A salt mist resistant transformer suitable for offshore wind power

Abstract

The application discloses a salt-fog-resistant transformer suitable for offshore wind power and relates to the technical field of transformer structures. The transformer comprises a transformer box, a mounting top cover and matched transformer elements, the mounting top cover is provided with an oil storage tank on one side through a supporting support, a detection assembly comprising a corrugated oil pillow, a scraper, a liquid level sensor and a single-through thin film baffle is arranged in the oil storage tank, a closed and isolated cavity is formed to isolate external salt-fog water vapor, and the oil level detection precision is ensured. The application further matches a salt-fog-resistant intelligent monitoring management and control system, and the transformer operation state monitoring, abnormality identification, trend prediction and autonomous protection regulation can be realized. The application solves the problems that the existing offshore wind power transformer oil storage tank is easy to be polluted, the oil pillow is aged and invalid, and the protection is insufficient, and improves the operation reliability and the full life cycle service life of the equipment in the offshore high-salt-fog environment.

Description

Technical Field ,

[0006] ,

[0001] The present invention relates to the technical field of transformer structures, and particularly to a salt - fog - resistant transformer applicable to offshore wind power. Background Art

[0002] Offshore wind power is an important direction for the transformation of China's energy structure. As the core equipment for electrical energy boosting and grid connection in wind farms, offshore wind power transformers operate in a marine environment with high salt fog, high humidity, and strong corrosion for a long time. The protection performance and operation stability of their oil conservators directly determine the overall service life of the transformer and the power supply safety of the wind farm.

[0003] The oil conservators supporting existing offshore wind power transformers mostly adopt a conventional structure of an open cabinet with a capsule - type oil pillow. During operation, salt fog and water vapor are easily introduced into the cabinet along with the breathing airflow and directly contact the transformer oil, resulting in oil pollution and aging failure of the capsule oil pillow; after long - term use, the protection performance of the moisture absorber rapidly decays, and there are no effective condition monitoring and intelligent control means, which cannot adapt to the unattended operation conditions of offshore wind power, easily cause equipment failures, and greatly increase the difficulty and cost of offshore operation and maintenance.

[0004] Existing improvement schemes mostly only optimize the anti - corrosion of a single structure, and do not form a systematic solution from root - cause salt fog isolation to full - life - cycle intelligent control, making it difficult to meet the actual needs of the long - term stable operation of offshore wind power transformers. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies in the prior art, and to propose a salt - fog - resistant transformer applicable to offshore wind power.

[0006] To achieve the above purpose, the present invention adopts the following technical scheme: A salt - fog - resistant transformer applicable to offshore wind power, including a transformer tank and transformer components installed inside the transformer tank. An installation top cover is installed at the top of the transformer tank. Multiple tap switches are installed at intervals in the middle of the installation top cover. Two high - voltage bushings are symmetrically installed on the installation top cover. A low - voltage bushing is installed on the installation top cover at the side of the high - voltage bushing. A support bracket is welded on one side of the installation top cover, and an oil conservator is installed on the support bracket. A detection component for detecting the working environment of the oil conservator is provided in the oil conservator; A salt - fog - resistant intelligent monitoring and control system is also provided. The salt - fog - resistant intelligent monitoring and control system includes a data acquisition unit, an edge computing and processing unit, a hierarchical warning unit, and an execution and regulation unit; The signal input terminal of the data acquisition unit is electrically connected to at least the signal output terminal of the liquid level sensing unit, and is used to synchronously acquire multi-dimensional operating status data during the operation of the transformer. The multi-dimensional operating status data includes at least oil level status data, oil temperature data, ambient salt spray concentration data, dehumidifier status data, corrugated oil conservator deformation data, and transformer load data. The signal input terminal of the edge computing processing unit is communicatively connected to the signal output terminal of the data acquisition unit, and is used to preprocess, extract features, identify abnormal states and predict operating trends of the acquired multi-dimensional operating status data, and generate corresponding control instructions based on the identification and prediction results. The signal input terminal of the hierarchical early warning unit is electrically connected to the signal output terminal of the edge computing processing unit, and is used to output early warning signals of the corresponding level according to the control instructions; The signal input terminal of the execution control unit is electrically connected to the signal output terminal of the edge computing processing unit, and the signal output terminal of the execution control unit is electrically connected to the matching execution components of the transformer, which is used to drive the corresponding execution components to complete the appropriate protection and control operation according to the control instructions.

[0007] Preferably, the detection component includes a corrugated oil conservator installed in the oil tank and a limiting collar installed on the corrugated oil conservator. An oil supply pipe is connected between the inner cavity of the oil tank and the transformer box. One end of the corrugated oil conservator is fixedly installed on the oil tank and forms a sealed cavity with the oil tank and the oil supply pipe.

[0008] Preferably, the oil tank has a detection slot, the limiting collar has a scraper that abuts against the side wall of the detection slot, a liquid level sensor is installed on one side of the scraper, a dehumidifier is installed on the oil tank, and a vent hole communicating with the dehumidifier is opened at one end of the detection slot of the oil tank.

[0009] Preferably, the scraper has two symmetrically arranged connecting holes, and a single-through diaphragm baffle is installed in the connecting hole; the conduction direction of the connecting hole is from the side of the detection groove near the corrugated oil conservator to the side of the detection groove where the breathing hole is opened, and the conduction direction of the single-through diaphragm baffle is consistent with the conduction direction of the connecting hole, which is used to restrict the reverse flow of transformer oil and condensate along the conduction direction.

[0010] Preferably, the data acquisition unit further includes a temperature sensing module, a salt spray concentration sensing module, a humidity sensing module, a deformation sensing module, a pressure sensing module, and a load acquisition module. The temperature sensing modules are respectively installed in the inner cavity of the transformer box and the inner cavity of the oil conservator, and are used to collect the oil temperature data of the transformer oil and the ambient temperature data. Both the salt spray concentration sensing module and the humidity sensing module are installed in the external environment of the transformer and in the breathing channel of the dehumidifier, and are used to collect environmental salt spray concentration data, environmental humidity data and salt spray humidity data in the breathing channel. The deformation sensing module is installed at the movable end of the corrugated oil conservator and is used to collect the deformation data of the corrugated oil conservator; the pressure sensing module is installed inside the sealed isolation cavity and is used to collect the internal pressure data of the sealed isolation cavity. The load acquisition module is electrically connected to the transformer's transformer components and is used to acquire real-time load data of the transformer.

