decision result as a query condition:

[0045] A1. Construct a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and on-site environment information code; each layer of the multidimensional hash index corresponds to the decision type, abnormal mode, bus model, key monitoring parameter, and on-site environment;

[0046] A2. Use the constructed multi-dimensional hash index to search layer by layer to obtain the final index result.

[0047] A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

[0048] Furthermore, the search module is used to construct a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and field environment information code.

[0049] A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window.

[0050] A12. If the CPU occupancy rate exceeds a preset first threshold or the memory occupancy rate exceeds a preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity;

[0051] A13. Calculate a data quality score based on the missing rate and the noise level. If the data quality score is lower than a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index;

[0052] A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

[0053] From the above, it can be seen that the intelligent decision-making method based on the bus status provided by the present invention transfers the complex decision-making basis explanation and operation guidance content generation process from the resource-limited edge computing node to the center-side system with higher computing and storage capabilities for offline completion. The center-side system pre-generates a large number of detailed decision-making basis explanation texts and targeted operation guidance content based on the possible status, abnormal mode, decision type, bus model and potential environmental factors of the power supply bus, and builds a multi-dimensional index library. The edge computing node stores the pre-generated content library and index library issued by the center side. When the status assessment model of the edge node outputs the power supply bus status decision result and related key information, the edge node uses this information as an index to quickly search and extract the corresponding pre-generated explanation and guidance content through the index library. The extracted content is sent to the terminal device of the on-site maintenance personnel through the communication interface for presentation.

[0054] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of an intelligent decision-making method based on bus status provided by an embodiment of the present invention.

[0056] Figure 2 A schematic structural diagram of an intelligent decision-making device based on bus status provided in an embodiment of the present invention.

[0057] Description of labels:

[0058] 100, acquisition module; 200, search module; 300, extraction module; 400, sending module. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0060] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0061] Reference Attachment Figure 1 The present invention provides an intelligent decision-making method based on bus status, comprising the following steps:

[0062] Obtaining the power supply bus status decision result; the power supply bus status decision result includes the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and on-site environment information code;

[0063] The power supply bus status decision result is used as a query condition to search in the local index library; the local index library stores a multi-dimensional index of the decision-making explanation text and maintenance operation instructions pre-generated by the central side system; the multi-dimensional index corresponds to the combination of power supply bus status, abnormal mode, decision type, bus model, and on-site environment information;

[0064] Based on the search results, the corresponding decision-making explanation text and maintenance operation guidance content are extracted from the local content library; the local content library stores the decision-making explanation text and maintenance operation guidance content pre-generated by the central side system;

[0065] The extracted decision-making basis explanation text and maintenance operation guidance content are formatted and packaged, and sent to the terminal equipment of the on-site maintenance personnel through the local communication interface.

[0066] The acquisition of power bus status decision results can be accomplished by edge computing nodes. Edge computing nodes receive sensor data, analyze the data using pre-set models, and generate results containing the decision type, anomaly pattern code, bus model code, key monitoring parameter values ​​or their codes, and on-site environmental information codes. A local index library and local content library are pre-generated by the central system and synchronized to the edge computing node, ensuring offline search capabilities. Multi-dimensional indexes are constructed using a combination of information such as decision type, anomaly pattern, bus model, key monitoring parameters, and on-site environmental conditions as index keys, pointing to detailed explanations and operational instructions stored in the local content library. The search process uses the acquired decision result information as input and performs a match within the local index library. Content extraction retrieves the corresponding text content from the local content library based on the specific storage location or identifier obtained from the index search. Formatting and packaging convert the extracted text content into a format suitable for display or broadcast on terminal devices, such as text messages or voice files, and transmits it via a local communication interface (such as Wi-Fi, Bluetooth, or a wired connection). This allows field maintenance personnel to receive and present detailed decision-making rationale and operational instructions on their terminal devices.

[0067] This solution works based on "pre-processing" on the central side and "fast search" on the edge side. During system offline or off-peak periods, the central system, acting as a powerful content generation engine, exhaustively or based on preset rules, generates detailed explanations of decision-making rationales and targeted operational guidance for various possible busbar states, anomaly modes, decision types, busbar models, and on-site environment combinations. This content is then structured and stored in a content library. Simultaneously, the central side constructs a multi-dimensional index library, associating this content with corresponding scenario dimensions (decision type, anomaly mode, busbar model, etc.). The central side then distributes and stores the generated content library and index library to resource-constrained edge computing nodes. When the edge computing node's state assessment model runs online and outputs specific busbar state decision results (including key information such as decision type and anomaly mode), the edge node does not perform complex text generation. Instead, it uses this decision information as multi-dimensional query criteria to perform high-speed searches within the locally stored index library. The search results point to the corresponding pre-generated explanations and guidance content in the content library. The edge node quickly extracts this content, performs simple formatting and packaging, and immediately transmits it to the on-site maintenance personnel's terminal via the local communication interface. On-site terminals receive and present this information in a friendly manner, thereby achieving instant, clear and targeted information push.

[0068] Specifically, this method addresses the technical issue of edge computing nodes lacking detailed explanations and targeted operational guidance for their decision results. First, the edge computing node analyzes the busbar status and generates a decision result containing a variety of key information. This result serves as a query input for rapid search within a locally stored multi-dimensional index library. The multi-dimensional index design enables the search process to comprehensively consider multiple factors, including decision type, anomaly pattern, busbar model, key parameters, and site environment, thereby locating the most relevant detailed information. Based on the index search results, pre-prepared explanations of the decision basis and targeted maintenance and operation instructions are extracted from the local content library. These details explain the reasons for the edge node's decision and the subsequent actions that field personnel should take. Finally, the extracted detailed information is processed and sent to the field personnel's terminal devices. By providing detailed decision-making basis and operational guidance at the edge, this method increases field personnel's trust in the edge node's decisions, reduces the need for manual review, and improves the efficiency and safety of field operations.

