A smart campus integrated operation and maintenance service management cloud platform control system

Abstract

This invention belongs to the field of smart education technology, specifically a smart campus integrated operation and maintenance service management cloud platform control system, comprising a four-layer distributed architecture: a data acquisition layer, a data processing layer, an intelligent analysis layer, and a decision execution layer. The intelligent analysis layer integrates an energy balance anomaly detection model based on the law of conservation of energy, a CNN+LSTM multi-dimensional feature fusion model based on deep learning, an energy efficiency ratio health assessment model, and a sensor health assessment model with multi-sensor redundancy cross-validation. By adopting the above technical solutions, this application can dynamically adjust the detection benchmark according to real-time meteorological conditions, adaptively identify hidden faults and correct sensor drift data, realize the transformation of anomaly detection mode from passive to active, significantly improve the efficiency of operation and maintenance work order processing and reduce unplanned equipment downtime.

Description

Technical Field

[0001] This invention belongs to the field of smart education technology, specifically a smart campus integrated operation and maintenance service management cloud platform control system. Background Technology

[0002] In the construction of smart campuses, multiple electronic systems such as power distribution and HVAC constitute the core of campus operation. Their efficient and stable operation is related to the teaching and research and the quality of life of teachers and students. With the surge in the number and complexity of electromechanical equipment, the traditional manual operation and maintenance mode can no longer meet the needs of refined management. The smart campus integrated operation and maintenance service management cloud platform control system has emerged to achieve centralized equipment monitoring, fault early warning and operation and maintenance decision support.

[0003] Existing smart campus operation and maintenance management platforms mainly rely on installing various sensors on equipment and setting fixed alarm thresholds in the background to detect anomalies. This detection logic played a certain role in ensuring safety in the early stages of the technology, but as application scenarios become more complex, its inherent technical limitations become increasingly apparent, revealing deep-seated structural defects.

[0004] The fixed threshold settings of existing operation and maintenance platforms are out of sync with the actual operating status of equipment, making it difficult to adapt to changes in environment and load. This leads to invalid alarms or missed anomalies, and the platforms are unable to detect hidden equipment faults. Furthermore, they are susceptible to false alarms and missed alarms due to sensor aging and electromagnetic interference, weakening the effectiveness of the early warning system and failing to meet the refined operation and maintenance needs of smart campuses. Therefore, the present invention provides a cloud platform control system for integrated operation and maintenance service management of smart campuses. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: a smart campus integrated operation and maintenance service management cloud platform control system, comprising: The data acquisition layer is used to collect operating parameters of the power distribution system, HVAC system, lighting system, and information communication system through a multimodal sensor network deployed at campus equipment nodes; The data processing layer is used to perform filtering, noise reduction, temporal alignment, and feature standardization on the collected raw data. The intelligent analysis layer is used to run a physical constraint-based energy balance anomaly detection model and a deep learning-based multi-dimensional feature fusion model to perform collaborative anomaly discrimination. The decision execution layer is used to generate differentiated operation and maintenance work orders based on the judgment results of the intelligent analysis layer; The energy balance anomaly detection model is based on the law of conservation of energy to construct physical constraint equations. It dynamically calculates the theoretical energy consumption value according to real-time meteorological conditions and compares the actual energy consumption value with the theoretical energy consumption value to identify hidden faults caused by equipment efficiency degradation and data distortion caused by sensor drift.

[0007] Preferably, the data acquisition layer includes a three-phase smart meter for the power distribution circuit, a platinum resistance temperature sensor for the HVAC system, a multi-functional electricity meter and illuminance sensor for the lighting circuit, and a current monitoring module for the information and communication equipment room. The multimodal sensor network transmits the acquired data to the edge computing gateway via RS485 / MODBUS or BACNET industrial bus.

[0008] Preferably, the filtering and denoising module of the data processing layer adopts a threshold denoising method based on wavelet transform, the number of decomposition layers is set to 5, and the wavelet basis function is selected as db4 wavelet; The time alignment module performs interpolation alignment with a 100-millisecond time window as the baseline; the feature normalization module uses the Z-Score normalization method to map the original feature values ​​to the standard normal distribution interval.

[0009] The data acquisition layer is configured to acquire in real time the current and voltage parameters of the power distribution circuit, the temperature parameters of the evaporator and condenser of the HVAC system, the power and illuminance parameters of the lighting circuit, and the load status parameters of the information and communication equipment through a multimodal sensor network deployed at key equipment nodes on campus. The data processing layer is configured to perform filtering, noise reduction, time alignment, and feature standardization on the acquired raw data. The intelligent analysis layer is configured to run a physical constraint-based energy balance anomaly detection model and a deep learning-based multi-dimensional feature fusion model to perform collaborative anomaly discrimination from the two dimensions of macroscopic energy conservation and microscopic equipment characteristics, respectively. The decision execution layer is configured to generate maintenance work orders for different fault levels based on the discrimination results of the intelligent analysis layer and push the maintenance work orders to maintenance management personnel through a visual interface.

