New energy transformer intelligent monitoring management system

Abstract

The invention discloses an intelligent monitoring management system for a new energy transformer, and belongs to the technical field of new energy transformers. A data preprocessing module; a time sequence modeling module; a fault trend prediction module; an early warning trigger module; a model updating module; and a control linkage module. Through the design of multi-source sensor fusion and edge computing integration, the system can collect multi-dimensional data of the operation state of the transformer in real time, localization processing and feature extraction are carried out through edge equipment, global state sensing and low-delay response are achieved, and the reliability of the system is improved. The effects of improving the monitoring precision and shortening the fault response time are achieved, the design of an attention mechanism is introduced through the improved LSTM network, the model can dynamically adjust weight distribution of different time steps, accurate extraction of key time sequence features is achieved, and the effects of improving the fault trend prediction precision and prolonging the early warning window period are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy transformers, and in particular to an intelligent monitoring and management system for new energy transformers. Background Art

[0002] As core equipment for smart grids and renewable energy power generation systems (such as wind power and photovoltaics), new energy transformers undertake critical functions such as voltage conversion, energy transmission, and system stability. With the advancement of the "dual carbon" goals and the rapid expansion of renewable energy installed capacity, transformers must adapt to complex operating environments (such as frequent load fluctuations and extreme weather conditions) and meet high reliability requirements.

[0003] However, traditional transformer monitoring systems still rely primarily on manual inspections, single-parameter monitoring, and offline diagnosis, resulting in problems such as a single monitoring dimension and delayed response. Existing transformer monitoring systems neglect the coordinated analysis of key indicators such as partial discharge intensity, dissolved gas content in oil, and ambient temperature and humidity. Transformer insulation degradation may manifest simultaneously as a sudden increase in partial discharge intensity and abnormal oil temperature, but a single-parameter monitoring system may only detect the increased oil temperature, misinterpreting it as a load overload rather than an insulation fault. Early fault signals are also extremely weak and susceptible to electromagnetic interference (low signal-to-noise ratio). Without cross-validation of multi-dimensional data, the system cannot detect potential hazards in a timely manner. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent monitoring and management system for new energy transformers to solve the problems raised in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: an intelligent monitoring and management system for new energy transformers, comprising: Data acquisition module, used to obtain voltage, current, temperature, partial discharge intensity and ambient temperature and humidity data during transformer operation; A data preprocessing module, used for standardizing the data and filling in missing values; A time series modeling module is used to construct the preprocessed data into a time series and extract time series features through an improved LSTM network, where the improved LSTM network has an attention configuration to enhance the model's focus on important time steps; The fault trend prediction module is used to calculate the parameter change trend within the next N time steps based on the extracted time series features. The trend prediction formula is as follows:

[0006] in, Represents the input vector at the current moment, Indicates the hidden state, Indicates the output prediction value at the next moment, represents the activation function; The early warning trigger module is used to generate an early warning signal when the predicted value deviates from the set threshold range; A model updating module is used to perform online incremental training on the improved LSTM network based on the latest collected data; The control linkage module is used to receive early warning signals and perform corresponding control operations, including but not limited to circuit breaker tripping, cooling device startup or remote alarm notification.

[0007] Preferably, the time series modeling module constructs input samples using a sliding window method, and each sample has historical data of T consecutive time points for training the improved LSTM network model. The attention configuration is based on the principle of adaptive weight distribution, and adjusts the contribution of each time step by scoring the importance of different time steps.

[0008] In this solution, the model update module preferably uses an online incremental learning algorithm. Each time new data is received, some weight parameters of the model are updated, and the weight coefficients in the attention configuration are adjusted to adapt the model to the new operating conditions. The update formula is as follows:

[0009] in, represents the learning rate, represents the gradient of the loss function with respect to the weight, Represent new input and label data respectively.

[0010] Preferably, the control linkage module includes a relay control unit and a communication interface unit. When a warning signal is received, the corresponding control instruction is matched according to a preset strategy table and corresponding actions are executed. The strategy table also includes priority sorting for different fault types to ensure that critical faults are responded to in a timely manner.

