Equipment updating and reconstruction effect evaluation method and system based on multi-modal data

Abstract

The embodiment of the invention provides an equipment updating and reconstruction effect evaluation method and system based on multi-modal data. The method comprises the following steps: collecting multi-source heterogeneous data of power equipment; performing feature fusion and cross-modal association analysis processing on the multi-source heterogeneous data to obtain fused multi-modal features; performing semantic analysis processing on the multi-source heterogeneous data, and constructing a knowledge graph; based on the multi-modal features, constructing a safety evaluation model of the power equipment updating and reconstruction project, and outputting an equipment health score; constructing an evaluation index system, and setting a weight for each index; and according to the data in the knowledge graph and the equipment health score, calculating the score of each index, and according to the score of each index and the weight of each index, obtaining the total score of the electrical equipment after updating and reconstruction. According to the technical scheme provided by the invention, the transformation of evaluation from static state, isolation to dynamic state and integration after the power equipment is transformed is realized, and the systematicness, the accuracy and the practicability of evaluation work are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of power equipment upgrading and renovation technology, and in particular relates to a method and system for evaluating the effect of equipment upgrading and renovation based on multimodal data. Background Technology

[0002] In the field of power equipment upgrading and renovation, evaluating the comprehensive benefits of the upgraded equipment is a crucial step in ensuring the safe and stable operation of the power system and optimizing resource allocation. Relevant known technologies have been widely applied in the industry for a long time. Among these, basic post-evaluation technologies mainly revolve around two core dimensions: equipment operating status monitoring and economic accounting. The former collects structured operating parameters such as voltage, current, and temperature from the equipment using sensors, combined with manual periodic inspection records of equipment appearance and operational responses, forming basic data on the equipment's operating status. The latter, based on financial indicators such as investment costs during the equipment upgrade process, savings in post-upgrade maintenance costs, and economic benefits from reduced energy consumption, employs static accounting methods, such as the payback period method and cost-benefit ratio method, to conduct a preliminary assessment of the economic viability of the equipment upgrade. These known technologies provide a basic data collection and analysis framework for the post-evaluation of power equipment upgrading and renovation, and are commonly used fundamental tools in the industry for conducting related evaluations.

[0003] The closest technical solution to this invention is the current industry-standard comprehensive evaluation method for power equipment retrofitting based on multi-source monitoring data. Its steps are as follows: First, data acquisition: Operating parameters, equipment appearance records, temperature distribution maps, and characteristic gas data of insulating oil are acquired through sensors, manual inspections, infrared thermal imagers, and oil chromatographs, respectively. Second, data preprocessing: Outlier removal and normalization are performed on structured data; noise reduction and target region extraction are performed on infrared spectra; and oil chromatographic data are calibrated. Third, single-dimensional evaluation: Performance is assessed by comparing operating parameters; economic efficiency is analyzed using static accounting; and the health status of the equipment is judged based on expert experience. Fourth, comprehensive evaluation: The single-dimensional results are simply weighted and summarized to generate a report.

[0004] This scheme has significant flaws, as evidenced by existing literature: Patent CN108763254A, "An Evaluation Method for the Effect of Power Equipment Retrofitting," while integrating operational and financial data, uses static weighted evaluation, failing to consider dynamic changes in equipment status. The article "Research on Post-Evaluation Methods for Power Equipment Retrofitting Projects," published in the 2020 issue 3 of *Power Automation Equipment*, points out that unstructured data relies on subjective expert judgment, resulting in poor repeatability and a lack of deep integration of multi-source data. Specifically, the shortcomings include: first, weak data fusion capabilities, fragmented multi-source data, a lack of a unified framework, and an inability to uncover data correlations; second, a lack of dynamic adaptability in the evaluation model, relying on fixed weights and methods, and an inability to track equipment status evolution; and third, low level of intelligence, requiring significant manual intervention for analysis, judgment, and report generation, leading to low efficiency and susceptibility to errors. These flaws result in insufficient reliability of the evaluation results, failing to provide strong support for accurate assessment and closed-loop optimization of power equipment retrofitting projects, highlighting the urgent need to construct a more advanced intelligent evaluation system. Summary of the Invention

[0005] The equipment upgrade and renovation effect evaluation method and system based on multimodal data provided in this application can realize the transformation of the evaluation of power equipment upgrades from static and isolated to dynamic and integrated, significantly improving the systematicness, accuracy and practicality of the evaluation work.