[0011] Preferably, the edge computing processing unit has a built-in data preprocessing module, a feature extraction module, an anomaly recognition module, and a trend prediction module. The data preprocessing module is used to perform outlier removal, data normalization and data completion on the collected multi-dimensional operational status data. The data normalization process adopts the minimum-maximum normalization method. The feature extraction module is used to extract features from the preprocessed multi-dimensional standard data. It extracts the time-series features, fluctuation features and correlation features of the data at different time scales using the sliding window method, and constructs a multi-dimensional feature matrix. The anomaly identification module identifies and classifies the multi-dimensional feature matrix based on a preset classification and identification model, and outputs the anomaly type, anomaly location and anomaly level of the transformer operation. The trend prediction module performs time-series fitting and prediction on the preprocessed multi-dimensional standard data based on the time-series prediction model, and outputs the operating status change trend and fault occurrence probability of the transformer within a preset time period.

[0012] Preferably, the anomaly identification module uses an optimized support vector machine model for classification and identification. The classification and identification model is pre-trained and optimized using historical operation datasets and fault datasets of offshore wind power transformers.

[0013] The decision function of the classification and recognition model is: In the formula, For the classification and recognition results, sgn() is the sign function, and n is the number of training samples. For Lagrange multipliers, For sample labels, For kernel function, Let x be the feature vector of the training sample, b be the multi-dimensional feature matrix to be identified, and b be the bias term. The trend prediction module uses a long short-term memory network model for time-series prediction. This model is pre-trained and its parameters are optimized using historical time-series operation datasets of offshore wind power transformers. The calculation process for the forget gate, input gate, and output gate of the time-series prediction model is as follows: Forget Gate Calculation:

[0014] Input gate calculation: ; Cell status update: ; Output gate calculation: ; In the formula, Let be the outputs of the forget gate, input gate, and output gate at time t, respectively, and σ() be the sigmoid activation function. These are the weight matrices for the corresponding gating. This is the output of the hidden layer at time t. The input data is at time t. These are the bias terms for the corresponding gating. Let t represent the candidate cell state at time t. Let be the cell state at time t, and tanh() be the hyperbolic tangent activation function. This is the output of the hidden layer at time t.

[0015] Preferably, the graded early warning unit has a built-in early warning level classification rule. The early warning level classification rule divides the early warning signal into four levels: alert early warning, general early warning, important early warning and emergency early warning, based on the abnormality level and the probability of failure. Different levels of early warning signals correspond to different early warning methods and early warning push paths.

[0016] The execution control unit is equipped with at least the following execution components: a transformer tap changer drive module, a cooling control module, a dehumidifier switching module, a load control module, and an emergency power-off module. The execution control unit drives the corresponding execution components to complete the protection and control operations of tap changer adjustment, cooling power control, standby dehumidifier switching, load limiting, and emergency power-off according to the level and type of control instructions.

[0017] Preferably, the salt spray resistant intelligent monitoring and control system further includes a cloud control platform, which is bidirectionally communicatively connected to the edge computing processing unit through a wireless communication module; the edge computing processing unit synchronously uploads the preprocessed operation status data, feature extraction results, anomaly recognition results and trend prediction results to the cloud control platform, and the cloud control platform is used for centralized storage, centralized analysis and centralized control of the operation data of multiple offshore wind power transformers, and iteratively optimizes the classification recognition model and the time series prediction model built in the edge computing processing unit based on the massive operation data, and issues the optimized model parameters to the corresponding edge computing processing unit to complete model update.

[0018] Preferably, lifting ring plates are symmetrically and fixedly arranged at the four corners of the top surface of the installation top cover, and lifting holes are formed in the plate bodies of the lifting ring plates; the transformer tank and the installation top cover are fixedly connected through a plurality of fastening bolts, and a sealing gasket is arranged between the connection surface of the transformer tank and the installation top cover to enhance the sealing and isolation performance of the inner cavity of the transformer tank.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this solution, the closed isolation cavity structure formed by the corrugated oil conservator, the oil storage tank and the oil pipeline replaces the open cabinet body and the capsule type oil conservator structure in the prior art, isolates the direct contact between the external salt spray, water vapor, dust and transformer oil, and solves the core problems of transformer oil pollution and capsule oil conservator aging and failure caused by salt spray intrusion in the prior art. The corrugated oil conservator can adaptively expand and contract with the change of oil temperature, balance the internal pressure of the transformer, and cooperate with the self-cleaning structure of the scraping plate on the inner wall of the detection tank to eliminate the interference of salt spray condensate and oil stain on the oil level detection, ensure the long-term stability of the oil level detection accuracy. At the same time, the single-pass film baffle restricts the reverse reflux of the oil liquid, stabilizes the oil liquid flow state in the detection area, and improves the long-term reliability of the oil level detection.

[0020] 2. The salt spray resistant intelligent monitoring and control system supporting this solution synchronously collects all-dimensional operation status data such as transformer oil level, oil temperature, environmental salt spray concentration, oil conservator deformation, operation load, etc. through the multi-dimensional data acquisition unit, and completes data preprocessing, feature extraction, abnormal state recognition and operation trend prediction through the edge computing processing unit. It can accurately identify typical faults of offshore wind power transformers such as salt spray intrusion, moisture absorber failure, and oil conservator aging, output warning signals of corresponding levels based on the hierarchical warning rules, and drive the execution components to complete the adapted protection and control operations, adapt to the unattended operation conditions of offshore wind power, reduce the probability of equipment failures, and reduce the frequency of on-site maintenance operations.

[0021] 3. This solution utilizes a bidirectional communication architecture between a cloud-based management platform and an edge computing processing unit to achieve centralized storage, analysis, and management of operational data from multiple offshore wind turbine transformers within a wind farm. Based on the massive amounts of on-site operational data and fault labeling data aggregated in the cloud, the anomaly identification model and time-series prediction model deployed at the edge are continuously iterated and optimized. This addresses the issues of insufficient adaptability to on-site operating conditions after model pre-training and the decline in identification and prediction accuracy due to equipment aging. The optimized model parameters can be sent to the corresponding edge devices for updates, enabling the model to continuously adapt to individual equipment differences, environmental changes, and equipment aging processes, ensuring the long-term stability and effectiveness of intelligent management and control functions.