[0010] Preferably, the energy balance anomaly detection model establishes a standard energy consumption model based on the thermal parameters of the building envelope of the campus buildings, and the thermal parameters include the thermal conductivity of the wall, the window-to-wall ratio, and the thickness of the insulation layer. The standard energy consumption model takes ambient temperature, relative humidity and solar radiation intensity as input boundary conditions and calculates the theoretical energy consumption value based on the first law of thermodynamics. The energy balance anomaly detection model in the intelligent analysis layer is configured to construct physical constraint equations based on the law of conservation of energy and the first law of thermodynamics. In the specific model construction process, a standard energy consumption model of the building is first established based on the thermal parameters of the building envelope. The standard energy consumption model uses outdoor meteorological parameters as input boundary conditions to calculate the theoretical energy consumption value under standard operating conditions. Then, the real-time collected actual energy consumption value is compared with the theoretical energy consumption value. When the actual energy consumption value deviates from the theoretical energy consumption value by more than a preset physical consistency threshold, the system determines that there is an abnormal operating condition. By decomposing the deviation value into positive deviation and negative deviation, two types of abnormal phenomena with different physical mechanisms can be identified: equipment efficiency degradation type latent fault (positive deviation) and sensor drift type data distortion (negative deviation).

[0011] When the actual power consumption of an HVAC system under rated operating conditions is significantly higher than the theoretical calculated value, it indicates that the air conditioning unit may have hidden faults such as heat exchanger blockage, insufficient refrigerant, or reduced compressor efficiency. Traditional detection methods based on fixed thresholds cannot capture this gradual performance degradation because the operating parameters of the equipment are often still within the normal range. However, the energy balance anomaly detection model of this invention can issue early warning signals in the early stage of the fault by comparing real-time energy consumption with a dynamic benchmark based on meteorological conditions.

[0012] Preferably, the intelligent analysis layer further includes a sensor health assessment model, wherein at least three sensors of the same type are deployed on the same physical node, and the output signals of the three sensors are cross-validated by calculating the covariance matrix; When the deviation of a sensor from other sensors exceeds three times the standard deviation of the covariance threshold, the sensor is determined to have zero drift or sensitivity decay. The sensor health assessment model is configured to use redundant data from multiple sensors for cross-validation. At least three sensors of the same type are deployed on the same physical node. By comparing the output signals of the three sensors, their covariance matrix is ​​calculated. When the deviation of a sensor from the other two sensors exceeds a certain multiple of the covariance threshold, the system determines that the sensor has experienced zero-point drift or sensitivity decay. The sensor health assessment model is further configured to establish the sensor's decay curve based on historical operating data, predict its future reliability trend, and perform adaptive weighted correction on the sensor's output signal during data fusion.

[0013] The deep learning-based multi-dimensional feature fusion model includes convolutional neural networks and long short-term memory networks; The convolutional neural network extracts local correlation features of sensor data from the spatial distribution dimension and identifies abnormal correlation patterns between multiple measurement points within the same subsystem. The Long Short-Term Memory network contains multiple hidden layers, each with 128 LSTM units, and extracts the changing trend features of device operating parameters from the time series dimension. The deep learning multi-dimensional feature fusion model in the intelligent analysis layer is configured to use convolutional neural networks and long short-term memory networks to perform collaborative feature extraction and anomaly detection on multi-source heterogeneous data. The convolutional neural network is configured to extract local correlation features of sensor data from the spatial distribution dimension and identify abnormal correlation patterns between multiple measuring points in the same subsystem. For example, when the power of the lighting circuit in a certain area increases abnormally, the convolutional neural network can analyze the data changes in adjacent areas to determine whether the anomaly is due to lamp aging or line leakage. The long short-term memory network is configured to extract the changing trend features of equipment operating parameters from the time series dimension and identify the time evolution law of equipment performance degradation. For example, by analyzing the operating data of the air conditioning system for several consecutive weeks, the long short-term memory network can identify the gradual decline trend of heat exchange efficiency and issue an early warning before it evolves into a significant fault.

[0014] The training process of the deep learning multi-dimensional feature fusion model includes: collecting a labeled dataset containing normal operating condition data and fault case data; The convolutional neural network is configured with a 3-layer convolutional structure and the long short-term memory network with a 2-layer hidden structure. The activation function is the ReLU function and the loss function is the cross-entropy loss function. Iterative training is performed using the Adam optimization algorithm, and the model parameters are then embedded in the cloud platform inference engine. The training process of the deep learning multi-dimensional feature fusion model follows these steps: A large amount of labeled datasets containing normal operating condition data and various fault case data were collected from the historical operation database of the smart campus. The labeled datasets are classified and labeled according to the fault type, including typical operating conditions such as equipment performance degradation, sensor drift, poor line contact, and sudden load changes. In the model building phase, the kernel size and number of layers of the convolutional neural network are set, the number of hidden neurons and time step of the long short-term memory network are set, the activation function is set to the ReLU function, and the loss function is set to the cross-entropy loss function. The model parameters are iteratively trained using the Adam optimization algorithm. The gradient is calculated and the weight parameters are updated using the backpropagation algorithm. After training, the model parameters are stored in the inference engine of the cloud platform. During the online operation phase, the model performs forward inference once every preset time interval and outputs the anomaly probability value of each device node.

