Compressor air system management and control system based on Internet of Things smart cloud platform

Abstract

The invention provides a compressor air system management and control system based on an Internet of Things smart cloud platform. The compressor air system management and control system comprises an edge computing unit, the smart cloud platform and an intelligent terminal. And the edge calculation unit collects the operation parameters of the compressor in real time through the Internet of Things sensor, processes the collected data, generates preprocessed data, and executes a high-priority local quick response control task. And the smart cloud platform receives the data uploaded by the edge computing unit, performs fault prediction on the running state of the compressor by using a neural network model, dynamically calculates and optimizes running parameters, and sends an optimization result to the edge computing unit to execute adjustment. The intelligent terminal interacts with the intelligent cloud platform and the edge computing unit to display the real-time operation state and the fault prediction result of the compressor; in an emergency, the intelligent terminal cooperates with the edge computing unit through direct communication to execute localized emergency management. The operation efficiency, the fault response capability and the energy consumption optimization level of the compressor system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of control and regulation systems, and in particular to a control system for a compressor air system based on an Internet of Things smart cloud platform. Background Art

[0002] Compressor air systems are an important part of industrial production and energy, and their operating status directly affects the performance and energy consumption of the overall equipment. With the development of IoT technology, more and more compressor systems are equipped with IoT sensors to collect operating parameters (such as pressure, temperature, vibration, current, etc.) in real time, and use cloud platforms to analyze data to improve operating efficiency and fault management capabilities. These systems usually rely on centralized data processing architectures to achieve basic predictive maintenance and energy efficiency optimization through analysis of historical and real-time data.

[0003] However, the existing technology has the following problems in the management and control of compressor air systems: first, the centralized data processing architecture leads to delays in tasks with high real-time requirements, especially when the network is interrupted or delayed, it is difficult to respond to emergencies; second, the accuracy of fault prediction and the optimization level of operating parameters are limited by simple rules or traditional algorithms, which cannot effectively process complex multi-dimensional dynamic data; in addition, the interaction between users and the system is often limited, lacking flexible control methods and real-time feedback mechanisms, which reduces the intelligence of the system.

[0004] Therefore, it is necessary to develop a new type of compressor air system control system based on the Internet of Things smart cloud platform. Summary of the invention

[0005] The present application provides a compressor air system control system based on the Internet of Things smart cloud platform to improve the operating efficiency, fault response capability and energy consumption optimization level of the compressor system.

[0006] The present application provides a control system for a compressor air system based on an IoT smart cloud platform, including:

[0007] The edge computing unit is used to collect the operating parameters of the compressor in real time through the configured Internet of Things sensors, wherein the operating parameters include pressure, temperature, vibration, and current; perform data denoising, anomaly detection, and pattern recognition on the collected operating parameters to generate pre-processed data; transmit the pre-processed data to the smart cloud platform; and execute high-priority local rapid response control tasks according to the pre-processed data to respond to emergencies;

[0008] A smart cloud platform is used to receive pre-processed data uploaded by the edge computing unit; use a pre-trained neural network model to perform fault prediction on the operating status of the compressor and generate a fault prediction result; based on the pre-processed data and the fault prediction result, dynamically calculate the optimized operating parameters of the compressor, the optimized operating parameters including pressure setting value, working cycle and energy consumption balance strategy; feed back the calculated optimized operating parameters to the edge computing unit for it to perform corresponding operation adjustments;

[0009] The smart terminal interacts with the smart cloud platform and the edge computing unit to display the real-time operating status information of the compressor and the fault prediction results generated by the smart cloud platform. In an emergency, it can perform localized emergency operation management and fault handling by directly communicating with the edge computing unit. It supports users to input control commands through the smart terminal and transmit them to the smart cloud platform or edge computing unit.

[0010] Furthermore, the neural network model used in the smart cloud platform includes an input layer, a spatiotemporal nested coding layer, a multi-scale feature extraction layer, a cyclic prediction layer, and a fault classification layer;

[0011] The input layer is used to receive the multidimensional time series data uploaded by the edge computing unit, and process the received data in segments according to a fixed time window to generate an initial input tensor; wherein the multidimensional time series data includes pre-processed pressure, temperature, vibration and current data;

[0012] The spatiotemporal nested coding layer is used to receive the initial input tensor provided by the input layer, and the spatiotemporal nested coding layer first uses a one-dimensional convolutional neural network to extract local trend features from the initial input tensor; the spatiotemporal nested coding layer then uses an attention mechanism to adjust the weights of the extracted local trend features to generate a spatiotemporal feature tensor;

[0013] The multi-scale feature extraction layer is used to receive the spatiotemporal feature tensor provided by the spatiotemporal nested coding layer, and extract short-term, medium-term and long-term time series features with different convolution kernel sizes through multiple parallel multi-scale convolution units; and combine the extraction results into a unified multi-scale feature tensor through a splicing operation;

[0014] The cyclic prediction layer is used to receive the data provided by the multi-scale feature extraction layer and combine the extraction results into a unified multi-scale feature tensor through a splicing operation; the cyclic prediction layer uses a double-layer long short-term memory network to process the received data and generate a prediction feature vector; wherein the first layer of the double-layer long short-term memory network learns short-term time dependencies, and the second layer learns long-term dependencies; each layer captures the time variation trend of the compressor operation state by dynamically updating the hidden state vector;

[0015] The fault classification layer is used to receive the prediction feature vector provided by the cyclic prediction layer, and use a fully connected neural network to perform linear transformation and nonlinear activation on the prediction feature vector to obtain a predicted probability distribution of the fault type.

[0016] This application has the following beneficial technical effects:

[0017] (1) Through the local processing capability of the edge computing unit, real-time preprocessing and anomaly detection of collected data are achieved, and high-priority rapid response control tasks can be executed in emergency situations, effectively reducing the system response lag caused by network delays or interruptions, and significantly improving the real-time performance and operation efficiency of the system. (2) The pre-trained neural network model on the smart cloud platform is used to predict the operating status, and the operating parameters of the compressor (including pressure setting value, working cycle and energy consumption balance strategy) are dynamically optimized by combining historical data and real-time data, which improves the accuracy of fault prediction and the intelligent level of operating parameter optimization, extends the life of the equipment and reduces energy consumption. (3) The smart terminal realizes the visualization of real-time operating status information and fault prediction results, and supports user input of control instructions. It can directly communicate with the edge computing unit in an emergency and perform local emergency operation management operations, ensuring the high reliability of the system and the flexibility of user control. (4) The layered collaborative architecture formed by the edge computing unit, the smart cloud platform and the smart terminal enables the system to efficiently switch between cloud intelligent analysis, local rapid processing and user interaction, which not only ensures the global optimization capability of complex tasks, but also enhances the local independent operation capability in emergency situations, and improves the robustness and stability of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of a compressor air system control system based on the Internet of Things smart cloud platform provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0019] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0020] The first embodiment of the present application provides a compressor air system control system based on the Internet of Things smart cloud platform. Figure 1 , which is a schematic diagram of the first embodiment of the present application. Figure 1 The first embodiment of the present application provides a compressor air system control system based on the Internet of Things smart cloud platform for detailed description.

[0021] The compressor air system management and control system based on the Internet of Things smart cloud platform includes an edge computing unit 101, a smart cloud platform 102, and a smart terminal 103.

[0022] The edge computing unit 101 is used to collect the operating parameters of the compressor in real time through the configured Internet of Things sensors, and the operating parameters include pressure, temperature, vibration, and current; perform data denoising, anomaly detection, and pattern recognition on the collected operating parameters to generate preprocessed data; transmit the preprocessed data to the smart cloud platform; and execute high-priority local rapid response control tasks based on the preprocessed data to respond to emergencies.

[0023] The edge computing unit 101 is one of the core components of the present invention. Its function is to collect, process and respond to real-time data during the operation of the compressor, provide high-quality data support for the smart cloud platform, and perform rapid local control in emergency situations. The unit is connected to the key components of the compressor through multiple IoT sensors, which are used to collect operating parameters, including but not limited to pressure sensors, temperature sensors, vibration sensors, and current sensors. After the sensor converts the collected analog signal into a digital signal, it is transmitted to the edge computing unit through the data interface. The data interface can be a wireless communication module, an industrial bus interface, or a standard wired connection module to ensure the real-time and accuracy of data transmission.

