Intelligent monitoring and adjusting system for moisture absorption equipment of large transformer

Abstract

The invention discloses an intelligent monitoring and adjusting system for moisture absorption equipment of a large transformer, which relates to the field of intelligent monitoring and is characterized in that multivariate data streams reflecting internal and external working conditions of the moisture absorption equipment are acquired in real time through a multi-dimensional sensor network; and converting the original data flow into a key feature set capable of better revealing the essence of the moisture absorption process by using a feature engineering technology. On the basis, the current health condition and working performance of the moisture absorbent are diagnosed in real time through an instant state evaluation model. And meanwhile, dynamic prediction is carried out by utilizing a dynamic prediction model and combining a historical feature sequence and a current state to predict the residual effective life of the moisture absorbent. And finally, fusing the instant evaluation result with the dynamic predicted life, and inputting the fused result into an intelligent decision engine, thereby generating a control instruction considering the current risk and the future trend. Therefore, the transformation of the moisture absorption equipment from passive maintenance to active prediction and intelligent management can be realized, and the reliability and economical efficiency of the operation of the transformer are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent monitoring, and more particularly, to an intelligent monitoring and regulation system for a large transformer moisture absorption device. BACKGROUND

[0002] As a pivotal device in the power system, the safe and stable operation of a large power transformer is the cornerstone of reliable power supply of the power grid. The insulating oil inside the transformer exchanges with the external air through the oil conservator during operation. To prevent moisture in the air from entering the insulating oil and damaging its insulation performance, a moisture absorption device (breather) is usually installed on the breathing channel of the oil conservator. The silica gel or other moisture absorbers filled in the moisture absorption device can effectively absorb the moisture in the air entering the transformer, thereby protecting the quality of the insulating oil. However, the moisture absorption capacity of the moisture absorber will gradually saturate and fail over time. Once the moisture absorber fails, the humid air will directly enter the transformer, causing irreversible damage to the device body, and even causing a major power accident in severe cases. Therefore, it is crucial to accurately monitor the working state of the moisture absorption device of the large transformer and make scientific decisions on its regulation and maintenance time to ensure the safety of the transformer and the entire power grid.

[0003] Traditional maintenance of the moisture absorption device mainly relies on periodic manual inspection, which determines whether the moisture absorber is saturated by observing the color change. This method is highly subjective and low in accuracy, and cannot reflect the real state of the moisture absorber in real time. The setting of the inspection period is often based on experience and is difficult to adapt to the dynamic changes of the actual operating conditions of the transformer, which may lead to waste due to premature replacement or safety hazards due to untimely replacement. To solve this problem, some improved schemes have emerged, such as monitoring the outlet humidity with a simple humidity sensor and alarming when the humidity exceeds a threshold. However, such schemes ignore the coupling effects of complex external environmental factors and the operating state of the transformer. In fact, the environmental temperature and humidity and the fluctuations in the load of the transformer will jointly affect the moisture absorption process, resulting in a significant nonlinearity and time-varying characteristics of the moisture absorption rate. This single threshold-based judgment method cannot reveal the comprehensive effects of dynamic factors such as the environment and the load on the saturation process of the moisture absorber, and it is also difficult to support more in-depth life prediction and intelligent maintenance decisions.

[0004] Therefore, there is an urgent need for an optimized intelligent monitoring and regulation system for a large transformer moisture absorption device. SUMMARY

[0005] To solve the above technical problems, the present application is proposed.

[0006] According to an aspect of the present application, an intelligent monitoring and regulation system for a large transformer moisture absorption device is provided, which comprises: An original data stream acquisition module is configured to acquire an original data stream of the large transformer moisture absorption device, wherein the original data stream includes an ambient temperature, an ambient relative humidity, an inlet temperature of the moisture absorber, an inlet relative humidity of the moisture absorber, an outlet temperature of the moisture absorber, an outlet relative humidity of the moisture absorber, and a transformer load current; A feature engineering and efficiency calibration module is configured to perform feature engineering and real-time efficiency calibration on the original data stream of the large transformer moisture absorption device to obtain a current feature set, wherein the current feature set includes an inlet absolute humidity, a current moisture absorption efficiency, a load change rate, and an ambient temperature; A state instant evaluation module is configured to perform instant evaluation on the moisture absorbent state based on the current feature set to obtain an instant state evaluation result; A life dynamic prediction module is configured to perform dynamic prediction on the remaining effective life after appending the current feature set to the end of a historical feature sequence to obtain a predicted remaining life; An intelligent decision module is configured to input the instant state evaluation result and the predicted remaining life into an intelligent decision engine to obtain a control instruction.

[0007] Compared with the prior art, the intelligent monitoring and adjusting system of the large transformer moisture absorption device provided by the present application can realize the transformation from passive maintenance to active prediction and intelligent management of the moisture absorption device, and significantly improve the reliability and economy of transformer operation. BRIEF DESCRIPTION OF DRAWINGS

[0008] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0009] Figure 1 FIG. 1 is a block diagram of an intelligent monitoring and adjusting system of a large transformer moisture absorption device according to an embodiment of the present application.

[0010] Figure 2A data flow diagram of the intelligent monitoring and regulation system of the large transformer moisture absorption equipment according to the embodiment of the present application.

[0011] Figure 3 A block diagram of the feature engineering and efficiency calibration module in the intelligent monitoring and regulation system of the large transformer moisture absorption equipment according to the embodiment of the present application.

[0012] Figure 4 A block diagram of the life dynamic prediction module in the intelligent monitoring and regulation system of the large transformer moisture absorption equipment according to the embodiment of the present application. DETAILED DESCRIPTION

[0013] Embodiments of the present disclosure will be described in more detail with reference to the drawings. Although certain embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are for exemplary purposes only and are not intended to limit the scope of protection of the present disclosure.

[0014] In view of the problems in the above background art, the present application provides an intelligent monitoring and regulation system of a large transformer moisture absorption equipment. Figure 1 A block diagram of the intelligent monitoring and regulation system of the large transformer moisture absorption equipment according to the embodiment of the present application. Figure 2 A data flow diagram of the intelligent monitoring and regulation system of the large transformer moisture absorption equipment according to the embodiment of the present application. As shown in Figure 1 and Figure 2 The intelligent monitoring and regulation system 100 of the large transformer moisture absorption equipment includes: an original data flow acquisition module 110, configured to acquire original data flow of the large transformer moisture absorption equipment, the original data flow including environmental temperature, environmental relative humidity, moisture absorber inlet temperature, moisture absorber inlet relative humidity, moisture absorber outlet temperature, moisture absorber outlet relative humidity, and transformer load current; a feature engineering and efficiency calibration module 120, configured to perform feature engineering and real-time efficiency calibration on the original data flow of the large transformer moisture absorption equipment to obtain a current feature set, the current feature set including inlet absolute humidity, current moisture absorption efficiency, load change rate, and environmental temperature; a state instant evaluation module 130, configured to perform instant evaluation of the moisture absorbent state based on the current feature set to obtain an instant state evaluation result; a life dynamic prediction module 140, configured to append the current feature set to the end of a historical feature sequence and perform dynamic prediction of the remaining effective life to obtain a predicted remaining life; and an intelligent decision module 150, configured to input the instant state evaluation result and the predicted remaining life into an intelligent decision engine to obtain a control instruction.