[0015] Preferably, the intelligent analysis layer further includes an energy efficiency ratio health assessment model, used to calculate the ratio of actual cooling capacity to input electrical power for the HVAC system, and compare the ratio with the energy efficiency ratio specified by the equipment manufacturer. When the actual energy efficiency ratio is lower than the preset ratio of the nominal value, it is determined that there is performance degradation; The energy efficiency ratio health assessment model performs sliding window smoothing on the energy efficiency ratio data and uses an exponentially weighted moving average algorithm to adjust the trend of change. The energy efficiency ratio health assessment model is configured to calculate the ratio of actual cooling capacity to input electrical power for HVAC systems, and compare this ratio with the energy efficiency ratio specified by the equipment manufacturer. When the actual energy efficiency ratio is lower than the preset ratio of the specified value, the system determines that there is performance degradation. In the specific calculation process, the actual cooling capacity of the air conditioner is calculated by measuring the evaporator inlet air temperature, outlet air temperature, and circulating air volume. By reading the power metering data, the input power of the air conditioner is obtained. The ratio of the actual cooling capacity to the input power is the real-time energy efficiency ratio. Since the performance degradation caused by latent faults is a gradual process, the decrease in energy efficiency ratio is often very slow. The change in a single measurement may be lower than a fixed threshold. This invention uses a sliding window to smooth the energy efficiency ratio data on the time axis, amplifying the slight trend of change, thereby achieving effective capture of latent faults.

[0016] Preferably, the specific rules for the decision execution layer to generate differentiated operation and maintenance work orders based on anomaly detection results are as follows: When the energy efficiency ratio is lower than 80% of the nominal value, a secondary priority maintenance work order is generated and pushed to the HVAC maintenance personnel; When the sensor health assessment score is below 60, a level 3 work order is generated. When the load rate exceeds 120% of the rated value, a level 1 alarm is triggered and the on-duty personnel are notified via SMS or telephone. At the decision-making and execution level, this invention is configured to generate differentiated maintenance work orders based on anomaly detection results. For HVAC equipment with a significant decrease in energy efficiency ratio, the system automatically generates maintenance work orders containing cleaning or refrigerant replenishment suggestions and prioritizes pushing them to qualified maintenance personnel. For detected sensor drift issues, the system automatically generates work orders containing sensor calibration or replacement suggestions. For sudden overload anomalies, the system immediately triggers a level one alarm and notifies on-duty personnel via SMS or telephone. This differentiated work order generation mechanism avoids the one-size-fits-all alarm push in traditional platforms and greatly improves the work efficiency of maintenance personnel.

[0017] Preferably, the control system is configured to support multi-subsystem collaborative optimization and establish a multi-subsystem joint energy model among the power distribution system, HVAC system, lighting system, and information communication system; The causal relationships between subsystems are determined based on the energy flow matrix, and multiple related abnormal events are fused and analyzed to generate a comprehensive diagnostic conclusion. The smart campus integrated operation and maintenance service management cloud platform control system of the present invention is configured to support the collaborative optimization of multiple subsystems. There is a coupling relationship of energy flow between the power distribution system, the HVAC system, the lighting system and the information and communication system. For example, when the load of the information and communication system increases, more heat will be generated, which will affect the cooling load of the HVAC system. By establishing a joint energy model of multiple subsystems, collaborative anomaly detection between systems can be achieved. When the abnormal pattern of one subsystem is causally related to the operating state of another subsystem, the system can automatically identify this relationship and integrate and analyze multiple abnormal events to generate more accurate diagnostic conclusions.

[0018] The beneficial effects of this invention are as follows: 1. The intelligent campus integrated operation and maintenance service management cloud platform control system described in this invention transforms anomaly detection from a passive mode relying on fixed thresholds to an active mode based on physical laws through an energy balance anomaly detection model. Since the benchmark value is dynamically calculated based on real-time meteorological conditions, the system's detection sensitivity is not affected by seasonal changes or load fluctuations. During the high-temperature period in summer, the system automatically raises the temperature alarm threshold to avoid a large number of invalid alarms; during the transitional season, the system automatically lowers the threshold to ensure that the sensitivity of anomaly detection does not decrease. This adaptive mechanism can fundamentally solve the logical disconnect between threshold setting and actual operating state in traditional technologies.

[0019] 2. The intelligent campus integrated operation and maintenance service management cloud platform control system described in this invention, through the synergistic effect of the energy efficiency ratio health assessment model and the deep learning multi-dimensional feature fusion model, can effectively identify latent faults. Experimental data shows that this invention can provide early warning of the performance degradation of air conditioning heat exchangers more than three weeks in advance, and can identify the aging of LED lamps with a power decrease of only 5%. Compared with the traditional fixed threshold method, the detection rate of latent faults is greatly improved.

[0020] 3. The intelligent campus integrated operation and maintenance service management cloud platform control system described in this invention can achieve adaptive compensation for sensor drift through a sensor health assessment model. In the event of slight sensor drift, the system can automatically identify and correct abnormal data to ensure data quality. In the event of severe sensor drift, the system can issue a replacement warning in a timely manner to avoid missed reports due to data distortion. This capability enhances the reliability and robustness of the entire warning system.

[0021] 4. The smart campus integrated operation and maintenance service management cloud platform control system described in this invention, through a differentiated work order generation mechanism and multi-subsystem collaborative optimization logic, elevates smart campus operation and maintenance management from a passive response to a proactive prevention level. Operation and maintenance personnel are no longer troubled by a massive number of invalid alarms, but can instead focus their attention on the equipment hazards that truly need to be addressed. Attached Figure Description

[0022] The invention will now be further described with reference to the accompanying drawings.

