An internet of things-based vehicle intelligent analysis system

Abstract

The present application relates to the technical field of vehicle analysis, and especially relates to a vehicle intelligent analysis system based on Internet of Things, a detection device of an air conditioner compressor of a subway vehicle is connected through Internet of Things networking connection of a management platform and energy consumption analysis module, running state data is generated into a local energy consumption feature sequence through a data feature extraction model constructed based on an improved TCN of embedded data causality selection; an energy consumption prediction module generates compressor prediction data through an improved GRU constructed compressor efficiency prediction model based on embedded thermodynamic constraints; a dynamic adjustment module is used to dynamically adjust the model parameters of the data feature extraction model and the compressor efficiency prediction model through a feedback model constructed based on Adam; the compressor prediction data is input into an air conditioner simulation model to generate air conditioner device simulation parameters, and fault analysis and maintenance are carried out according to the air conditioner device simulation parameters. The present application realizes predictive fault analysis and maintenance based on Internet of Things.

Description

Technical Field

[0001] This invention relates to the field of vehicle analysis technology, and in particular to a vehicle intelligent analysis system based on the Internet of Things. Background Technology

[0002] With the continuous development of urban economies, urban traffic has become extremely busy. More and more cities are adopting different rail transit modes, such as subways, light rail, or trams, as strategies and measures to solve urban traffic problems. Most rail transit vehicles are equipped with air conditioning, making the reliability analysis and fault diagnosis strategies for air conditioning increasingly important issues.

[0003] Most rail transit vehicles are equipped with Internet of Things (IoT) systems, which construct a closed-loop management system from perception to decision-making through real-time data acquisition, edge computing, and cloud collaboration. By deploying detection devices in the perception layer of the IoT to detect the vehicle's environment and status, it is possible to perceive various parameters of the vehicle's air conditioning system. Furthermore, by embedding AI chips, such as NVIDIA Jetson Xavier, into the IoT, fault analysis of the air conditioning system can be achieved.

[0004] However, at present, the fault analysis of air conditioning systems is still limited to decision diagnosis after the detection of data anomalies. Most of them adopt the fault tree reliability logic reasoning method, which can be mainly divided into the analytical fault tree reliability logic reasoning method based on minimal cut sets and the fault tree reliability logic reasoning method based on machine learning.

[0005] The analytical fault tree reliability logic reasoning method based on minimal cut sets obtains the minimal cut sets of each major event through Boolean algebra, performs quantitative and qualitative analysis and importance calculation on the basic events, and identifies the weak links in the events based on the importance results.

[0006] The fault tree reliability logic reasoning method based on machine learning uses a data-driven approach to deduce key causal factors from fault results, thereby identifying the weak links in the system.

[0007] Therefore, current predictive fault analysis and maintenance of rail vehicle air conditioning systems before data anomalies are detected is lacking. The application of artificial intelligence technology should be utilized to accurately analyze and predict relevant data from various parts and areas of equipment during subway operation. This would facilitate the discovery of potential anomalies, allow for the development of corresponding solutions in advance, facilitate predictive maintenance by staff, prevent equipment failures, effectively extend system lifespan, and improve subway vehicle operating efficiency.

[0008] In summary, how to combine the Internet of Things (IoT) system of rail vehicles with artificial intelligence technology to perform intelligent analysis of air conditioning compressors in order to achieve predictive fault analysis and maintenance is a technical problem that needs to be solved. Summary of the Invention

[0009] To this end, the present invention provides an IoT-based intelligent vehicle analysis system. By improving TCN, GRU and Adam, it realizes the mining of potential patterns in the IoT detection data of the air conditioning compressor of subway vehicles through the deep learning fusion model of artificial intelligence, determines the compressor prediction data, and then realizes predictive fault analysis and maintenance of subway vehicle air conditioning based on the prediction data.

[0010] To achieve the above objectives, this invention proposes an IoT-based intelligent vehicle analysis system. The management platform and detection device are connected via an IoT network. The detection device is used to collect operating status data and environmental data of the air conditioning compressor in subway vehicles. The management platform includes:

[0011] The energy consumption analysis module is used to generate local energy consumption feature sequences from the operating status data through a data feature extraction model, wherein the data feature extraction model is constructed based on an improved TCN with embedded data causal selection.

[0012] The energy consumption prediction module is used to generate compressor prediction data from the local energy consumption feature sequence through the compressor efficiency prediction model, which is based on an improved GRU with embedded thermodynamic constraints.

[0013] The dynamic adjustment module is used to dynamically adjust the model parameters of the data feature extraction model and the compressor efficiency prediction model through a feedback model, which is built based on Adam.

[0014] The simulation module is used to input the compressor prediction data and the environmental data into the air conditioning simulation model to generate simulation parameters of the air conditioning device. The air conditioning simulation model is built based on AMESim.

[0015] The fault analysis module is used to perform fault analysis and maintenance of the air conditioning system in subway vehicles based on the simulation parameters of the air conditioning components.

[0016] Furthermore, the energy consumption analysis module includes a decomposition unit, a causal selection unit, and a convolution operation unit;

[0017] The decomposition unit is used to generate multiple modes from each of the aforementioned operating state data through modal decomposition.

[0018] The causality judgment unit is used to select causal modes and causal environmental data from multiple modes and multiple environmental data through causality relationship testing.

