A new energy station unmanned operation and maintenance system based on multi-source data fusion

Abstract

The application discloses a new energy station unmanned operation and maintenance system based on multi-source data fusion, belongs to the technical field of new energy station operation and maintenance, and comprises a multi-source data acquisition module, a data preprocessing module, a multi-source data fusion module, an unmanned operation and maintenance execution module, a system management and control module and a data storage module: bidirectional data interaction is realized between modules through an Internet of Things communication module, a complete operation and maintenance closed loop is formed, the multi-source data fusion module comprises a feature extraction unit, a feature layer fusion unit, a decision layer fusion unit, a fusion weight adjustment unit and a fusion precision checking unit. By constructing a double-layer fusion architecture comprising a feature layer and a decision layer, introducing a fusion precision self-checking and weight dynamic adjustment mechanism, realizing a full-process automatic closed loop of multi-source data from acquisition, fusion, decision to execution and verification, and through reevaluation of the closed loop after disposal, the system ensures that operation and maintenance measures are real and effective, and the reliability of station operation is improved.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance technology for new energy power stations, and in particular to an unmanned operation and maintenance system for new energy power stations that integrates multi-source data. Background Technology

[0002] With the rapid development of the new energy industry, the construction scale of new energy power plants such as photovoltaic and wind power has continued to expand, and the number of power plants has increased year by year. Most of these power plants are deployed in remote outdoor areas with complex terrain and harsh natural environments, which poses great challenges to the operation and maintenance of these power plants. At present, the mainstream operation and maintenance mode of new energy power plants is still mainly based on manual inspection, single data monitoring, and semi-automated operation and maintenance. At the same time, the existing technical solutions have not solved the core technical problems such as deep integration of multi-source heterogeneous data, adaptive scenario-based weights, and self-verification of operation and maintenance closed loops, which are no longer able to meet the needs of efficient, accurate, and low-cost operation and maintenance of large-scale new energy power plants.

[0003] Existing new energy power plant operation and maintenance technologies generally suffer from numerous technical shortcomings: the manual inspection mode relies excessively on on-site operations by maintenance personnel, resulting in low efficiency, high costs, significant susceptibility to natural environmental influences, and blind spots in inspections, easily leading to missed or false diagnoses of equipment faults; the single data monitoring mode only monitors core equipment operating parameters from a single dimension, failing to achieve multi-source data linkage and fusion, resulting in severe data fragmentation and an inability to comprehensively assess the overall operating status of the power plant; while the semi-automated operation and maintenance mode introduces some automated equipment, fault identification, location, and handling still require manual intervention, resulting in a low degree of automation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a multi-source data fusion unmanned operation and maintenance system for new energy power stations, which can solve the above-mentioned problem.

[0005] To solve the above problems, the following technical solutions are provided: Design a multi-source data fusion unmanned operation and maintenance system for new energy power stations, including a multi-source data acquisition module, a data preprocessing module, a multi-source data fusion module, an unmanned operation and maintenance execution module, a system control module, and a data storage module: The various modules achieve bidirectional data interaction through an IoT communication module, forming a complete operation and maintenance closed loop. The multi-source data fusion module includes a feature extraction unit, a feature layer fusion unit, a decision layer fusion unit, a fusion weight adjustment unit, and a fusion accuracy verification unit. The feature extraction unit uses differentiated extraction methods to obtain feature vectors for different types of data. The feature layer fusion unit uses an improved CNN-LSTM neural network model to fuse the feature vectors based on spatial and temporal correlations, generating unified fused features. The decision layer fusion unit combines fuzzy inference algorithms and Bayesian estimation algorithms to generate the station's operating status based on the unified fused features. Based on the status assessment results and operation and maintenance instructions, the fusion weight adjustment unit dynamically adjusts the fusion weights of different data sources in the feature layer fusion unit and the decision layer fusion unit according to the current site type and operating scenario. The fusion accuracy verification unit calculates the accuracy of the fusion result in real time, and automatically corrects the parameters of the improved CNN-LSTM neural network model and the fusion weights when the accuracy exceeds a preset threshold, until the fusion error is ≤5%. The unmanned operation and maintenance execution module receives the operation and maintenance instructions and executes the handling. After the handling is completed, the multi-source data fusion module is triggered to re-evaluate the equipment status based on the re-collected data until the equipment returns to normal, forming a closed-loop verification.

