Oil liquid quality monitoring method and device and Internet of Things system
By constructing a multi -module combination of oil quality detection model, the problem of low oil parameter detection accuracy and low data processing efficiency are solved, and the efficient, accurate detection and rapid quality evaluation of oil parameters are realized.
Patent Information
- Application Number
- CN202411446292.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-16
- Publication Date
- 2025-01-28
AI Technical Summary
The detection accuracy of oil liquid parameters is low, and is affected by environmental factors and sensor accuracy errors, which affects the credibility of oil quality evaluation. It is difficult for the existing technology to efficiently process oil detection data, affecting detection efficiency.
Build the oil quality detection model, including multiple oil parameter processing modules and diagram neural network models. It is combined with Bitcn, Bigru, GAT diagram neural network models and Sarima models to perform efficient processing and prediction of oil parameters.
The efficient and accurate testing of oil parameters has been achieved, the credibility and detection efficiency of the quality evaluation of the oil quality are improved, and the quality level of the oil is quickly and accurately given.
Smart Images
Figure CN119378603A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of oil quality detection, and in particular to an oil quality monitoring method, device and Internet of Things system. Background Art
[0002] Oil quality testing has important guiding significance for oil production and application, and accurate and reliable test results of oil parameters (such as water content, viscosity, etc.) are the basis and support for oil quality evaluation. However, due to the interference of environmental factors and the accuracy error of the sensor itself, the accuracy of oil parameter detection is low, which in turn affects the credibility of oil quality evaluation. In addition, a large amount of data needs to be processed during the oil testing process, but efficient processing of these data has not yet been achieved, which in turn affects the efficiency of oil quality testing. Summary of the invention
[0003] Purpose of the invention: The first purpose of the present invention is to provide an oil quality monitoring method that can realize efficient and accurate detection of oil parameters, and then accurately and timely give oil quality evaluation results; the second purpose of the present invention is to provide an oil quality monitoring device; the third purpose of the present invention is to provide an oil quality monitoring Internet of Things system.
[0004] Technical solution: The oil quality monitoring method of the present invention comprises:
[0005] (1) Constructing oil quality detection model:
[0006] The oil quality detection model includes an oil parameter processing module 1-4, a GAT graph neural network model-WOA optimized Elman neural network model 4-5, a GAT graph neural network model-SARIMA model 3 and a GAT graph neural network model-GWO LightGBM model 2-3;
[0007] The outputs of multiple moisture sensors, multiple viscosity sensors, multiple pH sensors and multiple contamination sensors are respectively used as the input of oil parameter processing modules 1-4, and all of them are used as the input of LightGBM model 3 of GAT graph neural network model-GWO; the output of each oil parameter processing module is respectively used as the corresponding input of Elman neural network model 4 optimized by GAT graph neural network model-WOA, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO LightGBM model 2, GAT graph neural network model-GWO LightGBM model 3, GAT graph neural network model-WOA optimized Elman neural network model 4, GAT graph neural network model-SARIM The outputs of A model 3 and LightGBM model 2 of GAT graph neural network model-GWO are respectively used as the corresponding inputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA, and the outputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA are respectively used as the inputs of LightGBM model 2 of GAT graph neural network model-GWO, LightGBM model 3 of GAT graph neural network model-GWO, Elman neural network model 4 optimized by GAT graph neural network model-WOA and SARIMA model 3 of GAT graph neural network model-GWO; the quality of the detected oil is graded according to the three parameter values output by Elman neural network model 5 optimized by GAT graph neural network model-WOA;
[0008] The oil parameter processing module includes BiTCN neural network model 1-2, BiGRU neural network model 1-2, GAT graph neural network model-WOA optimized Elman neural network model 1-3, GAT graph neural network model-SARIMA model 1-2 and GAT graph neural network model-GWO LightGBM model 1;
[0009] The outputs of multiple parameter sensors are respectively used as the inputs of BiTCN neural network model 1, BiGRU neural network model 1 and LightGBM model 1 of GAT graph neural network model-GWO; the outputs of BiTCN neural network model 1 are respectively used as the inputs of BiTCN neural network model 2, GAT graph neural network model-SARIMA model 1 and Elman neural network model 1 optimized by GAT graph neural network model-WOA; the outputs of BiTCN neural network model 2 are respectively used as the inputs of GAT graph neural network model-SARIMA model 2 and Elman neural network model 1 optimized by GAT graph neural network model-WOA; the outputs of BiGRU neural network model 1 are respectively used as the inputs of BiGRU neural network model 2, GAT graph neural network model-SARIMA model 1 and G The input of the AT graph neural network model-WOA optimized Elman neural network model 2, the output of the BiGRU neural network model 2 is respectively used as the input of the GAT graph neural network model-SARIMA model 2 and the GAT graph neural network model-WOA optimized Elman neural network model 2, the output of the GAT graph neural network model-SARIMA model 1-2 is used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 3, the output of the LightGBM model 1 of the GAT graph neural network model-GWO is respectively used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 1-3, and the outputs of the GAT graph neural network model-WOA optimized Elman neural network model 1-3 are X, Y and Z respectively, constituting three parameter values.<X,Y,Z> , as the predicted values of multiple parameter sensor outputs.