[0015] In the intelligent monitoring and adjusting system of the above large transformer moisture absorption equipment, the original data stream acquisition module 110 is used to acquire the original data stream of the large transformer moisture absorption equipment, and the original data stream includes the environmental temperature, the environmental relative humidity, the moisture absorber inlet temperature, the moisture absorber inlet relative humidity, the moisture absorber outlet temperature, the moisture absorber outlet relative humidity and the transformer load current. It should be understood that the moisture absorption process of the moisture absorption equipment is affected by the coupling of the environmental conditions and the transformer operating state, the environmental temperature and humidity determine the water vapor content in the air, the inlet and outlet temperature and humidity of the moisture absorber directly reflect the working effect of the moisture absorbent, the transformer load current indirectly determines the air intake and frequency of the transformer by affecting the oil temperature change, and single parameter monitoring cannot capture these multi-dimensional interactions, and it is difficult to accurately reflect the real working condition of the moisture absorption equipment. Based on this, the present application realizes real-time collection of the original data stream which can fully represent the internal and external environment and the related operating state of the moisture absorption equipment through a multi-dimensional sensor network, covering environmental factors, inlet and outlet states of the moisture absorber and load conditions of the transformer, and providing complete original data basis for subsequent analysis. The execution effect is to realize real-time and continuous monitoring of the key parameters affecting the moisture absorption process, the obtained multi-element data stream can completely describe the environmental water vapor input, the working state of the moisture absorber and the driving force of the transformer breathing, and avoid the limitation of single data dimension, thereby providing comprehensive and reliable original data support for subsequent feature extraction and model analysis.

[0016] Particularly, in one possible embodiment, the implementation process of the original data stream acquisition module 110 is as follows: first, the selection and arrangement of sensors are performed. For the ambient temperature and the ambient relative humidity, a temperature and humidity integrated sensor suitable for industrial environment is selected and installed in a well-ventilated area around the transformer oil conservator, avoiding the direct heat dissipation surface of the equipment and the direct sunlight, so as to ensure that the collected environmental parameters can truly reflect the external atmospheric state of the transformer. For the inlet temperature and the inlet relative humidity of the moisture absorber, the same type of sensor is installed at the position close to the air inlet of the moisture absorber on the breathing pipe connected between the moisture absorber and the oil conservator, and the sensor probe is directed to the airflow entering direction, so as to accurately capture the temperature and humidity characteristics of the air before entering the moisture absorber. The collection of the outlet temperature and the outlet relative humidity of the moisture absorber is realized by installing the same type of sensor on the pipe at the outlet side of the moisture absorber, and the sensor is installed close to the outlet to ensure that the air state after being treated by the moisture absorber can be monitored. The collection of the transformer load current is realized by installing a current sensing device suitable for the transformer specification in the circuit at the low-voltage side of the transformer, so as to sense the dynamic change of the load current. Subsequently, all the sensors are connected to the local data acquisition terminal arranged near the transformer through shielded cables, and the terminal has an interface suitable for each sensor signal and can convert the sensor output signal into a processable digital form. The data acquisition terminal continuously collects the signals of each sensor according to the set period, so as to ensure that the dynamic change of each parameter can be captured in real time. The collected original data is sent to the remote monitoring server through an industrial transmission link, and a corresponding verification mechanism is used in the transmission process to ensure the integrity of the data. After receiving the data, the remote monitoring server adds a corresponding time mark to each group of data to clearly indicate the time of data collection. At the same time, the server performs preliminary effectiveness verification on the received data, and if the value of a certain parameter is obviously beyond the reasonable range, the data is marked as abnormal and recorded, and the continuous reception and storage of normal data are maintained. When storing, the original data stream is structured and arranged according to the device identification, time sequence and parameter type, so as to form a continuous original data stream.

[0017] In the intelligent monitoring and adjusting system of the large transformer moisture absorption equipment, the feature engineering and efficiency calibration module 120 is configured to perform feature engineering and real-time efficiency calibration on the original data stream of the large transformer moisture absorption equipment to obtain a current feature set, which includes inlet absolute humidity, current moisture absorption efficiency, load change rate and environmental temperature. It should be understood that the original data stream is an unprocessed original signal, which has scattered physical meaning and redundancy or interference. For example, the relative humidity cannot directly reflect the absolute content of water vapor due to the influence of temperature, and the features directly representing the essence of the moisture absorption process need to be extracted through conversion and calculation to reveal the internal correlation between the performance of the moisture absorption agent and the influencing factors. Specifically, the original data stream is converted into key features with clear physical meaning and analysis value. The influence of temperature on the representation of water vapor content is eliminated through absolute humidity calculation, and the current moisture absorption capacity of the moisture absorption agent is directly quantified through the moisture absorption efficiency. At the same time, the driving effect of the transformer breathing intensity on the moisture absorption process is reflected through the load change rate, and the environmental temperature is retained as a basic influencing factor to provide directly analyzable feature variables for subsequent state assessment and life prediction. The obtained current feature set realizes effective extraction and conversion of the original data stream. Among them, the inlet absolute humidity accurately reflects the absolute amount of water vapor entering the moisture absorber, the current moisture absorption efficiency directly reflects the real-time working performance of the moisture absorption agent, the load change rate quantifies the dynamic influence of the transformer breathing behavior on the moisture absorption process, and the environmental temperature retains the representation of the overall environmental conditions. These features provide high information density input for subsequent real-time state assessment and residual effective life prediction, significantly improve the ability to capture the essence of the moisture absorption process, and lay a foundation for accurate analysis and decision-making.

[0018] In particular, in one embodiment, Figure 3 A block diagram of the feature engineering and efficiency calibration module in the intelligent monitoring and adjusting system of the large transformer moisture absorption equipment according to the embodiments of the present application is shown in FIG. 12. As shown in FIG. 12, the feature engineering and efficiency calibration module 120 includes an inlet relative humidity calculation unit 121, an outlet absolute humidity calculation unit 122, a moisture absorption efficiency calculation unit 123 and a load change rate analysis unit 124. Figure 3 The inlet relative humidity calculation unit 121 is configured to input the inlet temperature of the moisture absorber and the inlet relative humidity of the moisture absorber into a water vapor saturation partial pressure model to obtain the inlet absolute humidity. The outlet absolute humidity calculation unit 122 is configured to input the outlet temperature of the moisture absorber and the outlet relative humidity of the moisture absorber into the water vapor saturation partial pressure model to obtain the outlet absolute humidity. The moisture absorption efficiency calculation unit 123 is configured to calculate the current moisture absorption efficiency based on the inlet absolute humidity and the outlet absolute humidity. The load change rate analysis unit 124 is configured to perform sliding window analysis on the transformer load current to obtain the load change rate.