[0019] The convolutional operation unit is used to extract and fuse features from the causal modal and causal environment data through a temporal convolutional network to generate the local energy consumption feature sequence.

[0020] Furthermore, the causal judgment unit is constructed based on the improved Granger Causality bidirectional causal relationship test, and is used to judge the modal and environmental data with causal relationships. It includes a time alignment subunit, a bidirectional constrained regression subunit, a bidirectional unconstrained regression subunit, a bidirectional causal statistics calculation subunit, and a screening subunit.

[0021] The time alignment subunit is used to align the timestamps of multiple modalities with the environmental data to generate a modal time series.

[0022] The bidirectional constrained regression subunit is used to calculate the first-direction constrained residual and the second-direction constrained residual of the modal time series and the environmental data respectively through the constrained regression model, wherein the constrained regression model also includes historical information, current value, autoregressive coefficient and lag order;

[0023] The bidirectional unconstrained regression subunit is used to calculate the first direction unconstrained residual and the second direction unconstrained residual by the modal time series to environmental data and the environmental data to modal time series, respectively, through the unconstrained regression model. The unconstrained regression model further includes the autoregressive coefficient, the lag order, the causality coefficient, and the lag coefficient.

[0024] The bidirectional causal statistics calculation subunit is used to calculate the first causal statistics and the second causal statistics by using statistical formulas for the first directional constrained residual, the first directional unconstrained residual, the second directional constrained residual, and the second directional unconstrained residual.

[0025] The filtering subunit is used to filter out the causal modality and causal environment data from the modality and the environment data based on the comparison result of the first causal statistic and the second causal statistic with the threshold.

[0026] In the above scheme, by modal decomposition and causal selection of operating status data, modal and environmental data can be better focused on key modalities and environmental factors with precise causal relationships, accurately eliminating interference from irrelevant data, adapting to changes in compressor operating conditions, reducing the prediction error of compressor prediction data (COP) in subsequent GRU predictions, advancing fault warning timeliness, and reducing maintenance costs. Especially in sudden scenarios where faults may occur, this module can reconstruct the feature space in a short time through dynamic causal adjustment, improving short-term prediction accuracy compared to fixed parameter models, and providing key technical support for subway vehicles to cope with peak passenger flow impacts.

[0027] Furthermore, the compressor prediction data includes compressor isentropic efficiency prediction data and mass flow rate prediction data, the operating status data includes intake and exhaust temperatures, isentropic efficiency, electrical parameters, mass flow rate and speed, and the energy consumption prediction module includes an isentropic efficiency constraint unit, an energy constraint unit and a loss function unit.

[0028] The isentropic efficiency constraint unit is used to calculate the maximum isentropic efficiency based on the isentropic efficiency and the intake and exhaust temperatures, and to construct an isentropic efficiency constraint term by using the maximum isentropic efficiency and the predicted isentropic efficiency data through the ReLU activation function.

[0029] The energy constraint unit is used to calculate the electric power based on the electrical parameters and the isentropic efficiency prediction data, calculate the mechanical power based on the mass flow prediction data, the rotational speed and the intake and exhaust temperatures, and construct an energy constraint term based on the isentropic efficiency, the electric power and the mechanical power.

[0030] The loss function unit is used to construct a total loss function based at least on the isentropic efficiency constraint and the energy constraint.

[0031] Furthermore, the operating status data also includes load rate, and the energy consumption prediction module also includes a mass flow continuity constraint unit and a prediction difference calculation unit;

[0032] The mass flow continuity constraint unit is used to calculate the mass flow rate based on the load rate, intake and exhaust temperatures, and rotational speed.

[0033] The difference constraint unit is used to calculate the prediction difference based on the mean square error between the predicted feature sequence and the true value of the input training parameters according to the efficiency prediction.

[0034] The loss function unit is used to construct the mass flow rate constraint term of the total loss function based on the rotational speed feature sequence, and to construct the prediction constraint term of the total loss function based on the prediction difference.

[0035] Furthermore, the compressor efficiency prediction model is configured with a bidirectional GRU layer, an attention layer, and a fully connected layer in sequence;

[0036] The bidirectional GRU layer is used to capture time-series features from the local energy consumption feature sequence and generate a time-series energy consumption feature sequence.

[0037] The attention layer is used to capture temporal features from the temporal energy consumption feature sequence and generate an enhanced temporal feature sequence.

[0038] The fully connected layer embeds the isentropic efficiency constraint unit, energy constraint unit, and load rate constraint unit to physically map the enhanced time-series feature sequence to generate the compressor prediction data.

[0039] Furthermore, the dynamic adjustment module includes a gradient calculation unit and an adjustment unit;

[0040] The gradient calculation unit is used to calculate the training gradients of the data feature extraction model and the compressor efficiency prediction model respectively according to the total loss function;

[0041] The adjustment unit is used to dynamically adjust the training gradient on the data feature extraction model and the compressor efficiency prediction model through the feedback model.

[0042] In the above scheme, thermodynamic constraints, including isentropic efficiency constraints, energy constraints, and continuity constraints, are used to ensure the thermodynamic rationality of the prediction results. Under extreme conditions, such as when environmental parameters are too extreme, the actual data of the air conditioning compressor of the subway vehicle used for model training may be insufficient. In this case, physical constraints serve as prior knowledge to guide the model to generate reasonable efficiency predictions even when data is lacking.

