Digital station building equipment state intelligent monitoring system and method

Abstract

The invention discloses a digital station building equipment state intelligent monitoring system and method. The system comprises a data acquisition module, a multi-dimensional data processing module, an LSTM neural network analysis module and a state evaluation module. The data acquisition module acquires multi-modal time series data through a multi-type sensor group; the multi-dimensional data processing module performs space-time alignment and feature fusion on the data to construct a composite feature vector; the LSTM neural network analysis module adopts an attention gating mechanism to fuse dual-scale features, and outputs equipment state probability distribution and a residual life prediction value; and the state evaluation module updates the posterior probability of the health state of the equipment based on the dynamic Bayesian network so as to realize graded early warning and maintenance decision. According to the invention, the accuracy and predictability of equipment state monitoring are improved, the requirements of the digital station building for accurate monitoring and predictive maintenance of the equipment state are met, and the intelligent level of operation and maintenance management is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment monitoring, and in particular to a digital station equipment status intelligent monitoring system and method. Background Art

[0002] With the rapid development of information technology and industrial automation, digital power stations, as critical infrastructure, are crucial for the stability and reliability of their equipment. Existing equipment status monitoring systems for digital power stations primarily rely on a single type of sensor to collect data and employ threshold comparisons or simple rules for fault diagnosis. However, this traditional approach has numerous limitations and cannot meet the demands of modern digital power stations for accurate equipment status monitoring and predictive maintenance. Summary of the Invention

[0003] The main purpose of this application is to provide a digital station equipment status intelligent monitoring system and method, aiming to address the deficiencies of the existing technology and improve the accuracy and efficiency of access control.

[0004] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a digital station equipment status intelligent monitoring system, comprising: A data acquisition module, comprising a multi-type sensor group, for synchronously collecting equipment operating parameters, environmental parameters and associated process parameters of the digital station building to form multimodal time series data; The multi-dimensional data processing module is used to perform spatiotemporal alignment and feature fusion on the multimodal time series data to construct a time domain feature map. , frequency domain characteristics and associated network characteristics The composite eigenvector of ; LSTM neural network analysis module, taking the composite feature vector As input, the attention gating mechanism is used to fuse the dual-scale features and output the device state probability distribution. Remaining life prediction value ; The status assessment module updates the posterior probability of equipment health status based on a dynamic Bayesian network to achieve hierarchical early warning and maintenance decision-making.

[0005] Optionally, the LSTM neural network analysis module includes: Short-term memory network , used to process the window length Time series data to capture high-frequency equipment failure characteristics; Long-term memory networks , used to process the window length Time series data to capture the slow degradation trend of equipment; Attention gating unit, used to calculate fusion weights And generate fusion features , the formula is:

[0006]

[0007] in, and are the hidden states of the short-term memory network and the long-term memory network, and is a learnable parameter; is the activation function; Dual-task output unit for parallel output of state classification probability distribution and remaining life prediction , the formula is:

[0008]

[0009] in, 、 、 、 are learnable parameters.

[0010] Optionally, the LSTM neural network analysis module adopts dual-task learning optimization, and its total loss function is the classification loss With regression loss Dynamic weighted sum of:

[0011]

[0012]

[0013] in, is the task weight, is the regularization coefficient, are network parameters, is the sample size, The actual remaining life value.

[0014] Optionally, the status assessment module includes: Multi-source fusion reasoning unit, used to fuse the state probability distribution output by the LSTM neural network analysis module , Remaining life prediction value and composite eigenvectors ,Update the posterior probability of the device health status through the dynamic Bayesian network; Multidimensional evaluation index construction unit, used to construct a comprehensive evaluation index system and generate a multidimensional evaluation vector; The hierarchical decision execution unit performs hierarchical processing on the multi-dimensional evaluation vector based on fuzzy decision rules to generate early warning response strategies corresponding to different risk levels.

[0015] Optionally, updating the posterior probability of the health status of the device through a dynamic Bayesian network specifically includes: Define the hidden state space ,in, represents a discrete time point; Constructing the state transition probability matrix ,in, Indicates the slave state Transfer to state probability; Update the posterior probability distribution through the particle filter algorithm:

[0016] in, is the number of particles, For the The weight of a particle, is the Dirac function, From time 1 to The observation sequence.