[0069] In some specific embodiments, the edge computing node detects that a partial discharge anomaly has occurred in a power supply bus, and the state assessment model of the edge node outputs a decision result: the decision type is "planned maintenance recommended", the anomaly mode is coded as "partial discharge anomaly", the bus model is coded as "a certain model of bus", the key monitoring parameter value is "partial discharge pulse amplitude exceeds the threshold", and the on-site environmental information is coded as "normal environment". This decision result is used as a query condition and searched in the local index library. The multi-dimensional index stored in the local index library maps information combinations such as "planned maintenance recommended", "partial discharge anomaly", "a certain model of bus", "partial discharge pulse amplitude exceeds the threshold", and "normal environment" to specific decision-making basis explanation text and maintenance operation guidance content stored in the local content library. Based on the search results, explanatory text is extracted from the local content library, such as "The amplitude of the partial discharge pulses detected on this busbar model has consistently exceeded the safety threshold, indicating a partial discharge anomaly. Planned maintenance is recommended." It also contains operational instructions, such as "For partial discharge anomalies on this busbar model, please follow the following inspection steps: 1. Wear insulating gloves and goggles; 2. Scan the busbar connection points using a partial discharge detector; 3. Record the discharge location and intensity; 4. Develop a detailed maintenance plan based on the recorded results." The extracted text is formatted as a text message and sent via local Wi-Fi to the on-site maintenance personnel's tablet computer. This allows on-site personnel to instantly access the reasons for their decisions and the specific operational steps, improving response speed and operational accuracy.

[0070] In some embodiments, the step of using the power bus status decision result as a query condition and searching in the local index library includes:

[0071] A1. Construct a multidimensional hash index based on the decision type, anomaly pattern code, busbar model code, key monitoring parameter values ​​or their codes, and on-site environment information code. Each layer of the multidimensional hash index corresponds to the decision type, anomaly pattern, busbar model, key monitoring parameter, and on-site environment. The hash function for each layer of the multidimensional hash index uses a low-computational-complexity hash algorithm.

[0072] A2. Using the constructed multidimensional hash index, we perform a layer-by-layer hash search based on the decision type to obtain a candidate index set. We then perform a second-layer hash search within the candidate index set based on the anomaly pattern encoding, until hash searches across all dimensions are completed to obtain the final index result.

[0073] A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

[0074] The step of constructing a multidimensional hash index maps the multiple dimensions of the power bus status decision results into a multi-layer hash structure, with each layer of hash index corresponding to a single dimension. Using a low-computational hash algorithm, this facilitates rapid index construction and access on resource-limited edge computing nodes, laying the foundation for subsequent rapid searches. The step of performing a layer-by-layer search using the constructed multidimensional hash index begins with a hash calculation to quickly locate the corresponding candidate index set. Within this candidate set, a hash search is then performed based on the next dimension, narrowing the search layer by layer. This effectively locates the index that matches the query criteria in the multidimensional space, improving search speed. If the final index result is empty, a fuzzy matching strategy is employed. If an exact match fails, the search process is re-executed by reducing the matching accuracy of the key monitoring parameter values ​​or their encodings, gradually relaxing the matching criteria until a matching index result is found or a preset minimum matching accuracy threshold is reached. This resolves the issue of exact searches failing due to real-time data fluctuations or inaccuracies, and improves the search success rate.

[0075] Specifically, this technical solution addresses the challenges of ensuring the speed of searching for decision-making explanations and maintenance instructions that match power bus status decision results on resource-constrained edge computing nodes, given the large size of local index libraries. It also addresses the issue of key monitoring parameter values ​​potentially fluctuating slightly or not fully matching the exact values ​​stored in the index, leading to search failures. First, a multidimensional hash index is constructed based on the multiple dimensions of the power bus status decision results, with each dimension corresponding to a layer of the index. A hash function with low computational overhead is selected. This transforms complex query conditions into a multi-layer hash search process. Next, using the constructed multidimensional hash index, a layer-by-layer hash search is performed according to the order of dimensions such as decision type and anomaly pattern encoding. At each layer, hash calculations are used to quickly locate the candidate set for the next layer, gradually narrowing the search scope and ultimately obtaining an index result that accurately matches the query. This layer-by-layer hash search significantly improves search efficiency for multi-dimensional queries. Furthermore, if an exact match search fails to find a corresponding index result, a fuzzy matching mechanism is activated. This mechanism gradually reduces the matching accuracy requirements for key monitoring parameter values ​​or their encodings, for example, allowing for numerical deviations within a certain range, and then re-executes the layer-by-layer hash search process. This process can be iterated until an index result that meets the fuzzy matching conditions is found, or a preset minimum matching accuracy threshold is reached. As a result, even if there are slight differences between the key monitoring parameter values ​​collected in real time and the exact values ​​in the index library, the system can find the relevant index through fuzzy matching, ensuring that the necessary decision-making basis and operational guidance can be obtained and provided to on-site maintenance personnel, improving the robustness and success rate of the search.

[0076] In some specific embodiments, when an edge computing node receives a power bus status decision result, for example, the decision type is "planned maintenance," the abnormality mode is coded as "001" (indicating a partial discharge abnormality), the bus model is coded as "B-500kV," the key monitoring parameter value is "partial discharge intensity = 150pC," and the on-site environment information is coded as "01" (indicating a normal environment), a five-layer multidimensional hash index is first constructed based on this information. The first layer corresponds to the decision type, the second layer corresponds to the abnormality mode, the third layer corresponds to the bus model, the fourth layer corresponds to the key monitoring parameters, and the fifth layer corresponds to the on-site environment. Each layer uses a simple hash function, such as a modulo operation on the code or value. A layer-by-layer search is then performed. First, a hash calculation is performed on "planned maintenance" to locate the corresponding position in the first-layer index, resulting in a candidate set. Next, a hash calculation is performed on "001" within the candidate set to locate the position in the second-layer index, further narrowing the scope. This process is repeated until the hash search is completed using "partial discharge intensity = 150pC" and "01" in all dimensions, attempting to find the final index entry. If the precise search fails to find a match, for example, if the index library stores an entry corresponding to "partial discharge intensity = 140pC," fuzzy matching is initiated. The matching accuracy of the partial discharge intensity is reduced, for example, by allowing a deviation of ±10pC, and the search is repeated. At this point, the search condition becomes "partial discharge intensity within the range of 140pC to 160pC." A layer-by-layer hash search is performed again. During the fourth-layer search, if an entry for "partial discharge intensity = 140pC" exists in the index library and its value is within the fuzzy matching range, this entry may be found as the final index result. Based on the found index, the corresponding decision-making explanation text and maintenance operation instructions are extracted.

