An intelligent maintenance method and system for a natural gas odorization device

Abstract

The present invention discloses an intelligent maintenance method and system for a natural gas odorizing device, the method comprising the following steps: obtaining the current operating parameters of the natural gas odorizing device from multiple sensors in real time; optimizing the current operating parameters to obtain the current operating status evaluation result of the natural gas odorizing device; based on a pre-established predictive maintenance model of the natural gas odorizing device, taking the current operating status evaluation result of the natural gas odorizing device as input, outputting the potential fault prediction result of the natural gas odorizing device; when the potential fault prediction result indicates the existence of a potential fault, generating maintenance suggestions for the potential fault according to the potential fault prediction result and the current operating status evaluation result. The present invention realizes the intelligentization and efficiency of the maintenance of the natural gas odorizing device by adopting technologies such as multi-sensor real-time data acquisition, optimization processing, predictive maintenance model and intelligent decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of maintenance and processing of a natural gas odorizing device, and in particular to an intelligent maintenance method and system for a natural gas odorizing device. Background Art

[0002] Natural gas odorization is a key link to ensure the safe use of natural gas, and the reliable operation of the odorization device is directly related to the safety of the entire natural gas supply system. The traditional maintenance method of natural gas odorization devices mainly relies on regular inspections and manual experience judgment, which has many shortcomings: the maintenance cycle is fixed and it is difficult to flexibly adjust according to the actual operating conditions, which may lead to untimely maintenance or excessive maintenance; manual judgment is easily affected by subjective factors and it is difficult to accurately predict potential failures; maintenance efficiency is low and it is difficult to achieve the optimal operating state of the equipment; it is impossible to fully utilize the equipment operation data and lacks data-driven scientific decision-making support.

[0003] In addition, with the continuous increase in natural gas usage and the continuous development of odorization device technology, the equipment operating environment has become increasingly complex, and traditional maintenance methods can no longer meet the maintenance needs of modern natural gas odorization devices. Some studies have attempted to introduce data analysis and artificial intelligence technology to improve maintenance methods, but there is still room for improvement in real-time, accuracy, and intelligence.

[0004] Therefore, there is an urgent need for a natural gas odorization device maintenance method that can perform real-time monitoring, accurate prediction, and intelligent decision-making to improve the operating efficiency and safety of the equipment, while reducing maintenance costs and extending the service life of the equipment. Summary of the invention

[0005] In view of this, an object of the present invention is to provide an intelligent maintenance method and system for a natural gas odorization device to solve the above technical problems.

[0006] To achieve the above objectives, in a first aspect, a smart maintenance method for a natural gas odorization device is provided, which comprises the following steps:

[0007] Obtain the current operating parameters of the natural gas odorization unit from multiple sensors in real time;

[0008] Optimizing the current operating parameters to obtain a current operating status evaluation result of the natural gas odorization device;

[0009] Based on a pre-established predictive maintenance model of a natural gas odorizing device, taking a current operating status evaluation result of the natural gas odorizing device as input, outputting a potential failure prediction result of the natural gas odorizing device;

[0010] When the potential fault prediction result indicates the existence of a potential fault, a maintenance suggestion for the potential fault is generated according to the potential fault prediction result and the current operating state evaluation result.

[0011] In a second aspect, an intelligent maintenance system for a natural gas odorization device is provided, comprising:

[0012] A data acquisition module, used to obtain the current operating parameters of the natural gas odorization device from multiple sensors in real time;

[0013] A data processing module, used to optimize the current operating parameters to obtain an evaluation result of the current operating state of the natural gas odorization device;

[0014] A fault prediction module is used to output a potential fault prediction result of the natural gas odorizing device based on a pre-established predictive maintenance model of the natural gas odorizing device and taking a current operating status evaluation result of the natural gas odorizing device as input;

[0015] The maintenance suggestion generating module is used to generate a maintenance suggestion for the potential fault according to the potential fault prediction result and the current operation status evaluation result when the potential fault prediction result indicates the existence of the potential fault.

[0016] According to a third aspect, a computer device is provided, comprising:

[0017] one or more processors;

[0018] a storage device for storing one or more programs,

[0019] When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent maintenance method for a natural gas odorization device as described in the first aspect.

[0020] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the intelligent maintenance method for a natural gas odorization device as described in the first aspect is implemented.

[0021] The above technical solution has the following beneficial technical effects:

[0022] The present invention realizes the intelligent and efficient maintenance of natural gas odorization device by adopting technologies such as multi-sensor real-time data acquisition, optimization processing, predictive maintenance model and intelligent decision-making. In terms of real-time performance, the operating parameters are acquired in real time through multiple sensors to realize real-time monitoring and evaluation of the operating status of the natural gas odorization device; in terms of accuracy, the optimization processing and predictive maintenance model are adopted to improve the accuracy of potential fault prediction; in terms of intelligence, maintenance suggestions are automatically generated based on data-driven methods to reduce human judgment errors and improve the scientific nature of maintenance decisions; in terms of prevention, preventive maintenance is achieved by predicting potential faults to avoid losses caused by sudden faults; in terms of economy, it optimizes maintenance strategies, reduces unnecessary maintenance operations, reduces maintenance costs, and improves equipment utilization efficiency; in terms of safety, it can timely discover and handle potential faults, improve the operational safety of the natural gas odorization device, and ensure the safety of natural gas use. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0024] Figure 1 is a flow chart of an intelligent maintenance method for a natural gas odorization device according to an embodiment of the present invention;

[0025] Figure 2 is a specific flow chart of step S20 of an embodiment of the present invention;

[0026] Figure 3 is a specific flow chart of step S30 of an embodiment of the present invention;

[0027] Figure 4 is a specific flow chart of step S33 of an embodiment of the present invention;

[0028] Figure 5 is a specific flow chart of step S34 of an embodiment of the present invention;

[0029] Figure 6 is a specific flow chart of step S40 of an embodiment of the present invention;

[0030] Figure 7 is a functional block diagram of an intelligent maintenance system for a natural gas odorization device according to an embodiment of the present invention;

[0031] Figure 8 It is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description. Embodiment 1

[0033] like Figure 1 As shown, an embodiment of the present invention provides an intelligent maintenance method for a natural gas odorization device, and the intelligent maintenance method includes the following steps:

[0034] S10: Acquire current operating parameters of the natural gas odorization device from multiple sensors in real time.

[0035] In this embodiment, various types of sensors are used to comprehensively monitor the operating status of the natural gas odorizing device. These sensors include: a pressure sensor installed on the odorizing agent delivery pipeline, used to measure the delivery pressure of the odorizing agent from the storage tank to the odorizing device; a flow sensor installed on the odorizing agent injection pipeline, used to measure the odorizing agent injection amount from the odorizing device to the natural gas main pipeline; a pressure sensor installed in the mixing chamber of the odorizing device, used to measure the mixing chamber pressure of the odorizing device; a liquid level sensor installed on the odorizing agent storage tank, used to measure the liquid level of the odorizing agent storage tank; a temperature sensor installed in the odorizing agent storage tank, used to measure the temperature of the odorizing agent storage tank; a vibration sensor installed on the main body shell of the odorizing device, used to measure the vibration frequency of the main body shell of the odorizing device; a gas concentration sensor installed on the natural gas output pipeline after odorization, used to measure the odorizing agent concentration in the natural gas after odorization; a voltage sensor installed at the power input end of the odorizing device, used to measure the power supply voltage of the odorizing device; a valve position sensor installed on the control valve of the odorizing device, used to measure the opening of the control valve of the odorizing device; and a time recording module integrated in the controller of the odorizing device, used to record the operating time of the odorizing device. These sensors collect data at a preset sampling frequency (e.g., once per second) and transmit the data in real time to the data processing module in the cloud server through the data acquisition module.

[0036] First, by using multiple types of sensors, comprehensive monitoring of key parts and parameters of natural gas odorization equipment is achieved, which goes beyond the coverage of traditional systems and provides more in-depth data support. Secondly, the diversification of monitoring points enables the system to capture minor anomalies, realize early fault detection, and improve the effect of preventive maintenance. In addition, data verification and complementarity between different sensors enhance the reliability and accuracy of data. For example, the data of pressure and flow sensors verify each other, which improves the accuracy of fault diagnosis. At the same time, the system optimizes the odorization process and improves energy efficiency by real-time monitoring of parameters such as odorant injection volume and mixing chamber pressure. Monitoring of parameters such as vibration frequency and power supply voltage also helps to timely discover potential damage factors and extend the life of equipment. The application of gas concentration sensors ensures the reasonable range of odorant concentration and ensures the safe use of natural gas. Combined with the time recording module, the system can analyze the relationship between equipment operation time and parameters, providing a basis for intelligent maintenance decisions. Long-term data analysis can establish accurate performance models and optimize operation and maintenance strategies. In addition, the monitoring of environmental factors such as tank temperature enables the system to automatically adjust parameters according to environmental changes and enhance the adaptability of equipment in different environments. The combination of these effects not only improves the operating efficiency and reliability of the natural gas odorization unit, but also enables a shift from passive maintenance to active prevention.

[0037] S20: Optimizing the current operating parameters to obtain a current operating status evaluation result of the natural gas odorizing device.