[0043] Furthermore, the air conditioning simulation model includes a compressor, an evaporator, and a condenser; the air conditioning device simulation parameters include control parameters and air conditioning performance indicators; and the fault analysis module includes an emergency adjustment unit for control parameters and a predictive fault analysis unit.

[0044] The emergency adjustment unit for control parameters is used to determine the target adjustment amount of the evaporator and the speed adjustment amount of the condenser in the control parameters in the air conditioning simulation model based on the compressor prediction data.

[0045] The predictive fault analysis unit is used to input the compressor prediction data and environmental data into the air conditioning simulation model to generate air conditioning performance indicators.

[0046] The fault analysis module is used to generate suggested strategies for fault analysis and maintenance of the air conditioning system in subway vehicles based on the air conditioning performance indicators.

[0047] Furthermore, the Internet of Things (IoT) has a data governance standard layer and an edge layer that communicate with each other;

[0048] The edge layer is used to access the operating status data and environmental data detected by the detection device, and to locally deploy the energy consumption analysis module, energy consumption prediction module and dynamic adjustment module;

[0049] The data governance standard layer is used to temporarily store and manage the operational status data locally and the environmental data in the cloud.

[0050] Furthermore, the Internet of Things also has a business middle platform layer that communicates with the data governance standard layer and the edge layer, and the business middle platform layer has a client interface and an emulation interface;

[0051] The business middleware layer is used to push the air conditioning device simulation parameters and fault analysis and maintenance strategies of the fault analysis module via WebSocket;

[0052] The simulation interface is used to deploy multiple simulation modules in a multi-line subway vehicle sandbox via a RESTful API.

[0053] In the above scheme, the AMESim simulation model is combined with the real-time system to build a digital twin platform and automatically generate adjustment suggestions.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. By improving TCN, GRU, and Adam, we have realized the use of deep learning fusion models of artificial intelligence to mine potential patterns in the IoT detection data of air conditioning compressors in subway vehicles, determine the compressor prediction data, and then realize predictive fault analysis and maintenance of subway vehicle air conditioning based on the prediction data.

[0056] 2. By using modal decomposition and causal selection of operational status data, modal and environmental data can be better focused on key modalities and environmental factors with precise causal relationships. This accurately eliminates interference from irrelevant data, adapts to changes in compressor operating conditions, reduces the prediction error of subsequent compressor prediction data (COP) by GRU, advances fault warning timeliness, and reduces maintenance costs. Especially in sudden scenarios where faults may occur, this module can reconstruct the feature space in a short time through dynamic causal adjustment. Compared with fixed parameter models, the short-term prediction accuracy is improved, providing key technical support for subway vehicles to cope with peak passenger flow impacts.

[0057] 3. By using thermodynamic constraints, including isentropic efficiency constraints, energy constraints, and continuity constraints, the reasonableness of the prediction results in thermodynamics is ensured. Under extreme conditions, such as when environmental parameters are too extreme, the actual data of the air conditioning compressor of the subway vehicle used for model training may be insufficient. In this case, physical constraints serve as prior knowledge to guide the model to generate reasonable efficiency predictions even when data is lacking.

[0058] 4. Combine the AMESim simulation model with the real-time system to build a digital twin platform and automatically generate adjustment suggestions. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the structure of the Internet of Things-based intelligent vehicle analysis system according to an embodiment of the present invention;

[0060] Figure 2 This is a flowchart illustrating the Internet of Things-based intelligent vehicle analysis system according to an embodiment of the present invention.

[0061] Figure 3 This is a schematic diagram of the model structure of the Internet of Things-based intelligent vehicle analysis system according to an embodiment of the present invention;

[0062] Figure 4 This is a schematic diagram of the Internet of Things (IoT) architecture of a vehicle intelligent analysis system based on the Internet of Things (IoT) according to an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0064] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0065] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0066] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0067] like Figures 1 to 4 As shown, this invention provides a vehicle intelligent analysis system based on the Internet of Things (IoT). By improving TCN, GRU, and Adam, it enables the mining of potential patterns in the IoT detection data of the air conditioning compressors of subway vehicles through a deep learning fusion model of artificial intelligence, determines the compressor prediction data, and then realizes predictive fault analysis and maintenance of subway vehicle air conditioning based on the prediction data.

[0068] like Figures 1 to 4 As shown, this embodiment proposes a vehicle intelligent analysis system based on the Internet of Things (IoT), including a management platform connected to a detection device for the air conditioning compressor of a subway car via an IoT network. The detection device is used to detect operating status data and environmental data. The management platform includes:

[0069] The energy consumption analysis module is used to generate local energy consumption feature sequences from the operating status data through a data feature extraction model, wherein the data feature extraction model is constructed based on an improved TCN with embedded data causal selection.

[0070] The energy consumption prediction module is used to generate compressor prediction data from the local energy consumption feature sequence through the compressor efficiency prediction model, which is based on an improved GRU with embedded thermodynamic constraints.

[0071] The dynamic adjustment module is used to dynamically adjust the model parameters of the data feature extraction model and the compressor efficiency prediction model through a feedback model, which is built based on Adam.