[0019] Specifically, the inlet relative humidity calculation unit 121 is configured to input the hygroscopic device inlet temperature and the hygroscopic device inlet relative humidity into a water vapor saturation pressure model to obtain the inlet absolute humidity. It should be understood that the relative humidity can only reflect the proportional relationship between the water vapor content in the air and the saturated water vapor content at the same temperature, and the value is significantly affected by the temperature. The absolute humidity is the core physical quantity for measuring the water vapor input in the hygroscopic process. Specifically, based on the water vapor saturation pressure model, the inlet temperature is combined to correct the inlet relative humidity, the inlet temperature and humidity parameters are converted into the inlet absolute humidity with clear physical meaning, the temperature interference is eliminated to accurately quantify the absolute amount of water vapor entering the hygroscopic device, and the actual water vapor content entering the hygroscopic device is objectively reflected. The accurate input benchmark is provided for subsequent calculation of the hygroscopic efficiency, the relative humidity characterization deviation caused by temperature fluctuation is avoided, and the accurate description of the input quantity of the hygroscopic process is ensured.

[0020] In particular, in one possible embodiment, the implementation process of the inlet relative humidity calculation unit 121 is as follows: first, the pre-processed hygroscopic device inlet temperature and inlet relative humidity data are extracted from the original data stream. These data have been verified for effectiveness, and abnormal values obviously beyond the reasonable range have been excluded to ensure that they can accurately reflect the air state at the inlet of the hygroscopic device. Subsequently, a water vapor saturation pressure model suitable for the temperature and humidity range of the transformer working environment is selected. The model is constructed based on the Antoine equation, and the saturation vapor pressure calculation under common atmospheric environment is adapted through preset equation parameters. The saturation water vapor partial pressure under the corresponding conditions can be derived according to the temperature value. In the specific calculation, the inlet temperature data is first converted into thermodynamic temperature, and then substituted into the temperature parameter term of the model to calculate the saturation water vapor partial pressure at the temperature through the built-in function relationship of the model. Then, combined with the inlet relative humidity data, the actual water vapor partial pressure of the air at the inlet is calculated according to the correlation between the relative humidity and the saturation vapor pressure, i.e. the relative humidity is the ratio of the actual water vapor partial pressure to the saturation water vapor partial pressure. Then, according to the physical definition of absolute humidity, i.e. the mass of water vapor contained in a unit volume of air, the actual water vapor partial pressure is converted into the inlet absolute humidity through the relationship between the actual water vapor partial pressure, the thermodynamic temperature and the general gas constant. In this process, the conversion coefficient built in the model is used to match the pressure unit and the mass unit to ensure the physical dimension of the calculation result. After the calculation is completed, the obtained inlet absolute humidity result is subjected to secondary verification. If the result deviates too much from the theoretically reasonable value derived based on the inlet temperature and humidity range, the data is returned to the model for recalculation to eliminate errors that may occur in the data transmission or model parameter calling process. Finally, the obtained inlet absolute humidity data is stored as the basic parameter of the current feature set, providing a reliable basis for evaluating the initial load of the hygroscopic agent.

[0021] Specifically, the outlet absolute humidity calculation unit 122 is configured to input the desiccant outlet temperature and the desiccant outlet relative humidity into a water vapor saturation pressure model to obtain the outlet absolute humidity. It should be understood that the desiccant outlet relative humidity is also affected by the desiccant outlet temperature, and the absolute amount of water vapor remaining after the desiccant treatment cannot be accurately reflected only by the relative humidity. Based on this, the application also corrects the outlet relative humidity by using the outlet temperature to convert it into the outlet absolute humidity by using the water vapor saturation pressure model, so as to accurately quantify the absolute amount of water vapor discharged after the desiccant treatment, truly reflect the removal effect of the desiccant on the water vapor, provide a reliable output parameter for subsequent calculation of the desiccant efficiency, and ensure accurate characterization of the output quantity of the desiccant process, thereby laying a data foundation for evaluating the working performance of the desiccant.

[0022] In particular, in one possible embodiment, the implementation process of the outlet absolute humidity calculation unit 122 is as follows: first, the desiccant outlet temperature and the desiccant outlet relative humidity data are extracted from the preprocessed original data stream. These data have been verified for effectiveness in the early stage, and the abnormal values beyond the reasonable physical range have been eliminated, so as to truly reflect the state of the air discharged from the desiccant after the desiccant treatment. Subsequently, a water vapor saturation pressure model compatible with the inlet absolute humidity calculation is selected. The model is constructed based on the Antoine equation, and the parameter setting covers the temperature and humidity interval of the working environment of the transformer desiccant device. The saturation water vapor pressure under the corresponding conditions can be accurately derived according to the temperature value. In the specific calculation process, the outlet temperature data are first converted into thermodynamic temperature, which is substituted into the temperature parameter item of the model, and the saturation water vapor pressure under the temperature is calculated by using the function relationship built in the model. Then, the actual water vapor pressure of the air at the outlet is calculated according to the physical definition of the relative humidity, i.e., the ratio of the actual water vapor pressure to the saturation water vapor pressure under the same temperature. Subsequently, the actual water vapor pressure is converted into the outlet absolute humidity by using the quantitative relationship between the actual water vapor pressure, the thermodynamic temperature, and the general gas constant according to the mass of the water vapor contained in the unit volume of air. In the conversion process, the matching between the pressure unit and the mass unit is realized by using the unit conversion coefficient built in the model, so as to ensure that the calculation result is physically dimensionally uniform and accurate. After the calculation is completed, the outlet absolute humidity result is subjected to secondary verification. If the deviation of the result from the theoretically reasonable value derived based on the outlet temperature and humidity range exceeds the preset limit, the original data are re-substituted into the model for recalculation, so as to eliminate the errors possibly occurring in the data transmission or model parameter calling process. Finally, the obtained outlet absolute humidity data are stored as the key parameters of the current feature set, thereby providing a reliable basis for evaluating the working performance of the desiccant.

[0023] Specifically, the moisture absorption efficiency calculation unit 123 is configured to calculate the current moisture absorption efficiency based on the inlet absolute humidity and the outlet absolute humidity. In one specific example of the present application, the moisture absorption efficiency calculation unit 123 is configured to calculate the current moisture absorption efficiency based on the inlet absolute humidity and the outlet absolute humidity according to the following formula: ; wherein, is the inlet absolute humidity, is the outlet absolute humidity, is the maximum function, is the current moisture absorption efficiency. That is, the current moisture absorption efficiency is obtained by dividing the difference between the inlet absolute humidity and the outlet absolute humidity by the inlet absolute humidity, and ensuring non-negativity by taking the maximum function, to quantitatively represent the actual moisture absorption capacity of the moisture absorbent under the current operating condition, which can intuitively and quantitatively reflect the real-time working performance of the moisture absorbent, providing a core index for real-time state evaluation, upgrading the evaluation of the performance of the moisture absorbent from qualitative description to quantitative analysis, and improving the accuracy of the evaluation.