[0006] In a first aspect, embodiments of this application provide a method for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, including:

[0007] Collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data;

[0008] Feature fusion and cross-modal correlation analysis were performed on equipment operation data, infrared spectral data and oil chromatographic data to obtain fused multimodal features;

[0009] Based on multimodal features, a safety assessment model for power equipment upgrading and renovation projects is constructed, and equipment health scores are output.

[0010] Semantic parsing is performed on unstructured text data, and a knowledge graph is constructed by combining it with device operation data;

[0011] Construct an evaluation indicator system and assign weights to each indicator; the evaluation indicator system includes safety indicators, efficiency indicators, economic indicators and environmental indicators;

[0012] The scores for efficiency, economic and environmental indicators are calculated based on the data in the knowledge graph. The score for safety indicators is obtained based on the equipment health score. The total score after the power equipment renovation and upgrading is obtained by weighted summation based on the scores and weights of each indicator.

[0013] In one optional implementation, feature fusion and cross-modal correlation analysis are performed on equipment operation data, infrared spectral data, and oil chromatographic data to obtain fused multimodal features, including:

[0014] Spatial features of infrared spectral data were extracted using a deep convolutional neural network.

[0015] The temporal characteristics of device operation data are processed using a long short-term memory network.

[0016] The spatial characteristics of infrared spectral data, the temporal characteristics of equipment operation data, and the correlation between oil chromatography data were established using an attention mechanism.

[0017] In one optional implementation, semantic parsing is performed on unstructured text data, and a knowledge graph is constructed by combining it with device operation data, including:

[0018] The BERT model is used to perform deep semantic analysis on unstructured text data to extract key data.

[0019] Align key data with equipment operation data to build a knowledge graph with equipment entities as the core nodes;

[0020] Key data includes historical fault records, maintenance work orders, and economic change data.

[0021] In one optional implementation, an evaluation index system is constructed, and a weight is assigned to each index, including:

[0022] Set initial weights for each indicator based on its importance;

[0023] The initial weights of each indicator are optimized using a reinforcement learning algorithm to obtain the weights of each indicator.

[0024] In one optional implementation, after optimizing the initial weights of each indicator using a reinforcement learning algorithm to obtain the weights of each indicator, the method further includes:

[0025] When abnormalities are detected in equipment operation data or oil chromatography data, the weights of each indicator are automatically adjusted to determine the final weight of each indicator.

[0026] In one alternative implementation, the sub-indicators of the performance index include: load capacity utilization, operating efficiency score, and reliability index.

[0027] Sub-indicators of environmental indicators include: reduction in carbon emissions;

[0028] The sub-indicators of the economic indicators include: the operation and maintenance cost saving rate, which is obtained from economic change data;

[0029] The score for each indicator is obtained by weighted summation of the sub-indicators of each indicator.

[0030] In one alternative implementation, the device health score is used as a score for a safety indicator.

[0031] Secondly, embodiments of this application provide a system for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, including:

[0032] The data acquisition module is used to collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data.

[0033] The feature fusion module is used to perform feature fusion and cross-modal correlation analysis on equipment operation data, infrared spectrum data and oil chromatography data to obtain fused multimodal features;

[0034] The safety assessment model construction module is used to build a safety assessment model for power equipment upgrading and renovation projects based on multimodal features, and output equipment health scores.

[0035] The knowledge graph construction module is used to perform semantic parsing on unstructured text data and construct a knowledge graph by combining it with device operation data.

[0036] The evaluation indicator system construction module is used to construct the evaluation indicator system and set weights for each indicator; the evaluation indicator system includes safety indicators, efficiency indicators, economic indicators and environmental indicators;

[0037] The scoring calculation module is used to calculate the scores of efficiency indicators, economic indicators and environmental indicators based on the data in the knowledge graph, obtain the score of safety indicators based on the equipment health score, and obtain the total score of the power equipment after upgrading and renovation by weighted summation based on the scores of each indicator and the weight of each indicator.

[0038] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method provided in embodiments of this application.

[0039] Fourthly, embodiments of this application provide a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed in a computer, causes the computer to perform the method provided in embodiments of this application.