[0077] In some embodiments, the specific steps in step A1 include:

[0078] A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window.

[0079] A12. If the CPU usage exceeds the preset first threshold or the memory usage exceeds the preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity;

[0080] A13. Calculate a data quality score based on the missing rate and noise level. If the data quality score falls below a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index by 1 minus the missing rate minus the noise level.

[0081] A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

[0082] When building a multidimensional hash index on an edge computing node, the node's computing resource status, including CPU and memory usage, is monitored. The quality of the sensor data used for index construction is measured over a specific time period, quantified as the data missing rate and noise level. If the edge computing node's CPU or memory resources are detected to be insufficient (i.e., the usage exceeds a set threshold), the system dynamically selects a hash function with lower computational overhead to replace the original one for each layer of the multidimensional hash index construction process, mitigating the index construction's consumption of limited computing resources. Furthermore, a data quality score is calculated based on the missing rate and noise level of the sensor data, reflecting the data's reliability. A low data quality score indicates potential issues with key monitoring parameter data. In this case, the multidimensional hash index construction method is adjusted to reduce the weight of the key monitoring parameter dimension in the index. This reduced weight ensures that subsequent searches can still perform effective matches even when the data quality of that dimension is low. Finally, the multidimensional hash index structure is regenerated using the adjusted hash function and dimension weight parameters. In this way, the index construction process can adapt to the resource limitations of edge computing nodes and the quality changes of input data, improving the efficiency of index construction and the robustness of search.

[0083] In some specific embodiments, the edge computing node continuously monitors its own CPU and memory usage. For example, the first threshold of CPU occupancy is set to 80%, and the second threshold of memory occupancy is set to 90%. At the same time, the number of missing sensor data and the variance in the past minute are counted, and the missing rate and noise level are calculated. For example, the third threshold of the data quality score is set to 0.6. If at a certain moment, the CPU occupancy reaches 85%, exceeding the first threshold, the system determines that resources are tight and replaces the SHA-1 hash function originally used with the faster FNV hash function. At the same time, if it is calculated that the missing rate of sensor data is 0.2 and the noise level is 0.3, the data quality score is 1-0.2-0.3=0.5, which is lower than the third threshold. At this time, the weight of the key monitoring parameter dimension is adjusted to 0.5. Subsequently, the system uses the FNV hash function and the adjusted dimension weight parameters to rebuild the multidimensional hash index for the latest data.

[0084] In some embodiments, the specific steps in step A13 include:

[0085] A131. When a preset data quality assessment model identifies suspected abnormal data in sensor data, relevant historical data is obtained and the similarity between the suspected abnormal data and the historical data is calculated. If the similarity is lower than a preset similarity threshold, the suspected abnormal data is determined to be true abnormal data and an abnormality alarm is generated. Otherwise, the suspected abnormal data is determined to be noise data. The data quality assessment model includes a missing rate assessment module, a noise level assessment module, and a data status assessment module. The missing rate assessment module counts the number of missing data in each sensor data within a preset time window and calculates the missing rate. The noise level assessment module calculates the variance of each sensor data within the preset time window and determines the noise level. The data status assessment module analyzes the sensor data change trend, identifies sudden changes in data, and marks them as suspected abnormal data.

[0086] A132. If an abnormality alert is generated, the abnormal data percentage is calculated based on the actual abnormal data. A data quality score is also calculated based on the missing rate, noise level, and abnormal data percentage. The data quality score = 1 - missing rate - noise level - abnormal data percentage. Otherwise, the data quality score is calculated based on the missing rate and noise level.

[0087] A133. If the data quality score is lower than the third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index.

[0088] The data quality assessment model is used to identify suspected anomalies in sensor data. Specifically, the missing rate assessment module counts sensor data loss within a specified time period and calculates the missing data percentage. The noise level assessment module quantifies the noise level by calculating the fluctuation (e.g., variance) of the sensor data within a specified time period. The data status assessment module analyzes the changing trends of the sensor data, detects any sudden changes, and marks these sudden changes as potential suspected anomalies. Furthermore, for data marked as suspected anomalies, the system retrieves relevant historical data and calculates the degree of similarity between the suspected anomaly data and the historical data. By comparing the calculated similarity with a preset similarity threshold, the system can distinguish whether the suspected data reflects anomalies reflecting the actual device status or is simply noise caused by environmental interference. If the similarity falls below the threshold, the data is confirmed to be a true anomaly and an anomaly alarm is triggered. Otherwise, it is determined to be noise data. This allows the system to more accurately identify anomalies that truly require attention.

[0089] Specifically, this technical solution improves the data quality assessment method for key monitoring parameters by introducing a mechanism for identifying true anomalies in sensor data. First, the data quality assessment model performs a preliminary analysis of the sensor data, identifying missing data, noise, and potential suspected anomalies. When the data status assessment module detects a suspected anomaly, the system does not simply deem it low-quality data. Instead, it further determines its nature by comparing it with historical data. This determination process effectively distinguishes data reflecting a true abnormal busbar state from simple noise interference. If a true anomaly is determined and an alarm is generated, the data quality score is calculated based on the percentage of truly abnormal data, in addition to the missing data rate and noise level. A higher percentage of abnormal data indicates a greater proportion of true anomalies in the data, and the data quality score is correspondingly lowered. If the data quality score is determined to be noise or no suspected anomalies are detected, the data quality score is calculated based solely on the missing data rate and noise level. The resulting data quality score more accurately reflects the reliability of the key monitoring parameter data. If the calculated data quality score falls below a preset threshold, the data quality of the key monitoring parameter dimension is insufficient for full trust, and the system accordingly reduces the weight of that dimension in the multidimensional hash index construction. Lowering the weight means that during subsequent index building and search processes, the dimension's impact on the final results is weakened, thereby reducing index bias or search errors caused by data quality issues. This approach avoids misclassifying data that truly reflects device anomalies as low-quality data and excessively reducing its weight. This ensures that the system can still effectively use this information to make decisions even when the data contains true anomalies, avoids ignoring important early warning signals, and improves the accuracy and robustness of intelligent decision-making.

