Oil product data analysis method and system based on deep learning

Abstract

The invention belongs to the technical field of data analysis, and discloses an oil product data analysis method and system based on deep learning. The method comprises the following steps of: acquiring real-time multi-source oil product data of an oil well by using a full-life-cycle data acquisition device; the method comprises the following steps: preprocessing real-time multi-source oil product data by using an edge computing gateway, and uploading the data to a cloud data center; in the cloud data center, according to the dynamic multi-dimensional feature space, using a multi-dimensional feature extraction model to extract real-time combined multi-dimensional features; analyzing the real-time merged multi-dimensional features by using an oil product data analysis model; and according to a real-time oil product data analysis result, generating a real-time mining, refining, storage and transportation strategy by using a mining, refining, storage and transportation strategy generation model. According to the invention, the problems of insufficient data processing capability, limited model depth and precision and lack of response mechanism in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the field of data analysis technology, and specifically relates to an oil product data analysis method and system based on deep learning. Background Art

[0002] Oil product data refers to the collection of various information related to oil and its products across the entire oil industry chain (from exploration, extraction, refining, storage, transportation, sales, and use). This data is typically multi-source and heterogeneous, with diverse sources, formats, and update frequencies. It also possesses complex physical, economic, and spatiotemporal relationships. Effective analysis and utilization of this data is crucial for optimizing oil production plans, improving refining efficiency, reducing storage and transportation costs, ensuring supply chain stability, and enhancing market responsiveness. With the development of the Internet of Things, big data, and artificial intelligence technologies, leveraging advanced data analysis techniques to enhance the intelligence of the oil industry has become an urgent need.

[0003] However, existing technologies still have many shortcomings, including: 1) Insufficient data processing capabilities: Traditional data analysis methods, such as simple statistical analysis, regression models, or rule-based systems, can often only process data from a single source or a few closely related data points. They struggle to effectively integrate heterogeneous data from multiple sources across different stages, such as mining, refining, storage, and transportation. Some rudimentary machine learning applications may perform feature selection, but these are typically based on a single objective or domain expert experience, lacking systematicity, automation, and intelligence. 2) Limited model depth and accuracy: Existing technologies use simple models, such as support vector machines, traditional neural networks, or ensemble learning, to predict oil production, quality, or demand. These models often struggle to capture complex patterns in high-dimensional, nonlinear, and strongly correlated oil data. They are also limited in their ability to handle the dynamics and long-term dependencies of time series data, resulting in low prediction accuracy and an inability to meet the needs of real-time decision-making. 3) Lack of response mechanism: Existing technologies only implement a single data analysis function and lack an optimized response mechanism for the entire life cycle. This results in low value of data analysis results and an inability to dynamically adjust production, storage, and transportation processes based on data analysis results. Summary of the Invention

[0004] In order to solve the problems of insufficient data processing capabilities, limited model depth and accuracy, and lack of response mechanism in the existing technology, the purpose of the present invention is to provide an oil data analysis method and system based on deep learning.

[0005] The technical solution adopted in the present invention is: A deep learning-based oil product data analysis method includes the following steps: Use full-lifecycle data acquisition devices to collect real-time multi-source oil product data from oil wells and transmit the real-time multi-source oil product data to the edge computing gateway within the communication range;

[0006] Use edge computing gateways to pre-process real-time multi-source oil product data and upload the pre-processed real-time multi-source oil product data to the cloud data center;

[0007] In the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built based on deep learning algorithms is used to extract the real-time merged multidimensional features of pre-processed real-time multi-source oil product data;

[0008] Use the oil product data analysis model built based on deep learning algorithms to analyze real-time merged multi-dimensional features and obtain real-time oil product data analysis results;

[0009] According to the real-time oil product data analysis results, the production, refining, storage and transportation strategy generation model built based on the reinforcement learning algorithm is used to generate the production, refining, storage and transportation strategy to obtain the real-time production, refining, storage and transportation strategy.

[0010] Furthermore, the full life cycle data acquisition device includes several stage data acquisition devices, and the engineering stage data acquisition device includes the mining end data acquisition device, the refining end data acquisition device, and the storage and transportation end data acquisition device;

[0011] The engineering real-time multi-source oil product data includes a number of real-time single-source oil product data, and the several engineering real-time single-source oil product data correspond one to one to a number of stage data acquisition devices, and the real-time single-source oil product data includes real-time mining process data, real-time refining process data and real-time inventory logistics data.