[0072] The simulation module is used to input the compressor prediction data and environmental data into the air conditioning simulation model to generate simulation parameters for the air conditioning components. The air conditioning simulation model is built based on AMESim.

[0073] The fault analysis module is used to perform predictive fault analysis and maintenance of the air conditioning system in subway vehicles based on the simulation parameters of the air conditioning components.

[0074] It should be noted that the management platform is a network-wide intelligent vehicle analysis platform for subway vehicles. Initial development has been completed for seven lines: Lines 1-8 (Batong Line), 2, 5, 6, 7, 10, and 15. It possesses intelligent vehicle status analysis capabilities and real-time transmission of subway vehicle information. Specifically, the management platform integrates an Internet of Things (IoT) architecture, which includes a front-end application layer (SaaS layer), a business middleware layer (PaaS layer), a data governance standard layer (DaaS layer), a cloud platform infrastructure layer (IaaS layer), and an edge layer. The edge layer achieves data access through the line-level operation and maintenance systems, supporting real-time collection of operational status and environmental data from the air conditioning compressor detection devices in subway vehicles. The data governance standards layer is used to integrate a large amount of structured, semi-structured and unstructured data from operational status data and environmental data. It provides a big data analysis ecosystem for real-time data acquisition, data aggregation and storage, data analysis, and data standards. Corresponding line sandboxes are set up for different lines, which are embedded with AI chips. The AI ​​chips are equipped with energy consumption analysis modules, energy consumption prediction modules and dynamic adjustment modules to generate compressor prediction data based on the Internet of Things and deep learning fusion models. This enables predictive maintenance of compressor and air conditioner failures, improving maintenance efficiency by more than 20%.

[0075] Specifically, the operating status data includes compressor speed, intake and exhaust temperatures, electrical parameters, vibration amplitude, load rate, and time-series data. The environmental data includes the temperature inside the vehicle compartment, the temperature outside the vehicle compartment, and the humidity inside the vehicle compartment.

[0076] Furthermore, the energy consumption analysis module includes a decomposition unit, a causal selection unit, and a convolution operation unit;

[0077] The decomposition unit is used to generate multiple modes from each of the aforementioned operating state data through modal decomposition.

[0078] The causality judgment unit is used to select causal modes and causal environmental data from multiple modes and multiple environmental data through causality relationship testing.

[0079] The convolutional operation unit is used to extract and fuse features from the causal modal and causal environment data through a temporal convolutional network to generate the local energy consumption feature sequence.

[0080] Specifically, such as Figure 3 As shown, the decomposition unit adopts the IVMD (Improved Variational Mode Decomposition) decomposition algorithm, which decomposes the original running state data into multiple modes (IMF). Compared with the traditional VMD, it solves the problems of its sensitivity to noise and fixed number of decomposition layers. The selection of the number of decomposition layers (K) and the penalty factor (α) of the IVMD decomposition algorithm is adaptively selected by the Adam dynamic adjustment module.

[0081] Furthermore, the temporal convolutional network has at least two residual block sub-units, each of which is configured with two sets of residual blocks consisting of an extended causal convolutional layer, a weight normalization layer, a ReLU activation function layer, and a regularization layer.

[0082] Specifically, the filter length of the dilated causal convolutional layer is 5. Model accuracy experiments using the mean squared error (MSE) loss term show that the TCN network with the dilated causal convolutional layer performs better in fitting the output data to the input data compared to the temporal convolutional network with a conventional dilated convolutional layer.

[0083] Therefore, the application of energy consumption analysis, energy consumption prediction, and dynamic adjustment modules can integrate historical data from the air conditioning system to construct a complete health assessment model for the system. This allows for real-time monitoring of the system's operation and lifespan, and also clarifies the maintenance cycle during operation. During operation, regular maintenance of the air conditioning system based on the prediction model prevents energy consumption from escalating due to equipment aging and severe wear, thus ensuring the long-term stable operation of the system.

[0084] Furthermore, the causal judgment unit is constructed based on the improved Granger Causality bidirectional causal relationship test, and is used to judge and focus on modal and environmental data with causal relationships. It includes a time alignment subunit, a bidirectional constrained regression subunit, a bidirectional unconstrained regression subunit, a bidirectional causal statistics calculation subunit, and a screening subunit.

[0085] The time alignment subunit is used to align the timestamps of multiple modalities with the environmental data to generate a modal time series.

[0086] The bidirectional constrained regression subunit is used to calculate the first-direction constrained residual and the second-direction constrained residual of the modal time series and the environmental data respectively through the constrained regression model, wherein the constrained regression model also includes historical information, current value, autoregressive coefficient and lag order;

[0087] The bidirectional unconstrained regression subunit is used to calculate the first direction unconstrained residual and the second direction unconstrained residual by the modal time series to environmental data and the environmental data to modal time series, respectively, through the unconstrained regression model. The unconstrained regression model further includes the autoregressive coefficient, the lag order, the causality coefficient, and the lag coefficient.

[0088] The bidirectional causal statistics calculation subunit is used to calculate the first causal statistics and the second causal statistics by using statistical formulas for the first directional constrained residual, the first directional unconstrained residual, the second directional constrained residual, and the second directional unconstrained residual.

