Pumping unit intelligent lubrication decision-making system and method based on multi-source data fusion

Abstract

The invention discloses a pumping unit intelligent lubrication decision making system and method based on multi-source data fusion, and the system obtains multi-source data, such as the rotating speed, the torque, the oil temperature and the oil pressure, and the viscosity, the pH value and the water content of lubricating oil, during the operation of a pumping unit through a multi-source data collection unit, and carries out the processing through a data transmission and preprocessing unit. Data features are deeply mined by means of an improved multi-layer perceptron feature extraction unit, feature fusion is carried out on data of an optimized Transform model fusion unit and a historical data storage unit, and a lubrication decision model construction unit constructs correlation models of lubricating oil viscosity, lubrication period and the like based on the fused features. And finally, the lubrication decision execution unit outputs and executes a decision instruction. The method executes the steps of data acquisition, processing, feature extraction, fusion, model construction and decision execution corresponding to each unit of the system. The problems of lubricating oil waste and insufficient lubrication are avoided, and the operation efficiency and the intelligent operation and maintenance level of the pumping unit in oilfield exploitation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of oil pumping unit lubrication decision-making, and in particular to an intelligent lubrication decision-making system and method for oil pumping units based on multi-source data fusion. Background Art

[0002] In the oil field, pumping units are core equipment, and their stable operation is crucial to ensuring crude oil production. With the expansion of oilfield production and the demand for intelligent development, traditional manual, empirical lubrication management and lubrication decision-making based on single data monitoring are no longer able to meet the efficient and accurate operation and maintenance requirements of modern oilfields. Pumping units operate under complex operating conditions, and their lubrication requirements are affected by multiple factors such as speed, torque, oil temperature, oil pressure, and the physical and chemical properties of the lubricant. This requires comprehensive multi-source data for scientific decision-making.

[0003] Existing pumping unit lubrication management technologies have numerous shortcomings. Firstly, at the data processing level, most systems rely on a single or limited set of data types, failing to fully capture the complex information present during pumping unit operation. For example, some systems only monitor speed or oil temperature, ignoring key parameters such as lubricant viscosity and pH. This leads to inaccurate assessments of the pumping unit's lubrication status. Furthermore, traditional data processing methods are limited in their ability to extract and fusion features, making it difficult to uncover potential connections between data and providing effective support for lubrication decision-making.

[0004] On the other hand, existing technologies lack intelligence and dynamic adaptability in lubrication decision-making. Decision-making methods based on fixed rules or empirical formulas are unable to adjust lubrication strategies in a timely manner based on the real-time operating conditions of the pumping unit and changes in lubricant status. This static decision-making model can easily lead to lubricant waste or insufficient lubrication, increasing the risk of equipment wear and reducing the service life and operating efficiency of the pumping unit. This makes it difficult to adapt to the development trend of intelligent oilfield production. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides an intelligent lubrication decision system and method for oil pumping units based on multi-source data fusion.

[0006] The technical solution adopted by the present invention is an intelligent lubrication decision system for oil pumping units based on multi-source data fusion, comprising:

[0007] Multi-source data acquisition unit, used to collect various data during the operation of the pumping unit, including the pumping unit's speed, torque, oil temperature, oil pressure, vibration data, and lubricating oil viscosity, pH, and water content data;

[0008] The data transmission and preprocessing unit transmits the data collected by the multi-source data acquisition unit, performs preliminary processing on the transmitted data, removes outliers and noise interference, and outputs the processed data to the improved multi-layer perceptron feature extraction unit;

[0009] The improved multi-layer perceptron feature extraction unit uses the improved multi-layer perceptron model to extract features from the data processed by the data transmission and preprocessing unit, and outputs feature vectors to the optimized Transformer model fusion unit through a multi-layer neural network structure with preset activation functions and weight parameters;

[0010] The Transformer model fusion unit is optimized. The optimized Transformer model is used to fuse the feature vector with the historical lubrication data features. This unit receives the historical lubrication data features, performs feature fusion through the self-attention mechanism and specific layer normalization operations, and outputs the fused features to the lubrication decision model construction unit.

[0011] The historical data storage unit is used to store various lubrication data and corresponding working condition data during the past operation of the pumping unit. This unit establishes a data interaction channel with the optimized Transformer model fusion unit and the lubrication decision model construction unit;

[0012] The lubrication decision model building unit builds a model for generating lubrication decisions based on the fusion features and historical data. This unit analyzes and processes the fusion features and historical data to build a lubrication decision model suitable for the current operating conditions of the pumping unit.

[0013] The lubrication decision execution unit generates specific lubrication decision instructions and transmits the instructions to the oil pumping unit lubrication equipment to perform corresponding lubrication operations. It is connected to the oil pumping unit lubrication equipment to send and execute lubrication decision instructions.

[0014] Furthermore, the improved multi-layer perceptron feature extraction unit, combined with the pumping unit vibration data V d The improved formula for feature extraction is: Among them, V d is the vibration data of the pumping unit; W v1 、W v2 b is the weight matrix set for vibration data feature extraction, and its dimension is determined according to the characteristics of vibration data and the number of neurons; v1 、b v2 、b v3 is the bias vector; σ is the preset activation function, a is an adjustable parameter; Represents a special matrix operation used to enhance the ability to extract complex features of vibration data, Yv The extracted feature vector of the pumping unit vibration is subsequently input into the optimized Transformer model fusion unit for comprehensive judgment of the lubrication status of the pumping unit.

[0015] Furthermore, the optimized Transformer model fusion unit considers the oil temperature T when fusing features. oil 、Oil pressure P oil The fusion formula of the features associated with lubrication decision is: F o =LN(MultiHeadAttention(Q o , K o , V o )+AddPositionEncoding(E o )), where Q o =W Qo E o , K o =W Ko E o , V o =W Vo E o , E o is an input vector containing the oil temperature, oil pressure data features and related historical data features; W Qo 、W Ko 、W Vo is the weight matrix set for the fusion of oil temperature and oil pressure data; MultiHeadAttention is a multi-head attention mechanism used to capture the correlation features between oil temperature, oil pressure data and other data; AddPositionEncoding is a position encoding function that adds position information to the input feature vector; LN is a layer normalization operation, F o is the fused feature vector related to oil temperature and oil pressure.

[0016] Furthermore, in the improved multi-layer perceptron feature extraction unit, the formula for further updating the weight matrices W1 and W2 is: i=1,2, where is the weight matrix before updating, is the updated weight matrix; α is the learning rate, which controls the step size of weight update; Loss is the loss function for the weight matrix gradient.