[0017] Optionally, the comprehensive evaluation index system includes: Failure probability index: ; Lifespan degradation indicators: ,in Design life for equipment; Feature deviation index: ,in is the Mahalanobis distance, is the health status feature template; The multi-dimensional evaluation vector is generated: .

[0018] Optionally, the hierarchical decision execution unit uses a fuzzy rule base To implement hierarchical warning, typical rules are: :IF is AND is AND is THENAlertLevelis in, and is a fuzzy set; Defuzzification using the centroid method: AlertLevel

[0019] in, is the rule activation degree, is a fuzzy set The center value of .

[0020] Optionally, the multi-type sensor group of the data acquisition module includes: Vibration sensor array, using a three-axis MEMS accelerometer to collect equipment vibration acceleration signals; Electrical parameter sensor group, including voltage transformer, current transformer and power analyzer, used to collect voltage, current, active power and reactive power; The environmental parameter sensor group includes a temperature and humidity sensor, an air pressure sensor, and a dust concentration sensor, which are used to collect ambient temperature, relative humidity, atmospheric pressure, and dust particle concentration.

[0021] In a second aspect, the present application provides a method for intelligently monitoring the status of equipment in a digital station building, characterized in that it includes the following steps: Through the synchronous collection of equipment operating parameters, environmental parameters and related process parameters of the digital station room by multiple types of sensor groups, multi-modal time series data is formed; Performing spatiotemporal alignment and feature fusion on the multimodal time series data to construct a composite feature vector including time domain features, frequency domain features, and associated network features; Utilizing an LSTM neural network analysis module, taking the composite feature vector as input, employing an attention gating mechanism to fuse dual-scale features, and outputting a device state probability distribution and a remaining life prediction value; The posterior probability of equipment health status is updated based on a dynamic Bayesian network, and hierarchical warning and maintenance decisions are generated by combining multi-dimensional evaluation indicators.

[0022] Through the above technical solutions, the beneficial effects of the present invention are as follows: This application realizes three-dimensional perception of the operating status of digital station equipment by synchronously collecting multimodal time series data through multi-type sensor groups; constructs composite feature vectors through spatiotemporal alignment and feature fusion, thereby improving the spatiotemporal consistency and feature representation capabilities of data processing; utilizes the dual-scale feature fusion and dual-task learning optimization of the LSTM neural network analysis module to accurately capture the high-frequency fault characteristics and slow-changing degradation trends of equipment, thereby improving the accuracy of state classification and remaining life prediction; based on the state assessment module of dynamic Bayesian network and fuzzy decision-making, it realizes probabilistic reasoning, multi-dimensional evaluation and graded warning of equipment health status, and constructs a full-process closed loop from data collection, feature analysis to intelligent decision-making. The overall technical solution significantly improves the accuracy, predictability and decision-making efficiency of digital station equipment status monitoring, effectively meets the needs of modern digital stations for accurate equipment status monitoring and predictive maintenance, reduces the risk of equipment failure, and improves the intelligence level of operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention, and the embodiments in the drawings do not constitute any limitation to the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A schematic diagram of the structure of a distributed hierarchical architecture of a system provided in one embodiment of the present application; Figure 2 A schematic diagram of the structure of a digital station equipment status intelligent monitoring system provided in one embodiment of the present application; Figure 3 A schematic diagram of the structure of the LSTM neural network analysis module provided in one embodiment of the present application; Figure 4 A schematic diagram of the structure of a status assessment module provided in one embodiment of the present application; Figure 5 A flowchart of a method for intelligently monitoring the status of equipment in a digital station provided in one embodiment of the present application; The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] In order to make the above-mentioned objectives, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0027] The digital station equipment status intelligent monitoring system of the present invention adopts a distributed layered architecture design. Figure 1 As shown, it mainly includes: Edge layer: Multiple sensor nodes and edge computing units deployed at the digital station site are responsible for raw data collection and preprocessing; Network layer: A hybrid communication network built based on industrial Ethernet and 5G / WiFi6 to achieve reliable data transmission; Platform layer: A data processing center deployed in the cloud, including data storage, feature extraction, model training, and status assessment modules; Application layer: Visual interface and intelligent decision support system for different user roles.