[0010] (2) Use the constructed oil quality detection model to monitor oil quality.
[0011] Furthermore, the oil quality grades are divided into level one, level two, level three, level four and level five.
[0012] Further,
[0013] Serial number Oil quality grade Three parameter values 1 Level 1 <0.7，0.8，0.9> 2 Level 2 <0.6，0.7，0.8 3 Level 3 <0.5，0.6，0.7> 4 Level 4 <0.4，0.5，0.6> 5 Level 5 <0.3，0.4，0.5>
[0014] The oil quality grade is determined according to the above table.
[0015] The oil quality monitoring device described in the present invention includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the oil quality monitoring method described is implemented.
[0016] The oil quality monitoring Internet of Things system described in the present invention is used to implement the oil quality monitoring method described, and includes a perception layer, an oil monitoring cloud platform and an application layer. The perception layer includes multiple oil parameter detection terminals, which are used to measure oil parameters and upload the measured oil parameter information to the oil monitoring cloud platform; the application layer includes an oil monitoring terminal, which can access the oil monitoring cloud platform, and the oil monitoring terminal software functions include communication parameter setting, oil parameter information analysis and processing, and oil quality detection model.
[0017] Furthermore, the oil parameter detection end includes a moisture sensor, a pH value sensor, a pollutant sensor, a temperature sensor, a viscosity sensor, a density sensor, a dielectric constant sensor, an STM32 microprocessor and a communication module. Each sensor is connected to the STM32 microprocessor through a corresponding signal conditioning circuit, and the communication module is connected to the STM32 microprocessor to upload the detected and measured oil parameter information to the oil monitoring cloud platform.
[0018] Furthermore, the communication module adopts a 4G communication module.
[0019] Furthermore, the application layer also includes an oil mobile phone APP monitoring terminal, through which users can monitor oil parameter information in real time.
[0020] Furthermore, the oil monitoring terminal accesses the oil monitoring cloud platform through the 5G network.
[0021] Furthermore, the oil monitoring terminal adopts an industrial control computer.
[0022] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:
[0023] (1) The convolutional neural network model in the TCN neural network model has the characteristics of parallel computing, which can effectively solve the problem of too long time spent on processing the input oil parameter information. It retains the causal expansion convolution and residual connection, learns the dependency of the input oil parameter information through the causal expansion convolution, and uses the residual connection to eliminate the gradient vanishing problem caused by the increase in the number of layers of the TCN neural network model; the singleness of the collected input oil parameter information of the TCN neural network model cannot encode the information from back to front, resulting in the inability to learn the association between the current feature item and the subsequent feature item. The BiTCN neural network model uses the forward input oil parameter information to learn the information dependency from front to back, and the reverse input oil parameter information to learn the information dependency from back to front. By combining the input oil parameter information learned in the two directions, the dependency of the input oil parameter information between the positive and negative propagation directions is obtained.