[0017] Furthermore, the formula for updating the weight of the attention mechanism by optimizing the Transformer model fusion unit is: j represents different attention heads, is the attention weight of the j-th head before updating, is the updated attention weight; γ is the learning rate, which is used to adjust the weight update amplitude; The attention loss function Lossatt is the attention weight of the j-th head The gradient of is calculated by back-propagating the loss of the model in the process of fusing pumping unit related features.

[0018] Furthermore, when the lubrication decision model building unit builds the model, the formula for the relationship between the pumping unit speed n, torque T and lubricating oil viscosity μ is: μ=f(n, T)=a1n 2 +a2T+a3, where a1, a2, and a3 are coefficients determined by regression analysis of historical data and the fusion features output by the optimized Transformer model fusion unit. This formula is used to determine the required viscosity of the lubricating oil based on the real-time speed and torque of the pumping unit when building a lubrication decision model and make lubrication decisions.

[0019] Furthermore, when the lubrication decision model building unit builds the model, the formula combining the relationship between the pH value of the lubricating oil, the water content w and the lubrication cycle t is: Among them, b1, b2, and b3 are parameters determined by a preset algorithm based on historical data and model fusion features. This formula is used to determine the lubrication cycle based on the pH and water content of the lubricating oil when constructing the lubrication decision model, providing a basis for the lubrication decision execution unit to determine the lubrication operation time interval.

[0020] The intelligent lubrication decision-making method for oil pumping units based on multi-source data fusion includes the following steps:

[0021] S1: Through the multi-source data acquisition unit, various sensors are used to collect data on the speed, torque, oil temperature, oil pressure, vibration, and viscosity, pH, and water content of the lubricating oil during operation;

[0022] S2: The data collected by the multi-source data acquisition unit is transmitted to the data transmission and pre-processing unit, which transmits the data and performs preliminary processing to remove outliers and noise interference;

[0023] S3: Input the data processed by the data transmission and preprocessing unit into the improved multi-layer perceptron feature extraction unit, use the improved multi-layer perceptron model to extract the potential features in the data, and generate a feature vector;

[0024] S4: Inputting the feature vector output by the improved multi-layer perceptron feature extraction unit and the historical lubrication data features provided by the historical data storage unit into the optimized Transformer model fusion unit, performing feature fusion using the optimized Transformer model, and obtaining fusion features;

[0025] S5: The lubrication decision model construction unit determines the model parameters by analyzing the fusion features output by the optimized Transformer model fusion unit and the historical data of the historical data storage unit, and constructs a lubrication decision model suitable for the current pumping unit operating condition;

[0026] S6: The lubrication decision execution unit generates a specific lubrication decision instruction based on the lubrication decision model constructed by the lubrication decision model construction unit;

[0027] S7: The lubrication decision execution unit transmits the generated lubrication decision instruction to the oil pumping unit lubrication device, and controls the lubrication device to perform corresponding lubrication operations.

[0028] Beneficial Effects: The present invention proposes an intelligent lubrication decision-making system and method for oil pumps based on multi-source data fusion. The system comprehensively acquires oil pump operation and lubricant status data through a multi-source data acquisition unit, covering equipment parameters such as speed, torque, oil temperature and oil pressure, as well as physical and chemical indicators such as lubricant viscosity, pH, and water content. Compared with the traditional single data monitoring mode, it can more completely reflect the lubrication needs of the oil pump. The improved multi-layer perceptron feature extraction unit and the optimized Transformer model fusion unit operate in coordination. The former deeply mines data features through a unique model formula, while the latter uses a self-attention mechanism and position encoding to effectively fuse multi-source features and capture long-term and short-term dependencies between data, breaking through the limitation of traditional data processing that is difficult to mine potential connections between data. In lubrication decision-making, the lubrication decision model construction unit uses fusion features and historical data, combined with specific formulas to accurately establish correlations such as speed torque and lubricant viscosity, lubricant pH water content and lubrication cycle, to generate dynamic and adaptive lubrication decisions, completely changing the traditional fixed rule decision-making model, avoiding lubricant waste and insufficient lubrication problems, reducing equipment wear, extending the service life of the pumping unit, and significantly improving the operating efficiency and intelligent operation and maintenance level of the pumping unit in oil field exploitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a diagram of the system unit composition of the present invention;

[0030] Figure 2 The figure is a flow chart of the method steps of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown in the figure, the intelligent lubrication decision system for oil pumping units based on multi-source data fusion includes:

[0033] A multi-source data acquisition unit is used to collect various data during the operation of the pumping unit, including the pumping unit's speed, torque, oil temperature, oil pressure, vibration data, and lubricating oil viscosity, pH, and water content data. This unit is connected to the pumping unit and lubricating oil storage device through multiple sensors to obtain comprehensive and real-time data;

[0034] Specifically, the multi-source data acquisition unit serves as the source of system data acquisition and uses a variety of high-precision sensors for data acquisition. In terms of pumping unit operation data acquisition, a speed sensor with an accuracy of ±0.1rpm is used to monitor the speed of the pumping unit, a torque sensor with a range of 0-5000N·m and an accuracy of ±1%FS is used to obtain torque data, a temperature sensor with a temperature range of -40℃-200℃ and an accuracy of ±0.5℃ is used to measure oil temperature, a pressure sensor with a pressure measurement range of 0-50MPa and an accuracy of ±0.25%FS is used to monitor oil pressure, and a vibration acceleration sensor can measure vibration data with a range of 0-50g (g is the acceleration of gravity) and an accuracy of ±2%. For lubricating oil data acquisition, a viscosity measurement range of 0-10000mm is provided. 2 The sensors include a viscometer with a measurement range of 0.500 rpm and an accuracy of ±1%, a pH sensor with a measurement range of pH 0-14 and an accuracy of ±0.01%, and a water content sensor that can accurately measure the water content of lubricating oil from 0-10% with an accuracy of ±0.1%. These sensors connect to the pumping unit and lubricating oil storage equipment via standardized interfaces, ensuring accurate and real-time data collection.

[0035] The significance of this unit's multi-dimensional data collection lies in its comprehensive reflection of the pumping unit's operating status and the actual condition of the lubricant. The pumping unit's speed and torque directly reflect its workload. High speed and high torque increase friction between equipment components, placing higher demands on the lubricant's anti-wear properties. Changes in oil temperature and pressure reflect the operating condition of the lubrication system. Excessively high oil temperature may indicate insufficient lubrication or heat dissipation issues, while abnormal oil pressure may indicate a clogged or leaky oil line. Lubricant parameters such as viscosity, pH, and water content determine the lubricating performance and service life of the lubricant. Improper viscosity can affect lubrication effectiveness, while imbalanced pH and increased water content can accelerate lubricant deterioration and equipment corrosion. Collecting this data provides a rich and critical information foundation for subsequent lubrication decisions, avoiding errors caused by missing data.