[0024] Specifically, the load change rate analysis unit 124 is configured to perform sliding window analysis on the transformer load current to obtain the load change rate. It should be understood that the dynamic change of the transformer load current will affect the winding loss and the oil temperature, and in turn change the air intake and frequency of the transformer. This dynamic change has time-varying characteristics on the driving force of the moisture absorption process, therefore, the present application captures the change trend by sliding window analysis to capture this dynamic influence. Specifically, the continuous transformer load current is segmented and statistically analyzed by sliding window, and the load current change rate in different windows is calculated, to quantitatively represent the dynamic change trend of the transformer load, which can accurately capture the short-term fluctuation characteristics of the transformer load, reflect the influence of the dynamic change of the breathing intensity of the transformer on the moisture absorption process, provide a key parameter for analyzing the correlation between the service life attenuation of the moisture absorbent and the load fluctuation, and improve the ability to characterize the dynamic influencing factors of the moisture absorption process.

[0025] In particular, in one possible embodiment, the load change rate analysis unit 124 is implemented as follows: first, continuous transformer load current data is extracted from the pre-processed raw data stream, which has passed the preliminary validity verification, eliminated abnormal jump values caused by sensor failure or transmission interference, and can truly reflect the load running state of the transformer at different times, providing continuous and reliable basic data for sliding window analysis. Subsequently, the sliding window parameters are set according to the time characteristics of transformer load change. The time span of the window needs to adapt to the common fluctuation period of transformer load, ensuring that meaningful short-term change trends can be captured; the sliding step needs to balance the continuity of analysis and computational efficiency, so that the adjacent windows can maintain data correlation, and the redundant calculation caused by too high overlap can be avoided. In the specific analysis process, the load current data is intercepted according to the set window parameters. For each intercepted window, the load current sequence in the window is extracted, the load current difference between the start time and the end time in the sequence is calculated, and the average change rate of the load current in the window is obtained combined with the time span of the window, which represents the overall change trend of the load in this time period. If the load current in the window shows an upward trend, the change rate is positive; if it shows a downward trend, the change rate is negative; if it remains basically stable, the change rate is close to zero. After the analysis is completed, the load change rate results of each window are checked for consistency. If the change rate of a window and the results of adjacent windows have a significant mutation, and exceed the normal load regulation range of the transformer, it is determined that the data of this window may be affected by instantaneous interference, and the window needs to be re-intercepted and corrected combined with the trend of the previous and subsequent data to ensure the continuity and reasonableness of the load change rate sequence.

[0026] In the intelligent monitoring and adjusting system of the large transformer moisture absorption equipment, the state real-time evaluation module 130 is configured to perform real-time evaluation of the state of the moisture absorption agent based on the current feature set to obtain a real-time state evaluation result. It should be understood that the state deterioration of the moisture absorption agent is the result of the coupling of multiple factors, and a single index or instantaneous state cannot accurately reflect the true health condition of the moisture absorption agent. The moisture absorption efficiency in the current feature set directly represents the moisture absorption capacity, and the combination of the inlet absolute humidity and the load change rate can reflect the potential risk. For example, in a high humidity environment, the decrease of the load of the transformer can cause the increase of the air absorption amount, which aggravates the burden of the moisture absorption agent. Therefore, a multi-condition and multi-threshold cooperative judgment is required to achieve accurate evaluation. In one specific example of the present application, the state real-time evaluation module 130 is configured to: if the current moisture absorption efficiency is lower than a first threshold value and the duration is more than 1 hour, determine that the real-time state is poor; if the current moisture absorption efficiency is lower than a second threshold value, determine that the real-time state is saturated and generate an emergency alarm; if the inlet absolute humidity is greater than a preset threshold value and the load change rate is less than a preset negative value, determine that the real-time state is attention; otherwise, determine that the real-time state is healthy. The first threshold value is 50%, and the second threshold value is 20%. Specifically, the present application establishes a multi-dimensional real-time evaluation mechanism by setting multi-level threshold values (50% and 20% corresponding to different deterioration degrees), duration constraints (to avoid misjudgment of instantaneous fluctuations), and composite conditions (the correlation between the inlet absolute humidity and the load change rate), and distinguishes the poor, saturated, attention, and healthy states of the moisture absorption agent in real time. In particular, the emergency alarm is triggered for the saturated state to avoid safety risks, and the potential risk state is captured to intervene in advance. In this way, real-time, dynamic, and accurate evaluation of the state of the moisture absorption agent is achieved. The multi-level threshold values and the composite conditions effectively avoid the limitations of single index judgment. The emergency alarm can quickly respond when the moisture absorption agent is saturated to prevent moist air from entering the transformer. The identification of the attention state can capture the potential deterioration trend in high-risk working conditions in advance. The determination of the healthy state avoids unnecessary maintenance intervention, which improves the timeliness and accuracy of the state evaluation of the moisture absorption equipment as a whole, provides a reliable state basis for subsequent intelligent decision-making, and ensures the safety and economy of the operation of the transformer.

[0027] In particular, in one possible embodiment, the implementation process of the state instant evaluation module 130 is as follows: first, the verified current feature set is extracted, which includes parameters such as inlet absolute humidity, current hygroscopic efficiency, load change rate and ambient temperature, which have been processed and verified in advance and can accurately reflect the current working environment and performance state of the desiccant, providing reliable input basis for instant evaluation. In the evaluation process, the judgment is carried out in turn according to the preset logic. First, it is detected whether the current hygroscopic efficiency is lower than the second threshold value, which corresponds to the critical state of the desiccant approaching complete saturation. If this condition is met, the instant state is directly determined to be saturated, and the emergency alarm mechanism is triggered, an alarm signal is sent to the monitoring center through the preset communication link, and an alarm record is stored locally to ensure that relevant personnel can know in time that the desiccant has lost effective moisture absorption capacity in an emergency. If the current hygroscopic efficiency is not lower than the second threshold value, it is further judged whether it is lower than the first threshold value and whether the duration of this state exceeds 1 hour. The time when the current hygroscopic efficiency first falls below the first threshold value is recorded by a time stamp, and the duration of this state is continuously accumulated. If the accumulated duration reaches the preset 1 hour, it means that the performance of the desiccant has significantly decreased and is in a continuous deterioration state, so the instant state is determined to be poor to reflect the serious deficiency of the current working efficiency of the desiccant. If the above two conditions are not met, it is further judged whether the inlet absolute humidity is greater than the preset threshold value and whether the load change rate is less than the preset negative value. The inlet absolute humidity greater than the preset threshold value indicates that the air moisture content entering the desiccant is high, and the load change rate less than the preset negative value means that the load of the transformer is decreasing, which may lead to an increase in the air intake of the transformer, and the combination of the two will exacerbate the burden of the desiccant. At this time, the instant state is determined to be attention to prompt the potential performance degradation risk. If the above conditions are not met, it is determined that the current working state of the desiccant is stable and the hygroscopic efficiency is within the normal range, and the instant state is determined to be healthy. During the entire evaluation process, the triggering conditions of each judgment condition are recorded in real time. If there is a conflict in the judgment logic, such as meeting multiple state conditions at the same time, the final state is determined in priority according to the emergency level, with the saturated state having the highest priority, followed by poor, attention and healthy. The instant state evaluation result obtained by evaluation will be stored in the corresponding data unit as the key basis for the intelligent decision module to generate control instructions, ensuring that the evaluation of the state of the desiccant can accurately guide the subsequent maintenance and adjustment operations.