[0089] The filtering subunit is used to filter out the causal modality and causal environment data from the modality and the environment data based on the comparison result of the first causal statistic and the second causal statistic with the threshold.

[0090] Specifically, the constrained regression model is:

[0091]

[0092] In the formula, Y t Y t-1 Representing the current value and the (t-1)th historical information respectively, a i Let ε represent the autoregressive coefficient for the i-th lag order, p represent the lag order, and ε represent the autoregressive coefficient for the ith lag order. t This indicates the first-direction constraint residual or the second-direction constraint residual. That is, when the current value and historical information are taken from modal time series, it is the first-direction constraint residual, and when the current value and historical information are taken from environmental data, it is the second-direction constraint residual.

[0093] Specifically, the unconstrained regression model is:

[0094]

[0095] In the formula, Y t Y t-i X represents the current value and the ti-th historical information, respectively. t-j a represents the ti-th historical information of the corresponding variable. i b j Let p and q represent the autoregressive coefficient of the i-th lag order and the causal coefficient of the j-th lag order, respectively, and let β represent the lag order and lag coefficient, respectively. t This indicates the unconstrained residual in the first direction or the unconstrained residual in the second direction.

[0096] Specifically, the sum of squared residuals is calculated for the first-direction constrained residuals or the second-direction constrained residuals at multiple time points to generate the first residual sum, and the sum of squared residuals is calculated for the first-direction unconstrained residuals or the second-direction unconstrained residuals at multiple time points to generate the second residual sum. The statistical formula is as follows:

[0097]

[0098] In the formula, RSS1 and RSS2 represent the first residual sum and the second residual sum, respectively; T represents the sample time; p and q represent the lag order and lag coefficient, respectively; and F is the first causal statistic and the second causal statistic.

[0099] Specifically, the filtering subunit is used to identify modal and environmental data whose at least one causal statistic, either the first causal statistic from the environmental data to the modal time series direction or the second causal statistic from the modal time series to the environmental data direction, is less than a threshold (preferably 0.15) as causal modal and causal environmental data.

[0100] In the above scheme, by modal decomposition and causal selection of operating status data, modal and environmental data can be better focused on key modalities and environmental factors with precise causal relationships, accurately eliminating interference from irrelevant data, adapting to changes in compressor operating conditions, reducing the prediction error of compressor prediction data (COP) in subsequent GRU predictions, advancing fault warning timeliness, and reducing maintenance costs. Especially in sudden scenarios where faults may occur, this module can reconstruct the feature space in a short time through dynamic causal adjustment, improving short-term prediction accuracy compared to fixed parameter models, and providing key technical support for subway vehicles to cope with peak passenger flow impacts.

[0101] Furthermore, the compressor prediction data includes compressor isentropic efficiency prediction data and mass flow prediction data; the operating status data includes intake and exhaust temperatures, isentropic efficiency, electrical parameters, mass flow rate, and rotational speed; the compressor efficiency prediction model has a ReLU activation function; and the energy consumption prediction module includes an isentropic efficiency constraint unit, an energy constraint unit, and a loss function unit.

[0102] The isentropic efficiency constraint unit is used to calculate the maximum isentropic efficiency based on the isentropic efficiency and the intake and exhaust temperatures, and to construct an isentropic efficiency constraint term by using the maximum isentropic efficiency and the predicted isentropic efficiency data through the ReLU activation function.

[0103] The energy constraint unit is used to calculate the electric power based on the electrical parameters and the isentropic efficiency prediction data, calculate the mechanical power based on the mass flow prediction data, the rotational speed and the intake and exhaust temperatures, and construct an energy constraint term based on the isentropic efficiency, the electric power and the mechanical power.

[0104] The loss function unit is used to construct a total loss function based at least on the isentropic efficiency constraint and the energy constraint.

[0105] Specifically, the constrained isentropic efficiency has a theoretical maximum value. If the efficiency value predicted by the model exceeds this upper limit, the loss function is forcibly corrected through a penalty term activated by ReLU to avoid generating results that violate physical laws. The calculation process for the maximum isentropic efficiency is as follows:

[0106]

[0107] In the formula, η max T represents the maximum isentropic efficiency. s T0 and T0 represent the intake temperature and exhaust temperature, respectively, and η mech This indicates isentropic efficiency.

[0108] Specifically, the calculation process for electrical power and mechanical power is as follows:

[0109]

[0110] P mech =m p (h0-h s )

[0111] In the formula, P in P mech Let U and I represent electrical power and mechanical power, respectively. UIcosφ represents electrical parameters, where U represents line voltage (a fixed value of 380V), I is the compressor operating current from input data, cosφ is the power factor (an empirical value of 0.82), and η... p For isentropic efficiency prediction data, m pFor mass flow prediction data, h s h0 and h0 represent the exhaust enthalpy and intake enthalpy, respectively, which are determined by the intake and exhaust temperatures and speeds through the refrigerant property table.