[0006] The aforementioned technical solution constructs a two-layer fusion architecture comprising a feature layer and a decision layer, and introduces a self-verification mechanism for fusion accuracy and a dynamic weight adjustment mechanism, achieving a fully automated closed-loop process for multi-source data from acquisition, fusion, decision-making to execution and verification. Compared to existing technologies that separate data fusion, fault diagnosis, and operation and maintenance execution, this system ensures the authenticity and effectiveness of operation and maintenance measures through a post-treatment re-evaluation closed loop, significantly improving the reliability of site operation and the level of intelligence in unmanned operation and maintenance.

[0007] Furthermore, the multi-source data acquisition module includes an equipment operation data acquisition unit, which acquires real-time operating parameters of photovoltaic modules, wind turbines, energy storage devices, converters, combiner boxes, and transformer substations. The acquisition frequency of photovoltaic module-related parameters is 10 seconds / time, the acquisition frequency of wind turbine-related parameters is 5 seconds / time, and the acquisition frequency of energy storage device and converter-related parameters is 8 seconds / time.

[0008] The above technical solution, by setting different high-frequency acquisition frequencies for different equipment characteristics, can accurately capture the slow power fluctuations of photovoltaic modules, the rapid operating condition changes of wind turbines, and the dynamic response processes of energy storage and converters. This provides high-time-resolution input data for subsequent multi-source data fusion and fault identification, effectively avoiding missed or misjudged status due to improper sampling frequency.

[0009] Furthermore, the multi-source data acquisition module also includes an environmental meteorological data acquisition unit, which collects light intensity, wind speed, wind direction, temperature, humidity, precipitation, snow thickness, dust concentration, and visibility at a frequency of 1 minute / time. The light intensity measurement range is 0-2000 W / m² with a measurement accuracy of ±5 W / m², and the wind speed measurement range is 0-60 m / s with a measurement accuracy of ±0.1 m / s.

[0010] The above technical solution integrates environmental factors unique to new energy power stations, such as dust concentration, snow thickness, and visibility, and combines them with wide-range, high-precision light and wind speed sensors to comprehensively characterize the external micro-meteorological environment in which the station equipment is located. This provides key operating condition correlation information for data analysis, enabling the system to distinguish between equipment malfunctions and performance degradation induced by environmental factors, thereby improving the accuracy of diagnosis.

[0011] Furthermore, the multi-source data acquisition module also includes a security environment data acquisition unit, which includes a high-definition network camera, an infrared thermal imaging detector, a sound sensor, and a vibration sensor; the high-definition network camera acquires video at 25 frames per second, the infrared thermal imaging detector has a temperature measurement range of -20℃ to 150℃ and a measurement accuracy of ±0.5℃; the anomaly recognition response time of the security environment data acquisition unit is ≤20s.

[0012] The aforementioned technical solution integrates visual, infrared thermal imaging, and acoustic and vibration multi-dimensional monitoring methods to construct a three-dimensional site security and equipment status perception system. The high-precision temperature measurement capability of the infrared thermal imaging detector can quickly detect potential hazards such as localized overheating of equipment, while the acoustic and vibration sensors can capture abnormal mechanical sounds and vibrations from equipment such as transformers and fans. Combined with a rapid response mechanism within 20 seconds, safety and environmental risks can be eliminated in their infancy.

[0013] Furthermore, the data preprocessing module includes a data format standardization unit, a noise removal unit, a missing value imputation unit, and a data normalization unit. The data format standardization unit converts structured data into JSON format, semi-structured data into XML format, and unstructured data into structured feature data after feature extraction. The noise removal unit uses an improved Kalman filter algorithm, and its filter coefficients are dynamically adjusted according to the data type. The missing value imputation unit dynamically selects linear interpolation, polynomial interpolation, or similarity matching imputation algorithm according to the proportion of missing data. The data normalization unit uses a min-max normalization algorithm to normalize the data to a fixed interval.

[0014] The aforementioned technical solution employs a hierarchical preprocessing strategy for multi-source heterogeneous data, forming a high-quality data cleaning pipeline encompassing format standardization, adaptive noise removal, and intelligent missing value imputation. In particular, the improved Kalman filter algorithm, which dynamically adjusts the filter coefficients based on data type, can filter out noise while preserving fault characteristics in the original signal to the greatest extent possible. Furthermore, the dynamically selected imputation algorithm ensures data integrity under harsh communication conditions, providing a solid data foundation for the fusion module.