[0022] In summary, this solution addresses the root cause of salt spray intrusion and oil tank aging failure in offshore wind power transformers from a hardware structure perspective. From an intelligent management perspective, it enables real-time monitoring of equipment operating status, anomaly warning, and autonomous protection and control. From a cloud-based iteration perspective, it ensures the long-term adaptability and effectiveness of intelligent management capabilities. Overall, it improves the reliability and safety of transformers in harsh offshore environments with high salt spray and high humidity, extends the equipment's lifespan, and adapts to the special operating requirements of offshore wind power scenarios. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall first-view three-dimensional structure proposed in this invention; Figure 2 This is a schematic diagram of the three-dimensional structure of the corrugated oil pillow proposed in this invention; Figure 3 This is a schematic diagram of the three-dimensional structure of the detection groove proposed in this invention; Figure 4 This is a schematic diagram of the three-dimensional structure of the sealing ring proposed in this invention; Figure 5 This is a three-dimensional structural diagram of region A proposed in this invention; Figure 6 This is a block diagram of the overall architecture of the salt spray resistant intelligent monitoring and control system proposed in this invention; Figure 7 This is a block diagram of the internal modules of the edge computing processing unit proposed in this invention; Figure 8 This is a block diagram of the core functions of the cloud-based management and control platform proposed in this invention; Figure 9 This is a flowchart of the closed-loop process for intelligent transformer management proposed in this invention.

[0024] The following are the components listed in the diagram: 1. Transformer box; 2. Mounting top cover; 3. Lifting ring plate; 4. Tap changer; 5. High-voltage bushing; 6. Oil conservator; 7. Dehumidifier; 8. Corrugated oil tank; 9. Limiting collar; 10. Detection groove; 11. Breathing hole; 12. Scraper; 13. Connecting hole; 14. Liquid level sensor; 15. Single-pass diaphragm baffle. Detailed Implementation

[0025] The technical solutions 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.

[0026] See Figures 1-9 This embodiment discloses a salt spray resistant transformer suitable for offshore wind power. It is mainly used in the step-up transformation stage of offshore wind farms. It is adapted to the high salt spray, high humidity, strong corrosion and strong vibration operating environment at sea, and has intelligent monitoring and autonomous protection and control capabilities throughout the entire life cycle.

[0027] This invention discloses a salt spray resistant transformer suitable for offshore wind power. The transformer box 1 has a transformer element fixedly installed inside its cavity. This transformer element includes core components such as an iron core, windings, and insulation parts, used to achieve voltage level conversion of electrical energy. A mounting cover 2 is fixedly and sealed to the top opening of the transformer box 1 by multiple sets of fastening bolts. A salt spray resistant sealing gasket is placed between the contact surfaces of the transformer box 1 and the mounting cover 2 to seal the connection gaps, preventing external salt spray and moisture from entering the inner cavity of the transformer box 1 and improving the overall sealing and protection performance.

[0028] Multiple tap changers 4 are installed at equal intervals along the length of the top surface of the mounting cover 2. The lower end of the tap changer 4 extends into the inner cavity of the transformer box 1 and is electrically connected to the tap of the winding to adjust the transformer ratio and adapt to the voltage fluctuation of the power grid. Two sets of high-voltage bushings 5 ​​are symmetrically fixed on both sides of the top surface of the mounting cover 2. The lower end of the high-voltage bushing 5 is electrically connected to the high-voltage winding of the transformer element, and the upper end is used to connect to the high-voltage transmission line. At least one set of low-voltage bushings is fixedly installed on the side of one set of high-voltage bushings 5 ​​on the top surface of the mounting cover 2. The lower end of the low-voltage bushing is electrically connected to the low-voltage winding of the transformer element, and the upper end is used to connect to the low-voltage output line of the offshore wind turbine.

[0029] A support bracket is welded and fixed to one side of the top surface of the top cover 2. An oil conservator 6 is fixed to the top surface of the support bracket by bolts. The inner cavity of the oil conservator 6 is connected to the inner cavity of the transformer box 1 by a stainless steel oil pipe. This is used to compensate for the volume change of the transformer oil due to temperature changes and to balance the internal pressure of the transformer box 1. The oil conservator 6 is equipped with a detection component for detecting the working environment of the oil conservator 6. The detection component specifically includes a corrugated oil tank 8, a limiting collar 9, a scraper 12, a liquid level sensor 14, a dehumidifier 7, and a single-pass diaphragm baffle 15. It should be further explained that the corrugated oil conservator 8 adopts an axially expandable metal corrugated structure, which has the properties of resisting salt spray corrosion, aging, and vibration, and is used to adapt to the operating conditions of offshore wind power. The fixed end of the corrugated oil conservator 8 is welded and sealed to the inner wall of the oil tank 6. The inner cavity of the corrugated oil conservator 8, the inner cavity of the oil tank 6, and the flow channel of the oil pipeline together form a sealed and isolated cavity for containing transformer oil. This sealed and isolated cavity is completely isolated from the outside atmosphere, which avoids direct contact between the high salt spray and high humidity air at sea and the transformer oil, and solves the problems of transformer oil contamination and oil conservator aging failure caused by salt spray intrusion. The corrugated oil conservator 8 realizes axial expansion and contraction deformation according to the temperature change of the transformer oil inside the sealed and isolated cavity: when the oil temperature rises, the transformer oil volume expands, compressing the corrugated oil conservator 8 to make it axially contract; when the oil temperature drops, the transformer oil volume contracts, and the corrugated oil conservator 8 axially extends to compensate for the volume change, thereby automatically balancing the pressure inside the sealed and isolated cavity and ensuring the stability of the transformer insulation performance.

[0030] More specifically, the movable end of the corrugated oil conservator 8 is fixedly fitted with a limiting collar 9 by an interference fit. The inner cavity of the oil tank 6 is provided with a detection groove 10 that communicates with the sealed isolation cavity. The detection groove 10 is a cylindrical groove set along the extension and retraction direction of the corrugated oil conservator 8. The outer wall of the limiting collar 9 is integrally formed with a scraper 12. The scraper 12 adopts an annular plate structure that matches the cross-sectional shape of the inner cavity of the detection groove 10. The outer edge of the scraper 12 is fitted with a wear-resistant sealing layer, which slides against the inner wall of the detection groove 10 by interference fit. The scraper 12 slides back and forth along the inner wall of the detection groove 10 as the corrugated oil conservator 8 extends and retracts axially. It can scrape off the oil stains, salt spray condensed water and impurities attached to the inner wall of the detection groove 10 in real time, avoid the deposits from blocking or interfering with the detection elements, and ensure the long-term stability of detection accuracy.

[0031] A liquid level sensor 14 is fixedly installed on the side of the scraper 12 facing the inner cavity of the detection tank 10. The liquid level sensor 14 constitutes a liquid level sensing unit, which is used to collect the oil level data in the detection tank 10 in real time, so as to provide feedback on the volume change of transformer oil and the operating status of the oil conservator. A dehumidifier 7 is fixedly installed on the outer wall of the oil tank 6. A vent 11 is opened at the end of the detection tank 10 near the dehumidifier 7. The two ends of the vent 11 are connected to the inner cavity of the detection tank 10 and the air inlet of the dehumidifier 7, respectively, to realize the safe breathing of the transformer. The dehumidifier 7 can dry and desalinate the air entering the oil tank 6, further reducing the impact of the external environment on the interior.