[0023] Figure 1 This is a structural block diagram of a smart campus integrated operation and maintenance service management cloud platform control system according to the present invention; Figure 2 This is a schematic diagram of the model architecture of the intelligent analysis layer in this invention. Detailed Implementation

[0024] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0025] like Figure 1 and Figure 2 As shown in the figure, a smart campus integrated operation and maintenance service management cloud platform control system according to an embodiment of the present invention includes a data acquisition layer, a data processing layer, an intelligent analysis layer and a decision execution layer. The four-layer distributed architecture system follows the core logic of perception-analysis-decision. The data acquisition layer is configured to acquire in real time the current and voltage parameters of the power distribution circuit, the evaporator and condenser temperature parameters of the HVAC system, the power and illuminance parameters of the lighting circuit, and the load status parameters of the information communication equipment through a multimodal sensor network deployed at key equipment nodes in the campus. In specific implementations, key nodes of the power distribution circuit are equipped with three-phase smart meters and open-type current transformers to achieve high-precision sampling of voltage, current, power factor, and harmonic content. The sampling accuracy is configured to reach the 0.5 standard. The evaporator and condenser of the HVAC system are equipped with platinum resistance temperature sensors (Pt100), with a temperature measurement accuracy of ±0.1℃. The lighting circuit is equipped with a single-phase multi-functional electricity meter and illuminance sensor, which can simultaneously collect the active power, apparent power, and desktop illuminance value of the measured area. The information communication equipment room is equipped with rack-mounted current monitoring modules, which can collect the load current and load power of servers, switches, and storage devices in real time.

[0026] In the hardware architecture design of the data acquisition layer, various types of sensors transmit the acquired data to the edge computing gateway through an industrial bus (RS485 / MODBUS or BACNET); The edge computing gateway is configured as an embedded controller with data caching, breakpoint resume and protocol conversion functions. It integrates a high-precision analog-to-digital converter and the sampling rate is configured to be no less than 1000 SPS. The gateway transmits the pre-processed data to the cloud platform data processing layer via 4G / 5G or fiber optic Ethernet. To ensure the reliability of data transmission, the system uses the MQTT protocol to build a communication link with message queue persistence, ensuring that data is not lost in the event of network interruption.

[0027] The data processing layer is configured to perform filtering, noise reduction, temporal alignment, and feature standardization on the acquired raw data; In a specific implementation, the filtering and denoising module adopts a threshold denoising method based on wavelet transform. This method first performs multi-scale wavelet decomposition on the original signal, with the number of decomposition layers set to 5, and the wavelet basis function selected is the db4 wavelet. In the wavelet domain, the high-frequency coefficients are thresholded by an adaptive threshold function, effectively eliminating random noise and power frequency interference introduced by the sensor. The denoised signal is recovered by wavelet reconstruction algorithm. The time alignment module is configured to solve the problem of data time axis misalignment caused by the difference in sampling period of different sensors. The system uses 100 milliseconds as the reference time window to interpolate and align the sampling data of each sensor to ensure that all data are on the same time reference when performing multi-source data fusion analysis. The feature standardization module adopts the Z-Score standardization method to map the original feature values ​​of different dimensions to the standard normal distribution interval with a mean of 0 and a variance of 1, providing input data with a unified format for subsequent intelligent analysis models.

[0028] The intelligent analysis layer is configured to run a physically constrained energy balance anomaly detection model and a deep learning-based multi-dimensional feature fusion model, performing collaborative anomaly discrimination from two dimensions: macroscopic energy conservation and microscopic device characteristics. The technical details of the energy balance anomaly detection model are as follows: Establish a standard energy consumption model for buildings based on the thermal parameters of the building envelope; During the model building process, it is necessary to accurately obtain geometric and material parameters such as the wall thermal conductivity (W / (m·K), window-to-wall ratio, and insulation layer thickness. These parameters are extracted from the building information model database or obtained through on-site thermal testing. The standard energy consumption model uses outdoor meteorological parameters as input boundary conditions. These meteorological parameters include, but are not limited to, ambient temperature, relative humidity, and solar radiation intensity. The calculation of the standard energy consumption model is based on the first law of thermodynamics and the basic equations of heat transfer. Specifically, the heat loss of the building's foundation is calculated using the formula Q=K×A×Δt; Where Q is the heat consumption, K is the heat transfer coefficient, A is the area of ​​the building envelope, and Δt is the indoor-outdoor temperature difference; Based on this, the internal heat source loads such as solar radiation heat gain, human body heat dissipation, and equipment heat dissipation are superimposed to obtain the theoretical energy consumption value under standard operating conditions.

[0029] The system compares the real-time collected actual energy consumption value with the theoretical energy consumption value. During the comparison process, the time delay effect caused by building thermal inertia is considered by setting the physical consistency threshold. The theoretical energy consumption value is corrected on the time axis by constructing a first-order hysteresis model. When the actual energy consumption value deviates from the theoretical energy consumption value by more than the physical consistency threshold, the system determines that there is an abnormal operating condition. Furthermore, by decomposing the deviation value into positive and negative deviations, it is possible to identify two types of abnormal phenomena with different physical mechanisms: equipment efficiency degradation-type latent faults and sensor drift-type data distortion.

[0030] Specifically, when the actual power consumption of an HVAC system under rated operating conditions is significantly higher than the theoretical calculated value, it indicates that the air conditioning unit may have hidden faults such as heat exchanger blockage, insufficient refrigerant, or reduced compressor efficiency. Traditional detection methods based on fixed thresholds cannot capture this gradual performance degradation because the operating parameters of the equipment are often still within the normal range. However, the energy balance anomaly detection model of this invention can issue early warning signals in the early stage of the fault by comparing real-time energy consumption with a dynamic benchmark based on meteorological conditions.