[0090] In some specific embodiments, consider an edge computing node monitoring the temperature sensor data of a bus connection point. Within a certain time window, the sensor data shows a sudden temperature rise. The data status assessment module marks this as a suspected anomaly. The system obtains the temperature history data of the connection point over the past period of time. Calculate the similarity between the current suspected abnormal data point and the historical data sequence (for example, using Euclidean distance or correlation coefficient). If the calculated similarity is lower than a preset similarity threshold (for example, a threshold of 0.1), the system determines that the temperature rise is a true anomaly and generates a temperature anomaly alarm. At the same time, the proportion of true abnormal data points in the time window to the total data points is counted, for example, 5%. Assume that the missing rate in the time window is 1% and the noise level (calculated based on variance) is 2%. Since an abnormal alarm is generated, the data quality score is calculated as 1-missing rate-noise level-abnormal data ratio = 1-0.01-0.02-0.05 = 0.92. If the preset third threshold is 0.95, then 0.92 is lower than 0.95, and the system determines that the data quality score of the temperature dimension is low, and its weight in the multidimensional hash index needs to be reduced. If the similarity is higher than the threshold (for example, the threshold is 0.1, and the calculated similarity is 0.8), the system determines that this temperature rise is noise. At this time, no abnormal alarm is generated, and the data quality score is calculated as 1-missing rate-noise level = 1-0.01-0.02 = 0.97. If the third threshold is 0.95, then 0.97 is higher than 0.95, and the weight of the temperature dimension remains unchanged or is adjusted according to normal rules. As a result, the system can more finely adjust the weights of the key monitoring parameter dimensions according to whether the data is a real abnormality or noise, thereby improving the adaptability of index construction and search.

[0091] In certain embodiments, the steps of formatting and packaging the extracted decision-making basis explanation text and maintenance operation instruction content and sending them to the terminal device of the on-site maintenance personnel via the local communication interface include:

[0092] Acquire the scene information sent by the terminal device; the scene information includes the scene environment brightness and the scene environment noise;

[0093] If the brightness of the on-site environment is lower than the preset fourth threshold, the extracted decision-making basis explanation text and maintenance operation instruction content are formatted and encapsulated into a broadcastable voice message, and the voice message is sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can listen to the voice message in a broadcasting manner;

[0094] If the on-site environmental noise is greater than the preset fifth threshold, the extracted decision-making basis explanatory text and maintenance operation guidance content will be formatted and encapsulated into readable text information, and the text information will be sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can read the text information by consulting.

[0095] Among them, obtaining the on-site information sent by the terminal device refers to receiving the environmental data collected by the terminal device. Specifically, the built-in light sensor and microphone of the terminal device can be used to collect the on-site environmental brightness and on-site environmental noise, and the data can be sent to the edge computing node wirelessly or wiredly.

[0096] The brightness of the on-site environment refers to the intensity of light on the site, and specifically lux can be used as the unit of measurement.

[0097] Among them, on-site environmental noise refers to the on-site sound intensity, which can be specifically measured in decibels (dB).

[0098] Among them, the preset fourth threshold refers to the brightness critical value used to determine whether the brightness of the on-site environment is suitable for reading text. It can be set according to the actual application scenario and human eye comfort, for example, set to 100 lux.

[0099] Among them, the preset fifth threshold refers to the noise critical value used to determine whether the on-site environmental noise is suitable for listening to speech, which can be set according to the actual application scenario and the human hearing threshold, for example, set to 70dB.

[0100] Among them, formatting the extracted decision-making basis explanation text and maintenance operation guidance content and encapsulating them into broadcastable voice information refers to converting the text content into audio data. Specifically, a text-to-speech (TTS) engine can be used to synthesize the text content into voice data, and the voice data can be encapsulated according to an audio format (such as WAV, MP3).

[0101] Among them, formatting and packaging the extracted decision-making basis explanation text and maintenance operation instruction content into accessible text information means organizing the text content into an easy-to-read format, which can be specifically packaged in plain text, HTML, PDF and other formats.

[0102] Among them, the local communication interface refers to the interface for data transmission between the edge computing node and the terminal equipment of the on-site maintenance personnel. It can be implemented by using short-range communication technologies such as Bluetooth, Wi-Fi Direct, USB interface or Ethernet interface.

[0103] Among them, listening to voice information through broadcasting means that after the terminal device receives the voice information, it plays the audio content through a speaker or headphones.

[0104] Reading text information by consulting means that after a terminal device receives the text information, the text content is displayed on the screen for the user to read.

[0105] This solution, combined with the technical solution of obtaining power bus status decision results, searching in the local index library, and extracting corresponding content from the local content library, enables the system to not only quickly and accurately obtain decision-making basis and operational guidance related to bus status, but also intelligently adjust the presentation of information based on the actual on-site environment. This combination solves the problem of information being difficult to effectively transmit in harsh on-site environments, improves the availability of information output by edge computing nodes and the efficiency of information reception by on-site personnel, thereby enhancing the practicality and reliability of the entire intelligent decision-making system.

[0106] Specifically, the solution first obtains current on-site environmental information, including brightness and noise levels, from the field maintenance personnel's terminal devices. After receiving this environmental data, the system evaluates it based on preset thresholds. If the brightness falls below a preset fourth threshold, indicating insufficient lighting, the system converts the previously extracted decision-making explanation text and maintenance operation instructions into voice format, encapsulates them, and sends them to the terminal device via a local communication interface. Upon receiving the voice information, the terminal device broadcasts it through an audio output device, allowing field personnel to obtain information aurally, eliminating the difficulty and visual fatigue associated with reading text in dimly lit environments. Conversely, if the noise level exceeds a preset fifth threshold, indicating a noisy environment, the system maintains the extracted text in text format, encapsulates it, and sends it to the terminal device via a local communication interface. Upon receiving the text information, the terminal device displays it on the screen, allowing field personnel to visually review the information, eliminating the difficulty and information loss associated with hearing voice in noisy environments. In this way, the system can dynamically select the most suitable information presentation format (voice or text) based on the actual situation of the on-site environment, ensuring that decision-making basis and operational guidance can be efficiently and reliably delivered to on-site maintenance personnel, overcoming the limitations of a single information format in different on-site environments.