[0015] Furthermore, the improved CNN-LSTM neural network model includes a CNN convolutional layer, an LSTM recurrent layer, and a fully connected layer connected in sequence. The CNN convolutional layer extracts the spatial correlation of features, the LSTM recurrent layer extracts the temporal correlation, and the fully connected layer maps to obtain the unified fused features. The decision layer fusion unit performs weighted fusion of the results of fuzzy inference and Bayesian estimation to obtain the site operation status assessment result. The fusion weight adjustment unit stores an initial fusion weight table corresponding to different site types and operation scenarios, and dynamically adjusts the fusion weights by calling the corresponding initial fusion weights in real time to identify the operation scenario.

[0016] The aforementioned technical solution defines a clear, layer-by-layer structure for the improved CNN-LSTM model, deeply integrating spatial multi-source correlation with temporal series evolution, resulting in more representative fusion features. Simultaneously, the fusion weight adjustment unit, through a stored prior weight table and real-time scene identification, automatically switches the weights of equipment and environmental data under different operating conditions. This ensures the model maintains optimal diagnostic performance in various scenarios, such as sunny weather conditions at photovoltaic power plants and high wind shear conditions at wind turbines, solving the problem of poor adaptability of fixed-weight models.

[0017] Furthermore, the unmanned operation and maintenance execution module includes a fault identification and location unit, an autonomous mobile inspection unit, and an emergency response unit. The fault identification and location unit combines a GPS positioning module and a site electronic map to achieve fault location with a positioning accuracy of ≤1m and a fault response time of ≤30s. The autonomous mobile inspection unit includes a drone inspection system and a ground inspection robot. The drone inspection system is equipped with high-definition acquisition equipment to collect fault details in areas with complex terrain. The ground inspection robot uses a SLAM navigation algorithm to achieve autonomous obstacle avoidance and is equipped with a handling device to complete the handling action. The emergency response unit automatically activates the emergency plan for preset extreme scenarios, and the effect is verified by the multi-source data fusion module after the handling is completed.

[0018] The aforementioned technical solution, through the collaborative operation of aerial drones and ground robots, overcomes the limitations of traditional single inspection methods. It can cover complex terrains such as mountains, water surfaces, and rooftops, and realizes an unmanned operation chain from macro-level inspection to detailed fault collection and simple on-site handling. Combined with sub-meter level fault location accuracy, rapid response within 30 seconds, and automatic effect verification after handling, it significantly shortens the average fault repair time.

[0019] Furthermore, the system management module includes a cloud management platform, a human-machine interface, and a permission management unit. The cloud management platform supports multi-terminal login, visualizes data through charts, and automatically generates operation and maintenance reports. The human-machine interface has fault alarm prompts and operation and maintenance progress reminders. The permission management unit sets three levels of permissions: administrator, operation and maintenance personnel, and viewer, with each level of permission corresponding to a different scope of operation.

[0020] The above technical solution enables remote centralized monitoring and control of the site cluster through a cloud-based management platform, and automatically generated operation and maintenance reports reduce the workload of manual statistics. A hierarchical access control system ensures the security and standardization of system operations, preventing operational risks caused by unauthorized operations, while a human-machine interface with voice control enhances the convenience of on-site operations.

[0021] Furthermore, the data storage module includes local storage and cloud backup. The local storage uses an industrial-grade solid-state drive with a capacity of ≥1TB to store real-time data and pre-processed data for the past 3 months. The cloud backup uses distributed cloud storage to store historical data and supports permanent data backup and rapid recovery. The data storage module uses encrypted storage.

[0022] The aforementioned technical solution, through a dual storage architecture of local hot data and cloud cold data, balances the read / write performance of high-frequency real-time data with the secure and permanent preservation of massive historical data. Industrial-grade solid-state drives ensure reliable data access in harsh environments, while the rapid recovery capabilities of distributed cloud storage guarantee no data loss in the event of extreme incidents.