[0032] Two sets of connecting holes 13 are symmetrically opened on the scraper 12. Each set of connecting holes 13 is fixedly installed with a single-pass diaphragm baffle 15. The conduction direction of the connecting hole 13 is from the side of the detection tank 10 near the corrugated oil conservator 8 to the side of the detection tank 10 with the breather hole 11. The conduction direction of the single-pass diaphragm baffle 15 is completely consistent with the conduction direction of the connecting hole 13. It only allows transformer oil and condensate to flow unidirectionally in the direction towards the breather hole 11, restricts its backflow, avoids the accumulation of oil-water mixture in the detection tank 10, stabilizes the oil flow state, and further improves the accuracy and stability of liquid level detection.

[0033] The top surface of the top cover 2 is symmetrically welded and fixed with lifting ring plates 3 at the four corners. Each set of lifting ring plates 3 has a lifting through hole for the overall lifting, transportation and offshore installation of the transformer. The symmetrical structure at the four corners can ensure uniform force during the lifting process, avoid equipment deformation, and adapt to the installation conditions of offshore wind power equipment.

[0034] The transformer proposed in this embodiment is also equipped with a salt spray resistant intelligent monitoring and control system. This system is suitable for unattended operation of offshore wind power and can realize real-time monitoring of transformer operation status, intelligent identification of anomalies, fault trend prediction and autonomous protection and control, thereby improving the reliability and maintenance efficiency of transformer operation in the offshore environment. The salt spray resistant intelligent monitoring and control system includes a data acquisition unit, an edge computing processing unit, a hierarchical early warning unit, an execution and control unit and a cloud management and control platform.

[0035] The signal input terminal of the data acquisition unit is electrically connected to at least the signal output terminal of the liquid level sensor 14, for synchronously acquiring multi-dimensional operating status data during the transformer's operation. In this embodiment, the data acquisition unit also includes a temperature sensing module, a salt spray concentration sensing module, a humidity sensing module, a deformation sensing module, a pressure sensing module, and a load acquisition module. The temperature sensing modules are installed at the bottom of the inner cavity of the transformer box 1, the inner cavity of the oil conservator 6, and the external environment of the transformer, respectively, to collect oil temperature data, oil conservator inner cavity temperature data, and ambient temperature data. The salt spray concentration sensing module and humidity sensing module are installed in pairs in the external environment of the transformer and in the breathing channel of the dehumidifier 7, respectively, to collect ambient salt spray concentration data, ambient humidity data, and salt spray humidity data in the breathing channel, thereby determining the deterioration state of the dehumidifier 7's protective performance. The deformation sensing module is installed between the movable end of the corrugated oil conservator 8 and the inner wall of the oil conservator 6, to collect axial deformation data of the corrugated oil conservator 8, thereby providing feedback on the operating status and aging degree of the oil conservator. The pressure sensing module is installed inside the sealed isolation cavity of the oil conservator 6, to collect internal pressure data of the sealed isolation cavity. The load acquisition module is electrically connected to the transformer's transformer components and incoming / outgoing line circuits, to collect real-time load data, voltage data, and current data of the transformer. It should be further explained that the data acquisition unit synchronously collects the above-mentioned multi-dimensional data according to the preset sampling frequency, and transmits the collected raw data to the edge computing processing unit through wired communication. The sampling frequency can be adaptively adjusted according to the transformer's operating status. Low-frequency sampling is used during steady-state operation, and high-frequency sampling is automatically switched when the load fluctuates or abnormal signs occur, taking into account both data integrity and device power consumption.

[0036] The edge computing processing unit employs an industrial-grade edge computing gateway, installed within the transformer's local control cabinet. Its signal input terminal communicates with the signal output terminal of the data acquisition unit, enabling preprocessing, feature extraction, anomaly identification, and operational trend prediction of the acquired multi-dimensional operational status data. Based on the identification and prediction results, it generates corresponding control commands. The edge computing processing unit integrates a data preprocessing module, a feature extraction module, an anomaly identification module, and a trend prediction module.

[0037] Normalization and data completion processes are used to eliminate data noise and dimensional differences, providing standardized data for subsequent feature extraction and model calculation.

[0038] Outlier removal employs the 3σ criterion, identifying and removing data that exceed the mean ± 3 times the standard deviation, thus avoiding the impact of acquisition errors and impulse interference on the accuracy of subsequent analysis. Data completion uses linear interpolation to fill in missing data resulting from outlier removal and sampling interruptions, ensuring the continuity of data sequence.

[0039] The data normalization process uses the min-max normalization method to map the original data with different dimensions and different value ranges to the interval [0,1].

[0040] The feature extraction module is used to extract features from the preprocessed multi-dimensional standard data. It uses a sliding window method to extract time-series features, fluctuation features, and correlation features at different time scales, constructing a multi-dimensional feature matrix. In this embodiment, the sliding window has multiple different time scales, including minute-level, hour-level, and day-level, corresponding to the feature extraction needs of short-term sudden anomalies, medium-term operational degradation, and long-term aging trends, respectively.

[0041] The time-series features include statistical characteristics such as mean, extreme values, variance, and rate of change of data in each dimension; fluctuation features include data fluctuation amplitude, fluctuation frequency, and peak interval; correlation features include coupling correlation characteristics between multi-dimensional data such as the correlation between oil temperature and load, the correlation between oil level and oil temperature, and the correlation between deformation data and pressure data, thereby comprehensively characterizing the transformer's operating status. After extraction, all features are arranged according to a preset dimensional order to construct a multi-dimensional feature matrix for model input.

[0042] The anomaly identification module classifies the multi-dimensional feature matrix based on a preset classification model, outputting the anomaly type, location, and level of the transformer operation. In this embodiment, the classification model is an optimized Support Vector Machine (SVM) model. This model has been pre-trained and optimized using historical operating datasets and fault datasets of offshore wind power transformers, and can identify various typical fault types of offshore wind power transformers, such as salt spray intrusion, oil tank aging, desiccant failure, abnormal oil level, abnormal oil temperature, and insulation degradation.