[0028] In the intelligent monitoring and adjusting system of the large transformer moisture absorbing device, the life dynamic prediction module 140 is configured to append the current feature set to the end of the historical feature sequence and perform residual effective life dynamic prediction to obtain the predicted residual life. It should be understood that the residual effective life of the moisture absorbing agent is affected by the long-term working condition cumulative effect, and the decay process presents significant nonlinearity and time-varying characteristics. The current feature at a single moment can only reflect the instantaneous state and cannot reveal the historical trend and dynamic law of life decay, and the historical feature sequence contains the performance evolution information of the moisture absorbing agent under different environmental conditions and load states. Based on this, the current feature and the historical sequence are fused to completely characterize the time sequence dependence of life decay and provide comprehensive time sequence feature support for accurate prediction.

[0029] In particular, in one specific embodiment, Figure 3 The block diagram of the life dynamic prediction module in the intelligent monitoring and adjusting system of the large transformer moisture absorbing device according to the embodiment of the present application is shown in FIG. 6. As shown in FIG. 6, the life dynamic prediction module 140 includes a feature sequence construction unit 141 configured to append the current feature set to the end of the historical feature sequence to obtain a feature time sequence; and a residual life prediction unit 142 configured to input the feature time sequence into a trained LSTM model to obtain a predicted residual life. Figure 3

[0030] Specifically, the feature sequence construction unit 141 is configured to append the current feature set to the end of the historical feature sequence to obtain a feature time sequence. It should be understood that the current feature set containing the inlet absolute humidity, the current moisture absorbing efficiency and other key features is time-series spliced with the historical feature sequence to form a feature time sequence containing a time dimension, the dynamic correlation information of the features changing over time is retained, the time sequence features of the historical working conditions and the current state are completely integrated, the dependence relationship between the features at different moments is effectively retained, the performance change trend of the moisture absorbing agent over time is objectively presented, the input data with continuous evolution characteristics is provided for the subsequent residual life prediction based on the time sequence model, and it is ensured that the model can capture the complete evolution trajectory of the performance of the moisture absorbing agent from the past to the present.

[0031] ​In particular, in one possible embodiment, the implementation process of the feature sequence construction unit 141 is as follows: first, the current feature set that has passed the final verification is extracted, which includes key features such as inlet absolute humidity, current moisture absorption efficiency, load change rate, and ambient temperature, each feature is accompanied by a collection timestamp accurate to milliseconds, and has passed the data integrity and rationality verification, ensuring that it can accurately reflect the current working state of the moisture absorption equipment and the associated environmental parameters. Subsequently, the historical feature sequence database is accessed, which stores the historical feature sequences in chronological order, each historical feature record contains the same feature dimensions as the current feature set, and is accompanied by a corresponding timestamp, which completely retains the evolution trajectory of each key feature of the moisture absorption equipment during the past operation process. By matching the corresponding historical feature sequence through the unique identification of the equipment, it is ensured that the extracted historical data is consistent with the monitoring object of the current equipment. Before performing the appending operation, the time stamp of the current feature set and the time stamp of the last record of the historical feature sequence are checked in time sequence, to confirm that the collection time of the current feature set is later than the time of the last record of the historical sequence, and the time interval between the two meets the data collection period rule, so as to avoid the time sequence logical error caused by time disorder. If the time stamp verification is passed, each feature data in the current feature set is appended to the end of the historical feature sequence in the same dimension order as the historical sequence, forming a new continuous sequence. After the appending is completed, the integrity of the newly generated feature time sequence is checked to confirm that each feature dimension in the sequence has no missing, the time stamp is continuous and has no repetition, if it is found that a feature has missing or abnormal time stamp, the data completion mechanism will be triggered to perform reasonable interpolation completion based on the evolution trend of adjacent features, to ensure the continuity and integrity of the feature time sequence. The finally formed feature time sequence integrates the feature information of the historical operation process and the current state, ensuring that the model can fully capture the internal correlation and rules of the evolution of the moisture absorption equipment features over time.

[0032] Specifically, the remaining life prediction unit 142 is configured to input the characteristic time sequence into a trained LSTM model to obtain a predicted remaining life. The attenuation process of the remaining effective life of the moisture absorbent has significant nonlinear and time-varying characteristics and is influenced by long-term and short-term historical working conditions. The LSTM model can selectively remember long-term feature correlations and suppress noise interference through a gating mechanism, is suitable for processing nonlinear time series data with long-term dependence, and can accurately mine the life attenuation law implied in the characteristic time sequence. Specifically, the trained LSTM model is used to perform deep time series analysis on the characteristic time sequence, and the predicted remaining life obtained through dynamic correlation learning of historical and current features can accurately reflect the sustainable working ability of the moisture absorbent under time series evolution. The effective capture of long-term and short-term feature dependence by the LSTM model makes the prediction results take into account both historical trends and current states, significantly improving the accuracy and timeliness of the remaining life prediction, providing a key future risk quantification index for the intelligent decision engine, and supporting the intelligent maintenance and management transformation from passive response to active prediction.