[0023] Furthermore, the data transmission unit of the multi-source data acquisition module adopts a dual-mode wireless transmission method of 5G and LoRa. The 5G transmission method is used to transmit data with high real-time requirements, while the LoRa transmission method is used to transmit data with lower real-time requirements. The operation and maintenance closed loop includes the entire process of acquisition, preprocessing, fusion, decision-making, execution, and verification.

[0024] The above technical solution, through the synergy of 5G and LoRa dual-mode communication, takes into account both the low-latency transmission of high-bandwidth real-time data such as high-definition video and high-frequency vibration, and the low-power wide-area coverage of small-data-volume long-cycle data such as meteorology and security, ensuring the reliability and economy of the entire site's IoT communication, and physically connecting the complete operation and maintenance closed loop from data collection to execution verification.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This multi-source data fusion-based unmanned operation and maintenance system for new energy power stations solves the problems of insufficient data fusion depth, poor scenario adaptability, and lack of closed-loop operation and maintenance in existing systems by adopting a technical solution that combines feature layer and decision layer fusion, adaptive fusion weight, and closed-loop self-verification. It realizes full-process automation and intelligence of the unmanned operation and maintenance system for new energy power stations. 2. This multi-source data fusion unmanned operation and maintenance system for new energy power stations integrates multi-dimensional data such as environmental meteorology and security with equipment operation data, and combines it with an improved CNN-LSTM model to accurately distinguish between environmental factors and equipment faults, significantly reducing false alarm rates and improving the accuracy of fault diagnosis. 3. This multi-source data fusion-based unmanned operation and maintenance system for new energy power stations achieves a complete unmanned closed loop of fault identification, location, handling, and verification through autonomous mobile inspections by drones and ground robots, as well as closed-loop verification based on data fusion after handling. This significantly reduces human intervention and improves the safety and efficiency of operation and maintenance in extreme environments. Attached Figure Description

[0026] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a block diagram of the overall structure of the multi-source data fusion unmanned operation and maintenance system for new energy power stations according to the present invention; Figure 2 This is a flowchart illustrating the workflow of the data preprocessing module of the present invention. Figure 3 This is a schematic diagram of the internal structure and process of the multi-source data fusion module of the present invention; Figure 4 This is a flowchart illustrating the workflow of the present invention, which integrates accuracy verification and model self-correction. Figure 5 This is a flowchart of the unmanned operation and maintenance execution module of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] like Figure 1 - Figure 5 As shown in the figure, this embodiment provides a multi-source data fusion-based unmanned operation and maintenance system for new energy power stations, including a multi-source data acquisition module, a data preprocessing module, a multi-source data fusion module, an unmanned operation and maintenance execution module, a system control module, and a data storage module. Each module achieves bidirectional data interaction through an IoT communication module, forming a complete operation and maintenance closed loop. The multi-source data fusion module includes a feature extraction unit, a feature layer fusion unit, a decision layer fusion unit, a fusion weight adjustment unit, and a fusion accuracy verification unit. The feature extraction unit uses differentiated extraction methods to obtain feature vectors from different types of data. The feature layer fusion unit uses an improved CNN-LSTM neural network model to fuse the feature vectors spatially and temporally to generate unified fused features. The decision layer fusion unit combines fuzzy inference algorithms and Bayesian estimation algorithms to generate site operation status assessment results and operation and maintenance instructions based on the unified fused features. The fusion weight adjustment unit dynamically adjusts the fusion weights of different data sources in the feature layer fusion unit and the decision layer fusion unit according to the current site type and operation scenario. The fusion accuracy verification unit... The system calculates the accuracy of the fusion results in real time and automatically corrects the parameters and fusion weights of the improved CNN-LSTM neural network model when the accuracy exceeds a preset threshold, until the fusion error is ≤5%. The unmanned operation and maintenance execution module receives operation and maintenance instructions and performs handling. After handling, the multi-source data fusion module is triggered to re-evaluate the equipment status based on the re-collected data until the equipment returns to normal, forming a closed-loop verification. The multi-source data acquisition module is responsible for acquiring raw data from various sensors and monitoring equipment in the site and uploading the data through the data transmission unit using a 5G+LoRa dual-mode wireless method. Data with high real-time requirements, such as high-frequency operating parameters of equipment and high-definition video streams, are transmitted through the 5G network to ensure low latency, while data with lower real-time requirements, such as routine environmental and meteorological monitoring values, are transmitted through the LoRa network to balance low power consumption and wide coverage. The transmission process uses the AES-256 encryption algorithm to ensure data security.