[0036] For example, during the operation of a pumping unit in an oil field, the multi-source data acquisition unit collected in real time the following data: the pumping unit speed suddenly increased from the normal 12rpm to 18rpm, the torque increased from 1500N·m to 2200N·m, the oil temperature rose from 50℃ to 65℃, and the lubricating oil viscosity increased from the original 150mm 2 / s down to 120mm 2The timely acquisition of this data allows the system to sense increases in the pumping unit's workload and changes in lubricant performance, providing a basis for subsequent adjustments to the lubrication strategy and avoiding equipment damage caused by untimely lubrication or mismatched lubricants.

[0037] The data transmission and preprocessing unit transmits the data collected by the multi-source data acquisition unit and performs preliminary processing on the transmitted data to remove outliers and noise interference so that the data meets the requirements of subsequent processing. This unit is connected to the multi-source data acquisition unit through a wired or wireless communication link and outputs the processed data to the improved multi-layer perceptron feature extraction unit;

[0038] Specifically, the data transmission and preprocessing unit undertakes the dual tasks of data transmission and preliminary processing. In terms of data transmission, multiple communication methods are supported, including wired (such as Ethernet, RS485) and wireless (such as 4G, 5G, Wi-Fi). For pumping units that are closer to the control center, Ethernet is used for data transmission, and its transmission rate can reach 100Mbps or even higher, which can quickly and stably transmit large amounts of data; for pumping units in remote areas, remote data transmission is achieved with the help of 4G or 5G networks to ensure real-time data. In the data preprocessing link, methods based on statistical principles are used to detect and remove outliers. For example, the 3σ principle is used to determine whether the data is abnormal. Data that deviates from the mean by more than 3 times the standard deviation is regarded as an outlier and eliminated; digital filtering techniques such as median filtering and mean filtering are used to remove noise interference in the data to ensure the quality of data input to subsequent units.

[0039] This unit is crucial for data transmission and preprocessing. Stable data transmission ensures that data acquired by the multi-source data acquisition unit reaches subsequent processing units promptly and accurately, preventing transmission delays or data loss from impacting the system's response speed and decision-making accuracy. Effective data preprocessing can remove interference from the data, ensuring that the data more accurately reflects the actual state of the pumping unit and lubricating oil. Without preprocessing, outliers can cause deviations in subsequent feature extraction, and noise interference can affect the model's accurate identification of data features, leading to errors in lubrication decisions. Preprocessed data can provide more reliable input for the improved multi-layer perceptron feature extraction unit, improving the stability and reliability of the entire system.

[0040] For example, during a data collection process for a pumping unit, the data transmission and preprocessing unit received data from a multi-source data acquisition unit. The oil pressure data showed an individual value of -5 MPa, which clearly did not conform to the actual situation (the normal oil pressure range is 0-50 MPa). This value was identified as an outlier using the 3σ principle and removed. Furthermore, the vibration data contained high-frequency noise fluctuations, which were processed using a median filter to smooth the data curve. This processed data more accurately reflected the actual operating status of the pumping unit, providing a reliable data foundation for subsequent feature extraction and lubrication decision-making, avoiding misjudgments caused by abnormal data and noise.

[0041] The improved multi-layer perceptron feature extraction unit uses the improved multi-layer perceptron model to extract features from the data processed by the data transmission and preprocessing unit and mine the potential feature information in the data. The unit inputs the data output by the data transmission and preprocessing unit and outputs the feature vector to the optimized Transformer model fusion unit through a multi-layer neural network structure with preset activation functions and weight parameters.

[0042] Specifically, the improved multi-layer perceptron feature extraction unit is the core unit for deep feature mining of data. This unit adopts a multi-layer neural network structure, which usually includes an input layer, multiple hidden layers and an output layer. The number of neurons in the input layer is determined according to the data dimension output by the data transmission and preprocessing unit. For example, if the input data contains 10 dimensions (such as multi-source data such as speed, torque, oil temperature, etc.), the input layer has 10 neurons. The number of hidden layers and the number of neurons have been optimized. Generally, 3-5 hidden layers are set, and the number of neurons in each hidden layer is between 20-50. The optimal structure is determined through multiple experiments and parameter adjustment. Each neuron is connected by a weight. The initial value of the weight adopts a random initialization method and is adjusted through the back-propagation algorithm during the training process. In terms of activation function, an improved nonlinear activation function is used to enhance the model's ability to extract complex data features.

[0043] The significance of this unit is to extract potential features that are valuable for lubrication decision-making from preprocessed data. Although the original multi-source data contains rich information, this information is relatively scattered and complex, making it difficult to use directly for lubrication decision-making. Through the improved multi-layer perceptron model, the data can be abstracted and features extracted layer by layer, converting low-level raw data into high-level, representative feature vectors. These feature vectors contain the inherent connections and patterns between the data. For example, they can extract the potential relationship between speed and torque changes and lubricant viscosity requirements, as well as the correlation characteristics between oil temperature, oil pressure and lubricant performance changes. The extracted feature vectors provide more valuable input for the subsequent optimized Transformer model fusion unit, which helps to improve the accuracy and effectiveness of lubrication decisions.

[0044] Taking the operating data of a particular oil pumping unit as an example, the data transmission and preprocessing unit outputs data such as speed, torque, oil temperature, and lubricant viscosity, which then enters the improved multi-layer perceptron feature extraction unit. After processing by the multi-layer neural network, this unit extracts a set of feature vectors. These feature vectors indicate that when the oil pumping unit's speed is between 10-15 rpm, the torque is between 1000-1800 N·m, and the oil temperature rises, the lubricant viscosity must remain within a certain range to meet lubrication requirements. Based on these extracted features, the system can more accurately determine the oil pumping unit's lubrication needs under current operating conditions, providing strong support for making appropriate lubrication decisions.

[0045] The Transformer model fusion unit is optimized. The optimized Transformer model is used to fuse the feature vectors extracted by the improved multi-layer perceptron feature extraction unit with the historical lubrication data features to capture the long-term and short-term dependencies between the data. This unit receives the output of the improved multi-layer perceptron feature extraction unit and the historical lubrication data features provided by the historical data storage unit, performs feature fusion through self-attention mechanism and specific layer normalization operations, and outputs the fused features to the lubrication decision model construction unit.