[0112] Specifically, as a particular implementation, the total loss function is:

[0113]

[0114] In the formula, Loss represents the total loss function, λ1 and λ2 represent the first weight and the second weight, respectively, and L mse 、(RELU(0,η p -η max )) Represent the predicted numerical MSE loss term, isentropic efficiency constraint term, and energy constraint term, respectively, where η max η represents the maximum isentropic efficiency. p For isentropic efficiency prediction data, ReLU represents the ReLU activation function, P in P mech Let represent electrical power and mechanical power, respectively. Wherein, when η p -η max When the value is greater than 0, the derivative of ReLU is 1, indicating the existence of an isentropic efficiency constraint term; otherwise, no isentropic efficiency constraint term exists. Therefore, in the early stages of model training, the predicted mechanical efficiency frequently exceeds the limit, and the penalty term can cause the predicted value to decrease more quickly, accelerating the model fitting speed.

[0115] Furthermore, the operating status data also includes load rate, and the energy consumption prediction module also includes a mass flow continuity constraint unit and a prediction difference calculation unit;

[0116] The mass flow continuity constraint unit is used to calculate the mass flow rate based on the load rate, intake and exhaust temperatures, and rotational speed.

[0117] The difference constraint unit is used to calculate the prediction difference based on the mean square error between the predicted feature sequence and the true value of the input training parameters according to the efficiency prediction.

[0118] The loss function unit is used to construct the mass flow rate constraint term of the total loss function based on the rotational speed feature sequence, and to construct the prediction constraint term of the total loss function based on the prediction difference.

[0119] Specifically, the calculation process for mass flow rate is as follows:

[0120]

[0121] In the formula, m0 is the mass flow rate, f is the load factor, and V s T is a fixed parameter for compressor displacement. sT0 and T0 represent the intake temperature and exhaust temperature, respectively, and n is the rotational speed.

[0122] Specifically, as another specific implementation, the total loss function is:

[0123]

[0124] In the formula, Loss represents the total loss function, λ1, λ2, and λ3 represent the first weight, second weight, and third weight, respectively, and L mse ,λ1(RELU(0,η p -η max )) Let η represent the prediction constraint, isentropic efficiency constraint, energy constraint, and mass flow rate constraint, respectively. max η represents the maximum isentropic efficiency. p For isentropic efficiency prediction data, ReLU represents the ReLU activation function, P in P mech These represent electrical power and mechanical power, respectively, with m0 being the mass flow rate and m p This is for mass flow prediction data.

[0125] Furthermore, the compressor efficiency prediction model is configured with a bidirectional GRU layer, an attention layer, and a fully connected layer in sequence;

[0126] The bidirectional GRU layer is used to capture time-series features from the local energy consumption feature sequence and generate a time-series energy consumption feature sequence.

[0127] The attention layer is used to capture temporal features from the temporal energy consumption feature sequence and generate an enhanced temporal feature sequence.

[0128] The fully connected layer embeds the isentropic efficiency constraint unit, energy constraint unit, and load rate constraint unit to physically map the enhanced time-series feature sequence to generate the compressor prediction data.

[0129] Furthermore, the dynamic adjustment module includes a gradient calculation unit and an adjustment unit;

[0130] The gradient calculation unit is used to calculate the training gradients of the data feature extraction model and the compressor efficiency prediction model respectively according to the total loss function;

[0131] The adjustment unit is used to dynamically adjust the training gradient on the data feature extraction model and the compressor efficiency prediction model through the feedback model.

[0132] Specifically, the training gradient is defined as:

[0133]

[0134] In the formula, For training gradients, Loss represents the total loss function, and θ represents the training parameters of the data feature extraction model and the compressor efficiency prediction model. The training parameters include convolution kernel weights, GRU weights, attention weights, etc.

[0135] In the above scheme, thermodynamic constraints, including isentropic efficiency constraints, energy constraints, and continuity constraints, are used to ensure the thermodynamic rationality of the prediction results. Under extreme conditions, such as when environmental parameters are too extreme, the actual data of the air conditioning compressor of the subway vehicle used for model training may be insufficient. In this case, physical constraints serve as prior knowledge to guide the model to generate reasonable efficiency predictions even when data is lacking.

[0136] Furthermore, the air conditioning simulation model includes a compressor, an evaporator, and a condenser; the air conditioning device simulation parameters include control parameters and air conditioning performance indicators; and the fault analysis module includes an emergency adjustment unit for control parameters and a predictive fault analysis unit.

[0137] The emergency adjustment unit for control parameters is used to determine the target adjustment amount of the evaporator and the speed adjustment amount of the condenser in the control parameters in the air conditioning simulation model based on the compressor prediction data.

[0138] The predictive fault analysis unit is used to input the compressor prediction data and environmental data into the air conditioning simulation model to generate air conditioning performance indicators.

[0139] The fault analysis module is used to generate suggested strategies for fault analysis and maintenance of the air conditioning system in subway vehicles based on the air conditioning performance indicators.

[0140] For example, during the evening rush hour, if the subway train predicts that the compressor's isentropic efficiency will decrease by 10%, the target adjustment amount of the evaporator in the emergency adjustment unit will be lowered to 9°C and the speed adjustment amount will be adjusted to 1700 RPM to limit the compressor speed to ≤90%, thus generating a bearing wear warning.

[0141] Furthermore, such as Figure 4 As shown, the Internet of Things (IoT) has a data governance standard layer and an edge layer that communicate with each other;

[0142] The edge layer is used to access the operating status data and environmental data detected by the detection device, and to locally deploy the energy consumption analysis module, energy consumption prediction module and dynamic adjustment module;

[0143] The data governance standard layer is used to temporarily store and manage the operational status data locally and the environmental data in the cloud.