[0038] Specifically, after receiving the sensor data in this step, the data processing module performs the following optimization processing: first, data cleaning is performed to remove outliers and noise; then, data standardization is performed to unify parameters of different dimensions to the same scale; then, feature extraction is performed to extract key features reflecting the state of the natural gas odorization device, such as parameter mean, variance, trend, etc.; finally, state evaluation is performed, and based on the processed data, a preset evaluation algorithm (such as fuzzy comprehensive evaluation method) is used to calculate the current operating state evaluation result, including the overall state score of the equipment and the state score of each subsystem. These operating parameters include any number of the following: odorant delivery pressure, odorant injection amount, mixing chamber pressure of the odorizer, odorant tank liquid level, odorant tank temperature, vibration frequency of the main shell of the odorizer, odorant concentration in the natural gas after odorization, power supply voltage of the odorizer, control valve opening of the odorizer, and operation time of the odorizer.

[0039] S30: Based on a pre-established predictive maintenance model of the natural gas odorizing device, taking the current operating status evaluation result of the natural gas odorizing device as input, outputting a potential fault prediction result of the natural gas odorizing device.

[0040] Specifically, in this step, the fault prediction module uses a pre-trained machine learning model (such as a long short-term memory network LSTM) as a predictive maintenance model. The model uses the current operating status evaluation results as input to predict potential faults that may occur in the future (for example, within 24 hours). The prediction results may include: potential fault type, fault probability, expected time of occurrence, and fault severity. The training data of the model comes from historical operating data and fault records. By learning from a large amount of data, the model can identify operating parameter patterns that may cause faults.

[0041] S40: When the potential fault prediction result indicates the existence of a potential fault, generating a maintenance suggestion for the potential fault according to the potential fault prediction result and the current operating status evaluation result.

[0042] Specifically, in this step, the maintenance suggestion generation module generates targeted maintenance suggestions based on the prediction results and the current status evaluation results. The specific steps are as follows: first determine the urgency of the potential fault; then determine the corresponding maintenance plan based on the fault type and severity; then optimize the maintenance plan based on the current operating status; finally generate detailed maintenance suggestions, including maintenance priority, specific maintenance operation steps, required tools and spare parts, estimated maintenance time, and safety precautions. This step presents the generated maintenance suggestions to the operator through the user interface, and can be automatically sent to the terminal device of the relevant maintenance personnel.

[0043] Through the above steps, the embodiment of the present invention realizes intelligent and preventive maintenance of the natural gas odorization device, effectively improves the operating efficiency and safety of the natural gas odorization device, and reduces maintenance costs. Embodiment 2

[0044] like Figure 2 As shown, in a specific embodiment of the present invention, step S20 optimizes the current operating parameters of the natural gas odorization device to obtain the current operating status evaluation result of the natural gas odorization device. This step specifically includes the following sub-steps:

[0045] S21: performing data cleaning on the current operating parameters of the natural gas odorizing device, wherein the data cleaning includes removing abnormal values ​​and noise to obtain the current operating parameters after cleaning.

[0046] Specifically, this step first uses the 3σ principle to identify and delete obvious outliers, and then uses the moving median filter method to remove sudden outliers. For high-frequency noise, the wavelet transform method is used for denoising. This step ensures that the data used in subsequent analysis has high reliability and accuracy.

[0047] The 3σ rule, also known as the rule of thumb, is a statistical outlier detection method. In a normal distribution, about 68% of the data fall within one standard deviation σ of the mean μ, about 95% of the data fall within 2σ, and about 99.7% of the data fall within 3σ. Therefore, if a data point falls outside the range of μ±3σ, it is considered an outlier. In practical applications, this range can be adjusted according to the specific situation, such as using 2.5σ or 3.5σ.

[0048] S22: Standardizing the current operating parameters after cleaning to obtain standardized current operating parameters.

[0049] Specifically, in this embodiment, the Z-score standardization method is used to unify parameters of different dimensions to a scale with a mean of 0 and a standard deviation of 1. The specific calculation formula is: Z = (X - μ) / σ, where X is the original data, μ is the sample mean, and σ is the sample standard deviation. This step allows different types of parameters to be directly compared and comprehensively analyzed.

[0050] S23: Analyze the standardized current operating parameters using a time series analysis method to extract trend characteristics and period characteristics of the standardized current operating parameters.

[0051] The trend feature refers to the trend characteristic parameter of the long-term change of the normalized current operating parameters of the natural gas odorization device over a period of time. This change is usually stable and continuous, and can be represented by parameters such as the long-term moving average or slope in the time series data. In this embodiment, the trend feature is extracted by the time series analysis method, and the long-term change trend in the operating state of the natural gas odorization device can be identified.

[0052] The periodic characteristics refer to the behavior of the normalized current operating parameters of the natural gas odorization unit showing repetitive fluctuations within a specific time interval. The periodic characteristics can be represented by parameters such as period length, amplitude, and phase in the time series data. In this implementation, the periodic characteristics are extracted by the time series analysis method, which can reveal the periodic change pattern in the operation of the natural gas odorization unit, such as weekly or monthly operation fluctuations.

[0053] Therefore, the trend characteristics refer to the changes of the standardized current operating parameters of the natural gas odorization device within the first time span (a longer time span), which includes the change direction, trend slope and trend inflection point. The period characteristics refer to the repetitive fluctuation behavior of the standardized current operating parameters of the natural gas odorization device within the second time span (a shorter specific time interval), which specifically includes multiple characteristics such as period length, amplitude, phase, frequency and waveform, and the first time span is greater than the second time span.

[0054] The direction of change refers to the overall change trend of the current operating parameter within the first time span, which may be rising, falling or stable. The trend slope refers to the rate of change of the current operating parameter over time, reflecting the rate and magnitude of trend change. It can be quantified by the slope of the long-term moving average or linear fit. The trend inflection point refers to the time point or position where the trend of the current operating parameter turns within the first time span, that is, the key node from rising to falling or from falling to rising. The cycle length refers to the duration of a complete cycle of the current operating parameter in the periodic fluctuation. For example, daily, weekly or monthly fluctuations. Amplitude refers to the maximum change amplitude of the operating parameter during the fluctuation process, that is, the difference between the maximum and minimum values ​​of the periodic fluctuation. Phase refers to the time offset of the periodic fluctuation, indicating the displacement of the fluctuation behavior on the time axis relative to a reference point. Frequency refers to the number of times the periodic fluctuation occurs in a unit time, that is, the frequency of the fluctuation. Waveform refers to the shape of the periodic fluctuation, which may be different types of fluctuation forms such as sine wave, square wave, triangle wave, etc.

[0055] Specifically, this embodiment adopts the seasonal and trend decomposition using Loess (STL) method based on local weighted regression, which is a time series analysis method used to understand the basic structure and characteristics of time series, such as trend, seasonality, periodicity and irregular fluctuations, and to predict future values ​​or control processes. In STL decomposition, seasonality refers to a pattern that repeats at fixed intervals in a time series, such as daily, weekly or annual periodic changes; residual refers to the irregular fluctuation part remaining in the time series after removing trend and seasonality.

[0056] The specific implementation process of S23 is performed in the following steps: First, data preparation is performed to ensure that the current operating parameters after standardization are equally spaced time series data. Then STL decomposition is applied, including setting a suitable seasonal cycle, estimating seasonal components and trend components using the Loess smoothing method, and calculating residuals. Then trend analysis is performed, and the long-term moving average is calculated based on the trend component obtained by STL decomposition to determine the direction and rate of change of the trend. Subsequently, periodic feature analysis is performed, and Fourier transform is performed on the detrended data. The results are analyzed and the main periodic components are identified, and the amplitude and phase of each main cycle are calculated. Then feature extraction is performed to extract trend features such as trend slope and trend inflection point from trend analysis, and extract features such as the length and amplitude of the main cycle from period analysis. Finally, the results are integrated, and the extracted trend features and period features are combined into a feature vector as input for subsequent analysis.

[0057] Among them, the Loess smoothing method is the abbreviation of Locally Weighted Scatterplot Smoothing, which is a non-parametric regression method used to fit a smooth curve between data points. The implementation idea is to select a set of neighboring points around each data point, assign weights, and then use the weighted least squares method to fit a low-order polynomial in the local area. This process is repeated for each data point, and finally a smooth curve is obtained. The advantage of the Loess method is that it can capture the nonlinear trend in the data well and is insensitive to outliers.

[0058] STL (Seasonal and Trend Decomposition using Loess) directly uses the Loess smoothing method as its core component. In the STL decomposition process, Loess is used to estimate the seasonal component, the trend component, and is continuously applied in an iterative optimization process. Specifically, STL estimates the seasonal pattern by applying Loess smoothing to the data points within each seasonal cycle, extracts the long-term trend by applying Loess smoothing to the remaining data after removing the seasonal component, and continuously improves these estimates in an iterative process. This approach gives STL the flexibility to capture complex patterns in time series, including those where seasonality or trend changes over time.

[0059] In the analysis of the operating parameters of the natural gas odorization unit, the seasonal and trend decomposition STL method based on local weighted regression can effectively separate the long-term trend, periodic changes and random fluctuations of the parameters. This decomposition can help better understand the operating status of the equipment, identify potential failure modes, and provide an important basis for predictive maintenance.