[0012] Furthermore, a full lifecycle data acquisition device is used to collect real-time multi-source oil product data of the oil well, and the real-time multi-source oil product data is transmitted to an edge computing gateway within the communication range, including the following steps:

[0013] Use the data acquisition device at each stage of the full life cycle data acquisition device to collect real-time single-source oil product data of the oil well and transmit it to the edge computing gateway within the communication range;

[0014] According to the OPC UA protocol, real-time multi-source oil product data consisting of several real-time single-source oil product data are written to the OPC UA instance in the OPC UA server.

[0015] Furthermore, an edge computing gateway is used to pre-process the real-time multi-source oil product data, and the pre-processed real-time multi-source oil product data is uploaded to the cloud data center, including the following steps:

[0016] Use the edge computing gateway to perform data cleaning, spatiotemporal alignment, and protocol conversion on real-time multi-source oil product data to obtain pre-processed real-time multi-source oil product data.

[0017] The pre-processed real-time multi-source oil product data is uploaded to the cloud data center through an encrypted secure channel.

[0018] Furthermore, the dynamic multidimensional feature space includes physical feature engineering, cost feature engineering, coupling feature engineering, and causal feature engineering;

[0019] Physical feature engineering includes parallel spatial feature engineering and temporal feature engineering;

[0020] Cost feature engineering includes parallel oil product value feature engineering, extraction cost feature engineering, refining cost feature engineering, storage and transportation cost feature engineering, and transaction cost feature engineering;

[0021] Coupled feature engineering is physical-cost feature engineering;

[0022] Causal feature engineering is causal chain-causal structure feature engineering;

[0023] The multidimensional feature extraction model is constructed based on the MPFEFN algorithm, and the multidimensional feature extraction model includes a physical feature extraction module, a cost feature extraction module, a coupling feature extraction module, a causal feature extraction module and a multidimensional feature merging module. The physical feature extraction module, the cost feature extraction module, the coupling feature extraction module and the causal feature extraction module correspond to the physical feature engineering, cost feature engineering, coupling feature engineering and causal feature engineering of the dynamic multidimensional feature space respectively, and the physical feature extraction module, the cost feature extraction module, the coupling feature extraction module and the causal feature extraction module are all connected to the multidimensional feature merging module, and the engineering multidimensional feature merging module is provided with an attention mechanism.

[0024] Furthermore, the oil product data analysis model is constructed based on the RF-MLP-MOSGA algorithm, and the oil product data analysis model includes a key feature screening module constructed based on the RF algorithm, an oil product data analysis module constructed based on the MLP algorithm, and an analysis result optimization module constructed based on the MOSGA algorithm, which are connected in sequence.

[0025] Furthermore, the mining, refining, storage and transportation strategy generation model is constructed based on the MPO-MOGRPO algorithm, and the mining, refining, storage and transportation strategy generation model includes a meta-strategy optimization module constructed based on the MPO algorithm and a mining, refining, storage and transportation strategy generation module constructed based on the MOGRPO algorithm, which are connected in sequence. The engineering mining, refining, storage and transportation strategy generation module is provided with an objective function set, an intelligent agent, a strategy network and an experience replay pool.

[0026] Furthermore, in the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built based on a deep learning algorithm is used to extract real-time merged multidimensional features of pre-processed real-time multi-source oil product data, including the following steps:

[0027] According to the dynamic multi-dimensional feature space, the pre-processed real-time multi-source oil product data is input into the multi-dimensional feature extraction model built based on the deep learning algorithm;

[0028] The physical feature extraction module, cost feature extraction module and coupling feature extraction module of the multi-dimensional feature extraction model are used to extract the real-time physical features, real-time cost features and real-time coupling features of the pre-processed real-time multi-source oil product data;

[0029] Based on real-time physical characteristics, real-time cost characteristics, and real-time coupling characteristics, a causal discovery algorithm is used to identify causal chains and obtain real-time causal chains.

[0030] Based on the causal model, the causal feature extraction module of the multidimensional feature extraction model is used to extract the real-time causal features corresponding to the real-time causal chain;

[0031] According to the dynamic attention weight value, the multidimensional feature merging module of the multidimensional feature extraction model is used to merge the real-time physical features, real-time cost features, real-time coupling features and real-time causal features to obtain real-time merged multidimensional features.