[0028] The modules of the above-mentioned systems realize data interaction through standardized interfaces, comply with the IEC61850 communication protocol and OPC UA data model, and ensure compatibility with the existing station automation system.

[0029] In an exemplary embodiment, Figure 2 As shown, a digital station equipment status intelligent monitoring system is provided, which includes a data acquisition module, a multi-dimensional data processing module, an LSTM neural network analysis module and a status evaluation module. Among them: The data acquisition module adopts a distributed multi-sensor collaborative architecture, which includes multiple types of sensor groups, which are used to synchronously collect the equipment operating parameters, environmental parameters and related process parameters of the digital station to form multi-modal time series data, so as to achieve three-dimensional perception of the operating status of the digital station equipment.

[0030] Among them, the multi-type sensor group includes: Vibration sensor array, using a three-axis MEMS accelerometer to collect equipment vibration acceleration signals; Electrical parameter sensor group, including voltage transformer, current transformer and power analyzer, used to collect voltage, current, active power and reactive power; The environmental parameter sensor group includes a temperature and humidity sensor, an air pressure sensor, and a dust concentration sensor, which are used to collect ambient temperature, relative humidity, atmospheric pressure, and dust particle concentration.

[0031] In an example embodiment, a three-axis MEMS accelerometer is installed in key locations such as the bearing seat and housing of rotating equipment such as transformers and compressors. The accelerometer captures the equipment's vibration acceleration signals in real time at a high sampling frequency of 16kHz, can detect tiny vibration changes at the 0.01g level, and effectively identify high-frequency impact signals generated by mechanical faults such as bearing wear and rotor imbalance.

[0032] LEMLV25-P voltage transformers and LA28-NP current transformers are deployed at the incoming line and equipment input ends of the power distribution system. They synchronously collect the effective values of voltage, current, and active power at a sampling frequency of 100 Hz. Together with a power analyzer, they calculate electrical characteristics such as power factor and harmonic distortion in real time, accurately reflecting the power quality and load status of the equipment.

[0033] Multi-dimensional environmental sensors are deployed in the station building, including SHT30 temperature and humidity sensors, MS5611 air pressure sensors, and dust concentration sensors. They collect environmental temperature, humidity, air pressure, and dust content at a frequency of 1Hz. Infrared thermal imagers are used to monitor the surface temperature distribution of the equipment to comprehensively evaluate the impact of environmental factors on equipment operation.

[0034] To ensure temporal consistency of multi-source data, a hardware-level synchronization network is built using the IEEE1588 precision clock protocol to provide a time reference for each sensor. This ensures precise temporal alignment of asynchronously sampled data, such as vibration, electrical, and environmental data, laying the foundation for subsequent multi-dimensional data fusion. Collected data is transmitted in real time to the edge computing unit via industrial-grade WiFi (IEEE802.11ac) or Ethernet, ensuring real-time and integrity.

[0035] The multi-dimensional data processing module converts the original multimodal time series data into high-value composite feature vectors through spatiotemporal alignment, multi-domain feature extraction, and correlation network analysis. The specific implementation is as follows: First, time and space alignment is performed. To address the asynchronous sampling characteristics of vibration signals (16kHz), electrical parameters (100Hz), and environmental data (1Hz), hardware clock synchronization technology is used to unify all data onto a 100Hz time grid. High-frequency vibration data is downsampled using a cubic spline interpolation algorithm, while low-frequency environmental data is upsampled. This ensures strict timestamp alignment of all parameters, with time errors within 10μs. A signal integrity check mechanism is incorporated into the interpolation process. By comparing the gradients of adjacent sampling points, anomalous interpolation points are automatically detected and corrected to ensure data quality.

[0036] In the time domain feature extraction stage, the root mean square value (RMS), crest factor (CF), and kurtosis value (Kurtosis) of the vibration signal are calculated. The RMS reflects the overall level of vibration energy, the crest factor sensitively identifies impact faults, and the kurtosis value effectively detects faults such as early pitting corrosion of bearings. The electrical parameters extract the effective values of voltage / current, frequency deviation, three-phase imbalance, and total harmonic distortion (THD). These indicators are calculated in real time using a sliding window (1 second) to capture sudden changes in equipment load and abnormal power quality. Environmental parameters are normalized, and parameters such as temperature and humidity are mapped to the [0, 1] range to eliminate dimensional effects.