[0011] Preferably, the data acquisition module includes multiple sensor nodes, which are respectively arranged in the transformer body, oil tank, bushing and cooling structure to achieve multi-dimensional operating status perception. The sensor nodes support wireless ad hoc networking function to facilitate rapid deployment and expansion of the monitoring range.

[0012] In this solution, the data preprocessing module uses the Z-score normalization method to normalize the original data, and the formula is as follows:

[0013] in, represents the mean, In addition, an outlier detection algorithm is used to automatically identify and remove abnormal data points that may affect the accuracy of the model.

[0014] Preferably, the control linkage module communicates with the on-site equipment through the Modbus protocol and uploads the warning information to the host computer platform through the MQTT protocol. At the same time, the module has an offline caching function, which saves key operation logs when the network is interrupted and automatically synchronizes them to the cloud server after the connection is restored.

[0015] This solution is preferred, where the system is deployed on an edge computing device, has local data processing capabilities, and supports data synchronization with cloud servers. The edge device has a built-in redundant power supply to ensure that it can continue to work for at least 2 hours in the event of a city power outage.

[0016] Preferably, the early warning trigger module is provided with a dynamic threshold mechanism, which adaptively adjusts the alarm threshold range according to the statistical results of historical data. At the same time, the module integrates the expert system knowledge base, which can provide more accurate early warning judgment basis under complex working conditions.

[0017] Preferably, the present solution adopts transfer learning technology in the training process of the improved LSTM network, using the monitoring data of the power equipment as the basis for pre-training, thereby accelerating the model convergence speed and improving the generalization performance.

[0018] Compared with the prior art, the technical effects and advantages of the present invention are: This intelligent monitoring and management system for new energy transformers, through the design of multi-source sensor fusion and edge computing integration, enables the system to collect multi-dimensional data on the transformer's operating status (voltage, current, temperature, partial discharge intensity, ambient temperature and humidity) in real time, and perform local processing and feature extraction through edge devices, achieving global status perception and low-latency response, thereby improving monitoring accuracy and shortening fault response time.

[0019] Through the design of Z-score normalization and outlier detection algorithms, raw data can eliminate dimensional differences and remove noise interference, achieving high-quality data input and improving model training stability and prediction reliability. By introducing an attention mechanism into the improved LSTM network, the model can dynamically adjust the weight distribution of different time steps, achieving accurate extraction of key time series features, improving the accuracy of fault trend prediction and extending the warning window.

[0020] Through the design of online incremental learning and strategy table priority sorting, the system can dynamically adjust model parameters according to new data and execute hierarchical control instructions, realizing adaptive working condition changes and efficient fault handling, thereby extending the model life cycle and reducing operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 This is a flow chart of an intelligent monitoring and management system for new energy transformers of the present invention. DETAILED DESCRIPTION

[0023] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art are not described to avoid confusion with the present invention.

[0024] Unless otherwise defined, the directions of up, down, left, right, front, back, inside and outside involved in this document are based on the directions of up, down, left, right, front, back, inside and outside shown in the figures of the present invention, and are explained here together.

[0025] This embodiment provides Figure 1 The intelligent monitoring and management system for new energy transformers shown in the figure includes: The data acquisition module is used to obtain voltage, current, temperature, partial discharge intensity, and ambient temperature and humidity data during transformer operation. Specifically, the distributed sensor network collects the transformer's electrical parameters, mechanical status (vibration, partial discharge), and environmental parameters in real time, such as: Voltage / current sensor: Hall effect sensor or current transformer (CT) monitors three-phase current and voltage waveforms in real time; Partial discharge sensor: High-frequency coupling capacitor or ultrasonic probe to detect partial discharge signals (such as 100kHz-1MHz frequency band); Environmental sensor: Temperature and humidity sensor (such as SHT31) monitors the environmental conditions around the transformer; In actual wind farms, ultrasonic sensors are installed on transformer bushings. When the local discharge intensity exceeds a threshold (e.g., 10pC), data collection is triggered and a timestamp is recorded. The data preprocessing module is used to standardize the data and fill in missing values. Specifically, the sliding average method or KNN algorithm is used to fill in missing data. For example, if the voltage data at a certain moment is missing, it is replaced by the average value of the previous three time steps. For example, if the current data within 10 seconds is lost due to a communication interruption during a data collection, the system automatically fills in the missing data with the sliding average value of adjacent time points. A time series modeling module is used to construct the preprocessed data into a time series and extract time series features through an improved LSTM network, where the improved LSTM network has an attention configuration to enhance the model's focus on important time steps; The fault trend prediction module is used to calculate the parameter change trend within the next N time steps based on the extracted time series features. The trend prediction formula is as follows:

[0026] in, Represents the input vector at the current moment, Indicates the hidden state, Indicates the output prediction value at the next moment, represents the activation function; The early warning trigger module is used to generate an early warning signal when the predicted value deviates from the set threshold range. Specifically, by comparing the predicted value with the threshold, it determines whether there is an abnormal trend. For example, if the predicted oil temperature will exceed 85°C (threshold) in the next three hours, an early warning is triggered. For example, if the transformer oil temperature is currently 75°C, and the model predicts that it will rise to 88°C in the next two hours, an "overheating risk" warning is triggered. A model update module is used to perform online incremental training on the improved LSTM network based on the latest collected data. For example, in a high-temperature summer environment, the system updates the model through incremental learning to increase the attention weight to oil temperature changes. The control linkage module receives warning signals and executes corresponding control actions, including but not limited to circuit breaker tripping, cooling unit activation, or remote alarm notification. Specifically, it matches preset control instructions (such as circuit breaker tripping and cooling unit activation) based on the warning level (e.g., Levels 1-3). It communicates with the relay control unit via the Modbus protocol to trigger physical device operations. If the system detects excessive partial discharge intensity (Level 2), it automatically activates the cooling fan and sends a text message to notify maintenance personnel.

[0027] In this embodiment, the time series modeling module constructs input samples using a sliding window approach, with each sample containing historical data from T consecutive time points for training the improved LSTM network model. The attention configuration is based on the principle of adaptive weight allocation, adjusting the contribution of each time step by assigning importance scores to different time steps. Specifically, a sliding window (with a window length of T = 24 hours) is used to construct time series samples, with each sample containing data from T historical time points. For example, assuming T = 24, the model will use the temperature data from the past 24 hours as input to predict the temperature trend over the next three hours. The sliding window construction and adaptive attention mechanism enable the model to dynamically adjust the weight allocation for different time steps, extracting key time series features and optimizing computational resources, thereby improving prediction accuracy and reducing model complexity. Through adaptive weight allocation (such as softmax normalization), the model assigns higher weights to recent or anomalous data points (e.g., a weight of 0.6 for the last three hours of data), improving prediction sensitivity. Low-weighted time steps (such as historical steady-state data) are simplified, reducing model computational overhead. The weight adjustment of the attention configuration is combined with the dynamic update of the sliding window. In wind farm scenarios (such as sudden changes in wind speed leading to a sudden increase in load), the model can prioritize processing new data and avoid the lag problem caused by the fixed weight allocation of traditional LSTM.

[0028] In this embodiment, the model update module uses an online incremental learning algorithm. Each time new data is received, some weight parameters of the model are updated, and the weight coefficients in the attention configuration are adjusted to adapt the model to the new operating conditions. The update formula is as follows:

[0029] in, represents the learning rate, represents the gradient of the loss function with respect to the weight, Representing new input and label data, respectively. Through online incremental learning and dynamic adjustment of attention weights, the model can adapt to changing operating conditions and optimize its performance, achieving adaptive learning and improved model generalization capabilities, extending the model's effective lifecycle and reducing the need for manual intervention.