[0033] In particular, in one possible embodiment, the implementation process of the remaining life prediction unit 142 is as follows: first, the integrity-verified feature time series is extracted, which is arranged in chronological order and contains historical and current data of key features such as inlet absolute humidity, current moisture absorption efficiency, load change rate, and ambient temperature. Each feature data is accompanied by a corresponding time stamp, and the standardization process has eliminated the dimensional differences of different features to ensure compliance with the input format requirements of the LSTM model. Subsequently, the pre-stored trained LSTM model is called, which has been trained based on a large number of historical feature time series and corresponding actual remaining life data of the moisture absorbent. The model structure includes an input layer, multiple LSTM hidden layers, and an output layer. The hidden layer captures the long-term feature dependency relationship through the gating mechanism, effectively identifies the implicit moisture absorbent life decay law in the feature time series, such as the nonlinear change of moisture absorption efficiency with environmental humidity fluctuations, the correlation between load change rate and life consumption rate, etc. During input, the feature time series is segmented according to the time step required by the model to ensure that each segment contains enough historical information to support prediction while maintaining the time sequence continuity of the sequence. The LSTM model processes the input feature time series layer by layer. The input layer receives the feature data and passes it to the hidden layer. The hidden layer filters the historical feature information to be retained through the forget gate, such as the cumulative effect of long-term low moisture absorption efficiency on life. The current feature updates the model state through the input gate, such as the short-term effect of instantaneous high inlet absolute humidity. The output gate generates the model state representation at the current time, and finally the output layer maps it to the predicted value of the remaining useful life. After prediction, the predicted remaining life is reasonably checked. Combined with the physical performance parameters of the moisture absorbent, such as the maximum moisture absorption capacity, typical life range, and historical prediction error range, if the predicted value exceeds the theoretical reasonable interval, the model re-prediction mechanism is triggered, the feature time series is re-input, and some parameters of the model are adjusted for secondary prediction to exclude accidental errors.

[0034] In particular, in another possible preferred embodiment, when the LSTM model captures the time series relationship, it may be difficult to effectively capture the instantaneous nonlinear interaction between features in the feature set, for example, at time t, a specific combination of inlet absolute humidity and load change rate may indicate a high-risk breath. It is worth noting that the rule-based expert system obtains the instantaneous state from the feature set, which is a high-level abstract information based on rules, and therefore can be used as state attention to guide the feature distribution within the feature set.

[0035] Specifically, a two-stage LSTM encoding mechanism is adopted when inputting the feature time series into the encoder of the LSTM model. The first-stage LSTM encoding mechanism is LSTM combined with a feature attention module. The feature set of a single time node is preliminarily associated and encoded by the LSTM, and then the dynamic interaction relationship among the features is learned and quantified by means of the internal feature attention mechanism, the influence of key combinations is strengthened, and a more relevant encoded feature vector reflecting the actual working condition is generated, so that the feature representation is more in line with the complex dynamics of the moisture absorption process. The second-stage LSTM encoding mechanism is LSTM combined with a time sequence attention module. The sequence of the relevant encoded feature vectors generated in the first stage is input into the LSTM to capture the long-term evolution law of the features over time, and the instantaneous state evaluation results of each time node are converted into a state vector. Through the time attention mechanism, the model is guided to focus on the key nodes that have a significant impact on the service life decay and to weaken the interference of irrelevant stable period data. Finally, an updated weighted hidden state vector sequence is generated to provide more accurate time sequence feature support for the remaining life prediction, so that the prediction result can better reflect the actual life trend of the moisture absorbent and better serve the subsequent intelligent decision-making.

[0036] First, each feature set is input into the LSTM model, so that the associated dimensions between samples are treated as time sequence dimensions for associated encoding, to obtain a plurality of associated encoded feature vectors Then, the attention vector is calculated , which is used to represent the importance of each feature to other features within the feature vector: Here, denotes the transpose symbol, denotes matrix multiplication, denotes the th associated encoded feature vector, and the associated encoded feature vector is in the form of a column vector, denotes the two-norm of denotes the th attention vector, that is, it is believed that the feature dimensions representing the inlet absolute humidity, the load change rate, etc. in the associated encoded feature vector have an associated impact on the moisture absorption process, such as, only when the inlet absolute humidity is very high, a dramatic load change rate (indicating deep inhalation) has a high risk.

[0037] Therefore, a feature internal attention mechanism is applied to the current feature set of each time node in the feature time series to generate an updated associated encoded feature vector, wherein the feature internal attention mechanism is used to learn and quantify the instantaneous interaction influence between the features in the current feature set. That is, the interaction weights between the features are dynamically learned through the self-attention association mapping of the associated encoded feature vector , and the interaction weights between the features are quantified through the self-attention association mapping of the associated encoded feature vector of the second norm to adaptively constrain the attention numerical distribution, and then the attention vector is multiplied by the associated encoding feature vector to update: ; wherein, represents element-level multiplication, represents the updated associated encoding feature vector.

[0038] That is, by learning and quantifying the instantaneous nonlinear interaction between each feature in the current feature set, the key association that is easily overlooked, such as the synergistic effect of high inlet absolute humidity and load sudden drop on the deterioration of the moisture adsorbent, is captured to enhance the representation ability of the feature. In this way, the dynamic interaction pattern between features can be effectively identified, for example, when the load of the transformer suddenly decreases in a humid environment, the interaction of the two on the adsorption process can be highlighted, so that the generated updated associated encoding feature vector more accurately reflects the complex dynamics of the moisture adsorbent working state, providing a more effective feature basis for subsequent time series modeling.

[0039] At the same time, when the sequence composed of the updated associated encoding feature vector is input to the encoder of the second-stage LSTM model, the LSTM model is used to capture the time series dependence, and the long-term dynamic association of the updated associated encoding feature vector in the time dimension is extracted, such as the continuous interaction of the adsorption efficiency change at different times and the environmental temperature and humidity fluctuations. In this way, the time series evolution law of the feature sequence is effectively mined, for example, the trend of gradually decreasing adsorption efficiency under continuous high humidity environment is captured, and the generated second-stage LSTM output hidden state vector sequence at each time node contains rich time series context information, providing reliable time series feature support for subsequent attention mechanism.

[0040] Therefore, by inputting the sequence composed of the updated associated encoding feature vector to the encoder of the second-stage LSTM model, the LSTM model is used to capture the time series dependence, and the long-term dynamic association of the updated associated encoding feature vector in the time dimension is extracted, such as the continuous interaction of the adsorption efficiency change at different times and the environmental temperature and humidity fluctuations. In this way, the time series evolution law of the feature sequence is effectively mined, for example, the trend of gradually decreasing adsorption efficiency under continuous high humidity environment is captured, and the generated second-stage LSTM output hidden state vector sequence at each time node contains rich time series context information, providing reliable time series feature support for subsequent attention mechanism.

[0041] Then, when the second-stage LSTM output hidden state vector sequence at each time node is obtained, the instantaneous state evaluation result corresponding to each time node is also converted into a state vector In other words, the real-time state assessment results, such as severe conditions and concerns, are transformed into vector forms that can be processed by the model. This allows state judgments based on expert experience to participate in the temporal attention calculation, achieving the integration of rule knowledge and data-driven models. The resulting state vectors can accurately convey the semantic information of the real-time state. For example, the state vector of a saturated and emergency alarm can clearly identify the high importance of this time point for life prediction, providing a structured state basis for subsequent attention guidance.

[0042] Then based on the corresponding vectors and An associative embedded attention representation is performed, that is, based on the hidden state vector of the second-stage LSTM output at each time point and the corresponding state vector, a temporal attention score is calculated to guide attention to key time points that have a significant impact on lifetime decay using the instantaneous state evaluation results: ;in, Indicates the first The second-stage LSTM outputs a hidden state vector. Represents the state vector. This represents vector addition. This represents the bias vector. This represents the time attention score.