[0031] To effectively combat sensor drift interference, this invention introduces a sensor health assessment model into the intelligent analysis layer. This model is configured to utilize redundant data from multiple sensors for cross-validation. At least three identical sensors are deployed on the same physical node, and their covariance matrix is ​​calculated by comparing the output signals of the three sensors. The formula for calculating the covariance matrix is ​​as follows: Cov(X,Y)=E[(X-μX)(Y-μY)]; Where Cov(X,Y) is the covariance of random variables X and Y; E is the expected value (mean); μX is the expected value of X, E[X]; μY is the expected value of Y, E[Y]. Where X and Y are the output signals of the two sensors; When the deviation of a sensor from the other two sensors exceeds a certain number of times the covariance threshold (usually set to 3 times the standard deviation), the system determines that the sensor has experienced zero drift or sensitivity decay. The sensor health assessment model is further configured to establish the sensor's decay curve based on historical operating data, predict its future reliability trend through exponential smoothing, and perform adaptive weighted correction on the sensor's output signal during the data fusion process. The logic of adaptive weighted correction is as follows: based on the sensor health assessment results, a weight coefficient is assigned to each sensor, and the weight coefficient is proportional to the health assessment score; When performing data fusion, a weighted average algorithm is used to calculate the fused signal value, thereby reducing the interference of abnormal sensors on the overall judgment.

[0032] The deep learning multi-dimensional feature fusion model in the intelligent analysis layer is configured to use convolutional neural networks and long short-term memory networks to perform collaborative feature extraction and anomaly detection on multi-source heterogeneous data. The convolutional neural network is configured to extract local correlation features of sensor data from the spatial distribution dimension and identify abnormal correlation patterns between multiple measurement points in the same subsystem. In the specific network architecture implementation, the convolutional layer uses 3×3 convolutional kernels, and the number of convolutional kernels increases from 32 in the first layer to 128 in the third layer. The activation function is the ReLU function, which is used to introduce non-linear transformation. The pooling layer uses 2×2 max pooling operation to reduce feature dimension and computational complexity. The fully connected layer contains 512 neurons and is used to integrate the local features extracted by the convolutional layer.

[0033] The spatial feature extraction logic of a convolutional neural network is as follows: The system organizes data from multiple sensors within the same subsystem into a two-dimensional feature matrix, where rows represent different sensor nodes and columns represent different time sampling points. The convolution kernel performs sliding convolution operations in the row direction (sensor node direction) and column direction (time direction) to extract spatial correlation features between adjacent sensor nodes and temporal correlation features between adjacent time points. For example, when the power of the lighting circuit in a certain area increases abnormally, the convolutional neural network can analyze the data changes in adjacent areas to determine whether the abnormality is due to aging of the lamps (local feature) or leakage in the circuit (related feature diffusion).

[0034] Long Short-Term Memory (LSTM) networks are configured to extract trend features of equipment operating parameters from a time-series dimension and identify the temporal evolution of equipment performance degradation. In a specific implementation, the Long Short-Term Memory (LSTM) network contains two hidden layers, each containing 128 LSTM units. The time step of the input sequence is set to 168 time points, corresponding to one week of sampled data (with hourly sampling intervals). Through the collaborative working mechanism of forget gates, input gates, and output gates, the LSTM network selectively retains or forgets historical information. By analyzing the continuous operation data of the air conditioning system for several weeks, the LSTM network can identify the gradual decline trend of heat exchange efficiency and issue an early warning before it evolves into a significant failure.

[0035] The training process of a deep learning multi-dimensional feature fusion model follows these steps: A large number of labeled datasets containing normal operating condition data and various fault case data were collected from the historical operation database of the smart campus. The labeled datasets were classified and labeled according to the fault type, including typical operating conditions such as equipment performance degradation, sensor drift, poor line contact, and sudden load changes. During the annotation process, experienced operations and maintenance engineers were invited to manually annotate the data to ensure annotation quality. The training dataset was configured to have no fewer than 100,000 samples, of which 60% were normal operating condition samples and 40% were various fault samples.

[0036] During the model building phase, the kernel size and number of layers of the convolutional neural network were set (3 convolutional layers, kernel size 3×3), the number of hidden layer neurons and time step of the long short-term memory network were set (2 hidden layers, 128 units per layer, time step 168), the activation function was set to ReLU, the loss function was set to cross-entropy loss, the optimizer adopted the Adam optimization algorithm, the learning rate was set to 0.001, the first moment estimation decay coefficient β1 was set to 0.9, and the second moment estimation decay coefficient β2 was set to 0.999.

[0037] The model parameters were iteratively trained using the Adam optimization algorithm. The training process employed the Mini-Batch gradient descent method with a batch size of 64, and a total of 100 epochs of training were performed. In each epoch, the training dataset is divided into multiple batches, which are then fed into the network sequentially for forward propagation computation to calculate the loss function value. The gradient is calculated and the weight parameters are updated using the backpropagation algorithm. During the training process, a validation dataset is used to monitor the model performance. When the value of the validation set loss function no longer decreases for 10 consecutive epochs, the learning rate decay strategy is activated to prevent overfitting.

[0038] After training is complete, the model parameters are stored in the inference engine of the cloud platform; During the online operation phase, the model performs forward inference once every preset time interval. The input is the feature vector after standardization by the data processing layer, and the output is the anomaly probability value of each device node. Device nodes with an anomaly probability value greater than 0.8 are judged as abnormal nodes, and the system automatically triggers the alarm process.