[0024] (2) The multi-level prediction model of the BiGRU neural network model and the BiTCN neural network model effectively captures the temporal and spatial sequence change data relationship of the oil parameter information and the long-term dependency between the characteristics of the oil parameter information. By improving the synergistic effect of the temporal and spatial fusion of the Elman neural network model, the oil parameter processing module has the dynamic performance of temporal dynamics, spatial dynamics and the temporal and spatial fusion of oil parameters, thereby improving the accuracy and generalization ability of oil parameter detection.
[0025] (3) The BiGRU neural network model can realize the forward and reverse processing of the oil parameter information time series data, and can extract more forward and reverse time series oil parameter information compared with the single GRU neural network model. By using the parallel channel method, it is ensured that the BiGRU neural network model can obtain the forward and reverse information of the forward accumulated oil parameters, and can also obtain the forward and reverse accumulated oil parameter information, and can extract more abundant oil parameter multiple change feature information from the input oil parameter information. After the BiTCN neural network model and the BiGRU neural network model extract the time and space characteristics of the corresponding oil parameters respectively, the feature vector learning operation of the output of the two network structures is carried out by improving the fusion of the Elman neural network model, so that the two branches that perform independent feature learning are integrated, thereby improving the robustness and dynamics of detecting oil parameter information.
[0026] (4) The BiTCN neural network model captures the spatial dependency of the oil parameter time series information, the BiGRU neural network model captures the temporal dependency of the oil parameter time series information, and the SARIMA model captures the oil parameter change trend information. The GAT graph neural network model adaptively learns the relationship weights of different types of data of the output oil parameter information of the BiTCN neural network model, the BiGRU neural network model, and the SARIMA model, mines their spatial and temporal characteristics, and realizes in-depth mining of the relationship between the data oil parameter information; the Elman neural network model optimized by WOA is used to achieve full fusion of the two types of information, thereby improving the accuracy and robustness of oil parameter information extraction and prediction.
[0027] (5) The GAT graph neural network model-GWO's LightGBM model is a series connection of the GAT graph neural network model and the GWO's LightGBM model. The GWO's LightGBM model uses a lightweight gradient boosting machine that constructs a histogram to efficiently process oil parameter information. Through LightGBM model residual fitting, the performance of the weak learner is gradually improved. At the same time, a leaf growth strategy is used to limit the depth of the tree, and iterative training is performed to split only the nodes with the largest gain. The GAT graph neural network model-GWO's LightGBM model utilizes the advantages of the GAT graph neural network model and the GWO's LightGBM model to reduce the error of oil parameter information and improve the prediction accuracy of oil parameter information. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 is a structural schematic diagram of an oil quality detection model in an embodiment of the present invention;
[0029] Figure 2 is a structural schematic diagram of an oil parameter processing module in an embodiment of the present invention;
[0030] Figure 3 This is an architecture diagram of an oil quality monitoring Internet of Things system provided by an embodiment of the present invention;
[0031] Figure 4 is a schematic diagram of the structure of the oil parameter detection end in an embodiment of the present invention;
[0032] Figure 5 Schematic diagram of the oil monitoring terminal software function in an embodiment of the present invention. DETAILED DESCRIPTION
[0033] The present invention will be further described below in conjunction with the accompanying drawings.
[0034] The embodiment of the present invention provides an oil quality monitoring method, comprising the following steps:
[0035] (1) Constructing oil quality detection model:
[0036] like Figure 1 As shown, the oil quality detection model includes oil parameter processing modules 1-4, GAT graph neural network model-WOA optimized Elman neural network model 4-5, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO LightGBM model 2-3;
[0037] The outputs of multiple moisture sensors, multiple viscosity sensors, multiple pH sensors and multiple contamination sensors are respectively used as inputs of oil parameter processing modules 1-4, and all of them are used as inputs of LightGBM model 3 of GAT graph neural network model-GWO; the outputs of each oil parameter processing module are respectively used as corresponding inputs of Elman neural network model 4 optimized by GAT graph neural network model-WOA, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO LightGBM model 2, GAT graph neural network model-GWO LightGBM model 3, GAT graph neural network model-WOA optimized Elman neural network model 4, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO. The outputs of LightGBM model 2 of network model-GWO are used as the corresponding inputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA, and the outputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA are used as the inputs of LightGBM model 2 of GAT graph neural network model-GWO, LightGBM model 3 of GAT graph neural network model-GWO, Elman neural network model 4 optimized by GAT graph neural network model-WOA and SARIMA model 3 of GAT graph neural network model-GWO; According to the three parameter values output by Elman neural network model 5 optimized by GAT graph neural network model-WOA, the detected oil quality is divided into level one, level two, level three, level four and level five, as shown in Table 1.