[0024] The edge computing unit 101 is equipped with a high-performance microprocessor or embedded chip, which can execute a variety of complex algorithms including data denoising, anomaly detection and pattern recognition. The data denoising module processes the collected sensor data based on filtering technology, such as using Kalman filtering or wavelet filtering, to eliminate environmental interference signals. The anomaly detection module determines whether the change of operating parameters exceeds the normal range through a preset threshold value, and establishes a dynamic threshold model based on historical data to further improve the accuracy of detection. The pattern recognition module uses feature extraction and classification technology to classify the operating status of the compressor to identify potential abnormal operating modes or fault characteristics. These processing steps generate preprocessed data to ensure the integrity and applicability of the data.

[0025] After data preprocessing, the edge computing unit 101 uploads the processed data to the smart cloud platform through a wireless communication module or industrial Ethernet. In order to cope with network delays or interruptions, the edge computing unit has a cache management function, which can temporarily store data and upload it in batches after the network is restored to avoid data loss. In addition, in order to further improve the robustness of the system, the unit can automatically switch to the local fast response mode in an emergency and directly execute preset control tasks, such as forcibly reducing operating pressure or stopping equipment operation. The fast response control task is based on a pre-set response rule table and is automatically triggered after logical judgment based on the operating parameters collected in real time.

[0026] The software architecture of the edge computing unit 101 adopts a modular design, mainly including a data acquisition module, a data processing module, a communication module and a rapid response module, and the modules interact with each other through a data bus. The data acquisition module is responsible for receiving raw data from sensors, the data processing module implements denoising, anomaly detection and pattern recognition, the communication module manages two-way communication with the smart cloud platform and smart terminals, and the rapid response module runs first in emergency situations to ensure the security and real-time performance of the system. In addition, the edge computing unit supports the remote update function of the firmware, so that the algorithm can be upgraded or the optimization function can be optimized as needed during operation to maintain long-term applicability and efficiency.

[0027] Through the above functions and architecture design, the edge computing unit 101 can not only realize efficient processing of real-time data, but also operate independently in emergency situations to ensure the reliability and safety of the compressor air system.

[0028] The smart cloud platform 102 is used to receive the pre-processed data uploaded by the edge computing unit; use the pre-trained neural network model to predict the fault of the operating status of the compressor and generate a fault prediction result; based on the pre-processed data and the fault prediction result, dynamically calculate the optimized operating parameters of the compressor, the optimized operating parameters including the pressure setting value, the working cycle and the energy consumption balance strategy; and feed back the calculated optimized operating parameters to the edge computing unit for it to perform corresponding operation adjustments.

[0029] The smart cloud platform 102 is an important component of the present invention for realizing global analysis and optimization control. It is responsible for receiving the pre-processed data uploaded by the edge computing unit, and analyzing, predicting and adjusting the operating status of the compressor based on advanced artificial intelligence algorithms and dynamic optimization technology. The platform consists of a data receiving module, an analysis module, an optimization module and a feedback module, which work together to provide efficient management and control functions.

[0030] During the data reception phase, the smart cloud platform receives pre-processed data uploaded by the edge computing unit through a stable network interface (such as 5G, industrial Ethernet or Wi-Fi). These data include time series of key parameters such as pressure, temperature, vibration, and current of the compressor. The received data is first temporarily stored by the data cache unit to handle possible network fluctuations. At the same time, the platform has a built-in data synchronization mechanism that can sort and merge data uploaded by different edge computing units according to timestamps to ensure the consistency and integrity of global data.

[0031] The analysis module is the core of the smart cloud platform, and is mainly used to perform fault prediction of the compressor operating status based on the received data. This module relies on a pre-trained neural network model to perform a fusion analysis of the compressor's historical data and real-time data. The neural network model is trained through a large data set of compressor operations before the platform is deployed, including common fault types (such as excessive pressure, abnormal temperature or abnormal vibration) and their corresponding operating modes. The platform will continuously learn and optimize the model during operation, dynamically update the model weights through the input of incremental data, and improve the accuracy and real-time performance of the prediction. When the data enters the analysis module, the platform first performs data normalization processing to standardize the range of each parameter to adapt to the input requirements of the neural network; then, the model generates a prediction result of the compressor fault based on the input data, including the fault type, probability of occurrence, and urgency.

[0032] After generating the fault prediction results, the optimization module further calculates the dynamic optimization operating parameters of the compressor based on these results, including pressure setpoints, working cycles, and energy consumption balance strategies. The optimization module uses a multi-objective optimization algorithm to dynamically adjust the operating parameters with the goals of maximizing energy efficiency, minimizing fault risks, and maintaining system stability. Specifically, the optimization module first determines the current operating priority (such as focusing on energy saving or fault avoidance) based on the fault prediction results, and then calculates the optimal combination of operating parameters through built-in decision rules or optimization models. The optimization results are stored in the platform database in real time and are available for query by other modules or terminals.

[0033] The feedback module sends the calculated optimized operating parameters to the edge computing unit through a stable communication link, so that it can perform corresponding operation adjustments. During the feedback process, the platform will ensure that the instructions successfully reach the edge unit through a confirmation mechanism, and resend them when necessary to ensure the accuracy of execution. In addition, the feedback module also supports synchronizing the optimization results to the smart terminal for users to view in real time.

[0034] The smart cloud platform is also highly scalable and compatible. It can be integrated with other industrial management systems through APIs, supports multiple communication protocols (such as MQTT, OPC UA, etc.), and is compatible with compressors of different brands and models. In addition, the platform supports multi-level permission management, provides hierarchical access functions for users of different roles, and provides a secure remote upgrade mechanism to achieve continuous optimization of models and algorithms.

[0035] Through the above functional design, the smart cloud platform 102 realizes global analysis and efficient optimization of the compressor operating status, while providing the system with powerful predictive capabilities and real-time control support, providing technical guarantees for the safety, stability and energy efficiency improvement of the compressor air system.

[0036] Furthermore, the neural network model used in the smart cloud platform includes an input layer, a spatiotemporal nested coding layer, a multi-scale feature extraction layer, a cyclic prediction layer, and a fault classification layer;

[0037] The input layer is used to receive the multidimensional time series data uploaded by the edge computing unit, and process the received data in segments according to a fixed time window to generate an initial input tensor; wherein the multidimensional time series data includes pre-processed pressure, temperature, vibration and current data;

[0038] The spatiotemporal nested coding layer is used to receive the initial input tensor provided by the input layer, and the spatiotemporal nested coding layer first uses a one-dimensional convolutional neural network to extract local trend features from the initial input tensor; the spatiotemporal nested coding layer then uses an attention mechanism to adjust the weights of the extracted local trend features to generate a spatiotemporal feature tensor;

[0039] The multi-scale feature extraction layer is used to receive the spatiotemporal feature tensor provided by the spatiotemporal nested coding layer, and extract short-term, medium-term and long-term time series features with different convolution kernel sizes through multiple parallel multi-scale convolution units; and combine the extraction results into a unified multi-scale feature tensor through a splicing operation;

[0040] The cyclic prediction layer is used to receive the data provided by the multi-scale feature extraction layer and combine the extraction results into a unified multi-scale feature tensor through a splicing operation; the cyclic prediction layer uses a double-layer long short-term memory network to process the received data and generate a prediction feature vector; wherein the first layer of the double-layer long short-term memory network learns short-term time dependencies, and the second layer learns long-term dependencies; each layer captures the time variation trend of the compressor operation state by dynamically updating the hidden state vector;

[0041] The fault classification layer is used to receive the prediction feature vector provided by the cyclic prediction layer, and use a fully connected neural network to perform linear transformation and nonlinear activation on the prediction feature vector to obtain a predicted probability distribution of the fault type.