[0043] That is, when converting vectors and When performing correlation, the focus is on the correlation at the level of distribution information representation, thereby obtaining the distribution information correlation matrix. And then Map and adjust the bias vector. After performing point addition, calculate the result. The inner product is used to obtain the time attention score. Thus, the state vector It can be used as a feature Global guidance of distribution, for example when in Instantaneous state at any moment When it is degraded, its embedding vector It will also be through vectors Information distribution association, via vector The mapping forms a state bias, which in turn guides the temporal attention score for states in a deteriorated state. or hidden states that may lead to degradation This allows for greater attention to time series information. For example, during the prediction process, the model will focus on periods three days prior when the moisture absorption efficiency was below 50% under sustained high humidity conditions. This increases the focus on key time points that drive lifespan degradation and enhances the relevance of time series modeling.

[0044] Then, the second-stage LSTM output hidden state vector sequence is point-multiplied based on the time attention score to generate an updated weighted hidden state vector sequence, which is more focused on the key driving information of life decay, for example, in continuous monitoring, the weighted sequence highlights the turning point of the beginning of the deterioration of the desiccant performance, providing a more accurate timing feature basis for the remaining life prediction. In this way, the judgment of the expert system can be combined with the timing modeling capability of the deep learning model to focus on the sequence feature distribution by guiding the timing attention allocation through the state features. For example, the current deterioration state is mainly caused by high humidity and heavy load inhalation for several hours two days ago, so the features of those moments are given a higher time attention score Moreover, by weakening the data of the stable healthy period that is close in time but irrelevant to the current problem, the anti-interference effect of the timing encoding can also be achieved.

[0045] Finally, the predicted remaining life is obtained based on the updated weighted hidden state vector sequence. Specifically, the updated weighted hidden state vector sequence processed by the two-stage LSTM encoding mechanism is used to complete the mapping from the feature sequence to the quantitative life value through the trained prediction model to accurately reflect the sustainable working time of the desiccant under the cumulative effect of the current and historical working conditions. Specifically, in the data preparation stage of training the prediction model, the historical raw data stream of the desiccant device is collected and converted into a historical feature set through feature engineering, spliced into a historical feature time sequence in chronological order, and the corresponding actual remaining life of the desiccant is recorded to form a sample pair of feature time sequence-actual remaining life. During preprocessing, outliers caused by sensor faults and transmission interference are cleaned up, features are standardized to eliminate dimension differences, and sequence length is unified according to fixed time steps, and short sequences are interpolated to complete. The prediction model adopts an input layer, an LSTM hidden layer, and an output layer architecture, integrating a two-stage encoding mechanism: the first stage strengthens the instantaneous interaction of features through feature internal attention, such as the synergistic effect of high inlet humidity and load sudden drop; the second stage combines the encoding vector of the immediate state evaluation result to focus on key timing nodes through time attention, such as periods when the desiccant efficiency is continuously below 50%. Training is supervised learning, and the sample is divided into training, validation, and test sets. The mean squared error is used as the loss function, the Adam optimizer is used for back propagation to adjust the parameters, the hyperparameters are dynamically optimized, and the early stopping mechanism is used to avoid overfitting. Finally, the MAE, RMSE, and other indicators of the test set are used for evaluation, the model needs to accurately capture the historical regularity, such as high humidity + load fluctuation corresponding to life shortening, and the error is within a reasonable range, which is qualified for training, and is used for actual remaining life prediction.

[0046] In the intelligent monitoring and adjusting system of the large transformer moisture absorption equipment, the intelligent decision module 150 is configured to input the real-time state evaluation result and the predicted remaining life into an intelligent decision engine to obtain a control instruction. It should be understood that, through the cooperative analysis of the real-time state and the predicted life by the intelligent decision engine, the application constructs a dynamic response mechanism in multiple scenarios, and forms a full-scenario response system covering emergency, early warning, near danger, and normal. The immediate regeneration in the case of emergency alarm can quickly block the risk of intrusion of humid air, the pre-regeneration alarm realizes the forward-looking planning of maintenance, the layered decision based on life and state effectively avoids excessive maintenance, such as unnecessary regeneration in a non-emergency state, and maintenance lag, such as not timely handling of the near danger state, significantly improves the timeliness and economy of the maintenance of the moisture absorption equipment, and ultimately guarantees the continuous reliability of the transformer operation through precise control instructions, and promotes the management upgrade from passive emergency to active prevention and control. In one specific example of the application, the intelligent decision module 150 is configured to: if an emergency alarm is detected, the control instruction is to perform regeneration; if no emergency alarm is detected, it is determined whether the predicted remaining life is less than a preset maintenance window period and the real-time state is deteriorating, and if so, the control instruction is a pre-regeneration alarm; if the predicted remaining life is greater than the preset maintenance window period or the real-time state is not deteriorating, it is determined whether the predicted remaining life is less than 1 day, and if so, the control instruction is to perform regeneration; and if not, the control instruction is to continue monitoring.

[0047] In particular, in one possible embodiment, the implementation process of the intelligent decision module 150 is as follows: first, the verified instant state evaluation result and the predicted remaining life data are extracted respectively. The instant state evaluation result contains the state identification of the current moisture absorbent, including saturation and emergency alarm, severe, attention or healthy, and the predicted remaining life is a quantitative result representing the working time of the moisture absorbent. Both of them are accompanied by corresponding time stamps to ensure the timeliness and relevance of the data, providing accurate input basis for the decision engine. After the intelligent decision engine starts, it first detects the instant state evaluation result to determine whether there is an emergency alarm. The emergency alarm corresponds to the saturation state of the moisture absorbent, at which time the moisture absorbent has lost its moisture absorption ability, and if not handled in time, it will lead to the direct invasion of moist air into the transformer. If an emergency alarm is detected, the engine immediately generates a control instruction to execute regeneration, which is transmitted to the regeneration module of the moisture absorbing device through the system control link to trigger the regeneration program, such as heating and drying the moisture absorbent. At the same time, the trigger time, state basis and instruction content are recorded in the instruction log to ensure the traceability of the operation. If no emergency alarm is detected, the engine enters the next judgment link, compares the predicted remaining life with the preset maintenance window period, and combines the instant state evaluation result to determine whether the current moisture absorption efficiency is lower than the first threshold and lasts for more than 1 hour. The preset maintenance window period is an advance preparation period set according to the transformer maintenance plan, which is used to reserve maintenance resource allocation time. If the predicted remaining life is less than the window period and the instant state is severe, it means that the performance of the moisture absorbent is continuously deteriorating and the remaining life is close to the maintenance cutoff point, so the engine generates a pre-regeneration alarm instruction, which is sent to the operation and maintenance management system through the communication module to prompt the relevant personnel to plan the maintenance work in advance, and at the same time, the pre-warning state is marked in the local system to continuously track the state change of the moisture absorbent. If the above conditions are not met, the engine further judges whether the predicted remaining life is less than 1 day. This judgment is for non-emergency but near-failure situations. If the remaining life is less than 1 day, even if the current state does not reach severe, it also needs to be handled immediately to avoid sudden failure risk. At this time, the engine generates an execution regeneration instruction to trigger the moisture absorbent regeneration program, and synchronously records the basis for generating the instruction, i.e. the critical value of the remaining life. If the predicted remaining life is not less than 1 day, it means that the current state of the moisture absorbent is stable and there is enough working time, so the engine generates a control instruction to continue monitoring. The instruction is issued to the data acquisition and evaluation system to maintain the original data acquisition frequency and state evaluation period, continuously track the state and life change of the moisture absorbent, and continue until the next decision condition is met. During the whole decision-making process, the engine records the judgment results of each step in the log. If there is a logical conflict, such as meeting multiple instruction trigger conditions at the same time, the final instruction is determined according to the priority, in which the emergency alarm is the highest, followed by the pre-regeneration alarm, the regeneration triggered by the short-term life, and the continue monitoring, to ensure the uniqueness and rationality of the decision.The generated control instructions are transmitted to the corresponding execution module after encryption, and are fed back to the monitoring center, realizing closed-loop management of instruction execution, and guaranteeing intelligent adjustment and maintenance of the transformer moisture absorption equipment.