[0038] Table 1 Correspondence between oil quality grade and three parameters
[0039] Serial number Oil quality grade Three parameter values 1 Level 1 <0.7，0.8，0.9> 2 Level 2 <0.6，0.7，0.8 3 Level 3 <0.5，0.6，0.7> 4 Level 4 <0.4，0.5，0.6> 5 Level 5 <0.3，0.4，0.5>
[0040] The oil quality grade is determined according to Table 1.
[0041] like Figure 2 As shown, the oil parameter processing module includes BiTCN neural network model 1-2, BiGRU neural network model 1-2, GAT graph neural network model-WOA optimized Elman neural network model 1-3, GAT graph neural network model-SARIMA model 1-2 and GAT graph neural network model-GWO LightGBM model 1;
[0042] The outputs of multiple parameter sensors 1-n are respectively used as inputs of BiTCN neural network model 1, BiGRU neural network model 1 and GAT graph neural network model-GWO's LightGBM model 1; the outputs of BiTCN neural network model 1 are respectively used as inputs of BiTCN neural network model 2, GAT graph neural network model-SARIMA model 1 and GAT graph neural network model-WOA optimized Elman neural network model 1; the outputs of BiTCN neural network model 2 are respectively used as inputs of GAT graph neural network model-SARIMA model 2 and GAT graph neural network model-WOA optimized Elman neural network model 1; the outputs of BiGRU neural network model 1 are respectively used as inputs of BiGRU neural network model 2, GAT graph neural network model-SARIMA model 1 and The input of the GAT graph neural network model-WOA optimized Elman neural network model 2, the output of the BiGRU neural network model 2 is used as the input of the GAT graph neural network model-SARIMA model 2 and the GAT graph neural network model-WOA optimized Elman neural network model 2 respectively, the output of the GAT graph neural network model-SARIMA model 1-2 is used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 3, the output of the LightGBM model 1 of the GAT graph neural network model-GWO is used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 1-3 respectively, and the outputs of the GAT graph neural network model-WOA optimized Elman neural network model 1-3 are X, Y and Z respectively, constituting three parameter values.<X,Y,Z> , as the predicted value output by multiple parameter sensors 1-n.
[0043] (2) Use the constructed oil quality detection model to monitor oil quality.
[0044] 1. Design of oil parameter processing module
[0045] The oil parameter processing module includes BiTCN neural network model 1-2, BiGRU neural network model 1-2, GAT graph neural network model-WOA optimized Elman neural network model 1-3, GAT graph neural network model-SARIMA model 1-2 and GAT graph neural network model-GWO LightGBM model 1.
[0046] 1) BiTCN neural network model design
[0047] BiTCN can extract the time series characteristics of oil parameter information from both the forward sequence and the reverse sequence, and fuse them in an additive way to obtain the bidirectional time series characteristics of the fused oil parameter information, which maximizes the deep and global feature information of the oil parameter information set. The TCN neural network model adds causal convolution and dilated convolution on the basis of ordinary one-dimensional convolution, prevents the leakage of oil parameter information from the future to the present, expands the receptive field, and uses residual connection to enhance the generalization of the TCN neural network model. The TCN neural network model has a good performance in the time series prediction problem of oil parameter information, and can process the input oil parameter information in parallel, greatly improving the processing speed of the input oil parameter information. The BiTCN neural network model is composed of multiple residual modules, each of which contains causal dilated convolution. The convolution operation is performed through causal dilated convolution to normalize the weights and normalize the input of the hidden layer, so as to solve the gradient vanishing problem of the TCN neural network model. The addition of Dropout can effectively solve the overfitting problem of the TCN neural network model for oil parameter information. The causal convolution only uses the input oil parameter information before the current moment for calculation, and at the same time restricts it from relying on the input oil parameter information in the future, thus avoiding the leakage of the input oil parameter information and having strict constraints. In order to solve the problem that traditional convolutional neural networks are not good at capturing the long-distance dependency information of the time-series input oil parameter information, the dilated convolution allows spaces between the input oil parameter information during convolution, that is, skipping part of the oil parameter information allows the convolution kernel to obtain a larger receptive field of the oil parameter information, thereby expanding the receptive field of the dilated convolution network.