[0042] The neural network model used in the smart cloud platform consists of multiple functional layers to achieve accurate analysis of the compressor's operating status and fault prediction. The model first receives multi-dimensional time series data uploaded by the edge computing unit through the input layer. These data include pre-processed pressure, temperature, vibration, and current information. The input layer segments the received data according to a fixed time window and converts each segment of data into an initial input tensor with a uniform dimension, providing a standardized input format for subsequent neural network processing.

[0043] After the input layer, the spatiotemporal nested coding layer processes the initial input tensor. This layer first uses a one-dimensional convolutional neural network to extract local trend features from the time series data. One-dimensional convolutional neural networks can capture short-term change patterns in time series, such as rapid fluctuations in compressor pressure or temperature. Next, the spatiotemporal nested coding layer uses the attention mechanism to adjust the weights of the extracted local trend features. The attention mechanism dynamically adjusts the weights of each feature according to its importance, so that information with significant influence at a specific time point or parameter dimension can be more prominently reflected in the spatiotemporal feature tensor. Through this processing, the spatiotemporal feature tensor generated by the spatiotemporal nested coding layer can more accurately reflect the spatiotemporal correlation between the compressor operating parameters.

[0044] The spatiotemporal feature tensor is then passed to the multi-scale feature extraction layer. This layer uses multiple parallel multi-scale convolutional units, each configured with a different convolution kernel size, to extract short-term, medium-term, and long-term time series features respectively. Short-term features capture rapidly changing operating modes, such as high-frequency fluctuations in vibration signals; medium-term features extract the changing trends of operating parameters on medium time scales, such as periodic fluctuations in pressure; and long-term features capture slowly changing operating modes, such as the gradual increase in temperature. The outputs of each convolutional unit are synthesized into a unified multi-scale feature tensor through a splicing operation, providing a comprehensive and multi-dimensional input for the subsequent cyclic prediction layer.

[0045] In the recurrent prediction layer, the multi-scale feature tensor is further processed. This layer consists of a two-layer long short-term memory network (LSTM), where the first layer is used to learn short-term time dependencies and capture rapid changes in compressor operating parameters, such as the changing pattern of vibration signals within seconds; the second layer focuses on learning long-term time dependencies and extracting operating trends over a longer time span, such as slow changes in temperature and current. By dynamically updating the hidden state vector, each layer can effectively integrate the dynamic information in the time series and provide a high-precision time series feature vector for fault prediction.

[0046] Finally, the fault classification layer receives the predicted feature vector generated by the cyclic prediction layer and further processes the feature vector using a fully connected neural network. The fully connected network first reduces or enhances the input features through linear transformation to adapt to the requirements of the classification task, and then enhances the classifier's ability to express complex patterns through nonlinear activation functions. The output of the fault classification layer is the predicted probability distribution of the fault type, such as the probability of occurrence of each fault type (such as abnormal pressure or excessive vibration). Through this probability distribution, the smart cloud platform can provide clear fault prediction results, providing a basis for compressor fault handling and optimized operation.

[0047] Through this layer-by-layer processing, the neural network model can extract the multi-dimensional features of the compressor operating parameters from the initial input tensor, accurately predict the fault type and provide its probability of occurrence. The structural design of this network combines the characteristics of spatiotemporal feature extraction, multi-scale feature fusion and long- and short-term time dependency modeling, providing reliable intelligent analysis and control support for the compressor air system.

[0048] Furthermore, the attention mechanism in the spatiotemporal nested encoding layer is used to adjust the weight relationship between different operating parameters in the initial input tensor, specifically including:

[0049] The operating parameter correlation score is calculated according to the following formula (1):

[0050]

[0051] Among them, e ij represents the attention correlation score between the i-th parameter and the j-th parameter; Z is the local trend feature vector, Where T is the time step, M is the number of operating parameters; α is the control coefficient of the prior weight, which is used to adjust the relative importance of physical characteristics and data-driven results; β is a hyperparameter that adjusts the decay rate of the relationship between parameters; represents the transpose of the vector Z[:,i]; ||·||2 represents the L2 norm of the vector;

[0052] According to the following formula (2), the attention weight is calculated:

[0053]

[0054] Among them, A ij is the attention weight of the i-th parameter to the j-th parameter;

[0055] According to the following formula 3, the spatiotemporal feature tensor is calculated:

[0056]

[0057] Among them, H[t,i] is the spatiotemporal feature tensor at time step t and parameter i; Z[t,j] is the eigenvalue of the jth parameter in the local trend feature vector Z at time step t.

[0058] This embodiment proposes a spatiotemporal nested coding layer based on the attention mechanism, which generates a more accurate spatiotemporal feature tensor by adjusting the weight relationship between operating parameters, thereby improving the dynamic analysis capability of the compressor operating status. The core of the whole scheme is to calculate the attention correlation score and attention weight based on the local trend characteristics in the initial input tensor, combined with the similarity and prior physical relationship between parameters, and finally use these weights to generate the spatiotemporal feature tensor.

[0059] First, the local trend feature vector Z is a matrix of shape T×M, where T represents the time step and M represents the number of operating parameters (such as pressure, temperature, vibration, and current). Z contains the feature information of each parameter at different time steps, and after being extracted by a one-dimensional convolutional network, it has local trend features. The spatiotemporal nested coding layer further optimizes these features through the following steps.

[0060] Formula (1) calculates the correlation score e between operating parameters ij , which is defined as follows:

[0061]

[0062] The correlation score consists of two parts: data-driven similarity measurement and prior physical properties.

[0063] Part I Indicates the cosine similarity of the i-th parameter and the j-th parameter in the local trend characteristics. This value is calculated by vector dot product and indicates the directional similarity between parameters. The value range is [-1,1]. For example, if Z[:,i] and Z[:,j] represent the trend characteristics of pressure and temperature, when the change trends of the two are similar, the score is close to 1.

[0064] Part 2 Physical prior knowledge is introduced to describe the decay behavior of the relationship between parameters with physical distance. For example, if the physical distance between pressure and temperature is close, the decay rate β controls the speed of its weight decay, and α controls the overall amplitude of the prior influence.

[0065] By combining these two parts, e ij It can simultaneously reflect the data-driven dynamic similarity and the inherent physical correlation between parameters.

[0066] Formula (2) is obtained by calculating the correlation score e ij Perform softmax normalization and calculate the attention weight:

[0067]

[0068] Among them, A ij is the normalized weight, which indicates the importance of the i-th parameter to the j-th parameter. The normalization operation of softmax ensures that the sum of the weights is 1, so that different parameters can be effectively weighted. The physical significance of this step is that, through the normalization operation, the relevance score of each parameter is converted into the influence weight on other parameters, highlighting important relationships and suppressing irrelevant features.

[0069] For example, when the correlation score between pressure and vibration is higher than the correlation score between pressure and temperature, softmax will assign higher weights to pressure and vibration, ensuring that vibration contributes more to pressure when generating spatiotemporal features later.

[0070] Formula (3) calculates the spatiotemporal feature tensor by weighted summation:

[0071]

[0072] Among them, H[t,i] represents the spatiotemporal feature value of time step t and parameter u. By using the attention weight A ij The local trend feature vector Z[t,j] is weighted to dynamically integrate the features of different parameters into a spatiotemporal feature with more global characteristics.

[0073] The significance of this step is to use the attention weight A ij To dynamically adjust the contribution of each parameter in generating spatiotemporal features. For example, at a certain time step, if the system operation shows that pressure and temperature have a high correlation, the weight A ij This will ensure that the trend of pressure has a greater impact on the eigenvalues ​​of temperature, thereby generating a space-time characteristic tensor that is more in line with reality.

[0074] Assume that there are 4 operating parameters M=4 (pressure, temperature, vibration, current) and the time step T=3. Assume that at a certain moment, the correlation score between pressure and temperature is high, while the correlation score with vibration is low. The attention mechanism will dynamically adjust these weights so that the pressure characteristics refer more to the trend characteristics of temperature and suppress the influence of vibration. The resulting spatiotemporal feature tensor H[t,i] not only contains the local trend of each parameter, but also incorporates the dynamic adjustment information of other parameters to it, providing accurate input for subsequent fault analysis.