[0060] Through these steps, the time series data can be comprehensively analyzed and key features that are helpful for predicting future trends can be extracted. This method can deeply understand the changing patterns of the operating parameters of the natural gas odorization unit and provide an important basis for subsequent fault prediction and maintenance decisions.

[0061] S24: Based on a preset normal operating parameter range, calculating the deviation degree of the standardized current operating parameter.

[0062] Specifically, this step pre-sets the normal operating range of each parameter, such as [μ-2σ, μ+2σ]. For each standardized current operating parameter, the standardized Euclidean distance between it and the center value of the preset normal range is calculated as a quantitative indicator of the degree of deviation. The greater the degree of deviation, the more the current parameter deviates from the normal operating state.

[0063] S25: generating a current operating state evaluation result of the natural gas odorizing device according to the trend characteristics, the period characteristics and the degree of deviation of the standardized current operating parameters.

[0064] Specifically, this embodiment adopts a weighted summation method to comprehensively consider trend characteristics, period characteristics and degree of deviation. Specifically, three weight coefficients w1, w2, and w3 (satisfying w1+w2+w3=1) are set, corresponding to trend characteristics, period characteristics and degree of deviation respectively. The normalized values ​​of these three characteristics are multiplied by the corresponding weights and then summed to obtain a comprehensive score between 0 and 100, indicating the current operating status. The higher the score, the better the operating status. This scoring result provides important input for subsequent potential fault prediction.

[0065] The advantages of this technical solution are: First, data quality is significantly improved through data cleaning and standardization. Secondly, the use of time series analysis methods to extract trend and periodic characteristics can not only capture the long-term change trend of the equipment's operating status, but also identify periodic fluctuations, which is crucial for predicting potential problems and optimizing maintenance plans. In addition, by calculating the degree of parameter deviation, the system can accurately quantify the difference between the current operating status and the normal range, providing an objective basis for timely intervention. Finally, by comprehensively considering trend characteristics, periodic characteristics and degree of deviation, the generated operating status evaluation results are more comprehensive and accurate, and can reflect multiple dimensions of the equipment status. This multi-dimensional, high-precision evaluation method not only improves the accuracy of fault prediction, but also identifies potential problems that are difficult to detect with traditional methods, thereby achieving a more proactive and accurate maintenance strategy and greatly improving the reliability and operating efficiency of the equipment. Embodiment 3

[0066] like Figure 3 As shown, in this embodiment, step S30 may specifically include the following steps:

[0067] S31: extracting features from the current running state evaluation result to obtain a current running feature vector;

[0068] S32: Input the current operation feature vector into a pre-trained predictive maintenance model, wherein the predictive maintenance model is a deep learning model having the following structure:

[0069] a long short-term memory network LSTM layer, whose input is a sequence of running feature vectors of the most recent N time points including the current running feature vector, and whose output is a feature representation that captures the temporal dependency;

[0070] A plurality of sequentially connected fully connected layers, specifically comprising: a first fully connected layer, whose input is the output feature representation of the LSTM layer; and one or more subsequent fully connected layers; each fully connected layer is composed of a plurality of neurons, and each neuron of each fully connected layer is connected to all neurons of an adjacent fully connected layer; the plurality of sequentially connected fully connected layers are used to perform nonlinear transformation on the input features and further feature extraction;

[0071] A softmax activation function output layer, whose input is the output features of the last fully connected layer, is used to generate the predicted probability distribution of each potential fault type;

[0072] S33: determining potential fault types and the probability of occurrence of each potential fault type according to the predicted probability distribution generated by the output layer of the soft maximum activation function;

[0073] S34: using a pre-established database of influencing factors of potential fault types, combining the potential fault types and the current operating status evaluation result, to evaluate the impact of each potential fault type;

[0074] S35: generating a potential fault prediction result of the natural gas odorizing device according to the potential fault type, the probability of occurrence of each potential fault type and the impact degree of each potential fault type.

[0075] The beneficial technical effects of this technical solution are: first, by combining the deep learning model of the LSTM layer and multiple fully connected layers, the system can capture both short-term and long-term time dependencies, greatly improving the modeling capabilities of complex, nonlinear equipment behaviors. This structure can not only identify immediate anomalies, but also predict potential long-term failure trends, far exceeding the capabilities of traditional prediction methods. Secondly, the introduction of the soft maximum activation function output layer enables the system to simultaneously predict multiple potential fault types and their probabilities, providing a more comprehensive risk assessment. In addition, by combining the pre-established database of influencing factors, the system can dynamically evaluate the actual impact of each fault. This multi-dimensional evaluation method significantly improves the accuracy and practicality of the prediction. Finally, by comprehensively considering the fault type, probability of occurrence and impact, the generated potential fault prediction results not only point out the possible problems, but also provide the severity and urgency of the problems, providing more accurate and targeted guidance for maintenance decisions. This prediction method not only greatly reduces the risk of unexpected equipment downtime, but also optimizes the allocation of maintenance resources and improves overall operational efficiency.

[0076] The following is a detailed description and example of each step:

[0077] In step S31, the current operating state of the natural gas odorization device is feature extracted. Specifically, the following features are extracted: odorant injection amount (L / h), delivery pressure (MPa), natural gas flow rate (m³ / h), odorant concentration in natural gas after odorization (mg / m³), odorant storage tank temperature (℃), vibration frequency of the main shell of the odorization device (Hz), mixing chamber pressure of the odorization device (MPa), odorant storage tank liquid level (%), power supply voltage of the odorization device (V), control valve opening of the odorization device (%), and operation time of the odorization device (h). In addition, some derived features or combined features are also extracted, such as the ratio of odorant injection amount to natural gas flow, the difference between mixing chamber pressure and delivery pressure, odorant storage tank liquid level change rate, standard deviation of vibration frequency, change rate of control valve opening, fluctuation amplitude of odorant concentration, stability index of power supply voltage, ratio of operation time to maintenance cycle, and odorization efficiency (odorant concentration / odorant injection amount). In this way, a total of 20 original features were extracted, providing a rich data foundation for subsequent dimensionality reduction and model training.

[0078] Next, the principal component analysis (PCA) method is used to reduce the 20 features to 10 main features to form the current running feature vector. The PCA dimension reduction process includes the following steps: first, all features are standardized to make their mean 0 and variance 1; then, the covariance matrix between the features is calculated; then, the covariance matrix is ​​decomposed to obtain the eigenvalues ​​and corresponding eigenvectors; then, the eigenvectors corresponding to the top 10 largest eigenvalues ​​are selected to form a projection matrix; finally, the original 20-dimensional feature data is multiplied by the projection matrix to obtain the reduced 10-dimensional feature vector. In this way, the most important information in the original data is retained, while the number of features is reduced, which helps to improve the training efficiency and generalization ability of the subsequent model. This 10-dimensional feature vector is the so-called current running feature vector, which will be used as the input of the predictive maintenance model.

[0079] It should be noted that although the dimension of the features is reduced by PCA, each principal component still contains the combined information of the original features. In practical applications, the physical meaning of these principal components needs to be further analyzed to ensure the interpretability and practicality of the model. For example, the weights of the original features in each principal component can be checked to identify the original features that contribute most to the principal component, thereby giving a physical interpretation of each principal component. This not only helps to understand the working principle of the model, but also provides a more intuitive basis for decision-making for equipment maintenance personnel. In addition, during the actual deployment of the model, these features and their dimensionality reduction results need to be regularly re-evaluated and updated to adapt to long-term changes in equipment operating conditions and potential new failure modes.

[0080] In step S32, a pre-trained predictive maintenance model is used. The LSTM layer of the model receives a sequence of operating feature vectors for the last 24 hours (N=24), with one data point per hour, so the input shape is (24, 10). Here, 24 means that the number of time steps is 24 hours, and 10 means the number of features per time step (i.e., 10 main features after PCA dimensionality reduction). The LSTM layer contains 128 hidden units and outputs a 128-dimensional feature representation. Subsequently, the predictive maintenance model contains three fully connected layers with 64, 32, and 16 neurons, respectively. Each layer uses a ReLU activation function and a random dropout rate of 0.3 to prevent overfitting. Finally, the soft maximum activation function output layer has 6 neurons, corresponding to 6 predefined potential fault types: odorant injection system fault, flow meter fault, pressure abnormality fault, control system fault, sensor fault, and vibration abnormality fault.

[0081] like Figure 4 As shown, step S33 may specifically include the following steps:

[0082] S331: Setting a prediction probability threshold;

[0083] S332: sorting the predicted probability distribution generated by the output layer of the soft maximum activation function to obtain a list of potential fault types and their corresponding predicted probabilities in descending order;

[0084] S333: Selecting a potential fault type with a predicted probability greater than the predicted probability threshold from the potential fault type list arranged in descending order as a focus object;

[0085] S334: For each potential fault type of the concerned object, calculate its relative probability, where the relative probability is the ratio of the predicted probability of the potential fault type to the sum of the predicted probabilities of the potential fault types of all concerned objects;

[0086] S335: Based on the relative probability, the potential fault type of the object of interest is classified into risk levels. Specifically, when the relative probability is greater than or equal to 0.6, the potential fault type is classified as a high risk level; when the relative probability is greater than or equal to 0.3 and less than 0.6, the potential fault type is classified as a medium risk level; when the relative probability is less than 0.3, the potential fault type is classified as a low risk level.