[0043] The decision function of the classification and recognition model is: In the formula, The classification result takes a value of +1 or -1, corresponding to normal and abnormal states respectively. For multi-classification scenarios, a one-to-one multi-classification strategy is used to construct multiple binary classifiers to achieve accurate differentiation of different abnormal types. sgn() is the sign function, which outputs +1 when the input value is greater than 0 and -1 when it is less than 0. n is the number of training samples. For Lagrange multipliers, only support vector correspondences Non-zero; The value is +1 for positive samples and -1 for negative samples. For the kernel function, this embodiment uses the radial basis function (RBF) to adapt to the nonlinear classification requirements of high-dimensional features. The kernel function expression is as follows: ; Let x be the feature vector of the training sample, b be the multi-dimensional feature matrix to be identified, and b be the bias term. The anomaly identification module outputs anomalies at four levels: minor, general, significant, and critical, providing a basis for subsequent tiered early warning and control measures. The trend prediction module performs time-series fitting and prediction on the preprocessed multi-dimensional standard data based on the time-series prediction model, outputting the operating status change trend and fault probability of the transformer within a preset time period. In this embodiment, the time-series prediction model is a Long Short-Term Memory (LSTM) network model, which can handle the dependencies of long-term time-series data, avoid the gradient vanishing problem of traditional recurrent neural networks, and adapt to the long-term trend prediction requirements of transformer operating status. The model is pre-trained and its parameters are optimized using historical time-series operating datasets of offshore wind power transformers. The calculation process of the forget gate, input gate, and output gate in the time series prediction model is as follows: Forget gate calculation: The forget gate is used to control the degree to which information from the cell state at the previous time step is forgotten. The calculation formula is as follows:

[0044] Input gate calculation: The input gate controls the degree to which new information at the current moment is stored in the cell state. It includes two parts: the input gate activation value and the candidate cell state. The calculation formula is as follows: ; Cell state update: Based on the calculation results of the forget gate and input gate, the cell state is updated to realize the transfer and update of temporal information. The calculation formula is as follows: ; Output gate calculation: The output gate controls the extent to which the current cell state is output to the hidden layer. It includes two parts: the output gate activation value and the hidden layer output. The calculation formula is as follows: ; In the above formula, Let be the outputs of the forget gate, input gate, and output gate at time t, respectively. σ() is the sigmoid activation function, which maps the input value to the interval [0,1] to achieve gated on / off control. These are the weight matrices corresponding to the forget gate, input gate, candidate cell state, and output gate, respectively. This is the output of the hidden layer at time t-1. The input is multi-dimensional time-series data at time t. These are the bias terms corresponding to the gating and candidate cell states, respectively. Let t represent the candidate cell state at time t. Let be the cell state at time t, and tanh() be the hyperbolic tangent activation function, which maps the input value to the interval [-1, 1]. This is the output of the hidden layer at time t.

[0045] In this embodiment, the trend prediction module can output the transformer operating status change trend at three time scales: 24 hours, 7 days, and 30 days, as well as the probability of fault occurrence within the corresponding time scale, providing data support for operation and maintenance planning and early protection and control.

[0046] The signal input terminal of the graded early warning unit is electrically connected to the signal output terminal of the edge computing processing unit, and is used to output early warning signals of corresponding levels according to control instructions. The graded early warning unit has built-in early warning level classification rules. The early warning level classification rules divide the early warning signals into four levels: alert early warning, general early warning, important early warning and emergency early warning, based on the anomaly level output by the anomaly identification module and the fault occurrence probability output by the trend prediction module. Different levels of early warning signals correspond to different early warning methods and early warning push paths.

[0047] In this embodiment, the warning corresponds to minor anomalies or low probability of failure, and is only displayed on the human-machine interface of the local control cabinet without being pushed to the operation and maintenance terminal; the general warning corresponds to general anomalies or low to medium probability of failure, and is pushed to the mobile terminal of the operation and maintenance personnel and the operation and maintenance platform via wireless communication; the important warning corresponds to important anomalies or medium to high probability of failure, and in addition to pushing the warning information, the sound and light alarm device is triggered simultaneously; the emergency warning corresponds to emergency anomalies or high probability of failure, and in addition to the above warning methods, it is pushed to the wind farm central control center simultaneously to trigger the emergency response process.

[0048] The signal input terminal of the execution control unit is electrically connected to the signal output terminal of the edge computing processing unit. The signal output terminal of the execution control unit is electrically connected to the corresponding execution components of the transformer, which are used to drive the corresponding execution components to complete the appropriate protection control operation according to the control instructions. In this embodiment, the execution components of the execution control unit include at least the transformer tap changer drive module, cooling control module, dehumidifier switching module, load control module and emergency power failure module.

[0049] The system includes the following modules: a tap changer drive module, which drives tap changer 4 to adjust the tap position and stabilize the output voltage according to the control command for abnormal voltage; a cooling control module, which adjusts the operating power and fan speed of the cooling system according to the control command for abnormal oil temperature to control the transformer oil temperature within the rated range; a dehumidifier switching module, which automatically switches to the standby dehumidifier according to the control command for deterioration of the protective performance of dehumidifier 7 to prevent salt spray intrusion caused by dehumidifier 7 failure; a load control module, which limits the operating load of the transformer according to the control command for transformer overload and insulation deterioration to prevent the fault from escalating further; and an emergency power-off module, which triggers the tripping of the high-voltage and low-voltage circuit breakers of the transformer according to the control command for emergency anomalies to achieve emergency power-off and ensure the safety of equipment and the power grid.

[0050] In this embodiment, the salt spray resistant intelligent monitoring and control system also includes a cloud-based control platform. The cloud-based control platform is deployed in the wind farm control center or cloud server and is bidirectionally connected to the edge computing processing unit through a 5G / 4G wireless communication module to realize the cloud-based aggregation and centralized control of transformer operation data.

[0051] The edge computing processing unit uploads preprocessed operational status data, feature extraction results, anomaly identification results, trend prediction results, and equipment operation and maintenance records to the cloud management platform according to a preset upload cycle. The cloud management platform has built-in data storage, centralized management and control, and model iteration and optimization modules, which are used to centrally store, analyze, and manage the operational data of multiple offshore wind power transformers in the wind farm. Based on massive operational data, it iteratively optimizes the classification and identification model and time series prediction model built into the edge computing processing unit, and sends the optimized model parameters to the corresponding edge computing processing unit to complete the model update.

[0052] The data storage module uses a distributed time-series database to store the massive amounts of uploaded transformer runtime sequence data, feature data, identification and prediction results, and operation and maintenance data. It supports rapid data retrieval and backtracking, providing a data foundation for centralized analysis and model iterative optimization.

[0053] The centralized management and control module is used to realize remote status monitoring, remote parameter configuration, operation and maintenance work order management and equipment life cycle management of all offshore wind power transformers in the wind farm. Operation and maintenance personnel can view the operating status of all transformers, abnormal early warning information and fault prediction results in real time through the visual interface of the centralized control center, and remotely issue parameter configuration instructions and operation and maintenance instructions to realize centralized management and control of offshore wind power transformers.