[0046] Specifically, the optimized Transformer model fusion unit is mainly responsible for fusing the feature vectors extracted by the improved multi-layer perceptron feature extraction unit with the historical lubrication data features. This unit is optimized based on the Transformer architecture and introduces a multi-head attention mechanism. Typically, 8-12 attention heads are set, each of which can capture the relationship between data features from different angles. In terms of layer structure, it contains multiple encoding layers, each of which consists of a multi-head attention sublayer and a feedforward neural network sublayer. The data is standardized through layer normalization operations to accelerate model training and improve stability. During the fusion process, the currently extracted feature vector is spliced ​​with the historical lubrication data features stored in the historical data storage unit as the input of the model. The attention weights between different features are calculated through the self-attention mechanism to achieve effective fusion of features from different sources.

[0047] The importance of this unit lies in its ability to fully utilize information from historical and current data and capture long-term and short-term dependencies between data. The lubrication status of the pumping unit is not only related to the current operating conditions and lubricant status, but also to historical operating conditions and lubrication records. By optimizing the Transformer model fusion unit, the features extracted under the current operating conditions can be fused with the lubrication data features under similar historical conditions to uncover a more comprehensive connection between the data. For example, by analyzing the performance change trends of lubricating oil and the effects of corresponding lubrication measures under the same speed, torque, and oil temperature conditions in historical data, and combining them with current features, the current lubrication needs can be judged more accurately, avoiding one-sided decisions caused by only considering current data and ignoring historical experience, and providing a more comprehensive and reliable basis for lubrication decisions.

[0048] For example, during one operation of a certain oil pump, the current feature vector extracted by the improved multi-layer perceptron feature extraction unit shows that the oil pump speed, torque, and oil temperature are within a specific range, and the viscosity of the lubricating oil has slightly decreased. The optimized Transformer model fusion unit fuses this feature vector with the historical lubrication data features under similar operating conditions in the historical data storage unit. Through multi-head attention mechanism analysis, it was found that under similar historical operating conditions, if the lubricating oil is not adjusted or the lubrication frequency is not increased in a timely manner, the equipment will subsequently suffer from increased wear. Based on the fused feature information, the system can make more reasonable lubrication decisions in advance, such as timely replenishing lubricating oil with appropriate viscosity or shortening the lubrication cycle, to ensure the normal operation of the oil pump and reduce the risk of equipment failure.

[0049] The historical data storage unit is used to store various lubrication data and corresponding working condition data during the past operation of the pumping unit, providing historical data support for the optimized Transformer model fusion unit and the lubrication decision model construction unit. This unit establishes a data interaction channel with the optimized Transformer model fusion unit and the lubrication decision model construction unit to provide the required historical data at any time;

[0050] Specifically, the historical data storage unit is used to store various types of lubrication data and corresponding working condition data during the past operation of the pumping unit. The unit uses a large-capacity database for storage, such as a relational database (MySQL, Oracle) or a non-relational database (MongoDB). The stored data includes historical operation data collected by the multi-source data acquisition unit (such as speed, torque, oil temperature, oil pressure, and vibration data at different time points), lubricating oil historical data (such as viscosity, pH, and water content change data over time), and records of each lubrication operation (such as lubrication time, lubricating oil model, lubrication method, etc.). Data storage is classified and managed according to time series and pumping unit number to facilitate quick query and call. At the same time, in order to ensure the security and integrity of the data, a data backup and recovery mechanism is adopted to back up the data regularly, and strict access control is set to prevent data loss and illegal access.

[0051] The storage of historical data in this unit is of great significance. Historical data is the basis for system learning and optimization. By analyzing and mining historical data, it is possible to discover the patterns and trends in the operation of the pumping unit, as well as the characteristics of lubricant performance changes and lubrication requirements under different operating conditions. For example, by analyzing historical data, it is possible to understand the changing patterns of the operating conditions and lubricant status of the pumping unit in different seasons and different mining stages, providing a reference for formulating more reasonable lubrication strategies. In addition, historical data can also be used to optimize the feature fusion process of the Transformer model fusion unit and the model training of the lubrication decision model construction unit, so that the system can continuously adapt to different operating conditions and environmental changes, and improve the accuracy and adaptability of lubrication decisions.

[0052] For example, a historical data storage unit for multiple pumping units in an oil field records years of operational data and lubrication records. During the hot summer months, historical data queries revealed that the viscosity of the lubricating oil in many pumping units decreased rapidly as the oil temperature rose, and that equipment components had worn out due to untimely lubrication. When the following summer arrived, the system, while processing the current pumping unit data, combined with historical data, could predict the impact of rising oil temperatures on lubricant performance and promptly adjust lubrication decisions, such as increasing lubricant cooling measures or replacing lubricants with better high-temperature performance. This prevented equipment failures that could have occurred without consulting historical data, improving the operational reliability and maintenance efficiency of the pumping units.

[0053] The lubrication decision model construction unit constructs a model for generating lubrication decisions based on the fusion features output by the optimized Transformer model fusion unit and the historical data of the historical data storage unit. This unit analyzes and processes the fusion features and historical data to determine model parameters, constructs a lubrication decision model suitable for the current operating conditions of the pumping unit, and outputs the model to the lubrication decision execution unit.

[0054] Specifically, the lubrication decision model construction unit constructs a model for generating lubrication decisions based on the fused features output by the optimized Transformer model fusion unit and the historical data from the historical data storage unit. This unit utilizes regression analysis and classification algorithms in machine learning to construct the model. In terms of regression analysis, a multivariate linear regression model or nonlinear regression model is established for continuous parameters such as lubricant viscosity and lubrication cycle. The model parameters are determined by training the fused features and relevant parameters from the historical data. For example, a nonlinear regression model is established for lubricant viscosity and parameters such as pump speed, torque, and oil temperature. Through training with a large amount of historical data, the functional relationship between each parameter and lubricant viscosity is found. In terms of classification algorithms, for classification problems such as lubrication methods (such as manual lubrication and automatic lubrication) and lubricant model selection, algorithms such as decision trees and support vector machines are used to construct classification models. The model is trained based on the fused features and historical data to determine the classification rules.