[0039] The energy efficiency ratio health assessment model is configured for HVAC systems to calculate the ratio of actual cooling capacity to input electrical power and compare this ratio with the energy efficiency ratio specified by the equipment manufacturer. When the actual energy efficiency ratio is lower than the preset ratio of the nominal value, the system determines that there is performance degradation; In the specific calculation process, the actual cooling capacity of the air conditioner is calculated by measuring the evaporator inlet air temperature, outlet air temperature, and circulating air volume. The formula for calculating the actual cooling capacity is as follows: ; Where Q is the actual cooling capacity (unit: W); ρ is the air density (kg / m³); V is the circulating air volume (m³ / s); Cp is the specific heat capacity of air (J / (kg·K)). Evaporator inlet air temperature (unit: °C); Evaporator outlet air temperature (unit: °C), during cooling > When the temperature difference is positive, the cooling capacity is also positive. By reading the power metering data, the input electrical work P (kW) of the air conditioner is obtained. The ratio of the actual cooling capacity to the input electrical work is the real-time energy efficiency ratio. Where COP is the energy efficiency ratio, Q is the actual cooling capacity, and P is the power input.

[0040] Because performance degradation caused by latent faults is a gradual process, the decrease in energy efficiency ratio (EWMA) is often very slow, and the change in a single measurement may be below a fixed threshold. To amplify subtle trends, a sliding window smoothing process is applied to the EWMA data over time. The width of the sliding window is set to 24 hours (i.e., 24 sampling points), and an exponentially weighted moving average (EWMA) algorithm is used. The smoothing coefficient α is set to 0.3. The recursive formula for the EWMA algorithm is as follows: ; in The smoothed energy efficiency ratio at the current time t; The measured energy efficiency ratio at the current time t; The smoothed energy efficiency ratio of the previous moment; α is the smoothing coefficient; t is the discrete-time index; In this way, the system can effectively identify aging LED lights with a power drop of only 5%, and the detection rate of latent faults is increased by more than 60% compared with the traditional fixed threshold method.

[0041] At the decision-making and execution level, this invention is configured to generate differentiated operation and maintenance work orders based on anomaly detection results. The work order generation logic follows these rules: For HVAC equipment with a significantly reduced energy efficiency ratio (below 80% of the nominal value), the system automatically generates a maintenance work order containing cleaning or refrigerant replenishment suggestions. The work order is set to a level two priority and is pushed to qualified HVAC maintenance personnel first. For detected sensor drift issues (sensor health assessment score below 60), the system automatically generates a work order containing recommendations for sensor calibration or replacement, with the work order priority set to level three; In the event of a sudden overload anomaly (load rate exceeding 120% of the rated value), the system immediately triggers a Level 1 alarm and notifies the on-duty personnel via SMS or telephone, while simultaneously generating an emergency repair work order. This differentiated work order generation mechanism avoids the uniform alarm push in traditional platforms, which helps improve the work efficiency of operation and maintenance personnel. Actual operation data shows that the application of this invention has increased the processing efficiency of operation and maintenance work orders by 40% and reduced unplanned equipment downtime by 55%.

[0042] Furthermore, the intelligent campus integrated operation and maintenance service management cloud platform control system of the present invention is configured to support the collaborative optimization of multiple subsystems, and there is a coupling relationship of energy flow between the power distribution system, the HVAC system, the lighting system and the information and communication system; For example, when the load on an information and communication system increases, more heat is generated, affecting the cooling load of the HVAC system. This invention can achieve coordinated anomaly detection between systems by establishing a multi-subsystem joint energy model. During the model construction process, an energy flow matrix is ​​established between each subsystem. The dimension of the energy flow matrix is ​​N×N, where N is the number of subsystems. The matrix element a(i,j) represents the proportion of energy contribution of subsystem j to subsystem i. For example, when the load on the information and communication system increases by 10%, the cooling load on the HVAC system will increase by approximately 5% according to the energy flow matrix. When performing anomaly detection, the system not only analyzes the operating status of each subsystem itself, but also analyzes the energy flow anomalies between subsystems. When an abnormal pattern in one subsystem is causally related to the operating state of another subsystem (the causal relationship is determined by the energy flow matrix), the system can automatically identify this relationship and integrate and analyze multiple abnormal events to generate more accurate diagnostic conclusions.

[0043] Example: This invention provides a comprehensive operation and maintenance service management cloud platform control system deployed in a smart campus of a university. The system is constructed with a four-layer distributed architecture according to the technical solution of this invention and deploys a multimodal sensor network including 862 sensor nodes.

[0044] Data acquisition layer: 42 three-phase smart meters are deployed in the power distribution circuit, 156 temperature sensors are deployed in the HVAC system, 328 multi-functional electricity meters and illuminance sensors are deployed in the lighting system, and 34 current monitoring modules are deployed in the information and communication system computer room.

[0045] Data processing layer: Deploys a wavelet transform-based filtering and denoising module, an interpolation-based temporal alignment module, and a Z-Score feature normalization module.

[0046] Intelligent Analysis Layer: Runs an energy balance anomaly detection model. The standard building energy consumption model is constructed based on the thermal parameters of the building envelope: wall thermal conductivity of 1.2 W / (m·K), window-to-wall ratio of 0.35, and insulation layer thickness of 50 mm. Runs a deep learning multi-dimensional feature fusion model. The convolutional neural network contains 3 layers of convolution and 2 layers of fully connected layers. The long short-term memory network contains 2 hidden layers with 128 units per layer. Runs an energy efficiency ratio health assessment model and a sensor health assessment model.

[0047] Decision execution layer: Generates differentiated operation and maintenance work orders, supporting collaborative optimization of multiple subsystems.