[0030] In this embodiment, the control linkage module includes a relay control unit and a communication interface unit. When an early warning signal is received, the corresponding control instruction is matched according to a preset strategy table and the corresponding action is executed. The strategy table also includes priority sorting for different fault types to ensure that critical faults are responded to in a timely manner. The structure of strategy table 1 is as follows: Warning level Fault type Control instructions Priority Level 1 Partial discharge exceeds the standard Start cooling device, SMS notification high Level 2 Abnormal oil temperature Adjust load, email notification middle Level 3 Abnormal vibration Record logs and wait for manual inspection Low Prioritization rules: Based on risk assessment of fault type (e.g., partial discharge risk > abnormal oil temperature > abnormal vibration). If both partial discharge and abnormal oil temperature are detected, the system prioritizes executing the Level 1 cooling device startup command.

[0031] In this embodiment, the data acquisition module includes multiple sensor nodes, which are respectively arranged in the transformer body, oil tank, bushing and cooling structure to achieve multi-dimensional operating status perception. The sensor nodes support wireless ad hoc networking functions to facilitate rapid deployment and expansion of the monitoring range. Specifically, the nodes automatically form a star or mesh network to ensure data transmission reliability. When in a photovoltaic power station in a remote mountainous area, the sensor nodes upload data to the base station through the LoRa ad hoc network without the need to lay optical fibers. Through the design of wireless ad hoc networking and multi-sensor node deployment, the system can flexibly expand the monitoring range and improve the integrity of data collection, achieving low-cost deployment and high-reliability perception, achieving the effect of reducing wiring costs and adapting to complex terrain. Specifically, the sensor nodes realize synchronous data transmission through wireless ad hoc networks (such as mesh topology), avoiding the high cost and construction difficulty of traditional optical fiber laying. Sensors are placed at key locations such as the transformer body, oil tank, and bushing, and multi-source data fusion (such as voltage, temperature, and partial discharge intensity) is used to improve fault diagnosis accuracy. This allows the node self-organizing network in mountainous wind farm scenarios to automatically repair broken links (for example, when a node fails, other nodes take a detour for transmission), ensuring that data is not lost.

[0032] In this embodiment, the data preprocessing module uses the Z-score normalization method to normalize the original data, and the formula is as follows:

[0033] in, represents the mean, represents the standard deviation. Furthermore, an outlier detection algorithm is employed to automatically identify and remove anomalous data points that could affect model accuracy. Specifically, if a data point deviates from the mean by more than three standard deviations, it is marked as an outlier. Isolated points are identified by randomly partitioning the data space. If a voltage sample value suddenly drops to 0 (due to sensor failure), the system detects the anomaly using the 3σ principle and removes the data point.

[0034] In this embodiment, the control linkage module communicates with field devices via the Modbus protocol and uploads warning information to the host computer platform via the MQTT protocol. Furthermore, the module features an offline caching function, saving key operation logs during network interruptions and automatically synchronizing them to the cloud server upon restoration. Specifically, the Modbus protocol is used for communication with industrial equipment such as relays and circuit breakers (e.g., sending control commands via a serial port); the MQTT protocol is used for cloud messaging (e.g., uploading warning information via TCP / IP). During network outages, key operation logs are stored locally on an SD card and synchronized to the cloud upon restoration. In the event of a network outage, the system caches the "cooling device start" command locally and uploads it to the cloud platform upon network restoration. The dual-protocol Modbus and MQTT communication and offline caching design enable the system to operate stably and ensure data integrity in a variety of network environments, achieving highly reliable communication and fault tolerance during network outages, thereby improving system availability and supporting extreme operating conditions. In the event of a typhoon-induced network outage, the system performs local control (e.g., cooling device start) via Modbus while caching logs, which are then uploaded to the cloud platform via MQTT upon network restoration.

[0035] In this embodiment, the system is deployed on edge computing devices, equipped with local data processing capabilities and supporting data synchronization with cloud servers. The edge devices have built-in redundant power supplies, including a built-in UPS (uninterruptible power supply) and lithium batteries, ensuring continuous operation for at least two hours during a utility power outage. In the event of a typhoon-induced utility power outage, the edge devices rely on the UPS to continue operating, complete data processing, and save critical logs.