[0040] The technical solution provided in this application has the following beneficial effects:

[0041] This invention utilizes deep learning-based multi-source heterogeneous data fusion technology. Through feature extraction and cross-modal correlation analysis, it deeply integrates heterogeneous data such as equipment operating parameters, infrared spectra, and text reports within a unified framework. This effectively solves the problems of data fragmentation and information silos in traditional methods, providing a comprehensive and consistent data foundation for evaluation. The dynamic evaluation model built using reinforcement learning algorithms can adaptively adjust the weights of evaluation indicators based on the real-time operating status and working conditions of the equipment. This overcomes the limitations of traditional fixed-weight evaluation models, making the evaluation results more closely reflect the actual operating conditions of the equipment and significantly improving the accuracy and dynamic adaptability of the evaluation. By parsing text feedback using natural language processing technology and combining it with semantic analysis to establish a user feedback-driven closed-loop optimization mechanism, the evaluation system possesses continuous self-evolution capabilities, effectively reducing human intervention and enhancing the repeatability and reliability of evaluation results. Furthermore, the equipment virtual mapping model built based on digital twin technology enables virtual pre-simulation and effect prediction of the transformation plan, providing forward-looking data support for decision-making, effectively avoiding resource waste caused by blind decision-making, and improving the scientific nature of technical transformation strategies. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the method for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, as provided in an embodiment of this application.

[0043] Figure 2 This is a schematic diagram of the structure of the equipment upgrade and renovation effect evaluation system based on multimodal data provided in the embodiments of this application. Detailed Implementation

[0044] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] To address the shortcomings in the post-evaluation of existing power equipment upgrade and renovation projects, such as insufficient multi-source heterogeneous data fusion capabilities, lack of dynamic adaptability in evaluation models, low levels of intelligence, and weak multimodal data collaborative analysis capabilities, which result in evaluation results lacking systematicity, dynamism, and reliability, and failing to support accurate assessment and closed-loop optimization throughout the project's entire lifecycle, this application provides a method for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data. Figure 1 This is a flowchart illustrating a method for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, provided in an embodiment of this application. The method can be executed by a system device for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data. The system can be implemented by software and / or hardware and can be configured in electronic devices such as computers.

[0046] like Figure 1 As shown, the technical solution provided in this application includes the following steps:

[0047] S110. Collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data.

[0048] Specifically, equipment operating data can be acquired through SCADA systems (Supervisory Control and Data Acquisition), online monitoring devices, and / or sensor networks. This includes electrical parameters such as voltage, current, power factor, load rate, and energy consumption; status parameters such as oil temperature and winding temperature; and environmental parameters such as ambient temperature, humidity, and carbon emissions. Oil chromatography data includes the concentrations of gases such as total hydrocarbons, acetylene, and hydrogen.

[0049] Among them, the monitoring data of the SCADA system, as the most core subset of equipment operation data, can include core operating data such as voltage, current, power, frequency, and load rate. Its monitoring data is the most critical and easily quantifiable part of the "equipment operating parameters".

[0050] Unstructured text data can include post-evaluation reports, operation logs, inspection records, maintenance work orders, etc.

[0051] S120. Perform feature fusion and cross-modal correlation analysis on equipment operation data, infrared spectrum data and oil chromatography data to obtain fused multimodal features.

[0052] First, the equipment operation data and oil chromatography data were cleaned, including removing outliers, imputing missing values, and normalizing. Then, the infrared spectral data were denoised and enhanced.

[0053] Specifically, a Long Short-Term Memory (LSTM) network is used to extract temporal features (such as "load curves every 5 minutes" and "oil temperature changes every hour") from equipment operation data (mainly monitoring data from the SCADA system). A Convolutional Neural Network (CNN) is used to extract spatial features (such as hotspot distribution patterns) from infrared spectral data. An attention mechanism is used to establish the correlation between the spatial features of infrared spectral data, the temporal features of equipment operation data, and oil chromatographic data, resulting in fused multimodal features. For example, when the ethylene concentration in the oil chromatographic data is abnormally high at 33.64 μL / L, combined with the operating data of a concurrent load rate of 85%, it can be accurately determined that there is a medium-to-high temperature overheating defect inside the main transformer, achieving collaborative diagnosis of multi-dimensional data.

[0054] S130. Based on multimodal features, construct a safety assessment model for power equipment upgrading and renovation projects, and output equipment health scores.

[0055] A safety assessment model for power equipment upgrading and renovation projects is constructed, including an input layer, a hidden layer, and an output layer. The input layer is used to receive multimodal features, the hidden layer adopts a 3-layer fully connected neural network, and the output layer outputs the equipment health score, with a value ranging from 1 to 100.

[0056] In the model training and optimization phase, pre-trained model parameters from substations and other similar equipment in other cities are first introduced based on transfer learning techniques. Then, the loss function is fine-tuned for the specific operating conditions of the substations in the target city using a domain adaptation algorithm.

[0057] L fine =L pre +λ×D MMD ;

[0058] Among them, L fine L represents the fine-tuned loss function. pre Let λ represent the pre-training loss function of the pre-trained model, and let D represent the regularization weight coefficients. MMD This represents the maximum mean difference, used to measure the difference in distribution between the source and target domains. The final output evaluation model can generate quantitative indicators such as equipment health scores and risk levels. Risk levels can be divided into high, medium, and low risks based on the equipment health scores; for example, an equipment health score ≥80 is considered low risk.