[0107] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0108] This solution is deployed on an edge computing node at a substation. Field maintenance personnel carry a ruggedized tablet computer as a terminal device. The tablet has a built-in light sensor and microphone. After the edge computing node generates a decision-making explanation text and maintenance operation instructions based on the busbar status, it prepares to send them to the tablet computer. At this point, the tablet detects the current on-site ambient brightness of 50 lux and the ambient noise level of 65 dB. The tablet transmits these values ​​to the edge computing node via Bluetooth. After receiving the data, the edge computing node determines whether the ambient brightness of 50 lux is below the preset fourth threshold of 100 lux. This determination is positive. The edge computing node further determines whether the ambient noise level of 65 dB is above the preset fifth threshold of 70 dB. This determination is negative. Because the brightness is below the threshold, the edge computing node invokes a text-to-speech engine to synthesize the text content into speech data, such as an MP3 file. The edge computing node then sends this MP3 file to the tablet computer via Bluetooth. After receiving the MP3 file, the tablet plays the audio content through its built-in speaker, allowing the field maintenance personnel to listen to the audio and receive the decision-making basis and operation instructions.

[0109] Through the above technical solution, this application solves the problem that on-site maintenance personnel have difficulty effectively perceiving and understanding received information in low-brightness or high-noise environments. By sending voice messages when the on-site brightness is below a preset threshold, and sending text messages when the on-site noise is above a preset threshold, the accessibility and reception efficiency of information in different on-site environments are improved, and the reliability of the delivery of decision-making basis and operational guidance to on-site personnel is enhanced.

[0110] In some embodiments, the specific steps in step A14 include:

[0111] A141. Monitor the remaining storage space of edge computing nodes;

[0112] A142. If the remaining storage space is less than the preset storage threshold, an incremental rebuild is performed. Specifically, a multidimensional hash index is constructed for the newly added data based on the adjusted hash function and dimension weights, and the newly added index is merged into the original multidimensional hash index.

[0113] A143. If the remaining storage space is greater than or equal to the storage threshold, a full reconstruction method is used. Specifically, the multidimensional hash index is rebuilt based on the adjusted hash function and dimension weights.

[0114] Monitoring the remaining storage space of an edge computing node refers to obtaining the currently available storage capacity of the edge computing node. This can be achieved by querying the file system status interface provided by the operating system, for example, by calling a system API to obtain the available space size of a specific storage volume.

[0115] The preset storage threshold refers to a pre-set storage space value or ratio. It can be determined based on the total storage capacity of the edge computing node, the estimated index size, and the minimum storage space required for system operation. For example, it can be set to 10% of the total storage space or a fixed number of bytes.

[0116] Incremental rebuilding doesn't rebuild the entire index from scratch. Instead, it only rebuilds the index based on newly generated or received data since the last index build or update. This can be achieved by maintaining a data processing timestamp or data version number, and only processing data with a timestamp later than the last index update time.

[0117] New data refers to data that arrives at the edge computing node after the last multi-dimensional hash index is built or updated. This can be achieved by using a data receiving queue or logging to mark or collect data that has not yet been covered by the current index.

[0118] Building a multidimensional hash index involves using a hash function to generate hash values ​​based on the data's multiple dimensional characteristics, and then storing the data items or their references in the index structure. This can be achieved using a multi-layer hash table or hash tree structure, with each layer corresponding to a data dimension and mapped using a corresponding hash function.

[0119] The adjusted hash function and dimension weights refer to the hash function and the importance of each dimension in index construction, which are dynamically selected or calculated based on the resource status of the edge computing node (such as CPU and memory utilization) and data quality (such as missing rate and noise level). This can be achieved by selecting a function with lower computational complexity from a preset set of hash functions, or by adjusting the combination of dimension hash values ​​or the distribution strategy of the hash table based on the results of data quality assessment.

[0120] Merging a new index into the original multi-dimensional hash index means integrating the index structure built for the new data with the existing multi-dimensional hash index structure. This can be achieved by using conflict resolution strategies such as chain addressing, open addressing, or rehashing, inserting the new index item into the corresponding position of the original index to resolve potential hash conflicts.

[0121] The full rebuild method involves building a new multidimensional hash index from scratch based on all relevant data using an adjusted hash function and dimension weights. This is achieved by iterating over all data items to be indexed, applying the adjusted hash function and dimension weights in sequence to calculate the index position, and then storing the data items or their references into the new index structure.

[0122] Rebuilding a multidimensional hash index refers to generating a new multidimensional hash index structure based on the current hash function and dimension weight parameters. This can be achieved by allocating new memory space or storage files and then mapping data items to the new index structure.

[0123] Specifically, when rebuilding a multidimensional hash index, this solution first monitors the remaining storage space on edge computing nodes. By obtaining information about currently available storage resources, the system determines whether it has the storage capacity required for a full rebuild. If the remaining storage space is below a preset storage threshold, indicating limited storage resources, the system then uses an incremental rebuild. In this incremental rebuild, the system does not process all historical data. Instead, it constructs a new index portion containing the newly received data since the last index update, using a hash function and dimension weights adjusted for resource availability and data quality. This newly constructed index portion is then merged into the existing multidimensional hash index. This approach avoids the large temporary storage space required to rebuild a new index for the entire dataset, significantly reducing storage overhead and enabling index updates even on edge nodes with limited storage resources. If the remaining storage space is greater than or equal to the preset storage threshold, indicating sufficient storage resources, the system then uses a full rebuild. In this full rebuild, the system constructs a new multidimensional hash index from scratch based on all relevant data, using a hash function and dimension weights adjusted for resource availability and data quality. This approach can ensure the integrity and consistency of the index structure when resources permit. By dynamically selecting the index reconstruction strategy based on the remaining storage space of the edge computing node, this solution effectively solves the problem of insufficient storage space that may result from rebuilding the entire index in a resource-constrained environment, and improves the system's adaptability and operational stability under different resource conditions. This solution of selecting the reconstruction strategy based on storage resources, combined with the solution of adjusting the hash function and dimension weights based on CPU, memory, and data quality, enables edge nodes to not only adapt to changes in computing resources and data quality to optimize the index structure, but also adapt to changes in storage resources to optimize the index update process, jointly improving the robustness and practicality of the system in edge environments.