[0029] In this embodiment, the multi-source data acquisition module includes an equipment operation data acquisition unit, which acquires real-time operating parameters of photovoltaic modules, wind turbines, energy storage devices, converters, combiner boxes, and transformer substations. The acquisition frequency of photovoltaic module-related parameters is 10 seconds / time, the acquisition frequency of wind turbine-related parameters is 5 seconds / time, and the acquisition frequency of energy storage device and converter-related parameters is 8 seconds / time.

[0030] In this embodiment, the multi-source data acquisition module also includes an environmental meteorological data acquisition unit, which collects light intensity, wind speed, wind direction, temperature, humidity, precipitation, snow thickness, dust concentration, and visibility. The acquisition frequency is 1 minute / time. The light intensity measurement range is 0-2000 W / m² with a measurement accuracy of ±5 W / m², and the wind speed measurement range is 0-60 m / s with a measurement accuracy of ±0.1 m / s.

[0031] In this embodiment, the multi-source data acquisition module also includes a security environment data acquisition unit, which includes a high-definition network camera, an infrared thermal imaging detector, a sound sensor, and a vibration sensor; the high-definition network camera acquires video at 25 frames per second, the infrared thermal imaging detector has a temperature measurement range of -20℃ to 150℃, and a measurement accuracy of ±0.5℃; the anomaly identification response time of the security environment data acquisition unit is ≤20s.

[0032] In this embodiment, the data preprocessing module includes a data format standardization unit, a noise removal unit, a missing value imputation unit, and a data normalization unit. The data format standardization unit converts structured data into JSON format, semi-structured data into XML format, and unstructured data into structured feature data after feature extraction. The noise removal unit uses an improved Kalman filter algorithm, with its filter coefficients dynamically adjusted according to the data type. The missing value imputation unit dynamically selects linear interpolation, polynomial interpolation, or similarity matching imputation algorithm based on the proportion of missing data. The data normalization unit uses a min-max normalization algorithm to normalize the data to a fixed interval. After the data enters the data preprocessing module, it is first processed by the data format standardization unit: converting structured data from relational databases into JSON format, converting semi-structured data such as XML and HTML into standard XML format, and extracting features from unstructured data such as video and audio and converting them into structured data. After the feature data is processed, the noise removal unit uses an improved Kalman filter algorithm to filter the data. The core of this algorithm is to dynamically adjust the filter coefficients according to the data type. For data with drastic changes such as current and vibration, a wider filter window is used to preserve abrupt changes, while for data with slow changes such as temperature and humidity, a narrower filter window is used to smooth the noise. The missing value imputation unit dynamically selects the algorithm according to the proportion of missing data in the current sequence: linear interpolation with low computational cost is used when the missing rate is less than 5%, polynomial interpolation is used when the missing rate is between 5% and 20%, and similarity matching imputation is initiated when the missing rate is higher than 20%, that is, the data segment most similar to the current working condition is found from the historical data and imputed. Finally, the data normalization unit uses the min-max normalization algorithm to uniformly map all features to a fixed interval, eliminating the influence of different units on subsequent model training. This series of preprocessing operations ensures that the data input to the fusion module has high quality, high consistency and strong integrity.

[0033] In this embodiment, the improved CNN-LSTM neural network model includes a CNN convolutional layer, an LSTM recurrent layer, and a fully connected layer connected sequentially. The CNN convolutional layer extracts the spatial correlation of features, the LSTM recurrent layer extracts the temporal correlation, and the fully connected layer maps to obtain unified fused features. The decision layer fusion unit performs weighted fusion of the results of fuzzy inference and Bayesian estimation to obtain the site operation status assessment result. The fusion weight adjustment unit stores an initial fusion weight table corresponding to different site types and operation scenarios, and dynamically adjusts the weights by calling the corresponding initial fusion weights in real time based on the operation scenario. The core of the multi-source data fusion module lies in the two-layer fusion architecture of the feature layer and the decision layer. The feature extraction unit uses statistical feature extraction, TF-IDF keyword extraction, and CNN pre-trained model methods for numerical sensor data, text alarm data, image and sound data, respectively, to generate fixed-dimensional features. The feature vector of degree is used. The feature layer fusion unit adopts an improved CNN-LSTM neural network model. This model consists of a CNN convolutional layer, an LSTM recurrent layer and a fully connected layer connected in sequence. The CNN convolutional layer slides the convolution kernel in the spatial dimension to effectively extract the spatial correlation between measurement points of each device and between device and environmental parameters. The LSTM recurrent layer uses a gating mechanism to capture the long-term dependency relationship and short-term mutation law of feature sequence in the time dimension. The fully connected layer maps the deep features extracted by the two into a unified fused feature expression. The decision layer fusion unit receives the unified fused feature and runs the fuzzy inference algorithm and the Bayesian estimation algorithm in parallel: fuzzy inference infers the probability of failure under different working conditions based on the expert knowledge base, while Bayesian estimation makes probability inference based on historical failure probability and current observation. The results of the two are weighted and fused to obtain the final site operation status assessment result and operation and maintenance instructions.