[0037] Frequency domain feature extraction uses a multi-level analysis method: a 1024-point fast Fourier transform (FFT) is performed on the vibration signal to calculate the power spectral density (PSD), extract the energy proportion of the main frequency and its 2-10 multiples, combine wavelet packet decomposition to extract the energy characteristics of each frequency band, and construct a 32-dimensional frequency domain feature vector; electrical parameters are analyzed for harmonic components through FFT to identify the harmonic interference characteristics generated by nonlinear devices such as inverters and rectifiers.

[0038] The correlation network construction unit calculates the Pearson correlation coefficient between parameters using a 60-second sliding window, constructing a time-varying correlation matrix. A correlation coefficient threshold of 0.7 is set to screen for strongly correlated parameter pairs. Topological features such as node degree centrality and clustering coefficient are then calculated to quantify the dynamic coupling relationship between equipment status parameters. For example, when a compressor experiences surge, the correlation coefficient between the main vibration frequency and the power parameter will increase sharply from 0.4 to 0.9 within 2 seconds. Detecting sudden changes in degree centrality can identify abnormal coupling in advance.

[0039] Finally, the time domain features, frequency domain features and associated network features are spliced into a composite feature vector Dimensional differences between features are eliminated through standardization. Furthermore, the SHAP value feature selection algorithm is introduced to dynamically select the top 30% of sensitive features contributing to device status classification. This approach preserves key information while reducing computational complexity, resulting in an optimal feature subset for input into the LSTM neural network. This entire processing flow achieves a layer-by-layer conversion from raw data to high-value features, providing structured data support for subsequent intelligent analysis.

[0040] The LSTM neural network analysis module adopts a dual-scale attention fusion architecture to achieve multi-time granularity feature extraction and intelligent analysis of device status. This module uses the composite feature vector output by the multi-dimensional data processing module As input, a parallel architecture consisting of a short-term memory network and a long-term memory network is constructed, such as Figure 3 As shown: The Short-term LSTM network uses a sliding window of 100 time steps (corresponding to 10 seconds of data) and a hidden layer dimension of 64. It focuses on capturing high-frequency fault characteristics and transient fluctuations of electrical parameters in equipment vibration signals. The network structure is as follows:

[0041] in, is the activation function, 、 is the weight matrix, is the bias term.

[0042] The Long-Term LSTM network uses a long window of 1000 time steps (corresponding to 100 seconds of data) and a hidden layer dimension of 128. It is used to learn slowly changing degradation characteristics such as device temperature drift and power trend changes, and effectively identify the long-term development pattern of progressive faults such as insulation aging and lubrication failure. The network structure is:

[0043] in, 、 is the weight matrix, is the bias term.

[0044] The fusion weight of the dual-scale hidden state is calculated through the attention gating unit. Specifically, the hidden state output by the short-term LSTM is concatenated with the long-term LSTM hidden state and then input into the fully connected layer. A 16-dimensional attention weight vector is generated through the Sigmoid activation function. This achieves dynamic weighted fusion of high-frequency impact features and long-term trend features, generating a 192-dimensional fusion feature vector, enabling the model to automatically focus on key features based on the device status. The formula is:

[0045]

[0046] When the device is in a rapidly changing state, the fusion weight Approaching 1, the model focuses on short-term characteristics; when the device is in the slow degradation stage, Approaching 0, the model focuses on long-term trend characteristics.

[0047] Fusion Features Input dual-task output unit, and output state classification probability distribution in parallel through dual-task output unit and remaining life prediction , the formula is:

[0048]

[0049] in, 、 、 、 are learnable parameters.

[0050] The LSTM neural network analysis module adopts dual-task learning optimization, and its total loss function is the classification loss With regression loss Dynamic weighted sum of:

[0051]

[0052]

[0053] in, is the dynamic weight, is the regularization coefficient, are network parameters, is the sample size, This is the actual remaining life value, obtained through historical fault data annotation.