[0124] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0125] A multidimensional hash index management module is deployed on an edge computing node. This module is responsible for building, updating, and maintaining the multidimensional hash index. When the index needs to be rebuilt based on the updated hash function and dimension weights, the module first calls the operating system interface to obtain the remaining storage space on the current file system mount point. For example, the available space obtained is 500MB. The preset storage threshold is set to 1GB. Because 500MB is less than 1GB, the system determines that the remaining storage space is insufficient for a full rebuild. At this point, the system triggers the incremental rebuild process. The system identifies new sensor data received since the last index update, such as busbar operating parameter data collected within the last hour. For this new data, a new multidimensional hash index component is constructed using a hash function (for example, switching from SHA-256 to MurmurHash) and dimension weights (for example, reducing the weight of key monitoring parameters) adjusted based on current CPU and memory usage and data quality assessment results. After construction is complete, this new index component is merged with the original multidimensional hash index currently stored locally. During the merge process, a chain address method is used to handle potential hash conflicts. This involves storing conflicting data item references in a linked list format within the hash bucket. If the remaining storage space detected is 2GB, which is greater than or equal to the 1GB threshold, the system triggers a full rebuild. At this point, the system traverses all historical and newly added data to be indexed, constructing a new multidimensional hash index structure from scratch using the adjusted hash function and dimension weights. Once this structure is complete, the existing index is replaced.

[0126] Through the above technical solution, this application solves the problem that simply rebuilding the entire index may require a large amount of storage space when rebuilding a multidimensional hash index based on the updated hash function and dimension weights on the edge computing node. This solution monitors the remaining storage space of the edge computing node and dynamically selects incremental reconstruction or full reconstruction based on whether the storage space is below a preset threshold. This effectively reduces the storage resource occupation of the index reconstruction process, avoids index reconstruction failures or impacts on the normal operation of the system due to insufficient storage space, and improves the reliability and continuous operation capability of the system in edge environments with limited storage resources.

[0127] In some embodiments, the specific steps in step A142 include:

[0128] When using the incremental reconstruction method, if the newly added index conflicts with the original multidimensional hash index, the conflict resolution strategy is dynamically adjusted based on the number and type of conflicts. The conflict resolution strategies include open addressing, chain addressing, and rehashing. The number of conflicts is calculated using a hash conflict detection algorithm, and the conflict types include data conflicts and structural conflicts. A data conflict refers to a storage location pointed to by the newly added index being occupied, and a structural conflict refers to a dimensional structure inconsistency between the newly added index and the original index.

[0129] When using incremental reconstruction, the system detects conflicts between the newly added index and the original multidimensional hash index. Conflict detection is performed using a hash conflict detection algorithm, which counts the number of conflicts. The system also identifies the type of conflict. These conflict types include data conflicts and structural conflicts. A data conflict occurs when the storage location pointed to by the newly added index is occupied. A structural conflict occurs when the dimensional structure of the newly added index is inconsistent with that of the original index. Based on the calculated number of conflicts and the identified conflict type, the system selects a conflict resolution strategy from a set of conflict resolution strategies. These include open addressing, chain addressing, and rehashing. This selection process is adjusted based on the number and type of conflicts.

[0130] Specifically, when the storage space of an edge computing node falls below a threshold, it uses an incremental reconstruction method to update the multidimensional hash index to save computing resources and storage space. During this process, the newly added index data needs to be merged into the existing index structure. The newly added index may conflict with the original index, affecting the efficiency and accuracy of index searches. This solution obtains conflict information by detecting and counting the number of conflicts and identifying the conflict type (data conflict or structural conflict). Based on this information, the system selects a conflict resolution strategy that is appropriate for the number and type of conflicts. For example, when the number of conflicts is small and primarily data conflicts, open addressing may be selected; when the number of conflicts is large or there are structural conflicts, chain addressing or rehashing may be selected. This approach of adjusting the strategy based on the conflict situation improves conflict handling efficiency and reduces unnecessary computational overhead. In resource-constrained environments at edge computing nodes, it can manage the index update process, ensure index validity, and ensure the performance of subsequent index-based search operations, thereby supporting power bus status decision-making.

[0131] In some specific embodiments, the remaining storage space of the edge computing node falls below a storage threshold. The system uses an incremental reconstruction approach to update the multidimensional hash index. When a busbar status decision result is generated and an index needs to be built and added to the local index repository, the system performs a hash calculation to determine the storage location of the newly added index in the multidimensional hash index. If the location is occupied, a conflict is detected. The hash conflict detection algorithm calculates the number of conflicts. For example, if two newly added indexes map to the same location, the number of conflicts is two. The system also identifies the type of conflict. If the conflict is caused solely by an occupied storage location, it is considered a data conflict. If the number or order of dimensions of the newly added index is inconsistent with the dimensional structure of the original index at that location, it is considered a structural conflict. The system selects a conflict resolution strategy based on the number and type of conflicts. For example, if the number of conflicts is less than or equal to five and the conflict type is a data conflict, the system uses open addressing, searching for the next available location to store the newly added index through linear probing. If a structural conflict is detected, regardless of the number of conflicts, the system uses a chain addressing method, creating a linked list at that location and adding the newly added index to the linked list. In this way, the system uses an appropriate conflict resolution strategy based on the conflict situation to complete the incremental update of the index.