[0059] In some embodiments, the safety assessment model for power equipment upgrading and renovation projects also has a dynamic update mechanism, which can automatically trigger model retraining when a new defect pattern is detected, thereby continuously improving the assessment accuracy.

[0060] Among them, the triggering condition for model retraining can be the detection of changes in data distribution:

[0061] D KL (P New ||P old )>θ update ;

[0062] Among them, D KL P represents the KL (Kullback-Leibler) divergence; a larger value indicates a greater distributional difference. New P represents the current data distribution represented by the newly collected data sample. old θ represents the historical data distribution during model training or the last update. update This represents the update threshold, used to determine whether a model update needs to be triggered.

[0063] S140. Perform semantic parsing on unstructured text data and construct a knowledge graph by combining it with equipment operation data.

[0064] Specifically, unstructured text data can first be cleaned, including removing special characters and stop words, and performing sentence segmentation and word segmentation.

[0065] A BERT-based named entity recognition model is used to adapt to the power industry corpus, and then power industry entities are identified. Entity types can include: equipment name, defect type, technical parameters, time information, etc. Then, deep semantic parsing is performed on unstructured text data to extract key data, including historical fault records, maintenance work orders, and economic change data. Historical fault records include professional descriptions such as "there is a medium-high temperature overheating defect inside the transformer", and economic change data includes "annual savings of 250,000 yuan in operation and maintenance costs".

[0066] Aligning key data with equipment operation data (mainly structured SCADA system monitoring data, as structured SCADA system monitoring data is the most core information and the most suitable form for building knowledge graphs) can be achieved by establishing a mapping relationship between text description time and operation data timestamps, and using a dynamic time warping algorithm to solve the problem of inconsistent time scales.

[0067] In the knowledge graph construction phase, specific equipment entities (such as the C phase of transformer No. 3 in City A) are used as core nodes. Multi-dimensional data, including equipment technical parameters (such as capacity 750MVA), historical fault records (such as abnormal chromatographic data events), and maintenance work orders, are integrated to construct a knowledge network with rich semantic relationships. The monitoring data from the structured SCADA system is incorporated into the knowledge graph through "indirect association"—for example, "hotspot defects" identified by infrared spectroscopy are associated with "synchronous current data" in the structured SCADA system's monitoring data through NLP-parsed post-evaluation report text (such as "hotspots detected in the infrared detector of transformer No. 3"), forming a knowledge link of "equipment defect operating parameters." For instance, the system automatically establishes a complete knowledge path of "C phase of transformer No. 3 - defect type - overheating defect - influencing parameters - total hydrocarbon content," while also associating relevant solutions and handling experience to form a traceable and reasonable equipment knowledge system, significantly improving the accuracy and efficiency of fault diagnosis.

[0068] In some embodiments, the dynamic update and maintenance mechanism can employ a dual strategy combining event-triggered and periodic-triggered approaches to ensure the timeliness of the knowledge graph. Event-triggered events may include: discovery of new defects, parameter exceedances, and creation of maintenance work orders; periodic-triggered events may include: quarterly assessments. When a critical event such as a new chromatographic data exceedance (e.g., a sudden increase in acetylene concentration) is detected, a real-time update of the knowledge graph is immediately triggered; simultaneously, a periodic update mechanism such as quarterly assessments is set up to comprehensively verify and supplement the knowledge system. Through this intelligent update method, the system can automatically discover and add new knowledge associations, such as important patterns like "correlation between acetylene growth trends and load fluctuations," continuously optimizing the knowledge base for equipment status assessment.

[0069] The relationship between equipment operation data, SCADA system monitoring data, and structured SCADA system monitoring data can be understood as a hierarchical relationship of "major category (equipment operation data) → core subset (SCADA system monitoring data) → formatted form of subset (structured SCADA system monitoring data)". For example:

[0070] Equipment operating parameters: "C-phase current (1200A), top oil temperature (65℃), infrared hot spot temperature (85℃), vibration amplitude (0.15mm / s)" of main transformer No. 3;

[0071] SCADA system monitoring data: including "C-phase current (1200A) and top oil temperature (65℃)" (real-time data collected by SCADA, including timestamps but not fully processed);

[0072] SCADA system structured operation data: organized into a table of data with the following format: "Time: 2025112410:00; Equipment: No. 3 main transformer; C-phase current: 1200A; Top layer oil temperature: 65℃".