[0054] In the above, the total loss function Balance dual-task optimization objectives through dynamic weights, dynamic weights It is used to adjust the priority of two types of tasks in the total loss function in real time. Its core design goal is to avoid imbalance in model training due to a task's loss value being too small or too large. The specific calculation logic is:

[0055] in, and are the classification loss and regression loss of the previous training batch respectively. This formula ensures that when the regression loss is high, such as when the equipment remaining life prediction error is large, the weight Automatically reduce, the model prioritizes the regression task; on the contrary, when the classification loss dominates, such as the abnormal state recognition accuracy decreases, Improve and strengthen training for classification tasks.

[0056] The status assessment module adopts the dynamic Bayesian network and fuzzy decision fusion architecture to achieve probabilistic reasoning and graded warning decision-making of equipment health status. Figure 4 As shown, this module specifically includes: Multi-source fusion reasoning unit, used to fuse the state probability distribution output by the LSTM neural network analysis module , Remaining life prediction value and composite eigenvectors ,Update the posterior probability of the device health status through the dynamic Bayesian network; Multidimensional evaluation index construction unit, used to construct a comprehensive evaluation index system and generate a multidimensional evaluation vector; The hierarchical decision execution unit performs hierarchical processing on the multi-dimensional evaluation vector based on fuzzy decision rules to generate early warning response strategies corresponding to different risk levels.

[0057] In the above, the posterior probability of the device health status is updated through the dynamic Bayesian network, specifically including: Define the hidden state space ,in, represents a discrete time point; Constructing the state transition probability matrix ,in, Indicates the slave state Transfer to state probability; Update the posterior probability distribution through the particle filter algorithm:

[0058] in, is the number of particles, For the The weight of a particle, is the Dirac function, From time 1 to The observation sequence.

[0059] The multi-dimensional evaluation index construction unit integrates the Bayesian inference results and LSTM output to build a three-dimensional evaluation index system, which includes: Failure probability index: ; Lifespan degradation indicators: ,in Design life for equipment; Feature deviation index: ,in is the Mahalanobis distance, It is the health status feature template.

[0060] After normalization, the failure probability index is directly taken, the life degradation index is normalized by the design life, and the characteristic deviation index is mapped by the maximum Mahalanobis distance threshold of the health state to generate a multidimensional evaluation vector. , providing a quantitative basis for subsequent hierarchical decision-making and realizing a multi-dimensional comprehensive evaluation of equipment status.

[0061] Hierarchical decision execution unit through fuzzy rule base To implement hierarchical warning, typical rules are: :IF is AND is AND is THENAlertLevelis in, and is a fuzzy set; Defuzzification using the centroid method: AlertLevel

[0062] in, is the rule activation degree, is a fuzzy set The center value of .

[0063] In an exemplary embodiment, the three-dimensional evaluation vector Each indicator in is divided into three fuzzy levels and described by the triangular membership function: Failure probability index : Low (L): =Triangle(0, 0, 0.3) Medium (M): =Triangle(0.2, 0.5, 0.8) High (H): =Triangle(0.7, 1, 1) Lifespan degradation index : Low (L): =Triangle(0, 0, 0.4) (remaining life > 60% of design life) Medium (M): =Triangle(0.3, 0.6, 0.9) (remaining life 10%-60% of design life) High (H): =Triangle(0.8, 1, 1) (remaining life < 10% of design life) Feature deviation index : Low (L): =Triangle(0, 0, 0.2) (Mahalanobis distance < 0.2σ) Medium (M): =Triangle(0.1, 0.3, 0.5) (Mahalanobis distance 0.2σ-0.5σ) High (H): =Triangle(0.4, 1, 1) (Mahalanobis distance > 0.5σ) According to the above definition of fuzzy sets, 27 fuzzy rules (3 indicators × 3 levels) are designed. Some typical rules are as follows:

[0064] Evaluate the vector at a certain moment For example, triggering rules :IF is AND is AND is THEN the warning level is level 2, and the rule activation degree is: .