[0048] 2) BiGRU neural network model design
[0049] The BiGRU neural network model is an improved bidirectional gated recurrent unit recurrent neural network (RNN) structure, which consists of two independent GRU neural network models, one GRU neural network model processes the oil parameter time series information in the forward direction, and the other GRU neural network model processes the oil parameter time series information in the reverse direction. The output is jointly determined by the states of the two GRU neural network models, and the predicted value of the BiGRU neural network model outputs the oil parameter information. At each moment, the input oil parameter information will simultaneously provide two GRU neural network models in opposite directions, and the predicted value is jointly determined by the outputs of these two unidirectional GRU neural network models. This bidirectional structure enables the BiGRU neural network model to capture the long-distance dependencies in the oil parameter time series information, thereby improving the performance of the model. The bidirectional gated recurrent unit (BiGRU) is introduced to capture the long-term dependencies between the oil parameter information features. Through the bidirectional processing method, the BiGRU neural network model can effectively process the time series data of the oil parameter information and better construct the time dependencies between the oil parameter information features, thereby improving the accuracy of the BiGRU neural network model.
[0050] 3) GAT graph neural network model-WOA optimized Elman neural network model design
[0051] The GAT graph neural network model-WOA optimized Elman neural network model is a series connection of the GAT graph neural network model and the WOA optimized Elman neural network model.
[0052] The GAT graph neural network model is a graph neural network model that uses the attention mechanism to learn the interaction between nodes. It introduces the attention mechanism, allowing nodes to give different attention weights to neighboring nodes of oil parameter information, so as to aggregate the oil parameter information of neighboring nodes more flexibly and finely. This flexibility enables the GAT graph neural network model to better capture the complex mutual influence and interaction patterns in oil parameter information. By paying different attention to the neighboring nodes of oil parameter information, the GAT graph neural network model considers the interactive relationship between oil parameter information nodes, thereby improving the accuracy of predicting the interactive relationship of oil parameter information. The GAT graph neural network model uses the multi-source information characteristics of oil parameters as heterogeneous nodes, and sets corresponding connection edges and initial connection weights based on different types of nodes to characterize the relationship between known oil parameter information and missing oil parameter information. Through the multi-level and multi-dimensional prediction of oil parameter information using the BiTCN neural network model, the BiGRU neural network model and the SARIMA model, the complex mapping relationship between the oil parameter information characteristics and the missing data is learned from multiple dimensions; the GAT graph neural network model combines the connection edge weights and aggregation functions to enable each updated node to aggregate the oil parameter feature information of adjacent nodes to form a more abstract and deeper expression of the oil parameter characteristics; finally, feature transformation is performed through the fully connected layer to achieve feature mapping from the high-dimensional features of the oil parameters to the missing data.
[0053] The GAT graph neural network model relies on the attention mechanism and uses the structural information of the oil parameters input by the graph to extract features. The GAT graph neural network model learns the association between the nodes of the input oil parameter information, thereby forming a more abstract and deeper feature expression of the oil parameter information, learning the heterogeneous information and time series information of the input oil parameter information, and extracting the features of the nodes with the feature X. i 'Node, the updated node feature X is obtained by weighted aggregation of neighbor node features i ", the GAT graph neural network model layer uses a fully connected layer to transform the graph features and obtain the final output result y. The calculation formula is as follows:
[0054] y=w D X i "+b D (1)
[0055] Among them, w D is the weight matrix of the fully connected layer, b D is the bias matrix of the fully connected layer.