[0087] In this example, first, the prediction probability threshold is set to 0.1 in step S331. Then, in step S332, assume that the prediction probability distribution of the model output is [0.05, 0.30, 0.40, 0.15, 0.05, 0.05], corresponding to 6 types of faults. The list of potential fault types and their prediction probabilities in descending order are: pressure abnormality fault (0.40), odorant injection system fault (0.30), control system fault (0.15), flow meter fault (0.05), sensor fault (0.05), and vibration abnormality fault (0.05).

[0088] In step S333, after sorting and screening, three objects of concern are obtained: abnormal pressure fault (0.40), odorant injection system fault (0.30) and control system fault (0.15). The predicted probabilities of these fault types all exceed the preset threshold of 0.1, and are therefore considered potential risks.

[0089] Step S334 involves the calculation of relative probabilities. Relative probabilities are calculated by dividing the predicted probability of each object of interest by the sum of the predicted probabilities of all objects of interest. The specific calculation process is as follows: First, the sum of the predicted probabilities is calculated (0.40 + 0.30 + 0.15 = 0.85); then, the relative probability of each fault type is calculated separately: 0.40 / 0.85 ≈ 0.47 for pressure abnormality fault, 0.30 / 0.85 ≈ 0.35 for odorant injection system fault, and 0.15 / 0.85 ≈ 0.18 for control system fault.

[0090] Finally, in step S335, the risk levels are divided according to the relative probability. The specific classification criteria are as follows: high risk (relative probability ≥ 0.6), medium risk (0.3 ≤ relative probability < 0.6), and low risk (relative probability < 0.3). According to this classification standard, the pressure abnormality fault (0.47) is a medium risk level, the odorant injection system fault (0.35) is a medium risk level, and the control system fault (0.18) is a low risk level. This risk assessment result provides clear priorities for equipment maintenance personnel and helps optimize preventive maintenance work.

[0091] like Figure 5 As shown, step S34 may specifically include the following steps:

[0092] S341: pre-establishing a potential fault type influencing factor database, wherein the potential fault type influencing factor database includes a plurality of potential fault types, each potential fault type corresponds to a plurality of influencing factors and their respective influencing degree weights;

[0093] Specifically, in step S341, a database of influencing factors of potential fault types is pre-established. The process identifies key influencing factors, such as safety risks, economic losses, environmental impacts, and repair difficulty, by analyzing historical fault data and equipment operation records. Use data-driven methods (such as principal component analysis or factor analysis) to determine the importance of each influencing factor and assign weights. For each potential fault type, establish evaluation criteria and quantification methods based on objective data, such as quantifying safety risks based on the possible casualties or equipment damage as a standard of 0-1, establishing a corresponding relationship between economic losses and scores based on historical data, establishing standards for environmental impacts based on pollutant emissions or the degree of environmental damage, and quantifying the repair difficulty by considering the time, resources, and technical complexity required for repair. Finally, establish a database structure to store this information.

[0094] S342: Determine the current state value of each influencing factor based on the current operating state evaluation result;

[0095] Specifically, in step S342, the current state values ​​of these factors are evaluated based on the current operating state. First, real-time operating data is collected, including parameters such as pressure readings, flow, temperature, vibration, etc. Then, the raw data is converted into a standardized score using predefined evaluation criteria. For example, for safety risks, the current pressure value is compared with the safety threshold; for economic losses, the degree of decline in production efficiency and potential downtime under the current abnormal state are estimated; for environmental impact, the ratio of possible pollutant emissions under the current state to environmental standards is calculated; for repair difficulty, the time and resources required for repair are estimated based on the current fault state and compared with historical data. Next, the current state value is adjusted using a time series analysis method taking into account historical data and trends. Finally, a machine learning model (such as a support vector machine or a random forest) is applied to comprehensively evaluate the state value of each influencing factor in combination with multiple relevant parameters, and the results are normalized to the range of 0-1.

[0096] The multiple related parameters here refer to that for each influencing factor, not only the directly related parameters are considered, but also other parameters that may indirectly affect the factor. For example: for safety risks, in addition to the pressure value, parameters such as temperature, flow, and equipment operating time may also be considered. For economic losses, multiple parameters such as production efficiency, energy consumption, and raw material utilization may be involved. For environmental impacts, parameters such as pollutant concentration, emissions, ambient temperature, and humidity may be included. For the difficulty of repair, parameters such as equipment operating time, historical maintenance frequency, and spare parts inventory status may be considered. By learning the complex relationship between these parameters, the machine learning model can more accurately evaluate the status of each influencing factor. Finally, the evaluation results output by the model are normalized to the range of 0-1 to ensure the comparability between different influencing factors.

[0097] S343: For each potential fault type, calculate its comprehensive impact score;

[0098] S344: according to a preset impact threshold, the comprehensive impact score of each potential fault type is divided into three levels: high, medium and low;

[0099] S345: Generate an impact assessment result list.

[0100] In this specific embodiment, in step S341, a database of influencing factors of potential fault types is pre-established. Taking the abnormal pressure fault as an example, its influencing factors include: safety risk (weight 0.4), economic loss (weight 0.3), environmental impact (weight 0.2) and repair difficulty (weight 0.1). These weights are obtained by analyzing historical data and using the principal component analysis method. In step S342, based on the current operating state, the current state values ​​of these factors are evaluated as: 0.8, 0.7, 0.6 and 0.5 (all normalized). These values ​​are obtained by integrating multiple real-time parameters through a machine learning model. In step S343, the comprehensive impact score is calculated: 0.4*0.8 + 0.3*0.7 + 0.2*0.6 + 0.1*0.5 = 0.7. In step S344, based on the preset evaluation criteria (high impact above 0.8, medium impact between 0.5 and 0.8, and low impact below 0.5), the impact of the abnormal pressure fault is determined to be medium. In step S345, similar calculations are performed on all potential fault types, and finally an impact assessment result list is generated.

[0101] In step S35, the potential fault type, occurrence probability and impact degree are comprehensively considered to generate a final potential fault prediction result.

[0102] In this step, the potential fault type, probability of occurrence and impact are comprehensively considered to generate the final potential fault prediction result. The comprehensive analysis method will use all the potential fault types identified in the previous steps as the basis, use the relative probability value calculated in step S33 as the probability of occurrence, and consider the various influencing factors evaluated in step S34, including safety risks, economic losses, environmental impacts and repair difficulties.

[0103] The steps of generating prediction results include data integration, risk matrix construction, fault location, risk level division, priority sorting and prediction result generation. First, all relevant data are integrated into a unified analysis framework, and then a two-dimensional risk matrix is ​​created, with the horizontal axis representing the probability of occurrence and the vertical axis representing the degree of impact. According to the probability of occurrence and the comprehensive degree of impact of each potential fault type, it is located in the corresponding position in the risk matrix. Then, according to the risk level division standard in step S335, the risk level of each potential fault type is determined, and priority is sorted based on the risk level and the specific degree of impact. Finally, based on the above information, detailed prediction results are generated for each potential fault type.

[0104] The final potential fault prediction results are presented in the form of a table or report, including a list of potential fault types, the probability of occurrence of each fault, the impact score of each aspect, the comprehensive risk level and other key information.

[0105] This detailed analysis and prediction approach allows maintenance resources to be allocated more targetedly, prioritizing high-risk and high-impact potential failures, thereby improving equipment reliability, reducing safety hazards, and maximizing operational efficiency. Embodiment 4

[0106] like Figure 6 As shown, the above step S40 may specifically include the following steps:

[0107] S41: determining a faulty component in the natural gas odorization device that needs maintenance according to the fault type, occurrence probability and impact degree in the potential fault prediction result, and the difference between the actual value and the expected value of the performance parameter in the current operation status evaluation result;

[0108] S42: Based on the faulty components requiring maintenance, combined with the operation plan of the natural gas odorization device and the maintenance resource availability information, recommend an optimal maintenance time and priority order;

[0109] S43: Generate maintenance operation instructions according to the potential fault type and the faulty component that needs maintenance, combined with historical maintenance records, including a location method, a test method, a repair method, and safety precautions for the faulty component;

[0110] S44: Listing tools, spare parts and materials required for performing maintenance according to the maintenance operation guide and the historical maintenance records;

[0111] S45: Predicting the degree of improvement in the performance of the natural gas odorization device after maintenance based on the historical maintenance records, the current operating status evaluation results, and the maintenance plan; wherein the maintenance plan includes: a list of faulty parts requiring maintenance, the recommended maintenance time and priority, the maintenance operation instructions, and a list of tools, spare parts, and materials required to perform the maintenance.

[0112] The advantages of the above technical solution include: first, by comprehensively considering the potential fault prediction results and the current operating status evaluation results, the system can accurately locate the faulty parts that need maintenance, greatly improving the pertinence and efficiency of maintenance. Secondly, by combining the operation plan and resource availability, the optimal maintenance time and priority are recommended, which not only maximizes the equipment operation time, but also optimizes the allocation of maintenance resources. Moreover, the system can automatically generate detailed maintenance operation instructions based on historical maintenance records, including positioning methods, testing methods, repair methods and safety measures, which greatly improves the standardization and safety of maintenance, and also reduces the dependence on highly skilled maintenance personnel. In addition, by pre-listing the required tools, spare parts and materials, delays and interruptions in the maintenance process are significantly reduced, and maintenance efficiency is improved. Finally, the system can predict the degree of performance improvement after maintenance, providing a quantitative basis for maintenance decisions. This comprehensive and intelligent maintenance management method not only greatly improves the reliability and operation efficiency of the natural gas odorization unit, but also achieves the optimal configuration of maintenance resources.