[0055] The significance of constructing a lubrication decision model in this unit lies in converting the fused feature information into specific lubrication decisions. Lubrication decisions for a pumping unit involve multiple aspects, including lubricant selection, lubrication intervals, and lubrication methods, requiring a comprehensive consideration of current operating conditions and historical experience. The constructed lubrication decision model can quickly and accurately calculate appropriate lubricant viscosity, recommended lubrication cycles, and optimal lubrication methods based on the input fused features. These results can meet the lubrication needs of the pumping unit under different operating conditions, avoiding irrational lubrication decisions caused by the subjectivity and limitations of human experience, improving the lubrication effectiveness and operating efficiency of the pumping unit, and reducing equipment maintenance costs.

[0056] For example, when the fusion features output by the optimized Transformer model fusion unit show that the current speed of the pumping unit is 15 rpm, the torque is 2000 N·m, the oil temperature is 60°C, and the lubricating oil viscosity has dropped to the critical value, the lubrication decision model construction unit calculates based on the trained model. Through the lubricating oil viscosity regression model, it is determined that the appropriate lubricating oil viscosity under the current working conditions is 180 mm 2 / s; using the lubrication cycle classification model, it determines that the lubrication cycle should be shortened to once every 8 hours; using the lubrication method classification model, it recommends automatic lubrication to ensure timely and accurate lubrication. Based on these decision results, the system can generate specific lubrication decision instructions, providing a scientific and reasonable lubrication plan for the pumping unit to ensure normal operation of the equipment.

[0057] The lubrication decision execution unit generates specific lubrication decision instructions based on the lubrication decision model constructed by the lubrication decision model construction unit, and transmits the instructions to the oil pump lubrication equipment to perform corresponding lubrication operations. The unit is connected to the lubrication decision model construction unit to receive the decision model, and is connected to the oil pump lubrication equipment to send and execute lubrication decision instructions.

[0058] Specifically, the lubrication decision execution unit, as the final link in the system, generates specific lubrication decision instructions based on the lubrication decision model constructed by the lubrication decision model construction unit and transmits these instructions to the pumping unit's lubrication equipment to execute the corresponding lubrication operation. This unit is connected to the lubrication decision model construction unit via a communication interface and receives the decision results output by the model, such as lubricant viscosity, lubrication cycle, and lubrication method. Different execution methods are used for different lubrication decisions. If the decision is to replace the lubricant, the valves and oil pumps in the lubricant delivery pipeline are controlled to drain the existing lubricant and inject new lubricant that meets the requirements. If the decision is to adjust the lubrication cycle, the lubrication equipment's timer is set and lubrication operations are performed according to the new cycle. If the decision is to change the lubrication method, such as switching from manual lubrication to automatic lubrication, the lubrication equipment's drive device and control system are controlled to make corresponding adjustments.

[0059] This unit's role is to translate lubrication decisions from theory into practical operations, directly impacting the pump's lubrication effectiveness. Accurate and timely execution of lubrication decision instructions ensures effective lubrication of the pump at the appropriate time and using the appropriate method, reducing friction and wear between equipment components and extending equipment life. Conversely, if the execution unit malfunctions or is inaccurate, even the most reasonable decisions will fail to achieve the desired lubrication effect, potentially leading to decreased equipment performance or even failure. Therefore, the lubrication decision execution unit must be highly reliable and accurate, capable of stably executing various lubrication decision instructions and providing real-time monitoring and feedback on the execution process.

[0060] For example, when the decision instruction generated by the lubrication decision model building unit is to automatically lubricate the pumping unit after 1 hour and replace the lubricating oil with a viscosity of 200mm 2 / s of new lubricant, the lubrication decision execution unit receives the instruction and first starts the timer to count down. After 1 hour, the execution unit controls the valve of the lubricant delivery pipeline to open and discharge the original lubricant into the recovery container, and at the same time starts the oil pump to pump the new 200mm 2 Lubricating oil with a viscosity of 100 / s is injected into the pumping unit's lubrication system. During this process, the execution unit monitors parameters such as the injection volume and pressure in real time to ensure accurate and sufficient injection of lubricating oil. This operation ensures timely and effective lubrication of the pumping unit, meeting its lubrication needs under current operating conditions and ensuring stable operation of the equipment.

[0061] Preferably, in the improved multilayer perceptron feature extraction unit, the improved multilayer perceptron model formula is: Y=σ((W1X+b1)⊙(W2X+b2)+b3), wherein X is the input data, i.e., the data transmitted and output by the preprocessing unit; W1 and W2 are weight matrices of different layers, whose dimensions are set according to the input data dimension and the number of neurons, and are used to perform weighted transformation on the input data; b1, b2, and b3 are bias vectors used to adjust the activation threshold of neurons; σ is a custom activation function, a is an adjustable parameter used to control the slope of the activation function; ⊙ represents the element-by-element multiplication operation, which enhances the model's ability to extract the interaction between different features. Y is the final extracted feature vector.

[0062] Specifically, in the intelligent lubrication system of the oil pump, this formula processes multi-source data (such as oil pump speed, torque, lubricant pH, etc.) output by the data transmission and preprocessing unit. The parameters in the formula are set according to the actual working conditions and data characteristics. For example, the dimension of the weight matrix is ​​determined according to the data dimension and the number of neurons, and the adjustment parameters of the activation function are selected through multiple experiments to obtain the optimal value. Its significance lies in the deep mining of the interactive relationship between data. Compared with the traditional multi-layer perceptron, it can more accurately extract key features related to lubricant performance under high-speed and high-torque conditions. For example, in an oil field, when the oil pump speed increases from 10rpm to 15rpm and the torque increases from 1200N·m to 1800N·m, the feature vector extracted by this formula provides an accurate basis for the subsequent judgment of the need for lubricant viscosity adjustment, thereby avoiding equipment wear due to insufficient lubrication.

[0063] Preferably, in the optimized Transformer model fusion unit, the formula used by the optimized Transformer model when fusing features is: F = LN (MultiHeadAttention (Q, K, V) + AddPositionEncoding (E)), where Q = W Q E, K = W K E, V = W V E, E are the input feature vectors, including the feature vectors output by the improved multi-layer perceptron feature extraction unit and the historical lubrication data features provided by the historical data storage unit; W Q 、W K 、W Vis a weight matrix used to generate the query vector Q, key vector K, and value vector V; MultiHeadAttention is a multi-head attention mechanism that calculates attention in parallel through multiple different heads to capture different aspects of the data; AddPositionEncoding is a position encoding function that adds position information to the input feature vector to distinguish features at different positions; LN is a layer normalization operation that normalizes the data to make model training more stable, and F is the fused feature vector.