[0144] Therefore, considering the differences in characteristics between high-frequency operational status data and low-frequency environmental data, a layered processing strategy is adopted to ensure efficient data storage, processing, and distribution. Specifically, local temporary storage management is achieved by deploying NVMe SSDs on edge nodes in the edge layer, while cloud-based temporary storage management is achieved by building a cloud-based relational database, Amazon Aurora, and a mapping table between storage device static attributes and environmental parameters.

[0145] Furthermore, such as Figure 4 As shown, the Internet of Things also has a business middle platform layer that communicates with the data governance standard layer and the edge layer. The business middle platform layer has a client interface and a simulation interface.

[0146] The business middleware layer is used to push the air conditioning device simulation parameters and fault analysis and maintenance strategies of the fault analysis module via WebSocket;

[0147] The simulation interface is used to deploy multiple simulation modules in a multi-line subway vehicle sandbox via a RESTful API.

[0148] Therefore, the visualization and interactive function of the simulation parameters of air conditioning devices in the Internet of Things is realized, and the prediction model, simulation model and Internet of Things structure are embedded into the existing subway vehicle management system.

[0149] In the above scheme, the AMESim simulation model is combined with the real-time system to build a digital twin platform and automatically generate adjustment suggestions.

[0150] In this embodiment, by improving TCN, GRU, and Adam, the IoT detection data of the air conditioning compressors in subway vehicles is fused using a deep learning model of artificial intelligence to mine potential patterns in the data, determine compressor prediction data, and thus achieve predictive fault analysis and maintenance of subway vehicle air conditioning based on the prediction data. Through modal decomposition and causal selection of operating status data, modal and environmental data can be better focused on key modalities and environmental factors with precise causal relationships, accurately eliminating interference from irrelevant data, adapting to changes in compressor operating conditions, reducing the prediction error of subsequent compressor prediction data (COP) by GRU, advancing fault warning timeliness, and reducing maintenance costs. Especially in sudden scenarios where faults may occur, this module can reconstruct the feature space in a short time through dynamic causal adjustment, improving short-term prediction accuracy compared to fixed parameter models, and providing key technical support for subway vehicles to cope with peak passenger flow impacts. Thermodynamic constraints, including isentropic efficiency constraints, energy constraints, and continuity constraints, ensure the thermodynamic rationality of the prediction results. In extreme conditions, such as when environmental parameters are too extreme, the actual data available for model training of the subway vehicle's air conditioning compressor may be insufficient. In such cases, physical constraints serve as prior knowledge, guiding the model to generate reasonable efficiency predictions even with limited data. The AMESim simulation model is combined with the real-time system to construct a digital twin platform, automatically generating adjustment suggestions.

[0151] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0152] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An Internet of Things-based vehicle intelligent analysis system, characterized in that, The management platform and the detection device are connected via an Internet of Things (IoT) network. The detection device is used to collect operating status data and environmental data of the air conditioning compressors in subway vehicles. The management platform includes: The energy consumption analysis module is used to generate local energy consumption feature sequences from the operating status data through a data feature extraction model, wherein the data feature extraction model is constructed based on an improved TCN with embedded data causal selection. The energy consumption prediction module is used to generate compressor prediction data from the local energy consumption feature sequence through the compressor efficiency prediction model, which is based on an improved GRU with embedded thermodynamic constraints. The dynamic adjustment module is used to dynamically adjust the model parameters of the data feature extraction model and the compressor efficiency prediction model through a feedback model, which is built based on Adam. The simulation module is used to input the compressor prediction data and the environmental data into the air conditioning simulation model to generate simulation parameters of the air conditioning device. The air conditioning simulation model is built based on AMESim. The fault analysis module is used to perform fault analysis and maintenance of the air conditioning system in subway vehicles based on the simulation parameters of the air conditioning components. The compressor prediction data includes compressor isentropic efficiency prediction data and mass flow rate prediction data. The operating status data includes intake and exhaust temperatures, isentropic efficiency, electrical parameters, mass flow rate, speed, and load rate. The dynamic adjustment module includes a gradient calculation unit and an adjustment unit. The gradient calculation unit is used to calculate the training gradients of the data feature extraction model and the compressor efficiency prediction model respectively based on the total loss function; The adjustment unit is used to dynamically adjust the training gradient on the data feature extraction model and the compressor efficiency prediction model through the feedback model. The energy consumption prediction module includes: an isentropic efficiency constraint unit, used to calculate the maximum isentropic efficiency based on the isentropic efficiency and the intake and exhaust temperatures, and to construct an isentropic efficiency constraint term using the maximum isentropic efficiency and the predicted isentropic efficiency data through a ReLU activation function; an energy constraint unit, used to calculate electrical power based on the electrical parameters and the predicted isentropic efficiency data, calculate mechanical power based on the predicted mass flow rate data, the rotational speed and the intake and exhaust temperatures, and construct an energy constraint term based on the isentropic efficiency, electrical power and mechanical power; and a loss function unit, used to construct a total loss function based at least on the isentropic efficiency constraint term and the energy constraint term. The total loss function is: ； or, ； In the formula, Represents the total loss function. These represent the first weight, the second weight, and the third weight, respectively. These represent the prediction numerical MSE loss term, the isentropic efficiency constraint term, and the energy constraint term, respectively. This represents the maximum isentropic efficiency. For isentropic efficiency prediction data, ReLU represents the ReLU activation function. These represent electrical power and mechanical power, respectively. For quality flow, For mass flow prediction data; The calculation process for the electrical power and mechanical power is as follows: ； In the formula, These represent electrical power and mechanical power, respectively. These represent electrical parameters, where U represents line voltage and I represents compressor operating current. For power factor, For isentropic efficiency prediction data, For mass flow prediction data, These represent the exhaust enthalpy and intake enthalpy, respectively, which are determined based on the exhaust and intake temperatures and rotational speed using a refrigerant property table. The calculation process for the maximum isentropic efficiency is as follows: ； In the formula, This represents the maximum isentropic efficiency. These represent the intake temperature and exhaust temperature, respectively. This indicates isentropic efficiency. 2.The Internet of Things based intelligent analysis system for vehicles according to claim 1, wherein, The energy consumption analysis module includes a decomposition unit, a causal selection unit, and a convolution operation unit. The decomposition unit is used to generate multiple modes from each of the aforementioned operating state data through modal decomposition. The causal selection unit is used to select causal modes and causal environmental data from multiple modes and environmental data through causal relationship testing; The convolutional operation unit is used to extract and fuse features from the causal modal and causal environment data through a temporal convolutional network to generate the local energy consumption feature sequence.