[0075] Through the above steps and formulas, the generation of each spatiotemporal eigenvalue is based on the dynamic characteristics of the operating data and combined with the physical prior relationship, ensuring that the generated spatiotemporal feature tensor has high credibility and accuracy, and can effectively support the intelligent analysis of the compressor operating status.

[0076] Furthermore, the multi-scale convolution unit in the multi-scale feature extraction layer includes three parallel convolution channels, including a short-term convolution channel, a medium-term convolution channel and a long-term convolution channel;

[0077] Among them, the convolution kernel size used by the short-term convolution channel is 3, which is used to capture the local characteristics of rapid fluctuations in the compressor operating parameters; the convolution kernel size used by the medium-term convolution channel is 7, which is used to extract the changing trend of medium frequency; the convolution kernel size used by the long-term convolution channel is 15, which is used to capture the slow changes and periodic characteristics of the compressor operating parameters.

[0078] The multi-scale feature extraction layer is constructed through three parallel convolution channels, namely the short-term convolution channel, the medium-term convolution channel and the long-term convolution channel. These channels use convolution kernels of different sizes to extract the characteristics of the compressor operating parameters at different time scales. The design of the multi-scale convolution unit aims to fully capture the dynamic characteristics of the operating parameters and provide rich information support for subsequent fault analysis and optimization.

[0079] First, the short-term convolution channel is configured with a convolution kernel size of 3. This means that in each convolution operation, the channel will slide the window in units of three consecutive time steps to perform local analysis on the operating parameter data. Due to the small receptive field of the convolution kernel, the short-term convolution channel focuses on extracting local features of rapid fluctuations, such as high-frequency changes in vibration signals or instantaneous current fluctuations. These features are usually closely related to the real-time dynamics of the compressor and are sensitive indicators of the health of the equipment.

[0080] The medium-term convolution channel uses a convolution kernel size of 7. Compared with the short-term convolution channel, its receptive field is larger and can capture the changing trend of operating parameters over a longer time range. For example, pressure and temperature data may show a smooth growth or decline trend on a medium time scale, and these trends are of great significance to the overall operating efficiency and potential risks of the equipment. The intermediate frequency features extracted by the medium-term convolution channel can reflect the stability and continuity of parameter changes.

[0081] The long-term convolution channel has a convolution kernel size of 15 and the largest receptive field, which is used to capture the slow changes and periodic characteristics of operating parameters. For example, the gradual increase in temperature during the operation of the compressor may take a long time to appear, while the periodic fluctuations in vibration characteristics reflect the inherent mechanical characteristics or potential failure modes of the equipment. The capture of these low-frequency features by the long-term convolution channel helps to identify chronic problems of equipment, such as aging of components or reduced energy efficiency.

[0082] In the multi-scale convolution unit, the three channels run in parallel, and each channel independently performs convolution processing on the input operating parameter data. The output of each channel is a feature mapping matrix, the time dimension of the matrix corresponds to the time step of the input data, and the feature dimension reflects the channel's multi-level abstraction of the input features. In order to generate a unified feature tensor, the system integrates the outputs of the three channels through a splicing operation, so that the final multi-scale feature tensor contains short-term, medium-term and long-term feature information at the same time.

[0083] Through such a multi-scale design, the multi-scale feature extraction layer can simultaneously capture the fast dynamics, stable trends and periodic behaviors of the compressor operating parameters, covering the different frequency characteristics of the operating data, and providing comprehensive and accurate feature input for subsequent prediction and classification. This hierarchical convolutional structural design ensures that the system can efficiently extract the multi-dimensional features of the operating parameters, making it highly adaptable and reliable when processing complex industrial data.

[0084] Furthermore, the first layer of the long short-term memory network of the recurrent prediction layer dynamically optimizes the hidden state according to the following formula (4):

[0085]

[0086] Among them, h t represents the hidden state vector at the current time step t, which is used to capture the dynamic characteristics of the compressor operating parameters in the short-term time dependency; t is the output gate, which is calculated by the output gate of the standard LSTM unit and is used to control the information in the hidden state of the current time step that needs to be output to the next step; c t is the unit state, indicating the memory state of the current time step; M is the number of operating parameters; is the convolution feature of the mth parameter at time step t; γ is the weight parameter.

[0087] The first layer of the LSTM network in the recurrent prediction layer achieves efficient modeling of short-term time dependencies through a dynamically optimized hidden state calculation method. The core of this solution is to use formula (4) to generate the hidden state vector of the current time step by combining the weighted fusion of the output gate, unit state and convolution features, thereby capturing the dynamic characteristics of the compressor operating parameters.

[0088] Formula (4) is defined as follows:

[0089]

[0090] In the formula, the hidden state h t It is the core output of the current time step t, which not only carries the system's memory of short-term time dependencies, but also provides input for the next time step.

[0091] The implementation logic of the entire solution is first based on the standard LSTM unit, which processes each time step in the input sequence in turn and combines the convolution features with the specific operating parameters of the current time step. Perform dynamic optimization. According to formula (4), the hidden state h t The generation not only relies on the memory mechanism of LSTM, but also integrates the real-time information from the multi-parameter convolution features, so that the output of each time step is more in line with the actual changes in the operating state of the compressor.

[0092] In the formula, the output gate o t It is calculated by the standard LSTM unit and is used to control which information in the hidden state of the current time step needs to be output to the next step. The calculation of the output gate is usually based on the input data and the hidden state of the previous time step, and is dynamically generated through a set of weight matrices and activation functions. For example, at a certain time step, if the input data indicates that the compressor is currently in a stable state, the output gate may filter out irrelevant short-term fluctuation information to ensure that the information transmitted has actual value.

[0093] Unit status c t It is the memory state of the current time step and is another core component of the LSTM unit, responsible for preserving important information between time steps. The unit state is updated through the combined action of the forget gate and the input gate. For example, if the input data shows that the change trend of a certain parameter of the compressor remains consistent over multiple time steps, the unit state will prioritize retaining this trend information to avoid losing long-term correlation.

[0094] In the formula, the convolution feature It is the feature value generated after performing a one-dimensional convolution operation on each operating parameter. Here M represents the number of operating parameters, such as pressure, temperature, vibration, and current. By performing local convolution processing on each parameter at time step t, local features that change rapidly can be extracted. By averaging the convolution features of M parameters, a mean feature term is obtained. Used to comprehensively reflect the overall dynamics of all parameters at the current time step.

[0095] The weight parameter γ controls the cell state c t and the balance between the convolution feature mean. In some cases, the dynamic characteristics of the compressor may rely more on historical memory (given by c t In this case, the value of γ is closer to 1; in scenarios that require real-time response to rapid changes (such as sudden pressure fluctuations), the value of γ may be lower, making the convolution feature mean contribute more to the hidden state. The weight parameters can be obtained based on experimental data or directly set using expert knowledge.

[0096] Finally, the hidden state ht By applying the tanh function to the above weighted results for nonlinear transformation, and combining the output gate o t The nonlinear transformation can limit the value range of the hidden state and ensure the numerical stability and expressiveness of the model. For example, at a certain time step, if the output gate indicates that a certain feature is of low importance, the contribution of the feature in the hidden state will be suppressed to avoid interfering with the prediction of the next time step.

[0097] For example, assuming that the operating parameters include pressure, temperature and vibration (M = 3), the convolution feature of the current time step t is By taking the average, we get =0.833. If the cell state c t = 0.9 and γ = 0.7, the weighted result is γ·c t +(1-γ)·0.833=0.7·0.9+0.3·0.833=0.882. Combine tanh and output gate o t , hidden state h t It is optimized to a value that better fits the current operating status of the compressor.

[0098] Through the application of the above calculation logic and formulas, the first layer of LSTM can effectively capture the short-term time dependency characteristics of the compressor operating parameters, laying the foundation for the long-term time modeling of the subsequent layers.