[0065] The hierarchical decision execution unit divides the failure probability, life degradation and feature deviation indicators into fuzzy levels by constructing a multi-dimensional fuzzy rule base, and combines the dynamic Bayesian network output to achieve accurate classification of equipment status. It has multi-indicator fusion reasoning capabilities and can comprehensively evaluate the health status of equipment; its hierarchical early warning mechanism and dynamic response strategy realize an automated closed loop from risk identification to operation and maintenance decision-making, significantly improving response efficiency; through the center of gravity defuzzification method and the rule base self-learning mechanism, the system can adapt to changes in equipment characteristics and operating scenarios.

[0066] Based on the same inventive concept, the present application also provides a method for intelligently monitoring the status of digital station equipment, which is used to implement the aforementioned intelligent monitoring system for digital station equipment status. The solution provided by this method is similar to the solution described in the aforementioned system. Therefore, the specific limitations of one or more embodiments of the method for intelligently monitoring the status of digital station equipment provided below can be found in the above-mentioned limitations on the intelligent monitoring system for digital station equipment status, and will not be repeated here.

[0067] In an exemplary embodiment, Figure 5 As shown, a method for intelligent monitoring of the status of digital station equipment is provided, comprising the following steps: Step S1, synchronously collecting equipment operating parameters, environmental parameters and related process parameters of the digital station through a multi-type sensor group to form multimodal time series data; Step S2, performing spatiotemporal alignment and feature fusion on the multimodal time series data to construct a composite feature vector including time domain features, frequency domain features and associated network features; Step S3, using an LSTM neural network analysis module, taking the composite feature vector as input, adopting an attention gating mechanism to fuse the dual-scale features, and outputting a device state probability distribution and a remaining life prediction value; Step S4: Update the posterior probability of the equipment health status based on the dynamic Bayesian network, and generate graded warning and maintenance decisions in combination with multi-dimensional evaluation indicators.

[0068] Using transformers in high-speed railway stations as monitoring targets, a triaxial MEMS accelerometer, voltage / current transformers, and temperature and humidity sensors were deployed to collect 24×3600 multimodal data points daily. The IEEE 1588v2 clock synchronization protocol was used to achieve spatiotemporal alignment of sensor sampling times with an error of ≤10μs. Cubic spline interpolation was used to unify the asynchronous data to a 100Hz time grid. A 1024-point FFT transform was performed on the vibration signal to extract the main frequency energy fraction (frequency domain feature). The effective value of the current signal was calculated (time domain feature). The Pearson cross-correlation coefficient (correlation network feature) was calculated for the temperature, humidity, and oil temperature data. A 13-dimensional composite feature vector (10 dimensions for main frequency energy, 1 dimension for current effective value, and 2 dimensions for cross-correlation coefficient) was constructed and fed into an LSTM neural network analysis module. This module consists of a short-term LSTM layer (window length 100, hidden layer 64 dimensions) and a long-term LSTM layer (window length 1000, hidden layer 128 dimensions). The dual-scale features were fused through an attention gating mechanism to output a probability distribution of the device state and a predicted remaining life. The state assessment module calculates the failure probability index based on the dynamic Bayesian network and characteristic deviation index , defined when >0.2 and When the value is greater than 2.5, a red warning is triggered and a shutdown and maintenance instruction is generated. The performance test results are as follows:

[0069] As can be seen from the table above, the digital station equipment status intelligent monitoring method of this application significantly improves the accuracy of high-speed rail transformer status monitoring through multimodal feature fusion and dual LSTM network analysis. It successfully identifies winding loosening faults that are missed by traditional methods. The average error of the remaining life prediction is less than 50 hours, and the false alarm rate is 0, providing a reliable solution for predictive maintenance of digital station equipment.

[0070] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0071] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0072] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0073] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. The digital station equipment status intelligent monitoring system is characterized by: include: A data acquisition module, comprising a multi-type sensor group, for synchronously collecting equipment operating parameters, environmental parameters and associated process parameters of the digital station building to form multimodal time series data; The multi-dimensional data processing module is used to perform spatiotemporal alignment and feature fusion on the multimodal time series data to construct a time domain feature map. , frequency domain characteristics and associated network characteristics The composite eigenvector of ; LSTM neural network analysis module, taking the composite feature vector As input, the attention gating mechanism is used to fuse the dual-scale features and output the device state probability distribution. Remaining life prediction value ; The status assessment module updates the posterior probability of equipment health status based on a dynamic Bayesian network to achieve hierarchical early warning and maintenance decision-making.