3. The vehicle intelligent analysis system based on the Internet of Things according to claim 2, characterized in that, The causal selection unit, constructed based on the improved Granger Causality bidirectional causality test, is used to determine modal and environmental data with causal relationships. It includes a time alignment subunit, a bidirectional constrained regression subunit, a bidirectional unconstrained regression subunit, a bidirectional causal statistics calculation subunit, and a screening subunit. The time alignment subunit is used to align the timestamps of multiple modalities with the environmental data to generate a modal time series. The bidirectional constrained regression subunit is used to calculate the first-direction constrained residual and the second-direction constrained residual of the modal time series and the environmental data respectively through the constrained regression model, wherein the constrained regression model also includes historical information, current value, autoregressive coefficient and lag order; The bidirectional unconstrained regression subunit is used to calculate the first direction unconstrained residual and the second direction unconstrained residual by the modal time series to environmental data and the environmental data to modal time series, respectively, through the unconstrained regression model. The unconstrained regression model further includes the autoregressive coefficient, the lag order, the causality coefficient, and the lag coefficient. The bidirectional causal statistics calculation subunit is used to calculate the first causal statistics and the second causal statistics by using statistical formulas for the first directional constrained residual, the first directional unconstrained residual, the second directional constrained residual, and the second directional unconstrained residual. The filtering subunit is used to filter out the causal modality and causal environment data from the modality and the environment data based on the comparison result of the first causal statistic and the second causal statistic with the threshold. 4.The Internet-of-Things based intelligent analysis system for vehicles according to claim 1, wherein, The compressor efficiency prediction model is configured with a bidirectional GRU layer, an attention layer, and a fully connected layer in sequence. The bidirectional GRU layer is used to capture time-series features from the local energy consumption feature sequence and generate a time-series energy consumption feature sequence. The attention layer is used to capture temporal features from the temporal energy consumption feature sequence and generate an enhanced temporal feature sequence. The fully connected layer embeds the isentropic efficiency constraint unit, energy constraint unit, and load rate constraint unit to physically map the enhanced time-series feature sequence to generate the compressor prediction data. 5.The Internet-of-Things based intelligent analysis system for vehicles according to claim 1, wherein, The air conditioning simulation model includes a compressor, an evaporator, and a condenser. The simulation parameters of the air conditioning components include control parameters and air conditioning performance indicators. The fault analysis module includes an emergency adjustment unit for control parameters and a predictive fault analysis unit. The emergency adjustment unit for control parameters is used to determine the target adjustment amount of the evaporator and the speed adjustment amount of the condenser in the control parameters in the air conditioning simulation model based on the compressor prediction data. The predictive fault analysis unit is used to input the compressor prediction data and environmental data into the air conditioning simulation model to generate air conditioning performance indicators. The fault analysis module is used to generate suggested strategies for fault analysis and maintenance of the air conditioning system in subway vehicles based on the air conditioning performance indicators. 6.The Internet of Things based intelligent analysis system for vehicles according to any one of claims 1 to 5, characterized in that, The Internet of Things (IoT) has a data governance standard layer and an edge layer that communicate with each other; The edge layer is used to access the operating status data and environmental data detected by the detection device, and to locally deploy the energy consumption analysis module, energy consumption prediction module and dynamic adjustment module; The data governance standard layer is used to temporarily store and manage the operational status data locally and the environmental data in the cloud. 7.The Internet of Things based intelligent analysis system for vehicles according to claim 6, wherein, The Internet of Things also has a business middle platform layer that communicates with the data governance standard layer and the edge layer. The business middle platform layer has a client interface and a simulation interface. The business middleware layer is used to push the air conditioning device simulation parameters and fault analysis and maintenance strategies of the fault analysis module via WebSocket; The simulation interface is used to deploy multiple simulation modules in a multi-line subway vehicle sandbox via a RESTful API.

Images