[0056] The Elamn neural network model is a typical dynamic recursive neural network for predicting oil parameter information. It adds a receiving layer to the hidden layer structure of the feedforward neural network represented by the BP neural network model. Its network model structure is mainly divided into an input layer, a hidden layer, a receiving layer and an output layer. The oil parameter information of the hidden layer at the past moment is fed back to the input of the hidden layer through the delay of the receiving layer, and the learning process information of the oil parameter information at the past moment is memorized, thereby improving the stability, adaptability and robustness of the Elamn neural network model in learning the oil parameter information. Therefore, it can handle the problem of dynamic sequence changes of oil parameter information related to time.
[0057] The Whale Optimization Algorithm (WOA) is a meta-heuristic optimization algorithm based on group simulation of humpback whale hunting behavior. The algorithm achieves the optimization purpose by simulating the hunting process of humpback whale groups in nature, such as searching, encircling, chasing and attacking. Since the Elman neural network model uses the gradient descent method in the optimization process, and the selection of initial values and thresholds is random, the training speed is slow and the prediction accuracy is low. Therefore, it is proposed to use WOA to optimize the initial values and thresholds of the Elman neural network model for training and prediction to make up for the defect of low prediction accuracy of a single model.
[0058] The algorithm flow and steps of WOA optimization Elamn neural network model are as follows:
[0059] ① Initialize the parameters of the Elamn neural network model and build the Elman neural network model;
[0060] ② Initialize the number of WOA whales, the maximum number of iterations, and the upper and lower bounds of the parameters;
[0061] ③ WOA continuously optimizes the learning rate, number of hidden layer nodes, weights and thresholds of the Elman neural network model through the processes of surrounding prey, updating the bubble network and searching for prey;
[0062] ④The Elman neural network model analyzes the optimized best individual, uses the optimized best weights and thresholds for training and predicts the oil parameter information.
[0063] 4) GAT graph neural network model-SARIMA model design
[0064] The GAT graph neural network model-SARIMA model is a GAT graph neural network model and a SARIMA model connected in series. The design method of the GAT graph neural network model refers to the relevant design process of the present invention.
[0065] ARIMA model is one of the most classic models in the analysis of oil parameter information time series. Its basic idea is to use the autocorrelation between oil parameter information time series as the basis, and to combine the autoregression, difference and moving average characteristics of oil parameter time series information to approximate and predict the oil parameter time series information value at that time with a certain mathematical model. When the ARIMA model is not sufficient to extract the periodicity in the oil parameter information, the seasonal difference autoregression moving average model is usually used. When the oil parameter information time series has periodicity, the seasonal difference is considered for the ARIMA model, and the SARIMA model is introduced to detect the oil quality parameters. This model takes into account the seasonal delay of oil parameter changes, the oil change cycle and the influence of seasons on oil quality, which further improves the accuracy of oil detection.
[0066] 5) GAT graph neural network model - GWO's LightGBM model 1 design
[0067] GAT graph neural network model-GWO's LightGBM model 1 is a series connection of the GAT graph neural network model and the LightGBM model.
[0068] The LightGBM model is a lightweight gradient boosting machine that uses a histogram to efficiently process oil parameter information. It gradually improves the performance of the weak learner through residual fitting of the LightGBM model. At the same time, it uses a leaf growth strategy to limit the depth of the tree and conducts iterative training to split only the nodes with the largest gain, thereby reducing errors and improving the prediction accuracy of oil parameter information.
[0069] The LightGBM model of GWO uses the gray wolf optimization algorithm (GWO) to optimize the parameters of the LightGBM model, overcome the local optimal problem of the LightGBM model, and improve the prediction accuracy of the model. The tree depth, number of cotyledons, minimum number of data, and learning rate in the LightGBM model are used as optimization objects, and the parameters of the GWO algorithm are initialized, including the value range of the parameters to be optimized, the initial position of the gray wolf population, the population size, the maximum number of iterations, and the fitness function. The GWO algorithm is used for optimization, and the position of the head wolf α is continuously updated until the iteration termination condition is met. The optimal parameter group is output to the LightGBM model, and the performance of the LightGBM model is comprehensively evaluated using the average relative error, correlation coefficient, root mean square error, average error, and average relative error mean indicators.