[0054] The model iteration and optimization module is used to continuously iterate and optimize the classification and recognition model (SVM) and time series prediction model (LSTM) deployed at the edge, based on the massive field operation data and fault labeling data aggregated in the cloud. This addresses the problems of insufficient generalization ability, decreased recognition accuracy, and increased prediction bias that occur in actual field conditions after model pre-training. It enables the model to continuously adapt to the individual differences of different transformers, changes in the marine environment, and the aging and deterioration process of equipment, ensuring the long-term stable performance of the model.

[0055] Before iterative optimization of the model, an iterative optimization dataset is first constructed, including a training set, a validation set, and a test set. The cloud management platform classifies and organizes all uploaded runtime data according to device number, runtime segment, and anomaly type. Abnormal and fault data are manually labeled using maintenance records and added to the original training dataset. At the same time, duplicate, invalid, and incorrectly labeled data are removed from the dataset, and the dataset is randomly divided into training, validation, and test sets in a ratio of 7:2:1.

[0056] For the iterative optimization of the classification and recognition model, each sample in the dataset contains a multi-dimensional feature matrix and corresponding labels, including normal state and various abnormal fault types; for the iterative optimization of the time series prediction model, each sample in the dataset contains multi-dimensional running data of continuous time series and corresponding future time period actual running state label data, which are used for supervised training of the model. The core of iterative optimization of the classification and recognition model lies in optimizing model parameters and updating the model based on newly added field datasets. The optimized parameters include Lagrange multipliers. The bias term b, the kernel width parameter γ, and the penalty coefficient C are used to control the model's fit and balance training error and generalization ability. The iterative optimization method employs grid search combined with cross-validation to find optimal parameters. The specific implementation process and related formulas are as follows: Constructing a parametric optimization grid: Pre-set the optimization range and search step size for the penalty coefficient C and the kernel function width parameter γ, and construct a two-dimensional parametric grid, where the optimization range of C is... The optimization range of γ is The search step size is set at logarithmic intervals to cover the optimal parameter range; Cross-validation training: For each (C,γ) parameter combination in the parameter grid, the model is trained on the training set using the k-fold cross-validation method. In this embodiment, k is set to 5. The training set is randomly divided into 5 subsets. Each time, 4 subsets are selected as training subsets and 1 subset is selected as validation subset. The model is trained and the classification accuracy of the validation set is calculated. Finally, the average of the 5 validation accuracies is taken as the evaluation index of the parameter set. Optimal parameter selection: The parameter combination (C, γ) with the highest average cross-validation accuracy is selected as the optimal parameters. Based on the optimal parameters and the complete training set, the classification and recognition model is retrained, and the optimized Lagrange multipliers are obtained. With the bias term b, complete the iterative training of the model; Model performance testing: The performance of the iteratively optimized model is tested using a test set. The model's classification accuracy, precision, recall, and F1 score are calculated. The formulas for calculating the evaluation metrics are as follows: Accuracy formula ; Precision formula ; Recall formula ; F1 value formula ; In the formula, TP is the number of true positives, that is, the number of abnormal samples correctly identified by the model; TN is the number of true negatives, that is, the number of normal samples correctly identified by the model; FP is the number of false positives, that is, the number of normal samples that the model misclassifies as abnormal; and FN is the number of false negatives, that is, the number of abnormal samples that the model misses as normal. Model update determination: When the test accuracy and F1 score of the iteratively optimized model are both better than the model currently deployed at the edge, the model iterative optimization is deemed effective, and the optimized model parameters (including optimal C, γ, ...) are updated. b) The data is transmitted wirelessly to the edge computing processing unit of the corresponding transformer. After receiving the data, the edge computing processing unit completes the local update of the model parameters. If the test performance is not better than the existing model, the existing model parameters are retained and the model is awaited for the next iteration optimization.

[0057] The core of the iterative optimization of the time series prediction model is the updating of model weights and bias parameters based on the newly added on-site time series dataset. Iterative training is performed using the Adam algorithm, an adaptive moment estimation algorithm based on backpropagation (BPTT), to minimize the model's prediction loss function and improve its prediction accuracy and generalization ability. The specific implementation process and related formulas are as follows: Loss function definition: The mean squared error (MSE) is used as the loss function for iterative training of the model to measure the deviation between the model's predicted values ​​and the actual values. The formula for the loss function is: In the formula, Loss is the loss value, and M is the time length of the training samples. The actual operating state value at time t. This represents the predicted value at time t output by the model. Gradient calculation: Based on the time backpropagation algorithm, the weight matrix of the loss function on all models is calculated backward along the time series. With bias term The gradient, and the formula for calculating the gradient is: In the formula, The gradient of the loss function with respect to the weight matrix W. Let W be the gradient of the loss function with respect to the bias term b, where W represents all weight matrices and b represents all bias terms. Adam optimization algorithm parameter update: The Adam optimization algorithm is used to iteratively update the model's weights and bias parameters. This algorithm can adaptively adjust the learning rate of each parameter to meet the training requirements of the LSTM model. The parameter update process is as follows: Updates to first-order moment and second-order moment estimates: ; ; Moment estimation bias correction: ; ; Parameter update: ; In the formula, This refers to all trainable parameters of the model, including the weight matrix and bias terms. The first moment estimate (mean) of the gradient at time t. The second moment estimate (variance) of the gradient at time t; , The attenuation coefficient for moment estimation is given in this embodiment. The value is 0.9. The value is 0.999; , These are the first and second moment estimates after bias correction, respectively; The learning rate is set to 0.001 in this embodiment, and a learning rate decay strategy is used during training; ϵ is a smoothing term with a value of To avoid the denominator being 0; These are the model parameters updated at time t; Iterative training termination condition: During the model iterative training process, after each training round, the validation loss value of the model is calculated using the validation set. When the validation loss value no longer decreases for several consecutive training rounds, the early stopping mechanism is triggered to terminate the training and avoid model overfitting. Training is also terminated when the number of training rounds reaches the preset maximum number of rounds.

[0058] Model performance testing: The performance of the iteratively optimized time series prediction model was tested using a test set. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination of the model were calculated. The evaluation index calculation formula is as follows: Mean Absolute Error Formula ; Root mean square error formula

[0059] Coefficient of determination formula

[0060] In the formula, This is the mean of the true values; the meanings of the other parameters are consistent with the loss function formula. Model update determination: When the MAE and RMSE of the iteratively optimized model are both lower than those of the model currently deployed at the edge, and When the model is closer to 1, the iterative optimization is deemed effective. All weight matrices, bias term parameters, and hyperparameters of the optimized model are then transmitted wirelessly to the edge computing processing unit of the corresponding transformer. Upon receiving the data, the edge computing processing unit completes the local update of the model. If the test performance is not better than the existing model, the existing model parameters are retained, and the model awaits the next iterative optimization.