[0032] Furthermore, an oil product data analysis model based on a deep learning algorithm is used to analyze the real-time merged multi-dimensional features to obtain real-time oil product data analysis results, including the following steps:

[0033] The real-time merging of multi-dimensional features is input into the oil product data analysis model built based on the deep learning algorithm;

[0034] Use the key feature screening module of the oil product data analysis model to extract real-time key features that merge multi-dimensional features in real time;

[0035] Based on the real-time key features, the oil product data analysis module of the oil product data analysis model is used to perform analysis and prediction to obtain the real-time oil product data analysis probability distribution;

[0036] The analysis result optimization module of the oil product data analysis model is used to optimize the probability distribution of real-time oil product data analysis to obtain real-time oil product data analysis results.

[0037] A deep learning-based oil product data analysis system is used to implement an oil product data analysis method. The system includes a cloud data center, several edge computing gateways, and several full-life cycle data acquisition devices. The engineering cloud data center is connected to several edge computing gateways respectively, and each engineering edge computing gateway is communicated with the full-life cycle data acquisition device within the communication range.

[0038] The beneficial effects of the present invention are: The present invention provides a deep learning-based oil product data analysis method and system. Through preprocessing on an edge computing gateway and a multidimensional feature extraction model (including physical, cost, coupling, and causal feature engineering) in a cloud data center, this method can systematically and structuredly integrate various data from oil wells, refineries, storage, and transportation, achieving data fusion and feature extraction from multi-source heterogeneous data, avoiding information silos, and providing comprehensive, high-quality input for subsequent analysis. The "causal feature engineering" and "coupling feature engineering" in the dynamic multidimensional feature space are specifically designed to capture interactions and causal relationships between data. This ensures that analysis results are not only based on correlations but also reflect inherent causal connections. This improves the model's interpretability and the depth of understanding of the system's dynamic behavior, enhancing the systematic, automated, and intelligent nature of oil product data analysis. The multidimensional feature extraction model and oil product data analysis model, constructed using a deep learning algorithm, can automatically learn complex nonlinear relationships and deep patterns in the data, extracting more representative and predictive key features, thereby significantly improving the accuracy and robustness of analysis and prediction, particularly when handling complex operating conditions and abnormal situations. Through the initial processing acceleration of edge computing and the real-time feature extraction, analysis and optimization process of the cloud center, it is possible to quickly respond to real-time data changes in the production, storage and transportation of oil products, generate real-time oil data analysis results and production, refining, storage and transportation strategies, support dynamic adjustment and instant decision-making, improve the flexibility and adaptability of the system, and realize the response mechanism.

[0039] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flowchart of the oil product data analysis method based on deep learning in the present invention.

[0041] Figure 2 This is a structural block diagram of the oil product data analysis system based on deep learning in the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1 like Figure 1As shown, this embodiment provides an oil product data analysis method based on deep learning, comprising the following steps:

[0044] S1: Use the full life cycle data acquisition device to collect real-time multi-source oil product data from the oil well and transmit the real-time multi-source oil product data to the edge computing gateway within the communication range;

[0045] The full life cycle data collection device includes several stage data collection devices. The engineering stage data collection device includes the mining end data collection device, the refining end data collection device, and the storage and transportation end data collection device.

[0046] The engineering real-time multi-source oil product data includes a number of real-time single-source oil product data, each of which corresponds to a number of stage data collection devices. The real-time single-source oil product data includes real-time mining process data, real-time refining process data, and real-time inventory logistics data.

[0047] Real-time production process data includes real-time wellhead dynamic data (three-phase motor extreme current, power harmonic distortion rate, casing pressure, tubing pressure), real-time downhole process data (bottomhole flow pressure, pump filling rate, dynamic liquid level, sand content), and real-time crude oil physical property data (density, viscosity, water content, pour point).

[0048] Real-time refining process data includes real-time process control data (distillation tower temperature curve, fractionation tower pressure, catalyst bed temperature), real-time oil product testing data (distillation range distribution, density, flash point, sulfur content), and real-time online analysis data (near infrared spectrum, mass spectrometry characteristic peaks, Fourier transform infrared);

[0049] Real-time inventory logistics data includes real-time storage tank monitoring data (liquid level, temperature, pressure, electrostatic potential), real-time pipeline transmission data (flow rate, magnetostriction signal, residual stress), real-time environmental interaction data (temperature and humidity, illumination, seismic wave spectrum), and real-time terminal sales data (fuel nozzle flow rate, sales transaction data, oil and gas recovery efficiency).