[0070] The design process of the GAT graph neural network model refers to the relevant design method of the present invention.
[0071] 2. Design of Oil Quality Detection Module
[0072] The oil quality detection module includes oil parameter processing modules 1-4, GAT graph neural network model-WOA optimized Elman neural network model 4-5, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO LightGBM model 2-3. Their design methods refer to the relevant design steps of the present invention.
[0073] An embodiment of the present invention further provides an oil quality monitoring device, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the oil quality monitoring method described in the embodiment of the present invention is implemented.
[0074] An embodiment of the present invention also provides an oil quality monitoring Internet of Things system, including a perception layer, an oil monitoring cloud platform and an application layer. The perception layer includes oil parameter detection terminals 1-n, which are used to measure oil parameters and upload the measured oil parameter information to the oil monitoring cloud platform. The application layer includes an oil monitoring terminal and an oil mobile phone APP monitoring terminal. The oil monitoring terminal accesses the oil monitoring cloud platform through a 5G network. The software functions of the oil monitoring terminal include real-time monitoring of oil parameters, communication parameter settings, oil parameter information analysis and processing, and oil quality detection models. Users monitor oil parameter information in real time through the oil mobile phone APP monitoring terminal. The structure of the oil quality monitoring Internet of Things system is as follows: Figure 3 shown.
[0075] 1. Design of oil parameter detection terminal
[0076] The oil parameter detection terminal based on the 4G communication module is used as the oil quality parameter information sensing terminal, and the oil parameter information is uploaded to the oil monitoring cloud platform through the 4G communication module. The oil parameter detection terminal includes a moisture sensor, a pH value sensor, a pollutant sensor, a temperature sensor, a viscosity sensor, a density sensor, a dielectric constant sensor, an STM32 microprocessor and a 4G communication module. Each sensor is connected to the STM32 microprocessor through a corresponding signal conditioning circuit, and the 4G communication module is connected to the STM32 microprocessor. The software of the oil parameter detection terminal mainly realizes the 4G communication module drive and the collection and preprocessing of oil quality parameters, which enhances the reliability, readability and portability of the program code. The structure of the oil parameter detection terminal is as follows: Figure 4 shown.
[0077] 2. Oil monitoring software design
[0078] The oil monitoring terminal is an industrial control computer. It mainly collects and processes oil quality parameters. It accesses the oil monitoring IoT cloud platform through the 5G network. The oil monitoring terminal realizes the functions of real-time monitoring of oil parameters, oil parameter information processing and oil quality detection module. The management software uses Microsoft Visual ++ 6.0 as the development tool. The oil monitoring terminal software functions are as follows: Figure 5 shown.
Claims
1. A method for monitoring oil quality, characterized in that: include: (1) Constructing oil quality detection model: The oil quality detection model includes an oil parameter processing module 1-4, a GAT graph neural network model-WOA optimized Elman neural network model 4-5, a GAT graph neural network model-SARIMA model 3 and a GAT graph neural network model-GWO LightGBM model 2-3; The outputs of multiple moisture sensors, multiple viscosity sensors, multiple pH sensors and multiple contamination sensors are respectively used as the input of oil parameter processing modules 1-4, and all of them are used as the input of LightGBM model 3 of GAT graph neural network model-GWO; the output of each oil parameter processing module is respectively used as the corresponding input of Elman neural network model 4 optimized by GAT graph neural network model-WOA, GAT graph neural network model-SARIMA model 3 and GAT graph neural network model-GWO LightGBM model 2, GAT graph neural network model-GWO LightGBM model 3, GAT graph neural network model-WOA optimized Elman neural network model 4, GAT graph neural network model-SARIM The outputs of A model 3 and LightGBM model 2 of GAT graph neural network model-GWO are respectively used as the corresponding inputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA, and the outputs of Elman neural network model 5 optimized by GAT graph neural network model-WOA are respectively used as the inputs of LightGBM model 2 of GAT graph neural network model-GWO, LightGBM model 3 of GAT graph neural network model-GWO, Elman