[0132] Please refer to Figure 2 , Figure 2 In some embodiments of the present invention, an intelligent decision-making device based on bus status is provided. The intelligent decision-making device based on bus status is integrated into a back-end control device in the form of a computer program and includes:

[0133] The acquisition module 100 is used to obtain the power supply bus state decision result; the power supply bus state decision result includes the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and field environment information code;

[0134] Search module 200 is used to use the power supply bus status decision result as a query condition and search in the local index library; the local index library stores a multi-dimensional index of the decision basis explanation text and maintenance operation instructions pre-generated by the central side system; the multi-dimensional index corresponds to the combination of power supply bus status, abnormal mode, decision type, bus model, and on-site environment information;

[0135] Extraction module 300, for extracting corresponding decision-making explanation text and maintenance operation guidance content from the local content library based on the search results; the local content library stores the decision-making explanation text and maintenance operation guidance content pre-generated by the central side system;

[0136] The sending module 400 is used to format and encapsulate the extracted decision basis explanation text and maintenance operation guidance content, and send them to the terminal device of the on-site maintenance personnel through the local communication interface.

[0137] In some embodiments, the search module 200 performs the following operations when searching the local index library using the power bus status decision result as a query condition:

[0138] A1. Construct a multidimensional hash index based on the decision type, anomaly pattern code, busbar model code, key monitoring parameter values ​​or their codes, and on-site environment information code. Each layer of the multidimensional hash index corresponds to the decision type, anomaly pattern, busbar model, key monitoring parameter, and on-site environment. The hash function for each layer of the multidimensional hash index uses a low-computational-complexity hash algorithm.

[0139] A2. Using the constructed multidimensional hash index, we perform a layer-by-layer hash search based on the decision type to obtain a candidate index set. We then perform a second-layer hash search within the candidate index set based on the anomaly pattern encoding, until hash searches across all dimensions are completed to obtain the final index result.

[0140] A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

[0141] In some embodiments, the search module 200 performs the following when constructing a multidimensional hash index based on the decision type, the abnormal mode code, the bus model code, the key monitoring parameter value or its code, and the field environment information code:

[0142] A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window.

[0143] A12. If the CPU usage exceeds the preset first threshold or the memory usage exceeds the preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity;

[0144] A13. Calculate a data quality score based on the missing rate and noise level. If the data quality score falls below a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index by 1 minus the missing rate minus the noise level.

[0145] A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

[0146] In some embodiments, the lookup module 200 calculates a data quality score based on the missing rate and the noise level, and reduces the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index if the data quality score is lower than a preset third threshold.

[0147] A131. When a preset data quality assessment model identifies suspected abnormal data in sensor data, relevant historical data is obtained and the similarity between the suspected abnormal data and the historical data is calculated. If the similarity is lower than a preset similarity threshold, the suspected abnormal data is determined to be true abnormal data and an abnormality alarm is generated. Otherwise, the suspected abnormal data is determined to be noise data. The data quality assessment model includes a missing rate assessment module, a noise level assessment module, and a data status assessment module. The missing rate assessment module counts the number of missing data in each sensor data within a preset time window and calculates the missing rate. The noise level assessment module calculates the variance of each sensor data within the preset time window and determines the noise level. The data status assessment module analyzes the sensor data change trend, identifies sudden changes in data, and marks them as suspected abnormal data.

[0148] A132. If an abnormality alert is generated, the abnormal data percentage is calculated based on the actual abnormal data. A data quality score is also calculated based on the missing rate, noise level, and abnormal data percentage. The data quality score = 1 - missing rate - noise level - abnormal data percentage. Otherwise, the data quality score is calculated based on the missing rate and noise level.

[0149] A133. If the data quality score is lower than the third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index.

[0150] In certain embodiments, the sending module 400 is used to format and encapsulate the extracted decision-making explanation text and maintenance operation instruction content and send them to the terminal device of the on-site maintenance personnel through the local communication interface.

[0151] Acquire the scene information sent by the terminal device; the scene information includes the scene environment brightness and the scene environment noise;

[0152] If the brightness of the on-site environment is lower than the preset fourth threshold, the extracted decision-making basis explanation text and maintenance operation instruction content are formatted and encapsulated into a broadcastable voice message, and the voice message is sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can listen to the voice message in a broadcasting manner;

[0153] If the on-site environmental noise is greater than the preset fifth threshold, the extracted decision-making basis explanatory text and maintenance operation guidance content will be formatted and encapsulated into readable text information, and the text information will be sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can read the text information by consulting.

[0154] In some embodiments, when the search module 200 is used to rebuild the multidimensional hash index according to the updated hash function and dimension weights, it performs the following operations:

[0155] A141. Monitor the remaining storage space of edge computing nodes;

[0156] A142. If the remaining storage space is less than the preset storage threshold, an incremental rebuild is performed. Specifically, a multidimensional hash index is constructed for the newly added data based on the adjusted hash function and dimension weights, and the newly added index is merged into the original multidimensional hash index.

[0157] A143. If the remaining storage space is greater than or equal to the storage threshold, a full reconstruction method is used. Specifically, the multidimensional hash index is rebuilt based on the adjusted hash function and dimension weights.

[0158] In some embodiments, when the remaining storage space is lower than a preset storage threshold, the search module 200 uses an incremental reconstruction method, specifically, constructing a multidimensional hash index for the newly added data based on the adjusted hash function and dimension weights, and merging the newly added index into the original multidimensional hash index.

[0159] When using the incremental reconstruction method, if the newly added index conflicts with the original multidimensional hash index, the conflict resolution strategy is dynamically adjusted based on the number and type of conflicts. The conflict resolution strategies include open addressing, chain addressing, and rehashing. The number of conflicts is calculated using a hash conflict detection algorithm, and the conflict types include data conflicts and structural conflicts. A data conflict refers to a storage location pointed to by the newly added index being occupied, and a structural conflict refers to a dimensional structure inconsistency between the newly added index and the original index.