[0036] In this embodiment, the early warning trigger module features a dynamic threshold mechanism that adaptively adjusts the alarm threshold range based on historical data statistics. Furthermore, the module integrates an expert system knowledge base, enabling more accurate early warning judgments under complex operating conditions. Through the collaborative design of dynamic thresholds and the expert system knowledge base, the early warning mechanism adapts to complex operating conditions and provides accurate judgments, achieving adaptive early warning and intelligent decision-making, reducing false alarm rates and improving targeted responses.

[0037] In this embodiment, the improved LSTM network training process employs transfer learning techniques, using monitoring data from power equipment as a pre-training basis to accelerate model convergence and improve generalization performance. Specifically, the LSTM model is pre-trained using monitoring data from other power equipment (such as generators and cables), and model parameters are fine-tuned on transformer data (e.g., freezing the convolutional layers and training only the LSTM layers). For example, the system pre-trains the model using generator vibration data and then fine-tunes it using transformer vibration data, accelerating convergence and improving generalization.

[0038] It should be noted that, in this article, relational terms such as one and two are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "including an element defined by ... does not exclude the presence of other identical elements in the process, method, article or device that includes the element."

[0039] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring and management system for new energy transformers, characterized in that: include: Data acquisition module, used to obtain data during transformer operation; A data preprocessing module, used for standardizing the data and filling in missing values; A time series modeling module is used to construct the preprocessed data into a time series and extract time series features through an improved LSTM network, where the improved LSTM network has an attention configuration to enhance the model's focus on important time steps; Fault trend prediction module, used to calculate the parameter change trend within the next N time steps based on the extracted time series features; The early warning trigger module is used to generate an early warning signal when the predicted value deviates from the set threshold range; A model updating module is used to perform online incremental training on the improved LSTM network based on the latest collected data; The control linkage module is used to receive early warning signals and perform corresponding control operations, including but not limited to circuit breaker tripping, cooling device startup or remote alarm notification.

2. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The time series modeling module constructs input samples using a sliding window method, and each sample has historical data of T consecutive time points for training the improved LSTM network model.

3. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The model update module uses an online incremental learning algorithm. Each time new data is received, some weight parameters of the model are updated and the weight coefficients in the attention configuration are adjusted to make the model adapt to new operating conditions.

4. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The control linkage module includes a relay control unit and a communication interface unit. When an early warning signal is received, the corresponding control instructions are matched according to a preset strategy table and corresponding actions are executed. The strategy table also includes priority sorting for different fault types to ensure that critical faults are responded to in a timely manner.

5. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The data acquisition module includes multiple sensor nodes, which are respectively arranged in the transformer body, oil tank, bushing and cooling structure to achieve multi-dimensional operating status perception. The sensor nodes support wireless ad hoc networking function.

6. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The data preprocessing module uses the Z-score standardization method to normalize the original data.

7. The intelligent monitoring and management system for new energy transformers according to claim 6 is characterized in that: The control linkage module communicates with the on-site equipment through the Modbus protocol and uploads the warning information to the host computer platform through the MQTT protocol.

8. The intelligent monitoring and management system for new energy transformers according to claim 7 is characterized in that: The system is deployed on edge computing devices, has local data processing capabilities, and supports data synchronization with cloud servers. The edge devices have built-in redundant power supplies to ensure that they can continue to work for at least 2 hours in the event of a mains power outage.

9. The intelligent monitoring and management system for new energy transformers according to claim 8 is characterized in that: The early warning trigger module is provided with a dynamic threshold mechanism, which adaptively adjusts the alarm threshold range according to the statistical results of historical data. At the same time, the module integrates the expert system knowledge base, which can provide more accurate early warning judgment basis under complex working conditions.

10. The intelligent monitoring and management system for new energy transformers according to claim 1 is characterized in that: The improved LSTM network is trained using transfer learning technology, which uses monitoring data of power equipment as a pre-training basis to accelerate model convergence and improve generalization performance.

Images