[0048] Comparative example: The comparative example uses a traditional fixed threshold alarm scheme. Other hardware configurations are the same as in the example. The anomaly detection logic is as follows: a high temperature alarm is triggered when the temperature exceeds 35°C, an overload alarm is triggered when the power exceeds 120% of the rated value, and no alarm is triggered when the equipment operating parameters are within the normal range.

[0049] Work order generation: All alarms are pushed uniformly without classification or priority distinction.

[0050] Table 1: Comparison of Embodiments of the Invention with Comparative Examples Lead time for latent fault warning (days) 0 (Unable to issue a warning) 21 Energy balance models enable proactive early warning Latent fault detection rate (%) 18 78 Energy efficiency ratio assessment model and deep learning work together to improve Sensor drift detection accuracy (%) 45 93 Multi-sensor redundancy cross-validation mechanism False alarm rate (%) 42 8.5 Physical constraint model eliminates environmental interference Maintenance work order processing efficiency (%) 58 85 Differentiated chemical single mechanism optimizes resource allocation Unplanned equipment downtime (hours / year) 1860 837 Proactive prevention mechanisms reduce downtime risk Multi-subsystem anomaly correlation identification capability none have Joint energy model enables cross-system diagnosis Alarm response time (minutes) 5 0.5 Edge computing and real-time inference work together to accelerate speed As can be seen from the comparison data table above, the energy balance anomaly detection model transforms anomaly detection from a passive mode that relies on fixed thresholds to an active mode based on physical laws. Since the benchmark value is dynamically calculated based on real-time meteorological conditions, the system's detection sensitivity will not be affected by seasonal changes or load fluctuations. During periods of high summer temperatures, the system automatically raises the temperature alarm threshold, preventing a large number of invalid alarms. During the transitional season, the system automatically lowers the threshold to ensure that the sensitivity of anomaly detection does not decrease. This adaptive mechanism can fundamentally solve the logical disconnect between the threshold setting and the actual operating state in traditional technology, and the false alarm rate is reduced significantly from 42% to 8.5%.

[0051] This invention, through the synergistic effect of an energy efficiency ratio health assessment model and a deep learning multi-dimensional feature fusion model, can effectively identify latent faults. Experimental data shows that this invention can provide early warning of the performance degradation of air conditioning heat exchangers up to 21 days in advance, and can identify the aging of LED lights with a power decrease of only 5%. The detection rate of latent faults has increased from 18% to 78%. This capability is of great value in ensuring the healthy operation of campus facilities throughout their entire life cycle.

[0052] The sensor health assessment model enables adaptive compensation for sensor drift. In the event of slight sensor drift, the system can automatically identify and correct abnormal data to ensure data quality. When a sensor experiences severe drift, the system can issue a timely replacement warning, avoiding missed reports due to data distortion. The accuracy of sensor drift identification has increased from 45% to 93%, enhancing the reliability and robustness of the entire warning system.

[0053] This invention elevates smart campus operation and maintenance management from a passive response to a proactive prevention level through a differentiated work order generation mechanism and multi-subsystem collaborative optimization logic. Maintenance personnel are no longer troubled by a massive number of invalid alarms, but can instead focus their attention on the actual equipment hazards that need to be addressed. Actual operation data shows that the processing efficiency of operation and maintenance work orders has increased from 58% to 85%, and the unplanned equipment downtime has decreased from 1860 hours / year to 837 hours / year.

[0054] In terms of specific signal flow and control logic, the technical solution of this invention can realize a complete closed loop from data perception to intelligent decision-making. When the multimodal sensor network collects raw data, the data flow first passes through the edge computing gateway for preprocessing. The preprocessed data is uploaded to the cloud platform data processing layer via the MQTT protocol. After the data processing layer completes filtering and noise reduction, time alignment and feature standardization, the standardized data is sent to the intelligent analysis layer. The energy balance anomaly detection model in the intelligent analysis layer first calculates the theoretical energy consumption value and compares it with the actual value to determine whether there is a macroscopic energy anomaly. The deep learning multi-dimensional feature fusion model performs collaborative feature extraction on multi-source data to determine whether there are micro-device anomalies. The energy efficiency ratio health assessment model calculates the real-time energy efficiency ratio of the HVAC system to determine whether there is performance degradation. The sensor health assessment model evaluates the reliability status of each sensor and performs adaptive weighted correction for abnormal sensors.

[0055] The output of the intelligent analysis layer includes the anomaly probability value, anomaly type classification, anomaly level assessment, and confidence score of each device node. This output data is sent to the decision execution layer, which generates differentiated operation and maintenance work orders based on the preset rule engine. Level 1 alarms are immediately notified to on-duty personnel via SMS or telephone, while Level 2 and Level 3 work orders are pushed to operation and maintenance personnel through a visual interface. The multi-subsystem joint energy model performs fusion analysis on cross-system anomaly events and generates a comprehensive diagnostic report. The end-to-end latency of the entire closed loop is controlled within 30 seconds, ensuring the timeliness of anomaly handling.

[0056] In terms of specific hardware deployment, the edge computing gateway of this invention uses an industrial-grade ARM Cortex-A72 processor with a main frequency of 1.8GHz, equipped with 4GB DDR4 memory and 32GB eMMC storage. The gateway runs a Linux operating system, deploys a Docker container engine, supports microservice architecture deployment, and the sensor network adopts a hybrid wired and wireless networking method. The wired part adopts RS485 / MODBUS bus, and the wireless part adopts LoRaWAN low-power wide area network.