[0061] In this embodiment, the iterative optimization cycle of the time series prediction model is synchronized with that of the classification and recognition model. At the same time, iterative optimization can be triggered according to seasonal changes and environmental conditions of offshore wind power, so that the model can continuously adapt to the complex operating conditions of offshore wind power.

[0062] The transformer in this embodiment, through its corrugated oil conservator 8 and sealed isolation cavity structure design, fundamentally isolates the transformer oil from external salt spray and water vapor, solving the problems of easy contamination of the oil tank 6, oil conservator aging and failure, and insufficient protection in existing offshore wind power transformers. The self-cleaning design of the scraper 12 and the flow stabilization design of the single-pass thin-film baffle 15 ensure long-term accurate and stable oil level detection. Simultaneously, the accompanying salt spray resistant intelligent monitoring and control system enables real-time monitoring of the transformer's operating status, intelligent anomaly identification, fault trend prediction, and autonomous protection and control, adapting to the unmanned operation requirements of offshore wind power. Through the model iteration and optimization mechanism of the cloud-based management platform, the intelligent analysis model can continuously adapt to changes in on-site operating conditions, ensuring the long-term recognition and prediction accuracy of the model and improving the operational reliability, safety, and lifespan of the transformer in high-salt-spray environments at sea.

[0063] Working principle: When using this invention, first complete the installation of the main equipment according to the transformer operation manual, match the signal transmission and display devices for each sensing element, the transformer element in the inner cavity of the transformer box 1 completes the voltage level conversion of electrical energy, and the tap changer 4, high voltage bushing 5 and low voltage bushing on the top cover 2 cooperate to realize circuit connection, transformation ratio adjustment and electrical energy input and output, and ensure the stable operation of the basic power transformation function of the equipment.

[0064] The support bracket on one side of the top cover 2 provides stable installation support for the oil conservator 6. The oil conservator 6 is connected to the inner cavity of the transformer box 1 through the oil supply pipe. The corrugated oil tank 8, together with the oil conservator 6 and the oil supply pipe, forms a sealed and isolated cavity. It adapts to the axial expansion and contraction with the temperature change of the transformer oil. When the oil temperature rises, the oil volume expands and compresses the corrugated oil tank 8. When the oil temperature drops, the oil volume contracts and the corrugated oil tank 8 extends to compensate, thereby automatically balancing the internal pressure of the transformer. At the same time, it isolates the direct contact between external salt spray, water vapor and transformer oil from the source.

[0065] During the expansion and contraction of the corrugated oil conservator 8, the limiting collar 9 moves synchronously. The scraper 12 on the limiting collar 9 slides back and forth against the inner wall of the detection tank 10, scraping away oil stains, salt spray condensate, and impurities adhering to the inner wall in real time, preventing deposits from interfering with the detection element. The liquid level sensor 14 on the scraper 12 collects the oil level data in the detection tank 10 in real time and converts the physical quantity of oil level into an electrical signal output. The single-pass thin film baffle 15 in the connecting hole 13 on the scraper 12 only allows the oil and condensate to flow unidirectionally toward the breather hole 11, restricting its backflow and stabilizing the oil flow state in the detection area. The detection tank 10 is connected to the dehumidifier 7 through the breather hole 11. The dehumidifier 7 dries and removes salt spray from the air entering the oil tank 6, enabling the transformer to breathe safely.

[0066] The equipment is equipped with a salt spray resistant intelligent monitoring and control system that operates synchronously. The data acquisition unit simultaneously collects multi-dimensional operating status data such as transformer oil level, oil temperature, ambient salt spray concentration, dehumidifier status, corrugated oil conservator deformation, and equipment load, and transmits them to the edge computing processing unit. The edge computing processing unit preprocesses and extracts features from the collected data, completes abnormal state identification and operating trend prediction, and generates corresponding control commands based on the identification and prediction results. The hierarchical early warning unit outputs early warning signals of corresponding levels according to the control commands, and the execution control unit simultaneously drives the tap changer drive module, cooling control module, dehumidifier switching module, and other supporting execution components according to the control commands to complete the appropriate protection and control operations.

[0067] The edge computing processing unit synchronously uploads the pre-processed operating data, feature extraction results, anomaly identification and trend prediction results to the cloud management platform. The cloud management platform centrally stores, analyzes and manages the operating data of multiple transformers in the wind farm. Based on the massive amount of aggregated field operating data and fault labeling data, it iteratively optimizes the analysis and prediction model deployed at the edge, and sends the optimized model parameters to the corresponding edge computing processing unit to complete the model update. This enables the intelligent management function to continuously adapt to changes in field conditions and ensures the long-term stable operation of equipment in the high salt spray and high humidity environment at sea.

[0068] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A salt spray resistant transformer suitable for offshore wind power, comprising a transformer box (1) and transformer elements installed inside the transformer box (1), a mounting cover (2) is installed on the top of the transformer box (1), multiple tap changers (4) are installed at intervals in the middle of the mounting cover (2), two high-voltage bushings (5) are symmetrically installed on the mounting cover (2), and low-voltage bushings are installed on the mounting cover (2) at the side end of the high-voltage bushings (5), characterized in that: A support bracket is welded to one side of the mounting top cover (2), and an oil tank (6) is installed on the support bracket. The oil tank (6) is equipped with a detection component for detecting the working environment of the oil tank (6). It is also equipped with a salt spray resistant intelligent monitoring and control system, which includes a data acquisition unit, an edge computing processing unit, a hierarchical early warning unit, and an execution control unit. The signal input terminal of the data acquisition unit is electrically connected to at least the signal output terminal of the liquid level sensing unit, and is used to synchronously acquire multi-dimensional operating status data during the operation of the transformer. The multi-dimensional operating status data includes at least oil level status data, oil temperature data, ambient salt spray concentration data, dehumidifier status data, corrugated oil conservator deformation data, and transformer load data. The signal input terminal of the edge computing processing unit is communicatively connected to the signal output terminal of the data acquisition unit, and is used to preprocess, extract features, identify abnormal states and predict operating trends of the acquired multi-dimensional operating status data, and generate corresponding control instructions based on the identification and prediction results. The signal input terminal of the hierarchical early warning unit is electrically connected to the signal output terminal of the edge computing processing unit, and is used to output early warning signals of the corresponding level according to the control instructions; The signal input terminal of the execution control unit is electrically connected to the signal output terminal of the edge computing processing unit, and the signal output terminal of the execution control unit is electrically connected to the matching execution components of the transformer, which is used to drive the corresponding execution components to complete the appropriate protection and control operation according to the control instructions.