[0073] S150. Construct an evaluation index system and assign weights to each index; the evaluation index system includes safety indicators, efficiency indicators, economic indicators and environmental indicators.

[0074] Specifically, first set initial weights for each indicator based on its importance. Alternatively, the analytic hierarchy process or expert experience method can be used. For example, the weights could be set as follows: safety indicators 40%, efficiency indicators 30%, economic indicators 20%, and environmental indicators 10%.

[0075] The initial weights of each indicator are optimized using reinforcement learning algorithms (such as Q-learning). When an anomaly is detected in a key indicator (e.g., total hydrocarbon content exceeding the standard to 160.60 μL / L), the system automatically increases the weight of the safety indicator to 50% and adjusts the weights of other indicators accordingly to ensure that the assessment focus matches the actual risk status of the equipment. This adaptive mechanism enables the evaluation system to respond in real time to changes in equipment status. For example, in the case of the transformer in City A, the system immediately triggered the weight adjustment strategy after detecting a sudden increase in acetylene.

[0076] Finally, the weights are normalized to obtain the final weights of each indicator.

[0077] S160. Calculate the scores of efficiency indicators, economic indicators and environmental indicators based on the data in the knowledge graph. Obtain the score of safety indicators based on the equipment health score. Calculate the total score of the power equipment after upgrading and renovation by weighted summation based on the scores and weights of each indicator.

[0078] Among them, the sub-indicators of the performance index include: load capacity utilization rate, operating efficiency score and reliability index;

[0079] Load capacity utilization rate E1 is the ratio of actual load to rated capacity, operating efficiency score E2 is the ratio of actual energy consumption to energy consumption design value, and reliability index E3 = (1 - number of failures / number of operating days) × 100%.

[0080] Sub-indicators of environmental indicators include: reduction in carbon emissions;

[0081] Carbon emission reduction En1 = (CE 改造前 -CE 改造后 / CE 改造前 ) × 100%, of which CE 改造前 Indicates the carbon emissions before the modification, CE 改造后 This indicates the carbon emissions after the modification.

[0082] The sub-indicators of the economic indicators include: the operation and maintenance cost saving rate, which can be obtained from economic change data;

[0083] Operation and maintenance cost savings rate C1 = (C 改造前 -C 改造后 / C 改造前 ) × 100%, where C 改造前 C represents the equipment operation and maintenance cost before the modification. 改造后 This indicates the maintenance cost of the upgraded equipment.

[0084] The equipment health score output from the safety assessment model for power equipment upgrading and renovation projects is directly used as the score for safety indicators.

[0085] The weights of each sub-indicator can be set according to the actual situation. The score of the corresponding indicator is obtained by summing the weighted sums of each sub-indicator and its weight. Based on the scores of each indicator and the final weight of each indicator, the total score after the power equipment renovation and upgrading is obtained by summing the weighted sums.

[0086] In some embodiments, evaluation results of similar renovation projects can be retrieved from the knowledge graph, and the rationality of the current score can be verified based on historical data. If the current total score deviates significantly from the total score of the evaluation results of historical renovation projects, manual review can be used for verification.

[0087] Furthermore, the system can analyze the long-term evolution trend of equipment status using an LSTM time-series prediction model. The prediction results show the scores of various evaluation indicators for the equipment within a preset time period (e.g., within 3 years), providing forward-looking data support for subsequent operation and maintenance decisions. The scoring calculation process fully considers the synergistic effect between various indicators, such as the indirect contribution of improved security performance to reduced operation and maintenance costs.

[0088] Furthermore, at the visualization level, a 3D radar chart can be used to visually present the comparative changes of various indicators before and after the upgrade, using different colors to indicate the scores of evaluation indicators such as safety (red) and efficiency (blue). The automatically generated assessment report not only includes key quantitative data (such as annual maintenance cost savings of 250,000 yuan), but also supports multi-level data drill-down analysis. Users can click on specific indicators to view the detailed analysis process (such as the breakdown of specific factors contributing to the improvement in safety scores). The report is output in PDF format and also provides an interactive web interface, making it easy for managers at different levels to quickly grasp the overall status of the equipment. For example, in the A City substation project, decision-makers clearly identified the greatest benefit contribution brought by the oil tanker upgrade through visualization tools.

[0089] In some embodiments, the equipment upgrade and renovation effect evaluation system based on multimodal data constructed according to the method of steps S110-S160 may further include the following operations:

[0090] S170. Establish a user feedback-driven closed-loop optimization mechanism, process user evaluation opinions through semantic analysis technology, continuously optimize the parameters of the evaluation system (including parameters in the safety assessment model of power equipment upgrading and renovation projects and the weights of each evaluation indicator, etc.), form a complete closed loop from problem discovery to solution improvement, and improve the accuracy of evaluation.