[0064] Specifically, during the operation of the oil pump, the formula fuses the feature vectors extracted by the improved multi-layer perceptron (including real-time speed, oil temperature, and other features) with the historical lubrication data features of the historical data storage unit. Parameters such as the number of heads and position encoding method of the multi-head attention mechanism are optimized to adapt to the characteristics of the oil pump data. Its core significance lies in capturing the long-term and short-term dependencies between data. For example, it can analyze the correlation between oil temperature changes and the decrease in lubricating oil viscosity under continuous high-temperature operation, as well as the lubrication measures under similar historical working conditions. In actual applications, when the continuous operation of a certain oil pump causes the oil temperature to continue to rise, the formula integrates historical data of similar high-temperature working conditions to predict changes in lubricating oil performance in advance, providing support for the system to make decisions to replenish high-viscosity lubricating oil in advance, thereby ensuring stable operation of the equipment.

[0065] Preferably, in the improved multi-layer perceptron feature extraction unit, the formula for further updating the weight matrices W1 and W2 is: i=1,2, where is the weight matrix before updating, is the updated weight matrix; α is the learning rate, which controls the step size of weight update; Loss is the loss function for the weight matrix The gradient of is calculated by the back propagation algorithm, so that the improved multi-layer perceptron model can continuously optimize the weights and improve the accuracy of feature extraction when extracting the features of the pumping unit operation data and lubricating oil related data.

[0066] Specifically, in the context of the massive amounts of data generated by long-term pumping unit operation, this formula dynamically updates the weight matrix using a backpropagation algorithm, calculating gradients based on the loss function. The learning rate is adjusted based on data trends and model training results to ensure the rationality of weight updates. This is crucial because it enables the model to adapt to dynamic changes in pumping unit operating conditions and lubricant status, continuously optimizing feature extraction capabilities. For example, when oilfield production fluctuates seasonally, pumping unit operating parameters differ significantly between winter and summer. By updating weights using this formula, the model can accurately extract lubricant demand characteristics for different seasons, avoiding seasonal factors that could lead to inaccurate lubrication decisions.

[0067] Preferably, in the optimized Transformer model fusion unit, the formula for updating the weight of the attention mechanism is: j represents different attention heads, is the attention weight of the j-th head before updating, is the updated attention weight; γ is the learning rate, which is used to adjust the weight update amplitude; The attention loss function Lossatt is the attention weight of the j-th head The gradient of the model is back-propagated in the process of fusing the relevant features of the pumping unit to optimize the attention weight, so that the optimized Transformer model can better fuse the feature information from different sources and improve the accuracy of judging the working condition and lubrication status of the pumping unit.

[0068] Specifically, when processing complex multi-source data fusion for oil pumping units, this formula calculates the gradient based on the attention loss function and updates the weights of each attention head. The learning rate is fine-tuned based on the model fusion results to ensure that the weight adjustments align with the changing patterns of data features. This allows the optimized Transformer model to better focus on integrating key features and improve the accuracy of determining the operating conditions of the oil pumping units. For example, if an oil pumping unit experiences abnormal vibration and the water content of the lubricating oil increases, this formula updates the attention weights, directing the model's attention to the correlation between the vibration and the change in the water content of the lubricating oil. This helps the system promptly identify equipment lubrication anomalies and initiate maintenance measures.

[0069] Preferably, when the lubrication decision model building unit builds the model, the formula for the relationship between the pumping unit speed n, torque T and lubricating oil viscosity μ is: μ=f(n, T)=a1n 2 +a2T+a3, where a1, a2, and a3 are coefficients determined by regression analysis of historical data and the fusion features output by the optimized Transformer model fusion unit. This formula is used to determine the required viscosity of the lubricating oil based on the real-time speed and torque of the pumping unit when building a lubrication decision model, so as to make lubrication decisions and ensure that the pumping unit is properly lubricated under different operating conditions.

[0070] Specifically, a formula for the relationship between pumping unit speed, torque, and lubricant viscosity is constructed and applied to the lubrication decision model construction unit. The coefficients in the formula are determined through regression analysis of a large amount of historical data and the features after the optimization Transformer model is integrated, fully considering the actual needs under different operating conditions. In actual pumping unit operation, different speed and torque combinations correspond to different friction conditions. This formula can accurately calculate the required lubricant viscosity. For example, when the pumping unit is in a heavy-load condition with high speed and high torque, this formula calculates that high-viscosity lubricant is required to enhance the lubrication effect. The system then decides to replace the lubricant with the appropriate viscosity, effectively reducing equipment wear and improving operating efficiency.

[0071] Preferably, when the lubrication decision model building unit builds the model, the formula combining the relationship between the pH value of the lubricating oil, the water content w and the lubrication cycle t is: Among them, b1, b2, and b3 are parameters determined by a preset algorithm based on historical data and model fusion features. This formula is used to determine the lubrication cycle based on the pH and water content of the lubricating oil when constructing the lubrication decision model, providing a basis for the lubrication decision execution unit to determine the lubrication operation time interval.

[0072] Specifically, a formula was established to correlate the pH value, water content, and lubrication cycle of the lubricating oil, which was then used to construct the lubrication decision model. The formula parameters were determined by analyzing historical data showing the relationship between changes in the lubricating oil's physical and chemical properties and the equipment's lubrication status. During the operation of the pumping unit, changes in the lubricating oil's pH value and water content can affect its lubrication performance and service life. This formula can scientifically calculate a reasonable lubrication cycle based on the lubricating oil's current pH value and water content. For example, if an imbalance in the lubricating oil's pH value and an increase in its water content are detected, the formula indicates that the lubrication cycle should be shortened. The system then promptly adjusts the lubrication strategy to avoid poor equipment lubrication due to decreased lubricating oil performance, thereby extending the service life of both the equipment and the lubricating oil.

[0073] Preferably, in the improved multi-layer perceptron feature extraction unit, the vibration data V of the pumping unit is combined with the d The improved formula for feature extraction is: Among them, V d is the vibration data of the pumping unit; W v1 、W v2 The weight matrix set for vibration data feature extraction, whose dimension is determined according to the characteristics of vibration data and the number of neurons; b v1 、b v2 、b v3 is the bias vector; σ is the preset activation function, a is an adjustable parameter; Represents matrix operations, which are used to enhance the ability to extract complex features of vibration data. vThe extracted feature vector of the pumping unit vibration is subsequently input into the optimized Transformer model fusion unit for comprehensive judgment of the lubrication status of the pumping unit.