2. The salt spray resistant transformer suitable for offshore wind power according to claim 1, characterized in that, The detection component includes a corrugated oil conservator (8) installed in the oil tank (6) and a limiting collar (9) installed on the corrugated oil conservator (8). An oil supply pipe is connected between the oil tank (6) and the inner cavity of the transformer box (1). One end of the corrugated oil conservator (8) is fixedly installed on the oil tank (6) and forms a closed cavity with the oil tank (6) and the oil supply pipe.

3. The salt spray resistant transformer suitable for offshore wind power according to claim 2, characterized in that, The oil tank (6) has a detection groove (10), and the limiting collar (9) has a scraper (12) that abuts against the side wall of the detection groove (10). A liquid level sensor (14) is installed on one side of the scraper (12). A dehumidifier (7) is installed on the oil tank (6). A breathing hole (11) communicating with the dehumidifier (7) is opened at one end of the detection groove (10) of the oil tank (6).

4. The salt spray resistant transformer suitable for offshore wind power according to claim 3, characterized in that, Two symmetrical connecting holes (13) are provided on the scraper (12), and a single-through thin film baffle (15) is installed in the connecting hole (13). The conduction direction of the connecting hole (13) is the side of the detection groove (10) near the corrugated oil tank (8) towards the side of the detection groove (10) where the breathing hole (11) is opened. The conduction direction of the single-through thin film baffle (15) is consistent with the conduction direction of the connecting hole (13) and is used to restrict the reverse flow of transformer oil and condensate along the conduction direction.

5. The salt spray resistant transformer for offshore wind power according to claim 3, characterized in that, The data acquisition unit also includes a temperature sensing module, a salt spray concentration sensing module, a humidity sensing module, a deformation sensing module, a pressure sensing module, and a load acquisition module. The temperature sensing modules are respectively installed in the inner cavity of the transformer box (1) and the inner cavity of the oil tank (6) to collect the oil temperature data of the transformer oil and the ambient temperature data. The salt spray concentration sensing module and the humidity sensing module are both installed in the external environment of the transformer and in the breathing channel of the dehumidifier (7) to collect environmental salt spray concentration data, environmental humidity data and salt spray humidity data in the breathing channel. The deformation sensing module is installed at the movable end of the corrugated oil tank (8) and is used to collect the deformation data of the corrugated oil tank (8); the pressure sensing module is installed inside the sealed isolation cavity and is used to collect the internal pressure data of the sealed isolation cavity. The load acquisition module is electrically connected to the transformer's transformer components and is used to acquire real-time load data of the transformer.

6. The salt spray resistant transformer for offshore wind power according to claim 1, characterized in that, The edge computing processing unit has a built-in data preprocessing module, feature extraction module, anomaly recognition module and trend prediction module; The data preprocessing module is used to perform outlier removal, data normalization and data completion on the collected multi-dimensional operational status data. The data normalization process adopts the minimum-maximum normalization method. The feature extraction module is used to extract features from the preprocessed multi-dimensional standard data. It extracts the time-series features, fluctuation features and correlation features of the data at different time scales using the sliding window method, and constructs a multi-dimensional feature matrix. The anomaly identification module identifies and classifies the multi-dimensional feature matrix based on a preset classification and identification model, and outputs the anomaly type, anomaly location and anomaly level of the transformer operation. The trend prediction module performs time-series fitting and prediction on the preprocessed multi-dimensional standard data based on the time-series prediction model, and outputs the operating status change trend and fault occurrence probability of the transformer within a preset time period.

7. The salt spray resistant transformer for offshore wind power according to claim 6, characterized in that, The anomaly identification module uses an optimized support vector machine model for classification and identification. The classification and identification model is pre-trained and optimized using historical operation datasets and fault datasets of offshore wind power transformers. The decision function of the classification and recognition model is: In the formula, For the classification and recognition results, sgn() is the sign function, and n is the number of training samples. For Lagrange multipliers, For sample labels, For kernel function, Let x be the feature vector of the training sample, b be the multi-dimensional feature matrix to be identified, and b be the bias term. The trend prediction module uses a long short-term memory network model for time-series prediction. This model is pre-trained and its parameters are optimized using historical time-series operation datasets of offshore wind power transformers. The calculation process for the forget gate, input gate, and output gate of the time-series prediction model is as follows: Forget Gate Calculation: Input gate calculation: ; Cell status update: ; Output gate calculation: ; In the formula, Let be the outputs of the forget gate, input gate, and output gate at time t, respectively, and σ() be the sigmoid activation function. These are the weight matrices for the corresponding gating. This is the output of the hidden layer at time t-1. The input data is at time t. These are the bias terms for the corresponding gating. Let t represent the candidate cell state at time t. Let be the cell state at time t, and tanh() be the hyperbolic tangent activation function. This is the output of the hidden layer at time t.

8. The salt spray resistant transformer for offshore wind power according to claim 1, characterized in that, The hierarchical early warning unit has built-in early warning level classification rules. The early warning level classification rules divide the early warning signals into four levels: alert early warning, general early warning, important early warning and emergency early warning, based on the anomaly level and the probability of failure. Different levels of early warning signals correspond to different early warning methods and early warning push paths. The execution control unit is equipped with at least the transformer tap changer drive module, cooling control module, dehumidifier switching module, load control module and emergency power cut-off module. The execution control unit drives the corresponding execution components to complete the protection and control operations of tap changer (4) adjustment, cooling power control, standby dehumidifier switching, load limiting and emergency power cut-off according to the level and type of control instructions.

9. The salt spray resistant transformer for offshore wind power according to claim 1, characterized in that, The salt spray resistant intelligent monitoring and control system also includes a cloud-based control platform, which is bidirectionally connected to the edge computing processing unit via a wireless communication module. The edge computing processing unit synchronously uploads the preprocessed operating status data, feature extraction results, anomaly identification results, and trend prediction results to the cloud-based control platform. The cloud-based control platform is used for centralized storage, analysis, and control of the operating data of multiple offshore wind power transformers. Based on massive amounts of operating data, iteratively optimizes the classification and identification model and time series prediction model built into the edge computing processing unit, and sends the optimized model parameters to the corresponding edge computing processing unit to complete the model update.

10. The salt spray resistant transformer for offshore wind power according to claim 1, characterized in that, The top surface of the mounting cover (2) is symmetrically fixed with lifting ring plates (3) at the four corners. The lifting ring plates (3) have lifting through holes. The transformer box (1) and the mounting cover (2) are fixedly connected by multiple sets of fastening bolts. A sealing gasket is provided between the connection surfaces of the transformer box (1) and the mounting cover (2).

Images