[0160] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0161] The foregoing description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent decision-making method based on bus status, characterized in that: The following steps are involved: Obtain the power supply bus status decision result; The power supply bus status decision result is used as a query condition to search in a local index library; the local index library stores a multi-dimensional index of decision-making explanation texts and maintenance operation guidance content pre-generated by the central side system; Based on the search results, the corresponding decision-making explanation text and maintenance operation guidance content are extracted from the local content library; The local content library stores the decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system; The extracted decision-making basis explanation text and maintenance operation guidance content are formatted and packaged, and sent to the terminal equipment of the on-site maintenance personnel through the local communication interface.

2. The intelligent decision-making method according to bus status according to claim 1, characterized in that: The steps of using the power supply bus status decision result as a query condition and searching in the local index library include: A1. Construct a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and on-site environment information code; each layer of the multidimensional hash index corresponds to the decision type, abnormal mode, bus model, key monitoring parameter, and on-site environment; A2. Use the constructed multi-dimensional hash index to search layer by layer to obtain the final index result. A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

3. The intelligent decision-making method according to bus status according to claim 2, characterized in that: The specific steps in step A1 include: A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window. A12. If the CPU occupancy rate exceeds a preset first threshold or the memory occupancy rate exceeds a preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity; A13. Calculate a data quality score based on the missing rate and the noise level. If the data quality score is lower than a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index; A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

4. The intelligent decision-making method according to bus status according to claim 3, characterized in that: The specific steps in step A13 include: A131. When a preset data quality assessment model is used to identify suspected abnormal data in sensor data, relevant historical data is obtained and the similarity between the suspected abnormal data and the historical data is calculated. If the similarity is lower than a preset similarity threshold, the suspected abnormal data is determined to be true abnormal data and an abnormality alarm is generated. Otherwise, the suspected abnormal data is determined to be noise data. A132. If the abnormal alarm information is generated, the abnormal data ratio is calculated based on the actual abnormal data, and the data quality score is calculated based on the missing rate, the noise level, and the abnormal data ratio. Otherwise, the data quality score is calculated based on the missing rate and the noise level. A133. If the data quality score is lower than the third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index.

5. The intelligent decision-making method according to bus status according to claim 1, characterized in that: The steps of formatting and encapsulating the extracted decision-making explanation text and maintenance operation instructions and sending them to the terminal device of the on-site maintenance personnel through the local communication interface include: Acquiring on-site information sent by a terminal device; the on-site information includes on-site environment brightness and on-site environment noise; If the brightness of the on-site environment is lower than a preset fourth threshold, the extracted decision-making basis explanation text and maintenance operation instruction content are formatted and encapsulated into a broadcastable voice message, and the voice message is sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can listen to the voice message in a broadcast manner; If the on-site environmental noise is greater than the preset fifth threshold, the extracted decision-making basis explanatory text and maintenance operation guidance content will be formatted and encapsulated into readable text information, and the text information will be sent to the terminal device of the on-site maintenance personnel through the local communication interface, so that the on-site maintenance personnel can read the text information by reference.

6. The intelligent decision-making method according to bus status according to claim 3, characterized in that: The specific steps in step A14 include: A141. Monitor the remaining storage space of edge computing nodes; A142. If the remaining storage space is less than the preset storage threshold, an incremental rebuild is performed. Specifically, a multidimensional hash index is constructed for the newly added data based on the adjusted hash function and dimension weights, and the newly added index is merged into the original multidimensional hash index. A143. If the remaining storage space is greater than or equal to the storage threshold, a full reconstruction method is adopted, specifically: the multidimensional hash index is rebuilt according to the adjusted hash function and dimension weights.

7. The intelligent decision-making method according to bus status according to claim 6, characterized in that: The specific steps in step A142 include: When using the incremental reconstruction method, if the newly added index conflicts with the original multidimensional hash index, the conflict resolution strategy is dynamically adjusted according to the number and type of conflicts; the conflict resolution strategy includes open addressing, chain addressing and rehashing; wherein, the number of conflicts is calculated by a hash conflict detection algorithm, and the conflict types include data conflicts and structural conflicts. The data conflict refers to that the storage location pointed to by the newly added index is already occupied, and the structural conflict refers to that the dimensional structure of the newly added index is inconsistent with that of the original index.

8. An intelligent decision-making device based on bus status, characterized in that: include: An acquisition module is used to obtain the power supply bus status decision result; A search module is used to use the power supply bus status decision result as a query condition to search in a local index library; the local index library stores a multi-dimensional index of decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system; An extraction module is used to extract the corresponding decision-making basis explanation text and maintenance operation guidance content from the local content library based on the search results; the local content library stores the decision-making basis explanation text and maintenance operation guidance content pre-generated by the central side system; The sending module is used to format and encapsulate the extracted decision-making basis explanation text and maintenance operation guidance content, and send them to the terminal equipment of the on-site maintenance personnel through the local communication interface.

9. The intelligent decision-making method according to bus status according to claim 8, characterized in that: The search module is used to perform the following operations when searching in the local index library using the power supply bus status decision result as a query condition: A1. Construct a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and on-site environment information code; each layer of the multidimensional hash index corresponds to the decision type, abnormal mode, bus model, key monitoring parameter, and on-site environment; A2. Use the constructed multi-dimensional hash index to search layer by layer to obtain the final index result. A3. If the final index result is empty, a fuzzy matching strategy is adopted. After reducing the matching accuracy of the key monitoring parameter value or its encoding, step A2 is re-executed until a matching index result is found or the preset matching accuracy threshold is reached.

10. The intelligent decision-making method according to bus status according to claim 9, characterized in that: The search module is executed when it is used to build a multidimensional hash index based on the decision type, abnormal mode code, bus model code, key monitoring parameter value or its code, and field environment information code: A11. Monitor the CPU and memory usage of edge computing nodes and calculate the sensor data loss rate and noise level within a preset time window. A12. If the CPU occupancy rate exceeds a preset first threshold or the memory occupancy rate exceeds a preset second threshold, the hash function originally used in each layer of the multidimensional hash index is updated to a hash function with lower computational complexity; A13. Calculate a data quality score based on the missing rate and the noise level. If the data quality score is lower than a preset third threshold, reduce the weight of the key monitoring parameter value or its encoding dimension in the multidimensional hash index; A14. Rebuild the multidimensional hash index based on the updated hash function and dimension weights.

Images