[0099] Furthermore, the internal state of the second-layer long short-term memory network of the recurrent prediction layer is propagated between time steps in a recursive manner, and the forgetting mechanism is used to dynamically filter irrelevant information within a too long time span, while retaining the long-term features that affect the current operating state.

[0100] The second layer of the LSTM network in the recurrent prediction layer recursively propagates its internal state between time steps, dynamically filters irrelevant information with the forgetting mechanism, and effectively retains long-term features that have a significant impact on the current operating state. The goal of this design is to solve the problem of long-term dependency modeling of complex time series data during compressor operation, ensure that key patterns in historical data can be captured, and eliminate redundant information to improve the accuracy of fault prediction and parameter optimization.

[0101] The core of the second-layer LSTM lies in the recursive propagation mechanism of its internal state, that is, at each time step, the network uses the hidden state and cell state of the previous time step as input, and combines the input data of the current time step to calculate the new hidden state and cell state. This mechanism enables LSTM to capture the changing trend of data in time series and maintain consistency between multiple time steps. Specifically, the propagation of LSTM's cell state between time steps is achieved through a set of dynamic gating mechanisms, including forget gate, input gate and output gate. Among them, the forget gate is responsible for filtering out historical information that is no longer relevant, the input gate determines which new information needs to be added to the current memory, and the output gate controls which information needs to be output as part of the hidden state.

[0102] The forgetting mechanism is a key component of the second-layer LSTM. It weights the previous time step value of the unit state through a dynamic weight matrix to determine which historical information should be retained and which should be discarded. For example, when the operating parameters of the compressor show periodic fluctuations in a certain period of time in the past, and the input data of the current time step no longer continues this periodic trend, the forgetting mechanism will automatically reduce the weight of information related to the periodicity, thereby gradually filtering it out of the unit state. This process allows the LSTM's memory capacity to always focus on the historical information most relevant to the current operating state, without causing memory redundancy due to too much irrelevant information.

[0103] At the same time, the second-layer LSTM integrates the data of the current time step into the unit state through the input gate, ensuring that the model can dynamically adapt to real-time changes in the operating state. The input gate filters the input data of the current time step and only allows important information to affect the unit state. For example, when the pressure data of the compressor rises slowly over a period of time, the input gate will increase the weight of pressure-related information based on this trend, making it occupy a larger proportion in the unit state. This dynamic weight allocation mechanism enables the second-layer LSTM to flexibly integrate long-term and short-term information, providing support for the modeling of the long-term operating characteristics of the compressor.

[0104] Finally, the cell state generates a new hidden state through nonlinear transformation at each time step. The hidden state contains dynamic memory of historical information and reflects the important features of the current time step. The hidden state, as the output of LSTM, is not only used to guide compressor fault prediction, but also provides necessary input for subsequent processing layers.

[0105] The steps to train the neural network model are as follows:

[0106] First, prepare multidimensional time series data as input, including preprocessed parameters such as pressure, temperature, vibration, and current. Each sample consists of a fixed time window, and the label is the corresponding fault category distribution or optimized parameter value. The data should be divided into training set, validation set, and test set, and normalized to standardize the parameter value range.

[0107] Then, a neural network model is defined, including an input layer, a spatiotemporal nested encoding layer, a multi-scale feature extraction layer, a cyclic prediction layer, and a fault classification layer. The weights of the model can be initialized using a common initialization method to ensure numerical stability.

[0108] Next, select an appropriate loss function, such as the cross entropy loss function for classification tasks, or the mean square error loss function for parameter optimization tasks. At the same time, set the optimizer, such as Adam, and configure a dynamic learning rate scheduler to enhance the training effect.

[0109] During the training process, the data is divided into small batches and input into the model. The output is calculated step by step through forward propagation, and then the loss is calculated with the true label. Finally, the weight is optimized through back propagation. At the end of each training cycle (epoch), the model performance is evaluated using the validation set to monitor whether overfitting occurs. If the performance of the validation set stops improving, you can consider terminating the training early.

[0110] Finally, after completing all training cycles, the final performance of the model is verified on the test set, and deployed and debugged in actual application scenarios. Through this process, the model can effectively capture the dynamic characteristics of compressor operation and provide support for real-time fault prediction and operation parameter optimization.

[0111] Furthermore, the intelligent cloud platform dynamically calculates the compressor optimization operating parameters by minimizing the objective function shown in the following formula (5): Implemented:

[0112]

[0113] in, To optimize the operating parameter vector, including the pressure setting value P set , working cycle T cycle and energy balance strategy S balance ; P represents the fault prediction result; λ1,λ2 are weight parameters; Energy(u) is the energy consumption estimation function of the compressor; FaultRisk(u,P) is the risk cost function based on the fault prediction result;

[0114] The energy consumption estimation function Energy(u) is implemented using the following formula (6):

[0115]

[0116] Among them, η eff is the efficiency factor of the compressor;

[0117] The risk cost function is implemented using the following formula (7):

[0118]

[0119] Where P[c] is the probability of occurrence of fault prediction category c; N is the total number of fault prediction categories; φ c (u) is the failure risk cost function, which is implemented using the following formula (8):

[0120] φ c (u) = α c ·|P set ―P opt,c |+β c ·|T cycle ―T opt,c | (8)

[0121] Among them, P set is the currently optimized pressure setting value; P opt,c is the optimal pressure value associated with fault category c; T cycle The current optimized working cycle; T opt,c is the optimal working cycle associated with fault category c; α c and β c is the sensitivity coefficient.

[0122] The smart cloud platform dynamically optimizes the objective function The objective function comprehensively considers the energy consumption and failure risk of the compressor, and achieves the pressure setting value P by weighing the impact of the two. set , working cycle T cycle Energy consumption balance strategy S balarce optimal settings, thereby improving the overall performance of the compressor system.

[0123] Objective Function The form is as follows:

[0124]

[0125] Among them, the optimization operation parameter vector represents the current compressor operating parameters; λ1 and λ2 are weight parameters used to adjust the relative importance of energy consumption and fault risk; Energy(u) is an energy consumption estimation function that evaluates the operating energy consumption of the compressor; FaultRisk(u,P) is a risk cost function based on the fault prediction results, reflecting the contribution of operating parameters to fault risk.

[0126] The energy consumption estimation function Energy(u) is calculated by the following formula (6):

[0127]

[0128] Here, the efficiency factor of the compressor is eff It is a proportional parameter that reflects the energy conversion efficiency of the compressor; P set is the pressure setting value, which indicates the target pressure maintained by the compressor during the current operating cycle; S balnnce It is an energy consumption balance strategy, which is used to dynamically adjust energy consumption distribution. The integral calculation represents the energy consumption in the whole working cycle T cycle For example, when P set =8bar, S balance =0.9 and η eff =0.85, if T cycle =10s, then the estimated energy consumption of this cycle is:

[0129]

[0130] Energy balance strategy S balance It is represented by a value (usually in the interval [0,1]) to dynamically adjust the compressor's operating energy consumption distribution and energy-saving priority. Specifically, this value reflects the trade-off between the compressor's performance and energy saving:

[0131] S balance =1 means that performance is given full priority. In this case, the compressor will provide the target pressure and operating parameters at the highest energy consumption to ensure that the performance indicators are achieved without considering the need for energy saving.

[0132] S balance = 0 means full priority is given to energy saving. In this case, the compressor may reduce the target pressure setting or reduce the operating cycle to minimize energy consumption, but there may be some impact on performance.

[0133] 0 balance <1 indicates a dynamic balance between performance and energy saving. balance = 0.9, the compressor considers performance with 90% weight and energy saving with 10%. Specifically, this may cause the compressor to slightly reduce the duty cycle frequency or adjust other parameters to reduce energy consumption while maintaining the target pressure.

[0134] The closer the value is to 1, the more emphasis is placed on performance; the closer the value is to 0, the more emphasis is placed on energy saving. For example, when the compressor is under high load or in an emergency, it may increase the S balance ​Set it close to 1 to ensure that the device performance meets the requirements. During light load or non-critical operation, the system can reduce S balance, Pay more attention to optimizing energy consumption to achieve economic operation.