2. The digital station equipment status intelligent monitoring system according to claim 1 is characterized in that: The LSTM neural network analysis module includes: Short-term memory network , used to process the window length Time series data to capture high-frequency equipment failure characteristics; Long-term memory networks , used to process the window length Time series data to capture the slow degradation trend of equipment; Attention gating unit, used to calculate fusion weights And generate fusion features , the formula is: in, and are the hidden states of the short-term memory network and the long-term memory network, and is a learnable parameter; is the activation function; Dual-task output unit for parallel output of state classification probability distribution and remaining life prediction , the formula is: in, 、 、 、 are learnable parameters.

3. The digital station equipment status intelligent monitoring system according to claim 1 is characterized in that: The LSTM neural network analysis module adopts dual-task learning optimization, and its total loss function is the classification loss With regression loss Dynamic weighted sum of: in, is the dynamic weight, is the regularization coefficient, are network parameters, is the sample size, The actual remaining life value.

4. The digital station equipment status intelligent monitoring system according to claim 3 is characterized in that: The status assessment module includes: Multi-source fusion reasoning unit, used to fuse the state probability distribution output by the LSTM neural network analysis module , Remaining life prediction value and composite eigenvectors ,Update the posterior probability of the device health status through the dynamic Bayesian network; Multidimensional evaluation index construction unit, used to construct a comprehensive evaluation index system and generate a multidimensional evaluation vector; The hierarchical decision execution unit performs hierarchical processing on the multi-dimensional evaluation vector based on fuzzy decision rules to generate early warning response strategies corresponding to different risk levels.

5. The digital station equipment status intelligent monitoring system according to claim 4 is characterized in that: Updating the posterior probability of the health status of the device through the dynamic Bayesian network specifically includes: Define the hidden state space ,in, represents a discrete time point; Constructing the state transition probability matrix ,in, Indicates the slave state Transfer to state probability; Update the posterior probability distribution through the particle filter algorithm: in, is the number of particles, For the The weight of a particle, is the Dirac function, From time 1 to The observation sequence.

6. The digital station equipment status intelligent monitoring system according to claim 5 is characterized in that: The comprehensive evaluation indicator system includes: Failure probability index: ; Lifespan degradation indicators: ,in Design life for equipment; Feature deviation index: ,in is the Mahalanobis distance, is the health status feature template; The multi-dimensional evaluation vector is generated: .

7. The digital station equipment status intelligent monitoring system according to claim 6 is characterized in that: The hierarchical decision execution unit is based on the fuzzy rule base To implement hierarchical warning, typical rules are: :IF is AND is AND is THENAlertLevelis in, and is a fuzzy set; Defuzzification using the centroid method: AlertLevel in, is the rule activation degree, is a fuzzy set The center value of .

8. The digital station equipment status intelligent monitoring system according to claim 1 is characterized in that: The multi-type sensor group of the data acquisition module includes: Vibration sensor array, using a three-axis MEMS accelerometer to collect equipment vibration acceleration signals; Electrical parameter sensor group, including voltage transformer, current transformer and power analyzer, used to collect voltage, current, active power and reactive power; The environmental parameter sensor group includes a temperature and humidity sensor, an air pressure sensor, and a dust concentration sensor, which are used to collect ambient temperature, relative humidity, atmospheric pressure, and dust particle concentration.

9. A method for intelligent monitoring of equipment status in a digital station building, characterized in that: The following steps are involved: Through the synchronous collection of equipment operating parameters, environmental parameters and related process parameters of the digital station room by multiple types of sensor groups, multi-modal time series data is formed; Performing spatiotemporal alignment and feature fusion on the multimodal time series data to construct a composite feature vector including time domain features, frequency domain features, and associated network features; Utilizing an LSTM neural network analysis module, taking the composite feature vector as input, employing an attention gating mechanism to fuse dual-scale features, and outputting a device state probability distribution and a remaining life prediction value; The posterior probability of equipment health status is updated based on a dynamic Bayesian network, and hierarchical warning and maintenance decisions are generated by combining multi-dimensional evaluation indicators.

Images