[0091] The evaluation system combines a general Transformer architecture with a knowledge graph. The knowledge graph serves as a pre-trained power industry dictionary, accurately identifying positive feedback such as "significant improvement in fire protection system renovation" (marked as +1 point) and improvement suggestions such as "delayed progress in oil tank renovation" (marked as -1 point). In the A City substation project, the system intelligently extracted 32 effective suggestions from over 200 text feedbacks, including 18 equipment improvement suggestions (such as "the main transformer foundation needs enhanced seismic design") and 14 process optimization suggestions (such as "acceptance documents need to include chromatographic data comparison charts"), forming a structured problem knowledge base.

[0092] Based on feedback analysis and dynamic parameter adjustment, the evaluation system has established a quantitative response mechanism. When more than 30% of user feedback mentions "insufficient economic assessment," a life-cycle cost (LCC) calculation module is automatically added to the assessment model, integrating financial models such as equipment residual value estimation and discount rate calculation. The model adopts a monthly iteration mechanism, and the effect is verified through A / B testing after each update. For example, in City A, the project showed that after 6 iterations, the accuracy of the economic assessment increased by a cumulative 3.2%, and user acceptance of cost predictions increased from 68% to 89%. Key improvements include: introducing a cost prediction algorithm for overhaul cycles and optimizing the energy efficiency conversion factor (from 0.78 to 0.82), etc.

[0093] Regarding the construction of a closed-loop management mechanism, the evaluation system has established a complete "evaluation-analysis-optimization-verification" workflow. For the 15 expert opinions collected regarding the City A project, the system automatically generated an optimization task list, including: ① adding load sensitivity analysis to the risk assessment module (responding to the suggestion to consider the impact of peak and valley loads); ② adding a comparison column for chromatographic data before and after the modification to the report template (responding to the need to strengthen data visualization). Tracking data after the new version model went live showed that expert satisfaction increased by 40%, average processing time was shortened by 25%, and significant improvements were made, particularly in the accuracy of defect warnings (false alarm rate decreased by 12%). The system also generates a closed-loop management report, detailing the processing status and effectiveness verification data for each suggestion.

[0094] The S180 deploys a lightweight edge computing engine, supporting real-time invocation of evaluation models at substation sites. It also employs federated learning technology to ensure secure data sharing and collaborative optimization across sites, thereby enhancing the model's generalization capabilities.

[0095] Specifically, for the special operating conditions of Phase C of the No. 3 main transformer in City A, this embodiment developed a highly optimized lightweight evaluation model. Deployed on edge computing devices at the substation site, this model can process key indicators such as chromatographic monitoring data (e.g., total hydrocarbon concentration of 160.60 μL / L) in real time, with computational latency controlled within 180ms. When an abnormal parameter is detected (e.g., acetylene concentration exceeding the 0.5 μL / L warning threshold), the system immediately triggers multi-level alarms: first, a red warning indicator is displayed on the local HMI interface; simultaneously, a voice prompt is broadcast through the station's broadcast system; and a work order containing specific abnormal parameters is automatically generated and pushed to the mobile terminal of maintenance personnel. This deployment method effectively solves the latency problem inherent in traditional cloud-based analysis. In the most recent oil chromatography anomaly event, the time from data acquisition to alarm triggering was only 152ms.

[0096] The federated learning architecture employs a hierarchical design, using the substation in City A as the core node, and uniting it with the substation in City B and four other similar 500kV substations to form a regional federated network. Each station transmits model parameters (such as convolutional kernel weights, LSTM unit states, and gradient information) via an AES-256 encrypted channel, strictly preserving the original data from leaving the station. In the most recent training cycle, the network exchanged parameters 127 times. The overheating defect identification feature parameters provided by the City A substation were adopted by other station models, resulting in a 12% improvement in overall identification accuracy (from 83% to 95%). Particularly noteworthy is the finding discovered through federated learning that "acetylene generation rate accelerates when load rate > 85%", which has been successfully applied to the optimization of early warning strategies for all member stations.