[0074] Specifically, vibration data contains information about equipment failures and lubrication status. The formula uses specialized operations and parameter settings specifically designed to extract complex features from vibration data. For example, specific operations are used to capture the correlation between changes in vibration frequency and amplitude, equipment component wear, and lubrication status. In practice, when a pumping unit experiences abnormal vibration, the vibration feature vector extracted by this formula helps the system determine whether insufficient lubrication is causing increased component wear. This allows for timely adjustments to lubrication decisions, such as increasing lubrication frequency or changing lubricating oil, to prevent equipment failure.

[0075] Preferably, the optimized Transformer model fusion unit considers the oil temperature T when fusing features. oil 、Oil pressure P oil The fusion formula of the features associated with lubrication decision is: F o =LN(MultiHeadAttention(Q o , K o , V o )+AddPositionEncoding(E o )), where Q o =W Qo E o , K o =W Ko E o , V o =W Vo E o , E o is an input vector containing the oil temperature, oil pressure data features and related historical data features; W Qo 、W Ko 、W Vo is the weight matrix set for the fusion of oil temperature and oil pressure data; MultiHeadAttention is a multi-head attention mechanism used to capture the correlation features between oil temperature, oil pressure data and other data; AddPositionEncoding is a position encoding function that adds position information to the input feature vector; LN is a layer normalization operation, F o It is the fused feature vector related to oil temperature and oil pressure. The feature vector is used by the lubrication decision model construction unit to build a more accurate lubrication decision model.

[0076] Specifically, the fusion formula of the Transformer model fusion unit is optimized for processing the correlation features of oil temperature, oil pressure and lubrication decision-making. In the oil pump lubrication system, oil temperature and oil pressure directly reflect the lubrication status. The formula accurately integrates oil temperature and oil pressure data with other operating conditions and historical data features by optimizing multi-head attention mechanisms and position encoding operations. For example, when the oil temperature rises and the oil pressure drops, the formula integrates the lubrication measures under similar abnormal operating conditions in the historical data, helping the system to quickly determine whether there may be problems with oil circuit blockage or deterioration of lubricating oil performance, and then generates accurate lubrication decision instructions such as checking the oil circuit and replacing the lubricating oil to ensure the normal operation of the oil pump lubrication system.

[0077] like Figure 2 As shown in FIG, the intelligent lubrication decision-making method for the oil pumping unit based on multi-source data fusion includes the following steps:

[0078] S1: Through the multi-source data acquisition unit, various sensors are used to collect multi-source data such as speed, torque, oil temperature, oil pressure, vibration data, and viscosity, pH, and water content of the lubricating oil during operation;

[0079] S2: The data collected by the multi-source data acquisition unit is transmitted to the data transmission and pre-processing unit, which performs preliminary processing such as data transmission and removal of outliers and noise interference;

[0080] S3: Input the data processed by the data transmission and preprocessing unit into the improved multi-layer perceptron feature extraction unit, use the improved multi-layer perceptron model to extract the potential features in the data, and generate a feature vector;

[0081] S4: Inputting the feature vector output by the improved multi-layer perceptron feature extraction unit and the historical lubrication data features provided by the historical data storage unit into the optimized Transformer model fusion unit, performing feature fusion using the optimized Transformer model, and obtaining fusion features;

[0082] S5: The lubrication decision model construction unit determines the model parameters by analyzing the fusion features output by the optimized Transformer model fusion unit and the historical data of the historical data storage unit, and constructs a lubrication decision model suitable for the current pumping unit operating condition;

[0083] S6: The lubrication decision execution unit generates a specific lubrication decision instruction based on the lubrication decision model constructed by the lubrication decision model construction unit;

[0084] S7: The lubrication decision execution unit transmits the generated lubrication decision instruction to the oil pumping unit lubrication device, and controls the lubrication device to perform corresponding lubrication operations.

[0085] In terms of data processing, traditional technologies rely on a single or small amount of data, making it difficult to fully reflect the lubrication needs of the pumping unit. However, the multi-source data acquisition unit of this system uses a variety of high-precision sensors to collect equipment data such as speed, torque, oil temperature, oil pressure, and other equipment data during the operation of the pumping unit, as well as physical and chemical indicators such as lubricating oil viscosity, pH, and water content, to achieve comprehensive acquisition of multi-source data. The data transmission and preprocessing unit ensures accurate and real-time data transmission, removes outliers and noise, and provides reliable data for subsequent processing. The improved multi-layer perceptron feature extraction unit and the optimized Transformer model fusion unit work together. The former deeply mines the potential features of the data, while the latter effectively fuses multi-source features and captures the long-term and short-term dependencies between data. Compared with traditional data processing methods, it can more accurately extract and utilize data value, solving the problem of insufficient data utilization.

[0086] In the lubrication decision-making process, traditional technologies are based on fixed rules or empirical formulas, lack dynamic adaptability, and can easily lead to lubricant waste or insufficient lubrication. In the present invention, the lubrication decision model construction unit is based on the fusion features output by the optimized Transformer model fusion unit and the historical data of the historical data storage unit, combined with specific formulas to establish correlations such as speed torque and lubricant viscosity, lubricant pH water content and lubrication cycle. The lubrication decision model constructed through regression analysis and classification algorithms can dynamically generate accurate lubrication decisions based on the real-time operating conditions of the pumping unit and changes in the lubricating oil state. For example, when changes in the pumping unit speed and torque and a decrease in lubricating oil viscosity are detected, the model can quickly calculate the appropriate lubricating oil viscosity, recommended lubrication cycle and optimal lubrication method. The lubrication decision execution unit accurately transmits the decision instructions to the pumping unit lubrication equipment for execution, ensuring that the pumping unit can be scientifically and reasonably lubricated under different operating conditions, completely changing the limitations of the traditional static decision-making model.

[0087] In addition, the historical data storage unit in the system provides rich historical experience for model training and optimization. By analyzing and learning from historical data, the system can continuously optimize model parameters and improve its adaptability to different working conditions and environmental changes. At the same time, the improved multi-layer perceptron feature extraction unit and the optimized parameter update mechanism in the Transformer model fusion unit enable the model to dynamically adjust according to new data, further improving the accuracy and reliability of the system. The intelligent design of the entire process from data collection and processing to decision execution enables this system and method to efficiently and accurately meet the lubrication needs of the pumping unit, reduce the risk of equipment wear, improve oilfield production efficiency and economic benefits, and provide strong technical support for the intelligent operation and maintenance of the pumping unit.