[0034] Crucially, the fusion weight adjustment unit stores an initial fusion weight table corresponding to different site types and operating scenarios. By identifying the operating scenario in real time, the system automatically calls the current optimal fusion weight from the weight table. This allows data sources with higher relevance to play a greater decision-making role in the diagnosis of various faults such as wind turbine blade icing, photovoltaic module hot spots, and energy storage system overcharging, thus achieving accurate diagnosis that is adaptive to the scenario.

[0035] In this embodiment, the unmanned operation and maintenance execution module includes a fault identification and location unit, an autonomous mobile inspection unit, and an emergency response unit. The fault identification and location unit combines a GPS positioning module and a site electronic map to achieve fault location with a positioning accuracy of ≤1m and a fault response time of ≤30s. The autonomous mobile inspection unit includes a drone inspection system and a ground inspection robot. The drone inspection system is equipped with high-definition acquisition equipment to collect detailed fault data in areas with complex terrain. The ground inspection robot uses a SLAM navigation algorithm to achieve autonomous obstacle avoidance and is equipped with a handling device to complete the handling action. The emergency response unit automatically activates emergency plans for preset extreme scenarios, and the effect is verified by a multi-source data fusion module after the handling is completed. The unmanned operation and maintenance execution module activates the fault identification and location unit based on the received operation and maintenance instructions. This unit analyzes alarm messages and GPS coordinates, and combines them with the high-precision electronic map of the site to mark the fault point on the map. The positioning accuracy can reach within 1 meter. The response time from alarm generation to location completion is no more than 30 seconds. After that, the autonomous mobile inspection unit commands the drone to fly to the site first to take macroscopic visible light and thermal imaging pictures and transmit detailed images back in real time. For areas with complex terrain or limited space, the ground inspection robot uses the SLAM navigation algorithm to autonomously plan the path and avoid obstacles. After arriving, it operates the onboard robotic arm, cleaning device and other handling devices to complete simple handling actions such as tightening bolts and cleaning up dust. If the system determines that it is an extreme scenario such as a fire or extreme weather, the emergency response unit will automatically activate the corresponding emergency plan, such as automatically disconnecting the circuit breaker or activating the fire protection system. After the response action is completed, the system will collect the equipment operating parameters again through the multi-source data acquisition module. The multi-source data fusion module will evaluate the response effect and confirm that the equipment status has been restored to normal. Only after this operation and maintenance loop is completed. If it has not been restored, a new operation and maintenance instruction will be generated and executed in a loop until the equipment is normal, thus realizing true closed-loop verification and self-optimization.

[0036] In this embodiment, the system management module includes a cloud management platform, a human-machine interface, and a permission management unit. The cloud management platform supports multi-terminal login, visualizes data through charts, and automatically generates operation and maintenance reports. The human-machine interface has fault alarm prompts and operation and maintenance progress reminders. The permission management unit sets three levels of permissions: administrator, operation and maintenance personnel, and viewer. Each level of permission corresponds to a different scope of operation. As the hub of human-machine interaction, the system management module's cloud management platform supports multi-terminal login in Web / APP form. It intuitively displays the site's operating status, fault distribution, and operation and maintenance statistics using various charts such as line charts, radar charts, and heat maps. The platform can automatically summarize the operation and maintenance data of the day and month to generate operation and maintenance reports, reducing the burden of manual reporting. The human-machine interface supports natural voice commands and can provide fault alarm prompts through voice and vision when there are new alarms, and push operation and maintenance progress reminders. The permission management unit strictly divides the permissions into three levels: administrator, operation and maintenance personnel, and viewer. Administrators can configure the system and manage users, operation and maintenance personnel can execute tasks and confirm alarms, and viewer permissions can only browse data and reports to prevent unauthorized operations.