[0097] The dynamic optimization mechanism employs a dual-timescale design: edge nodes synchronize model parameters with the provincial company's central server via VPN tunnels daily at midnight, with differential privacy technology (ε=0.5) ensuring data security during the synchronization process; the federated network performs global parameter aggregation every Sunday morning, using a weighted average algorithm (weights are allocated based on the data volume of each site). Practice shows that after absorbing the learning results from the B city substation regarding oil temperature mutation patterns, the A city substation model improved overheating defect identification accuracy by 8% and reduced false alarm rate by 3.2%. Simultaneously, the system records the performance increment of each optimization, automatically extending the aggregation cycle when the gain is <1% for three consecutive optimizations, achieving intelligent allocation of computing resources. The system has been running stably for six months, with the F1-score for key indicator identification at each site model consistently above 92%.

[0098] S190. Construct a digital twin of the equipment, integrate multimodal data and physical simulation, and predict the comprehensive score of different transformation schemes through virtual debugging to assist in formulating the optimal technical transformation strategy and improve the scientific nature of decision-making.

[0099] During the twin construction phase, the system integrated the equipment nameplate parameters of Phase C of the No. 3 main transformer in City A (including key specifications such as the ODFS-334000 / 500 model), nine years of historical operating data (covering 382 parameters such as total hydrocarbon content and load rate), and high-precision three-dimensional laser scanning data to construct a digital twin with holographic mapping capabilities. This twin receives on-site monitoring data through a real-time data interface, dynamically simulating equipment operating status, such as accurately reproducing the temperature distribution characteristics of transformer oil under different load conditions (oil temperature gradient simulation error <0.5℃), providing a high-fidelity simulation environment for subsequent scheme evaluation. For example, when analyzing abnormal chromatographic data (total hydrocarbons 160.60 μL / L), the twin can visualize the correlation between the oil flow velocity field and hotspot distribution.

[0100] In the virtual solution evaluation phase, based on the constructed digital twin, the system conducted multi-dimensional simulation analysis on three typical renovation schemes: Scheme A (complete replacement) modeling parameters included an initial investment of 17.5999 million yuan, an expected lifespan of 15 years, and an annual reduction in operation and maintenance costs of 250,000 yuan; Scheme B (partial repair) corresponded to an investment of 8 million yuan, an 8-year lifespan extension, and an annual reduction in operation and maintenance costs of 120,000 yuan; Scheme C (maintaining operation) required consideration of the risk of defect deterioration (annual failure probability increasing by 15%). Through 100,000 Monte Carlo simulations, the system quantitatively evaluated the techno-economic performance of each scheme throughout its entire lifecycle, particularly capturing the reliability advantage of Scheme A after the seventh year (failure rate less than 0.5% / year) and the potential hidden costs of Scheme C (annual unplanned power outage losses of 2.8 million yuan).

[0101] In the decision support output phase, the system automatically generates a structured comparative report, highlighting the analysis results across three dimensions: Economically, it emphasizes the net present value advantage (NPV = 3.2 million yuan) and shorter payback period (5.2 years) of Option A; in terms of technical reliability, it uses failure probability curves to illustrate the long-term performance differences among the options; and in terms of risk, it quantitatively assesses the potential losses of different options. The report employs dynamic visualization technology. When the user adjusts key parameters (e.g., changing the discount rate from 6% to 8%), the system updates the benefit indicators of all options in real time and annotates the sensitivity analysis results (e.g., the NPV change range for Option A is 2.8-3.5 million yuan). The final output is a decision matrix containing 12 core indicators, supporting multi-criteria decision-making through weighted scoring, providing a data-driven scientific basis for the transformation project in City A.

[0102] Figure 2 This is a schematic diagram of the structure of the equipment upgrade and renovation effect evaluation system based on multimodal data provided in the embodiments of this application, such as... Figure 2 As shown, the system includes:

[0103] The data acquisition module is used to collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data.

[0104] The feature fusion module is used to perform feature fusion and cross-modal correlation analysis on equipment operation data, infrared spectrum data and oil chromatography data to obtain fused multimodal features;

[0105] The safety assessment model construction module is used to build a safety assessment model for power equipment upgrading and renovation projects based on multimodal features, and output equipment health scores.

[0106] The knowledge graph construction module is used to perform semantic parsing on unstructured text data and construct a knowledge graph by combining it with device operation data.

[0107] The evaluation indicator system construction module is used to construct the evaluation indicator system and set weights for each indicator; the evaluation indicator system includes safety indicators, efficiency indicators, economic indicators and environmental indicators;

[0108] The scoring calculation module is used to calculate the scores of efficiency indicators, economic indicators and environmental indicators based on the data in the knowledge graph, obtain the score of safety indicators based on the equipment health score, and obtain the total score of the power equipment after upgrading and renovation by weighted summation based on the scores of each indicator and the weight of each indicator.