[0113] In some embodiments, step S41 specifically includes the following steps:

[0114] S411: using the fault severity scoring model, taking the fault type, occurrence probability and impact degree in the potential fault prediction results as input, calculating the severity score of each potential fault type;

[0115] S412: Calculating the deviation rate between the actual value and the expected value of each performance parameter according to the current operation status evaluation result;

[0116] S413: performing a weighted combination of the severity score and the deviation rate to obtain a comprehensive evaluation score;

[0117] S414: According to a preset threshold, objects whose comprehensive evaluation scores exceed the preset threshold are screened out and determined as faulty components that need maintenance.

[0118] In this embodiment, a pre-trained fault severity scoring model is first used. The model takes the fault type, probability of occurrence and impact degree in the potential fault prediction results as input to calculate the severity score of each potential fault type. Taking the odorant injection system as an example, the odorant pump fault type is considered, the probability of occurrence is 0.7, and the impact degree includes safety risk (8 points), economic loss (7 points), environmental impact (6 points) and repair difficulty (5 points), and the full score is 10 points. The fault severity scoring model combines these factors and calculates the severity score of the odorant pump fault as 8.5 (full score 10 points). Next, according to the current operating status evaluation results, the deviation rate between the actual value and the expected value of each performance parameter is calculated. For the odorant injection system, the actual value of the odorant flow is 5% lower than the expected value, the actual value of the odorant concentration is 3% lower than the expected value, and the actual value of the system pressure is 2% higher than the expected value. The average of these deviation rates is calculated to obtain an overall deviation rate of 4%. Then, the severity score (8.5) and the deviation rate (4%) are weighted and combined in a ratio of 6:4 to calculate the comprehensive evaluation score. The specific calculation is: Comprehensive evaluation score = 8.5 * 0.6 + (4 / 10) * 0.4 = 5.1 + 0.16 = 5.26. Finally, the threshold is set to 5. Since the comprehensive evaluation score of the odorant pump is 5.26, which exceeds the threshold of 5, the odorant pump is determined to be a faulty component that needs maintenance. In this way, the system can effectively identify and prioritize potential faulty components, thereby improving the overall reliability and performance of the natural gas odorization unit.

[0119] In some embodiments, step S42 may specifically include the following steps:

[0120] S421: Obtaining an operation plan of the natural gas odorization device, including a planned downtime and a continuous operation period;

[0121] S422: Obtaining maintenance resource availability information, including available maintenance personnel, spare parts inventory, and availability of special tools;

[0122] S423: for each faulty component requiring maintenance, determining a maintenance time window, wherein the time window is based on the urgency of the fault and the degree of impact on the operation of the odorizing device;

[0123] S424: Match the maintenance time window of each faulty component with the operation plan and the maintenance resource availability information to generate feasible maintenance time options;

[0124] S425: Calculating resource utilization efficiency based on the maintenance resource availability information, scoring the feasible maintenance time options according to the comprehensive evaluation score and the resource utilization efficiency, and selecting the maintenance time option with the highest score as the recommended optimal maintenance time;

[0125] S426: Determine the priority of the maintenance tasks based on the recommended optimal maintenance time and the comprehensive evaluation score.

[0126] In one example, the operation plan for the odorization device in the next month is first obtained, including two planned downtimes. Then the maintenance resource information is obtained, for example, there are three maintenance personnel available. For the odorizer pump, its maintenance time window is determined to be from next Tuesday to Thursday. This time window is matched with the operation plan and resource availability to generate feasible maintenance time options. The resource utilization efficiency is calculated, for example, choosing maintenance on Wednesday can maximize personnel utilization. Taking into account the evaluation score and resource efficiency, Wednesday is selected as the best maintenance time. Finally, according to the comprehensive evaluation score, the odorizer pump maintenance task is ranked first in the priority order.

[0127] In one example, suppose a natural gas odorization unit has three faulty components that need maintenance during a recent inspection: an odorant pump (comprehensive evaluation score of 8.5), a flow meter (comprehensive evaluation score of 6.2), and a pressure sensor (comprehensive evaluation score of 4.8). First, the operation plan is obtained (step S421), and it is learned that the next planned shutdown maintenance will be carried out in two months and last for 7 days, and the unit needs to continue to operate before then. Next, maintenance resource availability information is obtained (step S422). Currently, there are 3 maintenance personnel with relevant skills available, 2 spare parts for the odorant pump and flow meter, and 3 spare parts for the pressure sensor. Special maintenance tools are sufficient, but precision calibration equipment will not arrive until next week. Then, a maintenance time window is determined for each faulty component (step S423). Given the high evaluation score of the odorant pump, it is recommended to perform maintenance within two weeks; the flow meter can be maintained within one month; and the pressure sensor can wait until the planned shutdown for maintenance. Based on this information, feasible maintenance time options are generated (step S424). For the odorant pump, the maintenance time options include: the night shift on Wednesday next week, the Saturday and Sunday shift on Saturday next week, or the planned downtime period two months later. The options for the flow meter include: the Monday and Sunday shift three weeks later, or the planned downtime period. The pressure sensor only has the option of the planned downtime period. The resource utilization efficiency is then calculated and the options are scored (step S425). Considering the urgency of the odorant pump, the availability of personnel, and the arrival time of the precision calibration equipment, the Saturday and Sunday shift next week is recommended as the best maintenance time for the odorant pump. The maintenance of the flow meter is recommended to be performed on Monday three weeks later, because this can be staggered with the maintenance of the odorant pump and better allocate human resources. The maintenance of the pressure sensor maintains the recommendation to be performed during the planned downtime period. Finally, based on these recommended maintenance times and the comprehensive evaluation scores, the maintenance tasks are prioritized (step S426): 1. Odorant pump (Saturday and Sunday shift next week); 2. Flow meter (Monday and Sunday shift three weeks later); 3. Pressure sensor (during the planned downtime period two months later). This priority order takes into account both the severity of equipment failure and the effective use of resources, and develops a balanced and efficient maintenance plan for the natural gas odorization unit.

[0128] In some embodiments, step S43 specifically includes:

[0129] S431: Establish a maintenance knowledge base, which includes the fault characteristics, possible causes and corresponding maintenance methods of each potential fault type;

[0130] S432: Retrieving relevant maintenance information from the maintenance knowledge base according to the potential fault type and the faulty component that needs maintenance;

[0131] S433: Analyze successful maintenance cases of similar failures in combination with historical maintenance records, and extract maintenance experience from the successful maintenance cases;

[0132] S434: Generate maintenance operation guidance based on the maintenance information and the maintenance experience.

[0133] In this example, a knowledge base containing fault information of all components of the odorizing device is first established. For odorant pump failure, possible fault causes are retrieved from the knowledge base, such as seal aging, bearing wear, etc. Successful cases of similar failures in historical records are then analyzed to extract experience on how to properly remove and replace the seal. Finally, detailed maintenance operation instructions are generated, including the location method of the pump, the test method for the integrity of the seal, the replacement steps, and safety precautions during operation.

[0134] In some embodiments, step S44 specifically includes:

[0135] S441: Identify special tools and general tools required to perform maintenance according to the maintenance operation guide;

[0136] S442: Determine the type and quantity of spare parts required based on the faulty component that needs maintenance;

[0137] S443: Query the historical maintenance records to identify tools and spare parts used in similar maintenance tasks;

[0138] S444: Generate a tool and spare parts list, including: tool name, specification and quantity; spare parts name, model, quantity and storage location; how to obtain tools or spare parts.

[0139] In this embodiment, according to the maintenance operation instructions of the odorant pump, the special tools that need to be used, such as the pump body disassembly tool and the general tools such as the wrench set, are identified. It is determined that the required spare parts include new seals and bearings. By querying the historical records, it is found that the previous maintenance also used a special seal installation tool. Finally, a detailed list is generated, including the name, specification and quantity of all tools, such as "pump body disassembly tool, model XYZ 1 set"; spare parts information such as "seal model ABC 2 pieces, stored in warehouse area A shelf 12".

[0140] In some embodiments, step S45 specifically includes:

[0141] S451: Establishing a performance model of the natural gas odorization device, wherein the performance model is based on historical operating parameters and historical maintenance records, and describes the relationship between various operating parameters and the states of all components in the natural gas odorization device and the overall performance;

[0142] S452: Inputting the current operation status evaluation result into the performance model to obtain a current performance benchmark value;

[0143] S453: According to the maintenance plan, simulate and update corresponding parameters and states in the performance model;

[0144] S454: Use the updated performance model to predict the performance indicators after maintenance;

[0145] S455: Calculate the difference between the performance indicator after maintenance and the current performance benchmark value to obtain the expected performance improvement degree.

[0146] In this example, a machine learning-based performance model of the odorizer was first established, which takes into account the impact of the status of each component on the overall performance. The current operating status evaluation results were input into the model to obtain the current performance benchmark value, such as an odorization efficiency of 92%. Then, the state parameters of the odorizer pump in the model were updated according to the maintenance plan simulation. The updated model was used to predict the performance indicators after maintenance, and the expected odorization efficiency was 98%. Finally, the difference was calculated, and the expected performance improvement was 6 percentage points.