[0048] In summary, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment based on the embodiments of the present application is illustrated, which collects multi-dimensional data streams reflecting the internal and external working conditions of the moisture absorption equipment in real time through a multi-dimensional sensor network, and then converts the original data streams into a key feature set that can better reveal the essence of the moisture absorption process using feature engineering technology. Based on this, the current health status and working performance of the moisture absorption agent are diagnosed in real time through an instant state evaluation model. At the same time, the residual effective life of the moisture absorption agent is dynamically predicted by combining historical feature sequences and the current state using a dynamic prediction model. Finally, the instant evaluation result and the dynamically predicted life are fused and input into an intelligent decision engine, thereby generating control instructions that take into account the current risk and future trend. In this way, the transformation from passive maintenance to active prediction and intelligent management of the moisture absorption equipment can be realized, significantly improving the reliability and economy of transformer operation.

[0049] As described above, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment according to the embodiments of the present application can be implemented in various wireless terminals. In one possible implementation, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment according to the embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment can also be one of the many hardware modules of the wireless terminal.

[0050] Alternatively, in another example, the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment and the wireless terminal can also be separate devices, and the intelligent monitoring and adjustment system of the large transformer moisture absorption equipment can be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

Claims

1. An intelligent monitoring and regulating system 100 for a large transformer moisture absorption device, characterized in that, The method comprises the following steps: An original data stream acquisition module 110 is configured to acquire an original data stream of a large transformer moisture absorption device, wherein the original data stream comprises an ambient temperature, an ambient relative humidity, an inlet temperature of a moisture absorber, an inlet relative humidity of the moisture absorber, an outlet temperature of the moisture absorber, an outlet relative humidity of the moisture absorber, and a transformer load current; A feature engineering and efficiency calibration module 120 is configured to perform feature engineering and real-time efficiency calibration on the original data stream of the large transformer moisture absorption device to obtain a current feature set, wherein the current feature set comprises an inlet absolute humidity, a current moisture absorption efficiency, a load change rate, and an ambient temperature; A state instant evaluation module 130 is configured to perform moisture absorbent state instant evaluation based on the current feature set to obtain an instant state evaluation result; A life dynamic prediction module 140 is configured to append the current feature set to the end of a historical feature sequence and perform residual effective life dynamic prediction thereon to obtain a predicted residual life; An intelligent decision module 150 is configured to input the instant state evaluation result and the predicted residual life into an intelligent decision engine to obtain a control instruction.

2. The intelligent monitoring and regulating system of large transformer moisture absorption equipment according to claim 1, characterized in that, The feature engineering and efficiency calibration module 120 comprises: An inlet relative humidity calculation unit 121 is configured to input the inlet temperature of the moisture absorber and the inlet relative humidity of the moisture absorber into a water vapor saturation partial pressure model to obtain the inlet absolute humidity; An outlet absolute humidity calculation unit 122 is configured to input the outlet temperature of the moisture absorber and the outlet relative humidity of the moisture absorber into the water vapor saturation partial pressure model to obtain the outlet absolute humidity; A moisture absorption efficiency calculation unit 123 is configured to calculate the current moisture absorption efficiency based on the inlet absolute humidity and the outlet absolute humidity; A load change rate analysis unit 124 is configured to perform sliding window analysis on the transformer load current to obtain the load change rate.

3. The intelligent monitoring and regulating system of large transformer moisture absorption equipment according to claim 2, characterized in that, The moisture absorption efficiency calculation unit 123 is configured to calculate the current moisture absorption efficiency based on the inlet absolute humidity and the outlet absolute humidity according to the following formula: ; wherein, is the inlet absolute humidity, is the outlet absolute humidity, is the maximum function, is the current moisture absorption efficiency.

4. The intelligent monitoring and regulating system of large transformer moisture absorption equipment according to claim 1, characterized in that, The state instant evaluation module 130 is configured to: If the current moisture absorption efficiency is lower than a first threshold value and the duration exceeds 1 hour, determine that the instant state is poor; If the current moisture absorption efficiency is lower than a second threshold value, determine that the instant state is saturated and generate an emergency alarm; If the inlet absolute humidity is greater than a preset threshold value and the load change rate is less than a preset negative value, determine that the instant state is attention; Otherwise, determine that the instant state is healthy.

5. The intelligent monitoring and regulating system of large transformer moisture absorption equipment according to claim 4, characterized in that, The first threshold value is 50%, and the second threshold value is 20%.

6. The intelligent monitoring and regulating system of large transformer moisture absorbing device according to claim 1, characterized in that, The life dynamic prediction module 140 comprises: A feature sequence construction unit 141 is configured to append the current feature set to the end of a historical feature sequence to obtain a feature time sequence; A residual life prediction unit 142 is configured to input the feature time sequence into a trained LSTM model to obtain the predicted residual life.

7. The intelligent monitoring and regulating system of large transformer moisture absorption apparatus according to claim 1, characterized in that, The intelligent decision module 150 is configured to: If an emergency alarm is detected, the control instruction is to perform regeneration; If no emergency alarm is detected, determine whether the predicted residual life is less than a preset maintenance window period and the instant state is deteriorating, and if so, the control instruction is a pre-regeneration alarm; If the predicted residual life is greater than the preset maintenance window period or the instant state is not deteriorating, determine whether the predicted residual life is less than 1 day, and if so, the control instruction is to perform regeneration; Otherwise, the control instruction is to continue monitoring.

Images