[0135] The risk cost function FaultRisk(u,P) is calculated by the following formula (7):

[0136]

[0137] Here, P[c] is the probability of occurrence of category c in the fault prediction result; φ c (u) is the risk cost function of fault category c, which is used to evaluate the impact of current operating parameters on the specific fault risk.

[0138] N is the total number of fault categories.

[0139] Failure risk cost function φ c (u) is calculated by the following formula (8):

[0140] φ c (u) = α c ·|P set ―P opt,c |+β c ·|T cycle ―T opt,c |

[0141] Here, P opt,c is the optimal pressure setting value for fault category c. For example, to avoid abnormal vibration, the pressure setting value P opt,1 =7.5bar; T opt,c is the optimal duty cycle for fault category c. For example, for overtemperature faults, the optimal duty cycle T opt,2 =12s; α c and β c is the sensitivity factor, controlling the relative impact of pressure and cycle on the risk of failure.

[0142] Through the comprehensive objective function By minimizing calculations, the smart cloud platform can dynamically optimize the operating parameters of the compressor and achieve an effective trade-off between energy consumption and failure risk, thereby ensuring the safety, efficiency and stability of the compressor system.

[0143] The smart terminal 103 interacts with the smart cloud platform and the edge computing unit to display the real-time operating status information of the compressor and the fault prediction results generated by the smart cloud platform; in an emergency, it directly communicates with the edge computing unit to perform localized emergency operation management and fault handling; it supports users to input control commands through the smart terminal and transmit them to the smart cloud platform or the edge computing unit.

[0144] The intelligent terminal 103 is an important interface for the user to interact with the system in the present invention. It has display, communication, command input and emergency management functions, and provides users with means for real-time monitoring of the compressor operation status, fault response and control command transmission. The intelligent terminal is equipped with a high-resolution touch screen display, and its built-in hardware includes a processor, a communication module and an input module, which supports two-way data communication with the smart cloud platform and the edge computing unit.

[0145] The intelligent terminal is connected to the smart cloud platform and edge computing unit wirelessly or wired, and can receive and display the compressor's operating status information and fault prediction results generated by the smart cloud platform in real time. The operating status information includes multi-dimensional operating parameters such as pressure, temperature, vibration and current. This information is presented in a visual form, including line graphs, bar graphs and color-coded status indicators to reflect the dynamic changes and health status of the operating parameters. The fault prediction results are classified and displayed according to the fault type, predicted probability and urgency. Different colors and icons are used to identify the risk level of potential faults to help users quickly understand the current operating status of the system. The display interface supports multi-level view switching. Users can freely switch between the overall monitoring view and the single parameter detailed view to obtain the required system information in different scenarios.

[0146] In an emergency, when the smart cloud platform cannot respond in time, the smart terminal directly establishes a connection with the edge computing unit through the built-in communication module for point-to-point communication. Through this mechanism, the smart terminal can receive real-time data feedback from the edge computing unit and execute localized emergency management processes, such as prompting the user to start the compressor's standby mode, reduce operating pressure, or perform emergency shutdown operations. The instructions for emergency operations are guided by the built-in logical rules of the smart terminal and combined with user input confirmation to ensure the reliability and security of the decision. After executing the emergency processing, the smart terminal will also update the operation log in real time and feedback the current system status, so that users can grasp the latest operating conditions.

[0147] The smart terminal also supports users to pass control commands to the smart cloud platform or edge computing unit through the input interface. The input interface includes virtual buttons, sliders and voice recognition modules on the touch screen. Users can input commands to adjust operating parameters through simple and intuitive operations, such as modifying pressure setting values, adjusting working cycles or enabling energy-saving mode. After receiving the command, the smart terminal will first perform a legality check to ensure that the command is compatible with the current system operating status; then the command will be classified, and emergency commands will be sent directly to the edge computing unit to ensure a quick response, while adjustment commands will be transmitted to the smart cloud platform through the communication module, and the cloud will generate an adjustment plan based on the global optimization strategy and then send it to the edge computing unit.

[0148] In addition, the smart terminal has a user rights management function. Different users can be assigned different access and operation permissions according to their roles, such as only viewing status information, adjusting operating parameters, or performing emergency processing. The smart terminal also integrates local storage and logging functions to record operation history, system alarm information, and fault handling processes, and supports uploading this data to the cloud for further analysis.

[0149] Through the above functional design, the intelligent terminal 103 can not only provide users with comprehensive and intuitive system operation information, but also ensure the real-time and reliability of operations under complex operating environments, thereby realizing efficient interaction and intelligent control between users and the compressor air system.

[0150] Furthermore, the intelligent terminal displays the real-time operating status information of the compressor and the fault prediction results generated by the intelligent cloud platform through the following steps:

[0151] Receive real-time operating status data, including time series data of key parameters such as compressor pressure, temperature, vibration, current, as well as current operating mode and duty cycle;

[0152] The received data is parsed by the processing module of the intelligent terminal to generate dynamic visualization charts, including line charts to display time series data trends, status indicator lights to identify the current operating mode, and bar charts to display real-time changes in parameter values;

[0153] In the fault prediction result display, a graded display is generated according to the fault type, fault probability and urgency, and color coding is used to distinguish different levels of fault risks.

[0154] The smart terminal uses multiple steps to intuitively display the real-time operating status information of the compressor and the fault prediction results generated by the smart cloud platform, so that users can quickly understand the system operation status and identify potential problems in a timely manner. During the implementation process, the smart terminal receives real-time operating status data from the smart cloud platform through its communication module. These data include key parameters of the compressor, such as pressure, temperature, vibration, and time series information of current, as well as the current operating mode and working cycle. The received data is stored in the memory of the smart terminal in a structured form for use by subsequent processing modules.

[0155] The processing module of the intelligent terminal parses and processes the received data to ensure that the data format matches the display requirements of the visualization component. Time series data is used to generate dynamic line charts, which clearly show the changing trends of operating parameters in recent time, helping users to quickly discover fluctuations and anomalies. The current operating mode is indicated by status indicators. Indicators of different colors indicate the working status of the system, such as green for normal operation, yellow for precautions, and red for possible failures or abnormal conditions. At the same time, for the real-time changes in parameter values, the intelligent terminal displays them through bar charts, allowing users to intuitively compare the relative levels between different parameters, such as whether the current temperature is significantly higher than the pressure value within the normal range. The combination of these charts and indicators provides users with multi-level information presentation, which not only meets the overall monitoring needs, but also facilitates detailed analysis of single parameters.

[0156] When displaying fault prediction results, the intelligent terminal displays the prediction results in a graded manner according to fault type, probability of occurrence, and urgency. The fault type clearly indicates the possible problems predicted, such as bearing wear or motor overload. The probability of fault occurrence is intuitively displayed in percentage form to help users assess the possibility of risk. The urgency is presented in color-coded form, using a unified color standard to distinguish the level of fault risk, such as low risk in green, medium risk in yellow, and high risk in red. Through this grading mechanism, users can understand at a glance the potential problems facing the system and the urgent matters that need to be prioritized.

[0157] The layout of the display interface is optimized to ensure a good user experience on different device screen sizes. The interface supports multiple view switching. Users can monitor the trend and status of all parameters in the overall view, or select a single parameter or specific fault result for detailed analysis. To enhance interactivity, users can also directly obtain detailed explanations of relevant data by clicking on charts or indicator lights, such as the logic of fault prediction models, comparison of historical data and current data, etc.

[0158] Through this complete implementation process, the smart terminal can not only display the compressor operating status and fault prediction results in an intuitive, dynamic and hierarchical manner, but also help users quickly judge the system operating status through the integration of multi-dimensional data and clear presentation of charts, reduce fault risks and improve management efficiency.