[0088] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0089] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion is characterized by: include: Multi-source data acquisition unit, data transmission and preprocessing unit, improved multi-layer perceptron feature extraction unit, optimized Transformer model fusion unit, historical data storage unit, lubrication decision model construction unit, lubrication decision execution unit; The multi-source data acquisition unit is used to collect various data during the operation of the pumping unit, including the speed, torque, oil temperature, oil pressure, vibration data of the pumping unit and the viscosity, pH and water content data of the lubricating oil; The data transmission and preprocessing unit transmits the data collected by the multi-source data acquisition unit, performs preliminary processing on the transmitted data, removes abnormal values ​​and noise interference, and outputs the processed data to the improved multi-layer perceptron feature extraction unit; The improved multi-layer perceptron feature extraction unit utilizes an improved multi-layer perceptron model to extract features from the data processed by the data transmission and preprocessing unit, and outputs feature vectors to the optimized Transformer model fusion unit through a multi-layer neural network structure with preset activation functions and weight parameters.

2. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 1 is characterized in that: The optimized Transformer model fusion unit adopts the optimized Transformer model to fuse the feature vector with the historical lubrication data features. The unit receives the historical lubrication data features, performs feature fusion through a self-attention mechanism and a specific layer normalization operation, and outputs the fused features to the lubrication decision model construction unit. The historical data storage unit is used to store various types of lubrication data and corresponding working condition data during the past operation of the pumping unit. This unit establishes a data interaction channel with the optimized Transformer model fusion unit and the lubrication decision model construction unit; The lubrication decision model construction unit constructs a model for generating lubrication decisions based on the fusion features and historical data. The unit constructs a lubrication decision model suitable for the current working condition of the pumping unit by analyzing and processing the fusion features and historical data. The lubrication decision execution unit generates a specific lubrication decision instruction and transmits the instruction to the oil pumping unit lubrication device to execute the corresponding lubrication operation, and is connected to the oil pumping unit lubrication device to send and execute the lubrication decision instruction; The improved multi-layer perceptron feature extraction unit, combined with the pumping unit vibration data V d The improved formula for feature extraction is: Among them, V d is the vibration data of the pumping unit; W v1 、W v2 b is the weight matrix set for vibration data feature extraction, and its dimension is determined according to the characteristics of vibration data and the number of neurons; v1 、b v2 、b v3 is the bias vector; σ is the preset activation function, a is an adjustable parameter; Represents a special matrix operation used to enhance the ability to extract complex features of vibration data, Y v The extracted feature vector of the pumping unit vibration is subsequently input into the optimized Transformer model fusion unit for comprehensive judgment of the lubrication status of the pumping unit.

3. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 1 is characterized in that: The optimized Transformer model fusion unit considers the oil temperature T when fusing features. oil 、Oil pressure P oil The fusion formula of the features associated with lubrication decision is: F o =LN(MultiHeadAttention(Q o , K o , V o )+AddPositionEncoding(E o )), where Q o =W Qo E o , K i =W Ko E o , V o =W Vo E o , E o is an input vector containing the oil temperature, oil pressure data features and related historical data features; W Qo 、W Ko 、W Vo is the weight matrix set for the fusion of oil temperature and oil pressure data; MultiHeadAttention is a multi-head attention mechanism used to capture the correlation features between oil temperature, oil pressure data and other data; AddPositionEncoding is a position encoding function that adds position information to the input feature vector; LN is a layer normalization operation, F o is the fused feature vector related to oil temperature and oil pressure.

4. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 2 is characterized in that: In the improved multi-layer perceptron feature extraction unit, the formula for further updating the weight matrices W1 and W2 is: i=1,2, where is the weight matrix before updating, is the updated weight matrix; α is the learning rate, which controls the step size of weight update; Loss is the loss function for the weight matrix gradient.

5. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 3 is characterized in that: The formula for optimizing the Transformer model fusion unit and updating the weight of the attention mechanism is: j represents different attention heads, is the attention weight of the j-th head before updating, is the updated attention weight; γ is the learning rate, which is used to adjust the weight update amplitude; The attention loss function Lossatt is the attention weight of the j-th head The gradient of is calculated by back-propagating the loss of the model in the process of fusing pumping unit related features.

6. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 1 is characterized in that: When the lubrication decision model building unit builds the model, the formula for the relationship between the pumping unit speed n, torque T and lubricating oil viscosity μ is: μ=f(n, T)=a1n 2 +a2T+a3, where a1, a2, and a3 are coefficients determined by regression analysis of historical data and the fusion features output by the optimized Transformer model fusion unit. This formula is used to determine the required viscosity of the lubricating oil based on the real-time speed and torque of the pumping unit when building a lubrication decision model and make lubrication decisions.

7. The intelligent lubrication decision system for oil pumping units based on multi-source data fusion according to claim 1 is characterized in that: When the lubrication decision model building unit builds the model, the formula combining the relationship between the pH value of the lubricating oil, the water content w and the lubrication cycle t is: Among them, b1, b2, and b3 are parameters determined by a preset algorithm based on historical data and model fusion features. This formula is used to determine the lubrication cycle based on the pH and water content of the lubricating oil when constructing the lubrication decision model, providing a basis for the lubrication decision execution unit to determine the lubrication operation time interval.

8. The intelligent lubrication decision-making method for oil pumping units based on multi-source data fusion is characterized by: The following steps are involved: S1: Through the multi-source data acquisition unit, various sensors are used to collect data on the speed, torque, oil temperature, oil pressure, vibration, and viscosity, pH, and water content of the lubricating oil during operation; S2: The data collected by the multi-source data acquisition unit is transmitted to the data transmission and pre-processing unit, which transmits the data and performs preliminary processing to remove outliers and noise interference; S3: Input the data processed by the data transmission and preprocessing unit into the improved multi-layer perceptron feature extraction unit, use the improved multi-layer perceptron model to extract the potential features in the data, and generate a feature vector; S4: Inputting the feature vector output by the improved multi-layer perceptron feature extraction unit and the historical lubrication data features provided by the historical data storage unit into the optimized Transformer model fusion unit, performing feature fusion using the optimized Transformer model, and obtaining fusion features; S5: The lubrication decision model construction unit determines the model parameters by analyzing the fusion features output by the optimized Transformer model fusion unit and the historical data of the historical data storage unit, and constructs a lubrication decision model suitable for the current pumping unit operating condition; S6: The lubrication decision execution unit generates a specific lubrication decision instruction based on the lubrication decision model constructed by the lubrication decision model construction unit; S7: The lubrication decision execution unit transmits the generated lubrication decision instruction to the oil pumping unit lubrication device, and controls the lubrication device to perform corresponding lubrication operations.

Images