[0057] In terms of software architecture, the cloud platform adopts a microservice architecture. Each functional module (data acquisition, data processing, intelligent analysis, decision execution, and visualization) runs as an independent microservice on the Kubernetes container orchestration platform. Data storage adopts a hybrid architecture of time-series database InfluxDB and relational database PostgreSQL. The time-series database stores time-series data collected by sensors, while the relational database stores device ledgers, work order information, and model parameters. The inference engine of the intelligent analysis model adopts the inference framework optimized by TensorRT, which can achieve real-time inference on GPU acceleration cards.

[0058] In summary, this invention, through technological innovations in energy balance anomaly detection models, energy efficiency ratio health assessment models, sensor health assessment models, and deep learning multi-dimensional feature fusion models, constructs a global anomaly detection mechanism that can adapt to dynamic environmental changes, identify hidden faults, and combat sensor drift interference.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart campus integrated operation and maintenance service management cloud platform control system, characterized in that, include: The data acquisition layer is used to collect operating parameters of the power distribution system, HVAC system, lighting system, and information communication system through a multimodal sensor network deployed at campus equipment nodes; The data processing layer is used to perform filtering, noise reduction, temporal alignment, and feature standardization on the collected raw data. The intelligent analysis layer is used to run a physical constraint-based energy balance anomaly detection model and a deep learning-based multi-dimensional feature fusion model to perform collaborative anomaly discrimination. The decision execution layer is used to generate differentiated operation and maintenance work orders based on the judgment results of the intelligent analysis layer; The energy balance anomaly detection model is based on the law of conservation of energy to construct physical constraint equations. It dynamically calculates the theoretical energy consumption value according to real-time meteorological conditions and compares the actual energy consumption value with the theoretical energy consumption value to identify hidden faults caused by equipment efficiency degradation and data distortion caused by sensor drift.

2. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The data acquisition layer includes three-phase smart meters for power distribution circuits, platinum resistance temperature sensors for HVAC systems, multi-functional electricity meters and illuminance sensors for lighting circuits, and current monitoring modules for information and communication equipment rooms. The multimodal sensor network transmits the collected data to the edge computing gateway via an industrial bus.

3. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The filtering and denoising module of the data processing layer adopts a threshold denoising method based on wavelet transform; The timing alignment module performs interpolation alignment with a 100-millisecond time window as the baseline. The feature standardization module uses the Z-Score standardization method to map the original feature values ​​to the standard normal distribution interval.

4. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The energy balance anomaly detection model establishes a standard energy consumption model based on the thermal parameters of the building envelope of the campus buildings. The thermal parameters include the thermal conductivity of the wall, the window-to-wall ratio, and the thickness of the insulation layer. The standard energy consumption model uses ambient temperature, relative humidity, and solar radiation intensity as input boundary conditions, and calculates the theoretical energy consumption value based on the first law of thermodynamics.

5. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The intelligent analysis layer also includes a sensor health assessment model, which deploys at least three sensors of the same type on the same physical node and cross-validates the output signals of the three sensors by calculating the covariance matrix. When the deviation of a sensor from other sensors exceeds three times the standard deviation of the covariance threshold, the sensor is determined to have zero-point drift or sensitivity decay.

6. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The deep learning-based multi-dimensional feature fusion model includes convolutional neural networks and long short-term memory networks; The convolutional neural network extracts local correlation features of sensor data from the spatial distribution dimension and identifies abnormal correlation patterns between multiple measurement points within the same subsystem. The Long Short-Term Memory (LSTM) network contains multiple hidden layers, each with 128 LSTM units, and extracts the changing trend features of device operating parameters from a time series perspective.

7. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 6, characterized in that: The training process of the deep learning multi-dimensional feature fusion model includes: collecting a labeled dataset containing normal operating condition data and fault case data; The convolutional neural network is configured with a 3-layer convolutional structure and the long short-term memory network with a 2-layer hidden structure. The activation function is the ReLU function and the loss function is the cross-entropy loss function. The Adam optimization algorithm is used for iterative training, and the model parameters are then embedded in the cloud platform inference engine.

8. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The intelligent analysis layer also includes an energy efficiency ratio health assessment model, which is used to calculate the ratio of actual cooling capacity to input electrical power for HVAC systems and compare the ratio with the energy efficiency ratio specified by the equipment manufacturer. When the actual energy efficiency ratio is lower than the preset ratio of the nominal value, it is determined that there is performance degradation; The energy efficiency ratio health assessment model performs sliding window smoothing on the energy efficiency ratio data and uses an exponentially weighted moving average algorithm to adjust the trend of change.

9. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The specific rules for the decision execution layer to generate differentiated operation and maintenance work orders based on anomaly detection results are as follows: When the energy efficiency ratio is lower than 80% of the nominal value, a secondary priority maintenance work order is generated and pushed to the HVAC maintenance personnel; When the sensor health assessment score is below 60, a level 3 work order is generated. When the load rate exceeds 120% of the rated value, a Level 1 alarm is triggered and the on-duty personnel are notified via SMS or telephone.

10. The intelligent campus integrated operation and maintenance service management cloud platform control system according to claim 1, characterized in that: The control system is configured to support multi-subsystem collaborative optimization and establish a multi-subsystem joint energy model among the power distribution system, HVAC system, lighting system and information communication system. The causal relationships between subsystems are determined based on the energy flow matrix, and multiple related abnormal events are fused and analyzed to generate a comprehensive diagnostic conclusion.

Images