[0147] In some embodiments, the method further comprises step S50:

[0148] Execute the maintenance operation in the maintenance suggestion, and feed back the maintenance result to the predictive maintenance model for updating and optimizing the predictive maintenance model. Step S50 specifically includes: the predictive maintenance model is adjusted based on the fed-back maintenance result, including adjusting the time estimation parameters and the weights of relevant influencing factors to improve the prediction accuracy; the time estimation parameters refer to the weights and bias parameters in the LSTM layer; the weights of the relevant influencing factors refer to the weight parameters in the fully connected layer. In the fully connected layer, each input feature (including the running feature vector) has a corresponding weight parameter, which determines how the model weights each feature to generate the final prediction. The weights of the relevant influencing factors refer to the weight parameters in the fully connected layer, which reflect the importance of each input feature to the prediction of potential faults. Adjusting these weights can change the model's attention to different features, thereby improving the accuracy of the prediction.

[0149] In this embodiment, the maintenance personnel perform maintenance operations according to the generated maintenance suggestions, such as replacing the seal of the odorant pump at the recommended time. After the maintenance is completed, information such as the actual maintenance process, problems encountered, actual time and resources used are recorded. At the same time, the actual performance improvement of the odorizing device after maintenance is monitored. This information is fed back into the predictive maintenance model to adjust and optimize the model's predictive accuracy. For example, if the actual maintenance time is longer than predicted, the model will adjust the time estimation parameters accordingly; if the performance improvement is not as expected, the model will adjust the weights of the relevant influencing factors. Through this continuous feedback and optimization, the accuracy and practicality of the predictive maintenance model will continue to improve. Embodiment 5

[0150] like Figure 7 As shown, this embodiment provides an intelligent maintenance system 700 for a natural gas odorization device, which is deployed in a cloud server and includes:

[0151] A data acquisition module 710 is used to acquire the current operating parameters of the natural gas odorization device from multiple sensors in real time;

[0152] The data processing module 720 is used to optimize the current operating parameters to obtain the current operating status evaluation result of the natural gas odorization device;

[0153] A fault prediction module 730 is used to output a potential fault prediction result of the natural gas odorizing device based on a pre-established predictive maintenance model of the natural gas odorizing device and taking the current operating status evaluation result of the natural gas odorizing device as input;

[0154] The maintenance suggestion generating module 740 is used to generate a maintenance suggestion for the potential fault according to the potential fault prediction result and the current operating state evaluation result when the potential fault prediction result indicates the existence of the potential fault.

[0155] In some embodiments, the plurality of sensors include any of the following:

[0156] The pressure sensor installed on the odorant delivery pipeline is used to measure the delivery pressure of the odorant from the storage tank to the odorizing device;

[0157] A flow sensor installed on the odorant injection pipeline is used to measure the amount of odorant injected from the odorizing device to the natural gas main pipeline;

[0158] A pressure sensor installed in the mixing chamber of the odorizing device, used to measure the mixing chamber pressure of the odorizing device;

[0159] A liquid level sensor installed on the odorant storage tank is used to measure the liquid level of the odorant storage tank;

[0160] A temperature sensor installed in the odorant storage tank is used to measure the temperature of the odorant storage tank;

[0161] A vibration sensor installed on the main shell of the odorizing device is used to measure the vibration frequency of the main shell of the odorizing device;

[0162] A gas concentration sensor installed on the odorized natural gas output pipeline is used to measure the odorant concentration in the odorized natural gas;

[0163] A voltage sensor installed at the power input terminal of the odorizing device is used to measure the power supply voltage of the odorizing device;

[0164] A valve position sensor installed on the control valve of the odorizing device is used to measure the opening of the control valve of the odorizing device;

[0165] The time recording module integrated in the controller of the odorizing device is used to record the operating time of the odorizing device.

[0166] In some embodiments, the operating parameters include any of the following:

[0167] The delivery pressure, the odorant injection amount, the mixing chamber pressure of the odorizing device, the liquid level of the odorant storage tank, the temperature of the odorant storage tank, the vibration frequency of the main shell of the odorizing device, the odorant concentration in the natural gas after odorization, the power supply voltage of the odorizing device, the opening of the control valve of the odorizing device, and the operating time of the odorizing device.

[0168] In some embodiments, the data processing module 720 includes:

[0169] A data cleaning unit, used for cleaning the current operating parameters of the natural gas odorizing device, wherein the data cleaning includes removing abnormal values ​​and noise to obtain the current operating parameters after cleaning;

[0170] A standardization processing unit, used for performing standardization processing on the current operating parameters after cleaning to obtain standardized current operating parameters;

[0171] A time series analysis unit, used to analyze the standardized current operating parameters using a time series analysis method, and extract trend characteristics and period characteristics of the standardized current operating parameters;

[0172] a deviation calculation unit, configured to calculate the degree of deviation of the standardized current operating parameters based on a preset normal operating parameter range;

[0173] The evaluation result generating unit is used to generate the current operation state evaluation result of the natural gas odorizing device according to the trend characteristics, the period characteristics and the deviation degree of the standardized current operation parameters.

[0174] In some embodiments, the fault prediction module 730 includes:

[0175] A feature extraction unit, used to extract features from the current running state evaluation result to obtain a current running feature vector;

[0176] The deep learning model unit is used to input the current operation feature vector into a pre-trained predictive maintenance model, wherein the predictive maintenance model is a deep learning model having the following structure:

[0177] a long short-term memory network LSTM layer, whose input is a sequence of running feature vectors of the most recent N time points including the current running feature vector, and whose output is a feature representation that captures the temporal dependency;

[0178] A plurality of sequentially connected fully connected layers, specifically comprising: a first fully connected layer, whose input is the output feature representation of the LSTM layer; and one or more subsequent fully connected layers; each fully connected layer is composed of a plurality of neurons, and each neuron of each fully connected layer is connected to all neurons of an adjacent fully connected layer; the plurality of sequentially connected fully connected layers are used to perform nonlinear transformation on the input features and further feature extraction;

[0179] A softmax activation function output layer, whose input is the output features of the last fully connected layer, is used to generate the predicted probability distribution of each potential fault type;

[0180] a fault type determination unit, configured to determine potential fault types and the probability of occurrence of each potential fault type according to the predicted probability distribution generated by the soft maximum activation function output layer;

[0181] An impact assessment unit, configured to assess the impact of each potential fault type by using a pre-established database of impact factors of potential fault types, in combination with the potential fault type and the current operating state assessment result;

[0182] The prediction result generating unit is used to generate a potential fault prediction result of the natural gas odorizing device according to the potential fault type, the occurrence probability of each potential fault type and the influence degree of each potential fault type.

[0183] In some embodiments, the fault type determination unit includes:

[0184] A threshold setting module, used to set a prediction probability threshold;

[0185] A sorting module, used for sorting the predicted probability distribution generated by the output layer of the soft maximum activation function to obtain a list of potential fault types and their corresponding predicted probabilities in descending order;

[0186] An object of interest selection module, configured to select a potential fault type with a predicted probability greater than the predicted probability threshold from the potential fault type list arranged in descending order as an object of interest;

[0187] A relative probability calculation module, used to calculate the relative probability of each potential fault type of the concerned object, wherein the relative probability is the ratio of the predicted probability of the potential fault type to the sum of the predicted probabilities of the potential fault types of all concerned objects;

[0188] The risk level classification module is used to classify the risk level of the potential failure type of the object of concern based on the relative probability, specifically including:

[0189] When the relative probability is greater than or equal to 0.6, the potential fault type is classified as a high risk level;

[0190] When the relative probability is greater than or equal to 0.3 and less than 0.6, the potential fault type is classified as a medium risk level;

[0191] When the relative probability is less than 0.3, the potential fault type is classified as a low risk level.

[0192] In some embodiments, the impact assessment unit includes:

[0193] A database establishment module is used to pre-establish a potential fault type influencing factor database, wherein the potential fault type influencing factor database includes a plurality of potential fault types, each potential fault type corresponds to a plurality of influencing factors and their respective influencing degree weights;

[0194] A state value determination module, used to determine the current state value of each influencing factor based on the current operation state evaluation result;

[0195] The comprehensive impact degree calculation module is used to calculate the comprehensive impact degree score for each potential fault type, including:

[0196] Obtaining the influencing factors corresponding to the potential fault type and their influence degree weights from the potential fault type influencing factor database;

[0197] Multiply the influence weight of each influencing factor by the current state value of the influencing factor to obtain the single factor influence score;

[0198] Add up all single factor impact scores to get the comprehensive impact score of the potential fault type;

[0199] An impact level classification module is used to classify the comprehensive impact level score of each potential fault type into three levels: high, medium, and low according to a preset impact level threshold;

[0200] The assessment result generation module is used to generate an impact assessment result list, wherein the impact assessment result list includes potential fault types, the comprehensive impact degree scores and the levels corresponding to the comprehensive impact degree scores.