[0109] The execution process of the system part of this application embodiment is the same as that of the method part of the embodiment described above, and will not be repeated here.

[0110] This application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method provided in this application.

[0111] This application also proposes a computer storage medium storing a computer program, which, when executed by a processor, implements the method provided in this application.

[0112] Computer storage media may be simply referred to as media. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Dual Data SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus Direct RAM (RDRAM), Direct Memory Bus Dynamic RAM (DRDRAM), and Memory Bus Dynamic RAM (RDRAM). The various embodiments described in this specification are presented in a progressive manner, with reference allowed to each other for similar or identical parts. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for apparatuses, devices, and non-volatile computer storage media are described simply because they are substantially similar to the method embodiments; relevant details can be found in the descriptions of the method embodiments.

[0113] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, characterized in that, include: Collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data; The device operation data, the infrared spectrum data, and the oil chromatography data are subjected to feature fusion and cross-modal correlation analysis to obtain fused multimodal features; Based on the aforementioned multimodal features, a safety assessment model for power equipment upgrading and renovation projects is constructed, and an equipment health score is output. The unstructured text data is subjected to semantic parsing processing, and a knowledge graph is constructed by combining the device operation data; An evaluation index system is constructed, and a weight is assigned to each index; the evaluation index system includes safety indicators, efficiency indicators, economic indicators, and environmental indicators. The scores for efficiency, economic and environmental indicators are calculated based on the data in the knowledge graph. The score for safety indicators is obtained based on the equipment health score. The total score after the power equipment upgrade and renovation is obtained by weighted summation based on the scores and weights of each indicator.

2. The method according to claim 1, characterized in that, The device operation data, the infrared spectrum data, and the oil chromatography data are subjected to feature fusion and cross-modal correlation analysis to obtain fused multimodal features, including: The spatial features of the infrared spectral data are extracted using a deep convolutional neural network; The temporal characteristics of the device's operating data are processed using a long short-term memory network. The spatial characteristics of the infrared spectral data, the temporal characteristics of the equipment operation data, and the correlation between the oil chromatographic data are established through an attention mechanism.

3. The method according to claim 1, characterized in that, The unstructured text data undergoes semantic parsing processing, and a knowledge graph is constructed by combining it with the device operation data, including: The unstructured text data is subjected to deep semantic analysis using the BERT model to extract key data. Align the key data with the equipment operation data to construct a knowledge graph with the equipment entity as the core node; Key data includes historical fault records, maintenance work orders, and economic change data.

4. The method according to claim 1, characterized in that, Construct an evaluation index system and assign weights to each index, including: Set initial weights for each indicator based on its importance; The initial weights of each indicator are optimized using a reinforcement learning algorithm to obtain the weights of each indicator.

5. The method according to claim 4, characterized in that, After optimizing the initial weights of each indicator using a reinforcement learning algorithm to obtain the weights of each indicator, the process also includes: When abnormalities are detected in the equipment operation data or the oil chromatography data, the weights of each indicator are automatically adjusted to determine the final weights of each indicator.

6. The method according to claim 3, characterized in that, The sub-indicators of the performance index include: load capacity utilization, operating efficiency score, and reliability index; The sub-indicators of the environmental indicators include: carbon emission reduction; The sub-indicators of the economic indicators include: operation and maintenance cost savings rate, which is obtained from the economic change data; The score for each indicator is obtained by weighted summation of the sub-indicators of each indicator.

7. The method according to claim 6, characterized in that, The device health score is used as the score for the safety indicator.

8. A system for evaluating the effectiveness of equipment upgrades and renovations based on multimodal data, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous data from power equipment, including equipment operation data, infrared spectrum data, oil chromatography data, and unstructured text data. The feature fusion module is used to perform feature fusion and cross-modal correlation analysis on the equipment operation data, the infrared spectrum data and the oil chromatography data to obtain fused multimodal features; The safety assessment model construction module is used to construct a safety assessment model for power equipment upgrading and renovation projects based on the multimodal features, and output equipment health scores. The knowledge graph construction module is used to perform semantic parsing processing on the unstructured text data and construct a knowledge graph by combining it with the device operation data; The evaluation indicator system construction module is used to construct the evaluation indicator system and set weights for each indicator; the evaluation indicator system includes safety indicators, efficiency indicators, economic indicators and environmental indicators. The scoring calculation module is used to calculate the scores of efficiency indicators, economic indicators and environmental indicators based on the data in the knowledge graph, obtain the score of safety indicators based on the equipment health score, and obtain the total score of the power equipment after upgrading and renovation by weighted summation based on the scores of each indicator and the weight of each indicator.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images