neural network model 4 optimized by GAT graph neural network model-WOA and SARIMA model 3 of GAT graph neural network model-GWO; the quality of the detected oil is graded according to the three parameter values output by Elman neural network model 5 optimized by GAT graph neural network model-WOA; The oil parameter processing module includes BiTCN neural network model 1-2, BiGRU neural network model 1-2, GAT graph neural network model-WOA optimized Elman neural network model 1-3, GAT graph neural network model-SARIMA model 1-2 and GAT graph neural network model-GWO LightGBM model 1; The outputs of multiple parameter sensors are respectively used as the inputs of BiTCN neural network model 1, BiGRU neural network model 1 and LightGBM model 1 of GAT graph neural network model-GWO; the outputs of BiTCN neural network model 1 are respectively used as the inputs of BiTCN neural network model 2, GAT graph neural network model-SARIMA model 1 and Elman neural network model 1 optimized by GAT graph neural network model-WOA; the outputs of BiTCN neural network model 2 are respectively used as the inputs of GAT graph neural network model-SARIMA model 2 and Elman neural network model 1 optimized by GAT graph neural network model-WOA; the outputs of BiGRU neural network model 1 are respectively used as the inputs of BiGRU neural network model 2, GAT graph neural network model-SARIMA model 1 and G The input of the AT graph neural network model-WOA optimized Elman neural network model 2, the output of the BiGRU neural network model 2 is respectively used as the input of the GAT graph neural network model-SARIMA model 2 and the GAT graph neural network model-WOA optimized Elman neural network model 2, the output of the GAT graph neural network model-SARIMA model 1-2 is used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 3, the output of the LightGBM model 1 of the GAT graph neural network model-GWO is respectively used as the input of the GAT graph neural network model-WOA optimized Elman neural network model 1-3, and the outputs of the GAT graph neural network model-WOA optimized Elman neural network model 1-3 are X, Y and Z respectively, constituting three parameter values.<X,Y,Z> , as the predicted values of multiple parameter sensor outputs. (2) Use the constructed oil quality detection model to monitor oil quality.
2. The oil quality monitoring method according to claim 1, characterized in that: The oil quality grades are divided into grade one, grade two, grade three, grade four and grade five.
3. The oil quality monitoring method according to claim 2, characterized in that: The oil quality grade is determined according to the above table.
4. An oil quality monitoring device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the oil quality monitoring method according to any one of claims 1 to 3 is implemented.
5. An oil quality monitoring Internet of Things system, used to implement the oil quality monitoring method according to any one of claims 1 to 3, characterized in that: It includes a perception layer, an oil monitoring cloud platform and an application layer. The perception layer includes multiple oil parameter detection terminals, which are used to measure oil parameters and upload the measured oil parameter information to the oil monitoring cloud platform; the application layer includes an oil monitoring terminal, which can access the oil monitoring cloud platform. The software functions of the oil monitoring terminal include communication parameter setting, oil parameter information analysis and processing, and oil quality detection model.
6. The oil quality monitoring Internet of Things system according to claim 5 is characterized in that: The oil parameter detection end includes a moisture sensor, a pH sensor, a contaminant sensor, a temperature sensor, a viscosity sensor, a density sensor, a dielectric constant sensor, an STM32 microprocessor and a communication module. Each sensor is connected to the STM32 microprocessor through a corresponding signal conditioning circuit, and the communication module is connected to the STM32 microprocessor to upload the detected and measured oil parameter information to the oil monitoring cloud platform.
7. The oil quality monitoring Internet of Things system according to claim 6 is characterized in that: The communication module adopts 4G communication module.
8. The oil quality monitoring Internet of Things system according to claim 5, characterized in that: The application layer also includes an oil mobile phone APP monitoring terminal, through which users can monitor oil parameter information in real time.
9. The oil quality monitoring Internet of Things system according to claim 5, characterized in that: The oil monitoring terminal accesses the oil monitoring cloud platform through the 5G network.
10. The oil quality monitoring Internet of Things system according to claim 5, characterized in that: The oil monitoring end adopts industrial control computer.