[0159] Furthermore, the specific implementation of the smart terminal performing localized emergency operation management and fault handling by directly communicating with the edge computing unit in an emergency includes the following steps:

[0160] When a network interruption or cloud command delay is detected, the smart terminal directly establishes a point-to-point communication connection with the edge computing unit through the built-in communication module to ensure the real-time transmission of critical data;

[0161] Receive the compressor's operating status and emergency description data from the edge computing unit, including current pressure, temperature, vibration over-limit information, and the timestamp and type of abnormal events;

[0162] According to the emergency information received, trigger the local preset emergency processing process, including reducing the operating pressure, switching to the standby compressor or forced shutdown;

[0163] After executing the emergency command, the emergency processing results are displayed to the user through the terminal feedback interface, including the start time of the emergency processing, the executed operation steps and the current system status;

[0164] The execution records of localized processing are stored in the log system of the smart terminal and uploaded to the smart cloud platform for global analysis after the network is restored.

[0165] In an emergency, the smart terminal directly communicates with the edge computing unit to implement specific steps for localized emergency operation management and troubleshooting to ensure the safety and reliability of the system when the cloud is unavailable or delayed. First, when the smart terminal detects a network interruption or the cloud command cannot respond in time, its built-in communication module will automatically switch to point-to-point communication mode and establish a direct connection with the edge computing unit. This communication connection can use wireless protocols (such as Wi-Fi Direct) or wired industrial interfaces (such as Modbus) to ensure the real-time and reliability of data transmission. Through this mechanism, the smart terminal can bypass the reliance on the cloud and obtain key data in emergency situations directly from the edge computing unit.

[0166] The data received by the smart terminal from the edge computing unit includes the current operating status of the compressor and the description of the emergency situation. These data are transmitted in a structured form, covering the real-time values ​​of operating parameters such as pressure, temperature, vibration, and clear abnormal event information, such as the specific parameters that exceed the limit, the timestamp that triggers the abnormality, and the description of the abnormality type. The processing module in the smart terminal will parse this data, match it with the built-in preset rule library, and quickly determine what treatment measures need to be taken for the current emergency situation.

[0167] After confirming the emergency, the smart terminal triggers the localized emergency handling process, which is pre-stored in the terminal's control logic, including reducing the operating pressure to reduce the equipment load, switching to the backup compressor to maintain production continuity, or forcing shutdown to protect equipment and environmental safety in extremely high-risk situations. Each emergency handling operation is executed by the smart terminal sending control instructions to the edge computing unit to ensure the real-time and accuracy of the operation.

[0168] After the emergency treatment is completed, the intelligent terminal displays the treatment results to the user through its feedback interface. These results are presented in a clear format, including the start time of the emergency treatment, the specific steps that have been executed, and the current system status (such as the operating pressure has dropped to a safe range or the backup compressor has been enabled). The user can view this feedback information through the touch screen and can also perform further operations as needed.

[0169] At the same time, the smart terminal will store all localized processing execution records in the log system. These records include not only the original data and processing results of abnormal events, but also detailed information about the operation process, such as the sending time and execution confirmation status of each instruction. After the network is restored, the smart terminal will automatically upload these log data to the smart cloud platform for global analysis and further optimization of the system's control strategy.

[0170] Through the above steps, the smart terminal can rely on direct communication with the edge computing unit to respond and handle emergency events in a timely manner in the event of network anomalies or cloud command delays, which not only ensures the operational safety of the compressor air system, but also provides complete data support for subsequent system optimization.

[0171] Furthermore, the smart terminal supports the user to input control instructions, and the specific implementation of transmitting to the smart cloud platform or edge computing unit includes the following steps:

[0172] The user inputs control commands through the input interface of the intelligent terminal, and the input interface includes touch screen menu selection, virtual keyboard input or voice command entry;

[0173] The intelligent terminal parses and verifies the legality of the control commands input by the user, including checking the command format, parameter range, and whether it is compatible with the current operating status;

[0174] If the command is verified, the intelligent terminal selects the target receiving end according to the command type:

[0175] The instructions for adjusting operating parameters are sent to the smart cloud platform through the communication module, and the cloud platform optimizes the instructions globally and then sends them to the edge computing unit;

[0176] Send emergency control instructions directly to the edge computing unit to ensure real-time execution of instructions;

[0177] After the command is executed, the intelligent terminal receives and displays the execution results in real time, including confirmation that the command has been adopted, the update status of the current compressor operating parameters, and whether the execution is successful;

[0178] If the instruction fails to pass the verification, the intelligent terminal generates a prompt message and feeds back the specific error cause to the user.

[0179] Smart terminals support user input control commands through a complete input and processing mechanism, and selectively transmit them to the smart cloud platform or edge computing unit according to the command type to ensure efficient execution and feedback of the commands. During implementation, users can issue control commands through a variety of input interfaces of smart terminals. These interfaces include touch screen menus, where users select operation options by clicking or sliding; virtual keyboards, where users can directly enter specific parameter values; and voice recognition modules, where users input control commands through voice. The input interface of smart terminals supports a variety of interaction methods to adapt to different usage scenarios and user preferences.

[0180] When the user completes the command input, the smart terminal will parse and verify the legality of the input command. The parsing process converts the command into a standardized format that the system can recognize, including the target, operation type, and related parameters of the parsed command. The legality verification step includes checking whether the format of the command is correct, whether the parameter value is within the preset range, and whether the command is compatible with the current operating status of the compressor. For example, if the current system is running in emergency mode, the input adjustment command may be marked as illegal. Through these checks, the smart terminal can filter out invalid commands that may cause system failures.

[0181] For instructions that pass the verification, the smart terminal selects the appropriate target receiver according to its type. If the instruction belongs to the category of adjusting operating parameters, such as modifying the pressure setting value or changing the working cycle, the smart terminal will send the instruction to the smart cloud platform through the communication module. The smart cloud platform performs global optimization processing on the instructions, and after comprehensively considering the overall operating status and historical data of the system, it generates a specific execution plan, and then sends the optimized parameters to the edge computing unit for execution. For emergency control instructions, such as immediate shutdown or switching to a standby compressor, the smart terminal will directly send the instructions to the edge computing unit through the communication module to ensure that the instructions can take effect immediately to respond to emergencies.

[0182] After the command is executed, the smart terminal receives feedback from the edge computing unit or the smart cloud platform through the communication module, and displays the execution results to the user in real time. The display content includes whether the command has been successfully adopted, the latest status of the current compressor operating parameters, and the actual execution effect of the command. For example, if the user modifies the pressure setting value, the smart terminal will display the new pressure value and the stability of the parameters after adjustment in actual operation. If the command fails to pass the verification, the smart terminal will generate a prompt message to clearly inform the user of the specific cause of the error, such as "the parameter is out of the allowable range" or "the command is incompatible with the current operating mode", so that the user can correct the input in time.

[0183] Through the above implementation steps, the intelligent terminal can not only flexibly receive the control instructions input by the user, but also ensure the accurate transmission and efficient execution of the instructions through analysis, verification and target selection, while providing a comprehensive feedback mechanism to help users understand the system status and operation results. This mechanism provides reliable user interaction support for the intelligent operation of the compressor air system.

[0184] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A compressor air system management and control system based on the Internet of Things smart cloud platform, characterized in that: include: The edge computing unit is used to collect the operating parameters of the compressor in real time through the configured IoT sensors, wherein the operating parameters include pressure, temperature, vibration, and current; perform data denoising, anomaly detection, and pattern recognition on the collected operating parameters to generate pre-processed data; Transmit the pre-processed data to the smart cloud platform; According to the preprocessed data; Based on the pre-processed data, high-priority local rapid response control tasks are executed to deal with emergencies; Smart cloud platform, used to receive pre-processed data uploaded by edge computing units; Use a pre-trained neural network model to predict the operating status of the compressor and generate a fault prediction result; Based on the preprocessed data and the fault prediction results, dynamically calculating the optimized operating parameters of the compressor, the optimized operating parameters including the pressure setting value, the working cycle and the energy consumption balance strategy; Feedback the calculated optimized operating parameters to the edge computing unit so that it can make corresponding operating adjustments; Smart terminals interact with the smart cloud platform and edge computing unit to display the real-time operating status information of the compressor and the fault prediction results generated by the smart cloud platform; in emergency situations, they directly communicate with the edge computing unit to perform localized emergency operation management and fault handling; Support users to input control commands through smart terminals and transmit them to the smart cloud platform or edge computing unit.

2. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 1 is characterized in that: The neural network model used in the smart cloud platform includes an input layer, a spatiotemporal nested coding layer, a multi-scale feature extraction layer, a cycle prediction layer, and a fault classification layer; The input layer is used to receive the multidimensional time series data uploaded by the edge computing unit, and process the received data in segments according to a fixed time window to generate an initial input tensor; wherein the multidimensional time series data includes pre-processed pressure, temperature, vibration and current data; The spatiotemporal nested coding layer is used to receive the initial input tensor provided by the input layer, and the spatiotemporal nested coding layer first uses a one-dimensional convolutional neural network to extract local trend features from the initial input tensor; the spatiotemporal nested coding layer then uses an attention mechanism to adjust the weights of the extracted local trend features to generate a spatiotemporal feature tensor; The multi-scale feature extraction layer is used to receive the spatiotemporal feature tensor provided by the spatiotemporal nested coding layer, and extract short-term, medium-term and long-term time series features with different convolution kernel sizes through multiple parallel multi-scale convolution units; and combine the extraction results into a unified multi-scale feature tensor through a splicing operation; The cyclic prediction layer is used to receive the data provided by the multi-scale feature extraction layer and combine the extraction results into a unified multi-scale feature tensor through a splicing operation; the cyclic prediction layer uses a double-layer long short-term memory network to process the received data and generate a prediction feature vector; wherein the first layer of the double-layer long short-term memory network learns short-term time dependencies, and the second layer learns long-term dependencies; each layer captures the time variation trend of the compressor operation state by dynamically updating the hidden state vector; The fault classification layer is used to receive the prediction feature vector provided by the cyclic prediction layer, and use a fully connected neural network to perform linear transformation and nonlinear activation on the prediction feature vector to obtain a predicted probability distribution of the fault type.

3. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 2 is characterized in that: The attention mechanism in the spatiotemporal nested encoding layer is used to adjust the weight relationship between different operating parameters in the initial input tensor, specifically including: The operating parameter correlation score is calculated according to the following formula (1): Among them, e ij represents the attention correlation score between the i-th parameter and the j-th parameter; Z is the local trend feature vector, Where T is the time step, M is the number of operating parameters; α is the control coefficient of the prior weight, which is used to adjust the relative importance of physical characteristics and data-driven results; β is a hyperparameter that adjusts the decay rate of the relationship between parameters; represents the transpose of the vector Z[:,i]; ||·||2 represents the L2 norm of the vector; According to the following formula (2), the attention weight is calculated: Among them, A ij is the attention weight of the i-th parameter to the j-th parameter; According to the following formula 3, the spatiotemporal feature tensor is calculated: Among them, H[t,i] is the spatiotemporal feature tensor at time step t and parameter i; Z[t,j] is the eigenvalue of the jth parameter in the local trend feature vector Z at time step t.

4. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 2 is characterized in that: The multi-scale convolution unit in the multi-scale feature extraction layer includes three parallel convolution channels, including a short-term convolution channel, a medium-term convolution channel and a long-term convolution channel; Among them, the convolution kernel size used by the short-term convolution channel is 3, which is used to capture the local characteristics of rapid fluctuations in the compressor operating parameters; the convolution kernel size used by the medium-term convolution channel is 7, which is used to extract the changing trend of medium frequency; the convolution kernel size used by the long-term convolution channel is 15, which is used to capture the slow changes and periodic characteristics of the compressor operating parameters.

5. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 2 is characterized in that: The first layer of the long short-term memory network of the recurrent prediction layer dynamically optimizes the hidden state according to the following formula (4): Among them, h t represents the hidden state vector at the current time step t, which is used to capture the dynamic characteristics of the compressor operating parameters in the short-term time dependency; t is the output gate, which is calculated by the output gate of the standard LSTM unit and is used to control the information in the hidden state of the current time step that needs to be output to the next step; c t is the unit state, indicating the memory state of the current time step; M is the number of operating parameters; is the convolution feature of the mth parameter at time step t; γ is the weight parameter.

6. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 2 is characterized in that: The internal state of the second-layer long short-term memory network of the cyclic prediction layer is propagated between time steps in a recursive manner, and the forgetting mechanism is used to dynamically filter irrelevant information within a too long time span, while retaining the long-term features that affect the current operating state.

7. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 1 is characterized in that: The intelligent cloud platform dynamically calculates the compressor optimization operating parameters by minimizing the objective function shown in the following formula (5): Implemented: in, To optimize the operating parameter vector, including the pressure setting value P set , working cycle T cycle and energy balance strategy S balance ; P represents the fault prediction result; λ1,λ2 are weight parameters; Energy(u) is the energy consumption estimation function of the compressor; FaultRisk(u,p) is the risk cost function based on the fault prediction result; The energy consumption estimation function Energy(u) is implemented using the following formula (6): Among them, η eff is the efficiency factor of the compressor; The risk cost function is implemented using the following formula (7): Where P[c] is the probability of occurrence of fault prediction category c; N is the total number of fault prediction categories; φ c (u) is the failure risk cost function, which is implemented using the following formula (8): f c (u)=a c ·|p set ―p opt,c |+b c ·|T cycle ―T opt,c | (8) Among them, p set is the currently optimized pressure setting value; p opt,c is the optimal pressure value associated with fault category c; T cycle The current optimized working cycle; T opt,c is the optimal working cycle associated with fault category c; α c and β c is the sensitivity coefficient.

8. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 1 is characterized in that: The smart terminal displays the real-time operating status information of the compressor and the fault prediction results generated by the smart cloud platform through the following steps: Receive real-time operating status data, including time series data of key parameters such as compressor pressure, temperature, vibration, current, as well as current operating mode and duty cycle; The received data is parsed by the processing module of the intelligent terminal to generate dynamic visualization charts, including line charts to display time series data trends, status indicator lights to identify the current operating mode, and bar charts to display real-time changes in parameter values; In the fault prediction result display, a graded display is generated according to the fault type, fault probability and urgency, and color coding is used to distinguish different levels of fault risks.

9. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 1 is characterized in that: The specific implementation of the smart terminal performing localized emergency operation management and fault handling by directly communicating with the edge computing unit in an emergency includes the following steps: When a network interruption or cloud command delay is detected, the smart terminal directly establishes a point-to-point communication connection with the edge computing unit through the built-in communication module to ensure the real-time transmission of critical data; Receive the compressor's operating status and emergency description data from the edge computing unit, including current pressure, temperature, vibration over-limit information, and the timestamp and type of abnormal events; According to the emergency information received, trigger the local preset emergency processing process, including reducing the operating pressure, switching to the standby compressor or forced shutdown; After executing the emergency command, the emergency processing results are displayed to the user through the terminal feedback interface, including the start time of the emergency processing, the executed operation steps and the current system status; The execution records of localized processing are stored in the log system of the smart terminal and uploaded to the smart cloud platform for global analysis after the network is restored.

10. The control system of the compressor air system based on the Internet of Things smart cloud platform according to claim 1 is characterized in that: The specific implementation of the intelligent terminal supporting the user to input control instructions and transmit them to the smart cloud platform or edge computing unit includes the following steps: The user inputs control commands through the input interface of the intelligent terminal, and the input interface includes touch screen menu selection, virtual keyboard input or voice command entry; The intelligent terminal parses and verifies the legality of the control commands input by the user, including checking the command format, parameter range, and whether it is compatible with the current operating status; If the command is verified, the intelligent terminal selects the target receiving end according to the command type: The instructions for adjusting operating parameters are sent to the smart cloud platform through the communication module, and the cloud platform optimizes the instructions globally and then sends them to the edge computing unit; Send emergency control instructions directly to the edge computing unit to ensure real-time execution of instructions; After the command is executed, the intelligent terminal receives and displays the execution results in real time, including confirmation that the command has been adopted, the update status of the current compressor operating parameters, and whether the execution is successful; If the instruction fails to pass the verification, the intelligent terminal generates a prompt message and feeds back the specific error cause to the user.

Images