[0037] In this embodiment, the data storage module includes local storage and cloud backup. The local storage uses an industrial-grade solid-state drive with a capacity of ≥1TB to store real-time data and pre-processed data for the past 3 months. The cloud backup uses distributed cloud storage to store historical data and supports permanent data backup and fast recovery. The data storage module uses encrypted storage.

[0038] In this embodiment, the data transmission unit of the multi-source data acquisition module adopts a dual-mode wireless transmission method of 5G and LoRa. The 5G transmission method is used to transmit data with high real-time requirements, while the LoRa transmission method is used to transmit data with lower real-time requirements. The operation and maintenance closed loop includes the entire process of acquisition, preprocessing, fusion, decision-making, execution, and verification.

[0039] The multi-source data fusion unmanned operation and maintenance system for new energy power stations proposed in this embodiment works as follows: After the system is powered on, the multi-source data acquisition module continuously acquires on-site data from various sensors and cameras. After dual-mode encrypted transmission, the data enters the data preprocessing module for normalization and cleaning. The cleaned high-quality data is then sent to the multi-source data fusion module. After feature extraction, improved CNN-LSTM feature layer fusion, and decision layer fusion of fuzzy inference and Bayesian estimation, the system outputs accurate status assessment results and operation and maintenance instructions under the guarantee of dynamic weights and accuracy self-verification mechanisms. The unmanned operation and maintenance execution module is responsible for the efficient execution of instructions and triggers a second assessment to complete the closed-loop verification. The system management module provides a visual interactive interface, and the data storage module ensures the security and persistence of the data. Through the efficient and orderly collaborative operation of each module, the entire process of automation and intelligence of new energy power stations from data perception to unmanned operation and maintenance execution is realized.

[0040] In the description of this invention, it should be understood that the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two elements; they can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-source data fusion unmanned operation and maintenance system for new energy power stations, characterized in that, It includes a multi-source data acquisition module, a data preprocessing module, a multi-source data fusion module, an unmanned operation and maintenance execution module, a system management and control module, and a data storage module. The various modules achieve bidirectional data interaction through an IoT communication module, forming a complete operation and maintenance closed loop. The multi-source data fusion module includes a feature extraction unit, a feature layer fusion unit, a decision layer fusion unit, a fusion weight adjustment unit, and a fusion accuracy verification unit. The feature extraction unit uses differentiated extraction methods to obtain feature vectors for different types of data. The feature layer fusion unit uses an improved CNN-LSTM neural network model to fuse the feature vectors based on spatial and temporal correlations, generating unified fused features. The decision layer fusion unit combines fuzzy inference algorithms and Bayesian estimation algorithms to generate the station's operating status based on the unified fused features. Based on the status assessment results and operation and maintenance instructions, the fusion weight adjustment unit dynamically adjusts the fusion weights of different data sources in the feature layer fusion unit and the decision layer fusion unit according to the current site type and operating scenario. The fusion accuracy verification unit calculates the accuracy of the fusion result in real time, and automatically corrects the parameters of the improved CNN-LSTM neural network model and the fusion weights when the accuracy exceeds a preset threshold, until the fusion error is ≤5%. The unmanned operation and maintenance execution module receives the operation and maintenance instructions and executes the handling. After the handling is completed, the multi-source data fusion module is triggered to re-evaluate the equipment status based on the re-collected data until the equipment returns to normal, forming a closed-loop verification.

2. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 1, characterized in that: The multi-source data acquisition module includes an equipment operation data acquisition unit, which acquires real-time operating parameters of photovoltaic modules, wind turbines, energy storage devices, converters, combiner boxes, and transformer substations. The acquisition frequency of photovoltaic module-related parameters is 10 seconds / time, the acquisition frequency of wind turbine-related parameters is 5 seconds / time, and the acquisition frequency of energy storage device and converter-related parameters is 8 seconds / time.

3. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 2, characterized in that: The multi-source data acquisition module also includes an environmental meteorological data acquisition unit, which collects light intensity, wind speed, wind direction, temperature, humidity, precipitation, snow thickness, dust concentration, and visibility at a frequency of 1 minute / time. The light intensity measurement range is 0-2000 W / m² with a measurement accuracy of ±5 W / m², and the wind speed measurement range is 0-60 m / s with a measurement accuracy of ±0.1 m / s.

4. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 3, characterized in that: The multi-source data acquisition module also includes a security environment data acquisition unit, which includes a high-definition network camera, an infrared thermal imaging detector, a sound sensor, and a vibration sensor; the high-definition network camera acquires video at 25 frames per second, the infrared thermal imaging detector has a temperature measurement range of -20℃ to 150℃ and a measurement accuracy of ±0.5℃; the anomaly recognition response time of the security environment data acquisition unit is ≤20s.

5. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 4, characterized in that: The data preprocessing module includes a data format standardization unit, a noise removal unit, a missing value imputation unit, and a data normalization unit. The data format standardization unit converts structured data into JSON format, semi-structured data into XML format, and unstructured data into structured feature data after feature extraction. The noise removal unit uses an improved Kalman filter algorithm, and its filter coefficients are dynamically adjusted according to the data type. The missing value imputation unit dynamically selects linear interpolation, polynomial interpolation, or similarity matching imputation algorithm according to the proportion of missing data. The data normalization unit uses a min-max normalization algorithm to normalize the data to a fixed interval.

6. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 5, characterized in that: The improved CNN-LSTM neural network model includes a CNN convolutional layer, an LSTM recurrent layer, and a fully connected layer connected in sequence. The CNN convolutional layer extracts the spatial correlation of features, the LSTM recurrent layer extracts the temporal correlation, and the fully connected layer maps to obtain the unified fused features. The decision layer fusion unit performs weighted fusion of the results of fuzzy inference and Bayesian estimation to obtain the site operation status assessment result. The fusion weight adjustment unit stores an initial fusion weight table corresponding to different site types and operation scenarios, and dynamically adjusts the fusion weight by calling the corresponding initial fusion weights in real time to identify the operation scenario.

7. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 6, characterized in that: The unmanned operation and maintenance execution module includes a fault identification and location unit, an autonomous mobile inspection unit, and an emergency response unit. The fault identification and location unit combines a GPS positioning module and a site electronic map to achieve fault location with a positioning accuracy of ≤1m and a fault response time of ≤30s. The autonomous mobile inspection unit includes a drone inspection system and a ground inspection robot. The drone inspection system is equipped with high-definition acquisition equipment to collect detailed fault data in areas with complex terrain. The ground inspection robot uses a SLAM navigation algorithm to achieve autonomous obstacle avoidance and is equipped with a handling device to complete the handling action. The emergency response unit automatically activates emergency plans for preset extreme scenarios, and the effect is verified by the multi-source data fusion module after the handling is completed.

8. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 7, characterized in that: The system management module includes a cloud management platform, a human-machine interface, and a permission management unit. The cloud management platform supports multi-terminal login, visualizes data through charts, and automatically generates operation and maintenance reports. The human-machine interface has fault alarm prompts and operation and maintenance progress reminders. The permission management unit sets three levels of permissions: administrator, operation and maintenance personnel, and viewer, with each level of permission corresponding to a different scope of operation.

9. The unmanned operation and maintenance system for new energy power stations based on multi-source data fusion according to claim 8, characterized in that: The data storage module includes local storage and cloud backup. The local storage uses an industrial-grade solid-state drive with a capacity of ≥1TB to store real-time data and pre-processed data for the past 3 months. The cloud backup uses distributed cloud storage to store historical data and supports permanent data backup and fast recovery. The data storage module uses encrypted storage.

10. A multi-source data fusion unmanned operation and maintenance system for new energy power stations according to any one of claims 1-9, characterized in that: The data transmission unit of the multi-source data acquisition module adopts 5G and LoRa dual-mode wireless transmission. The 5G transmission mode is used to transmit data with high real-time requirements, while the LoRa transmission mode is used to transmit data with lower real-time requirements. The operation and maintenance closed loop includes the entire process of acquisition, preprocessing, fusion, decision-making, execution, and verification.

Images