[0201] In some embodiments, the maintenance suggestion generating module 740 includes:

[0202] a faulty component determination unit, for determining a faulty component in the natural gas odorization device that needs maintenance according to the fault type, occurrence probability and impact degree in the potential fault prediction result, and the difference between the actual value and the expected value of the performance parameter in the current operation status evaluation result;

[0203] A maintenance time recommendation unit, for recommending an optimal maintenance time and priority order based on the faulty components requiring maintenance, combined with an operation plan of the natural gas odorization device and maintenance resource availability information;

[0204] A maintenance operation guide generating unit, configured to generate a maintenance operation guide according to the potential fault type and the faulty component requiring maintenance, in combination with historical maintenance records, which includes a location method, a test method, a repair method, and safety precautions for the faulty component;

[0205] A tool spare parts list generating unit, used to list tools, spare parts and materials required to perform maintenance according to the maintenance operation instructions and the historical maintenance records;

[0206] A performance improvement prediction unit is used to predict the performance improvement degree of the natural gas odorization device after maintenance based on the historical maintenance records, the current operating status evaluation results, and the maintenance plan; wherein the maintenance plan includes: a list of faulty parts that need maintenance, the recommended maintenance time and priority, the maintenance operation instructions, and a list of tools, spare parts and materials required to perform the maintenance.

[0207] In some embodiments, the system further includes a model updating module for executing the maintenance operations in the maintenance recommendations and feeding back the maintenance results to the predictive maintenance model for updating and optimizing the predictive maintenance model.

[0208] In some embodiments, the faulty component determination unit includes:

[0209] A severity scoring module is used to calculate the severity score of each potential fault type by using the fault severity scoring model and taking the fault type, occurrence probability and impact degree in the potential fault prediction results as input;

[0210] A deviation rate calculation module, used to calculate the deviation rate between the actual value and the expected value of each performance parameter according to the current operating state evaluation result;

[0211] A comprehensive evaluation module, used for weighted combination of the severity score and the deviation rate to obtain a comprehensive evaluation score;

[0212] The faulty component screening module is used to screen out objects whose comprehensive evaluation scores exceed the preset threshold according to the preset threshold, and determine them as faulty components that need maintenance.

[0213] In some embodiments, the maintenance time recommendation unit includes:

[0214] An operation plan acquisition module is used to obtain the operation plan of the natural gas odorization device, including the planned downtime and continuous operation time period;

[0215] A resource availability acquisition module is used to obtain maintenance resource availability information, including available maintenance personnel, spare parts inventory, and availability of special tools;

[0216] A time window determination module, for determining a maintenance time window for each faulty component requiring maintenance, wherein the time window is based on the urgency of the fault and the degree of impact on the operation of the odorizing device;

[0217] A maintenance time matching module, used to match the maintenance time window of each faulty component with the operation plan and the maintenance resource availability information to generate feasible maintenance time options;

[0218] An optimal time selection module, configured to calculate resource utilization efficiency based on the maintenance resource availability information, score the feasible maintenance time options according to the comprehensive evaluation score and the resource utilization efficiency, and select the maintenance time option with the highest score as the recommended optimal maintenance time;

[0219] The priority determination module is used to determine the priority of the maintenance tasks based on the recommended optimal maintenance time and the comprehensive evaluation score.

[0220] In some embodiments, the maintenance operation instruction generating unit includes:

[0221] A knowledge base building module is used to build a maintenance knowledge base, which contains the fault characteristics, possible causes and corresponding maintenance methods of each potential fault type;

[0222] A maintenance information retrieval module, used to retrieve relevant maintenance information from the maintenance knowledge base according to the potential fault type and the faulty component that needs maintenance;

[0223] A maintenance experience extraction module is used to analyze successful maintenance cases of similar failures in combination with historical maintenance records, and extract maintenance experience from the successful maintenance cases;

[0224] The operation guide generation module is used to generate a maintenance operation guide based on the maintenance information and the maintenance experience.

[0225] In some embodiments, the tool spare parts list generating unit includes:

[0226] a tool identification module, for identifying special tools and general tools required to perform maintenance according to the maintenance operation instructions;

[0227] A spare parts determination module, used to determine the type and quantity of spare parts required based on the faulty parts that need maintenance;

[0228] A history query module, used to query the historical maintenance records and identify tools and spare parts used in similar maintenance tasks;

[0229] The list generation module is used to generate a tool and spare parts list, including: tool name, specifications and quantity; spare parts name, model, quantity and storage location; and how to obtain tools or spare parts.

[0230] In some embodiments, the performance improvement prediction unit includes:

[0231] A performance model building module is used to build a performance model of the natural gas odorization device, wherein the performance model is based on historical operating parameters and historical maintenance records, and describes the relationship between various operating parameters and the status of all components in the natural gas odorization device and the overall performance;

[0232] A current performance evaluation module, used to input the current operating state evaluation result into the performance model to obtain a current performance benchmark value;

[0233] A model updating module, used for simulating and updating corresponding parameters and states in the performance model according to the maintenance plan;

[0234] A post-maintenance performance failure prediction module is used to predict the performance indicators after maintenance using the updated performance model;

[0235] The performance improvement calculation module is used to calculate the difference between the performance index after maintenance and the current performance benchmark value to obtain the expected performance improvement degree.

[0236] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0237] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the above methods is implemented.

[0238] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. Of course, there are other forms of readable storage media, such as quantum memory, graphene memory, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunication signals.

[0239] The present invention also provides a computer device. The computer device of the embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided by the present invention.

[0240] Reference below Figure 8 , which shows a schematic diagram of the structure of a computer device 800 suitable for implementing an embodiment of the present invention. Figure 8 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0241] like Figure 8 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0242] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed.

[0243] In particular, according to the embodiments disclosed in the present invention, the process described in the main step diagram above can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the main step diagram. In the above embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the central processing unit 801, the above functions defined in the system of the present invention are executed.

[0244] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0245] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0246] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be arranged in a processor. The names of these units do not, in some cases, limit the units themselves.

[0247] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent maintenance method for a natural gas odorization device, characterized in that: The following steps are involved: S10: acquiring current operating parameters of the natural gas odorization device from multiple sensors in real time; S20: Optimizing the current operating parameters to obtain a current operating status evaluation result of the natural gas odorizing device; S30: Based on a pre-established predictive maintenance model of the natural gas odorizing device, taking the current operating status evaluation result of the natural gas odorizing device as input, outputting a potential fault prediction result of the natural gas odorizing device; S40: When the potential fault prediction result indicates the existence of a potential fault, generating a maintenance suggestion for the potential fault according to the potential fault prediction result and the current operating state evaluation result; Wherein, step S20 specifically includes: S21: performing data cleaning on the current operating parameters of the natural gas odorization device, wherein the data cleaning includes removing abnormal values ​​and noise to obtain the current operating parameters after cleaning; S22: performing standardization processing on the current operating parameters after cleaning to obtain standardized current operating parameters; S23: Analyze the standardized current operating parameters by using a time series analysis method, and extract the trend characteristics and period characteristics of the standardized current operating parameters; the trend characteristics refer to the changes of the standardized current operating parameters of the natural gas odorization device within a first time span, including the change direction, trend slope and trend inflection point; the period characteristics refer to the repetitive fluctuation behavior of the standardized current operating parameters of the natural gas odorization device within a second time span, including multiple of the period length, amplitude, phase, frequency and waveform, and the first time span is greater than the second time span; the time series analysis method adopts the seasonal and trend decomposition method STL based on local weighted regression; S24: Based on the preset normal operating parameter range, calculating the deviation degree of the standardized current operating parameter; wherein, for each standardized current operating parameter, calculating the standardized Euclidean distance between the standardized current operating parameter and the center value of the preset normal operating parameter range as a quantitative indicator of the deviation degree; S25: according to the trend characteristics, the period characteristics and the degree of deviation of the standardized current operating parameters, and three weight coefficients corresponding to the trend characteristics, the period characteristics and the degree of deviation, respectively, a weighted summation method is used to generate a current operating state evaluation result of the natural gas odorizing device; Wherein, step S30 specifically includes: S31: extracting features from the current running state evaluation result to obtain a current running feature vector; S32: Input the current operation feature vector into a pre-trained predictive maintenance model, wherein the predictive maintenance model is a deep learning model having the following structure: a long short-term memory network (LSTM) layer, whose input is a sequence of running feature vectors of the most recent N time points including the current running feature vector, and whose output is a feature representation that captures the temporal dependency; A plurality of sequentially connected fully connected layers, specifically comprising: a first fully connected layer, whose input is the output feature representation of the LSTM layer; and one or more subsequent fully connected layers; each fully connected layer is composed of a plurality of neurons, and each neuron of each fully connected layer is connected to all neurons of an adjacent fully connected layer; the plurality of sequentially connected fully connected layers are used to perform nonlinear transformation on the input features and further feature extraction; A softmax activation function output layer, whose input is the output features of the last fully connected layer, is used to generate the predicted probability distribution of each potential fault type; S33: determining potential fault types and the probability of occurrence of each potential fault type according to the predicted probability distribution generated by the output layer of the soft maximum activation function; S34: using a pre-established database of influencing factors of potential fault types, combining the potential fault types and the current operating status evaluation result, to evaluate the impact of each potential fault type; S35: generating a potential fault prediction result of the natural gas odorization device according to the potential fault type, the probability of occurrence of each potential fault type and the impact degree of each potential fault type; Wherein, step S40 specifically includes: S41: determining a faulty component in the natural gas odorization device that needs maintenance according to the fault type, occurrence probability and impact degree in the potential fault prediction result, and the difference between the actual value and the expected value of the performance parameter in the current operation status evaluation result; S42: Based on the faulty components requiring maintenance, combined with the operation plan of the natural gas odorization device and the maintenance resource availability information, recommend an optimal maintenance time and priority order; S43: Generate maintenance operation instructions according to the potential fault type and the faulty component that needs maintenance, combined with historical maintenance records, including a location method, a test method, a repair method, and safety precautions for the faulty component; S44: Listing tools, spare parts and materials required to perform maintenance according to the maintenance operation guide and the historical maintenance records; S45: Predicting the degree of improvement in the performance of the natural gas odorization device after maintenance based on the historical maintenance records, the current operating status evaluation results, and the maintenance plan; wherein the maintenance plan includes: a list of faulty parts requiring maintenance, the recommended maintenance time and priority, the maintenance operation instructions, and a list of tools, spare parts, and materials required to perform the maintenance.

2. The method according to claim 1, characterized in that The plurality of sensors include any of the following: The pressure sensor installed on the odorant delivery pipeline is used to measure the delivery pressure of the odorant from the storage tank to the odorizing device; A flow sensor installed on the odorant injection pipeline is used to measure the amount of odorant injected from the odorizing device to the natural gas main pipeline; A pressure sensor installed in the mixing chamber of the odorizing device, used to measure the mixing chamber pressure of the odorizing device; A liquid level sensor installed on the odorant storage tank is used to measure the liquid level of the odorant storage tank; A temperature sensor installed in the odorant storage tank is used to measure the temperature of the odorant storage tank; A vibration sensor installed on the main shell of the odorizing device is used to measure the vibration frequency of the main shell of the odorizing device; A gas concentration sensor installed on the odorized natural gas output pipeline is used to measure the odorant concentration in the odorized natural gas; A voltage sensor installed at the power input terminal of the odorizing device is used to measure the power supply voltage of the odorizing device; A valve position sensor installed on the control valve of the odorizing device is used to measure the opening of the control valve of the odorizing device; The time recording module integrated in the controller of the odorizing device is used to record the operating time of the odorizing device.

3. The method according to claim 1, characterized in that Step S33 specifically includes: S331: Setting a prediction probability threshold; S332: sorting the predicted probability distribution generated by the output layer of the soft maximum activation function to obtain a list of potential fault types and their corresponding predicted probabilities in descending order; S333: Selecting a potential fault type with a predicted probability greater than the predicted probability threshold from the potential fault type list arranged in descending order as a focus object; S334: For each potential fault type of the concerned object, calculate its relative probability, where the relative probability is the ratio of the predicted probability of the potential fault type to the sum of the predicted probabilities of the potential fault types of all concerned objects; S335: Based on the relative probability, classify the potential fault types of the object of concern into risk levels.

4. The method according to claim 1, characterized in that Step S34 specifically includes: S341: pre-establishing a potential fault type influencing factor database, wherein the potential fault type influencing factor database includes a plurality of potential fault types, each potential fault type corresponds to a plurality of influencing factors and their respective influencing degree weights; S342: Determine the current state value of each influencing factor based on the current operating state evaluation result; S343: For each potential fault type, calculate its comprehensive impact score, including: Obtaining the influencing factors corresponding to the potential fault type and their influence degree weights from the potential fault type influencing factor database; Multiply the influence weight of each influencing factor by the current state value of the influencing factor to obtain the single factor influence score; Add up all single factor impact scores to get the comprehensive impact score of the potential fault type; S344: according to a preset impact threshold, the comprehensive impact score of each potential fault type is divided into three levels: high, medium and low; S345: Generate an impact assessment result list, wherein the impact assessment result list includes potential fault types, the comprehensive impact degree scores, and levels corresponding to the comprehensive impact degree scores.

5. The method according to claim 1, characterized in that The method further comprises step S50: The maintenance operation in the maintenance suggestion is performed, and the maintenance result is fed back to the predictive maintenance model for updating and optimizing the predictive maintenance model.

6. The method according to claim 5, characterized in that Step S50 specifically includes: The predictive maintenance model is adjusted based on the feedback maintenance results, including adjusting the time estimation parameters and the weights of relevant influencing factors; the time estimation parameters refer to the weights and bias parameters in the LSTM layer; the weights of the relevant influencing factors refer to the weight parameters in the fully connected layer.

7. An intelligent maintenance system for a natural gas odorization device, characterized in that: It includes: A data acquisition module, used to obtain the current operating parameters of the natural gas odorization device from multiple sensors in real time; A data processing module, used to optimize the current operating parameters to obtain an evaluation result of the current operating state of the natural gas odorization device; A fault prediction module is used to output a potential fault prediction result of the natural gas odorizing device based on a pre-established predictive maintenance model of the natural gas odorizing device and taking a current operating status evaluation result of the natural gas odorizing device as input; A maintenance suggestion generating module, configured to generate a maintenance suggestion for the potential fault according to the potential fault prediction result and the current operating state evaluation result when the potential fault prediction result indicates the existence of the potential fault; Wherein, the data processing module includes: A data cleaning unit, used for cleaning the current operating parameters of the natural gas odorizing device, wherein the data cleaning includes removing abnormal values ​​and noise to obtain the current operating parameters after cleaning; A standardization processing unit, used for performing standardization processing on the current operating parameters after cleaning to obtain standardized current operating parameters; A time series analysis unit, for analyzing the standardized current operating parameters by using a time series analysis method, and extracting trend characteristics and period characteristics of the standardized current operating parameters; the trend characteristics refer to changes in the standardized current operating parameters of the natural gas odorization device within a first time span, including a change direction, a trend slope, and a trend inflection point; the period characteristics refer to repetitive fluctuation behaviors of the standardized current operating parameters of the natural gas odorization device within a second time span, including multiple of a period length, an amplitude, a phase, a frequency, and a waveform, and the first time span is greater than the second time span; the time series analysis method adopts a seasonal and trend decomposition method STL based on local weighted regression; a deviation calculation unit, configured to calculate the degree of deviation of the standardized current operating parameters based on a preset normal operating parameter range; wherein, for each standardized current operating parameter, a standardized Euclidean distance between the standardized current operating parameter and a central value of the preset normal operating parameter range is calculated as a quantitative indicator of the degree of deviation; An evaluation result generating unit, configured to generate an evaluation result of the current operation state of the natural gas odorizing device by using a weighted summation method according to the trend characteristics, the period characteristics and the degree of deviation of the standardized current operation parameters, and three weight coefficients corresponding to the trend characteristics, the period characteristics and the degree of deviation respectively; Wherein, the fault prediction module includes: A feature extraction unit, used to extract features from the current running state evaluation result to obtain a current running feature vector; A deep learning model unit is used to input the current operation feature vector into a pre-trained predictive maintenance model, wherein the predictive maintenance model is a deep learning model having the following structure: a long short-term memory network (LSTM) layer, whose input is a sequence of running feature vectors of the most recent N time points including the current running feature vector, and whose output is a feature representation that captures the temporal dependency; A plurality of sequentially connected fully connected layers, specifically comprising: a first fully connected layer, whose input is the output feature representation of the LSTM layer; and one or more subsequent fully connected layers; each fully connected layer is composed of a plurality of neurons, and each neuron of each fully connected layer is connected to all neurons of an adjacent fully connected layer; the plurality of sequentially connected fully connected layers are used to perform nonlinear transformation on the input features and further feature extraction; A softmax activation function output layer, whose input is the output features of the last fully connected layer, is used to generate the predicted probability distribution of each potential fault type; a fault type determination unit, configured to determine potential fault types and the probability of occurrence of each potential fault type according to the predicted probability distribution generated by the soft maximum activation function output layer; An impact assessment unit, configured to assess the impact of each potential fault type by using a pre-established database of impact factors of potential fault types, in combination with the potential fault type and the current operating state assessment result; A prediction result generating unit, used for generating a potential fault prediction result of the natural gas odorizing device according to the potential fault type, the probability of occurrence of each potential fault type and the impact degree of each potential fault type; Wherein, the maintenance suggestion generation module includes: a faulty component determination unit, for determining a faulty component in the natural gas odorization device that needs maintenance according to the fault type, occurrence probability and impact degree in the potential fault prediction result, and the difference between the actual value and the expected value of the performance parameter in the current operation status evaluation result; A maintenance time recommendation unit, for recommending an optimal maintenance time and priority order based on the faulty components requiring maintenance, combined with an operation plan of the natural gas odorization device and maintenance resource availability information; A maintenance operation guide generating unit, configured to generate a maintenance operation guide according to the potential fault type and the faulty component requiring maintenance, in combination with historical maintenance records, which includes a location method, a test method, a repair method, and safety precautions for the faulty component; A tool spare parts list generating unit, used to list tools, spare parts and materials required to perform maintenance according to the maintenance operation instructions and the historical maintenance records; A performance improvement prediction unit is used to predict the performance improvement degree of the natural gas odorization device after maintenance based on the historical maintenance records, the current operating status evaluation results, and the maintenance plan; wherein the maintenance plan includes: a list of faulty parts that need maintenance, the recommended maintenance time and priority, the maintenance operation instructions, and a list of tools, spare parts and materials required to perform the maintenance.

Images