[0050] Use a full lifecycle data acquisition device to collect real-time multi-source oil product data from oil wells and transmit the real-time multi-source oil product data to an edge computing gateway within communication range, including the following steps:

[0051] S1-1: Use the data acquisition device at each stage of the full life cycle data acquisition device to collect real-time single-source oil product data of the oil well and transmit it to the edge computing gateway within the communication range;

[0052] S1-2: Writing real-time multi-source oil product data consisting of several real-time single-source oil product data to an OPC UA instance in an OPC UA server according to the Open Platform Communications Unified Architecture (OPC UA) protocol includes the following steps:

[0053] S1-2-1: Build the corresponding OPC UA information model based on the physical information of the data acquisition device at all stages of the full life cycle data acquisition device;

[0054] S1-2-2: Deploy the OPC UA server in the edge computing gateway and create a corresponding OPC UA information model instance in the address space of the OPC UA server based on the OPC UA information model;

[0055] S1-2-3: According to the OPC UA protocol, the real-time multi-source oil product data composed of several real-time single-source oil product data is written to the OPC UA instance in the OPC UA server;

[0056] S2: Use the edge computing gateway to preprocess the real-time multi-source oil product data and upload the preprocessed real-time multi-source oil product data to the cloud data center, including the following steps:

[0057] S2-1: Use the edge computing gateway to perform data cleaning, spatiotemporal alignment, and protocol conversion on real-time multi-source oil product data to obtain pre-processed real-time multi-source oil product data;

[0058] Data cleaning rules: Outlier processing: using the Laida criterion combined with reservoir physical property constraints (e.g., porosity 0.1-0.4);

[0059] Missing value filling: spatial interpolation based on reservoir geological model;

[0060] Spatiotemporal alignment method: Time axis: Uses Precision Time Protocol (PTP) clock synchronization (IEEE1588) to achieve ±1ms level synchronization;

[0061] Spatial axis: Establish a three-dimensional geological grid in the WGS84 coordinate system (resolution 50m×50m×10m);

[0062] Protocol conversion standards: Edge computing gateway: supports the OPC UA industrial protocol and converts the OPC UA protocol into a format that can be recognized by the cloud data center;

[0063] Cloud data center: unified conversion to JSON Schema v4.0 format;

[0064] S2-2: Upload pre-processed real-time multi-source oil product data to the cloud data center through an encrypted secure channel;

[0065] S3: In the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built with deep learning algorithms is used to extract real-time merged multidimensional features of pre-processed real-time multi-source oil product data;

[0066] Dynamic multidimensional feature space includes physical feature engineering, cost feature engineering, coupling feature engineering, and causal feature engineering;

[0067] Physical feature engineering includes parallel spatial feature engineering and temporal feature engineering;

[0068] Cost feature engineering includes parallel oil product value feature engineering, extraction cost feature engineering, refining cost feature engineering, storage and transportation cost feature engineering, and transaction cost feature engineering;

[0069] Coupled feature engineering is physical-cost feature engineering;

[0070] Causal feature engineering is causal chain-causal structure feature engineering;

[0071] The multidimensional feature extraction model is built based on the Multi-scale Parallel Feature Extraction Network (MPFEFN) algorithm and includes a physical feature extraction module, a cost feature extraction module, a coupling feature extraction module, a causal feature extraction module, and a multidimensional feature merging module. The physical feature extraction module, the cost feature extraction module, the coupling feature extraction module, and the causal feature extraction module correspond to the physical feature engineering, cost feature engineering, coupling feature engineering, and causal feature engineering of the dynamic multidimensional feature space, respectively. The physical feature extraction module, the cost feature extraction module, the coupling feature extraction module, and the causal feature extraction module are all connected to the multidimensional feature merging module, and the engineering multidimensional feature merging module is equipped with an attention mechanism.

[0072] In the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built with a deep learning algorithm is used to extract the real-time merged multidimensional features of pre-processed real-time multi-source oil product data. The steps include:

[0073] S3-1: According to the dynamic multi-dimensional feature space, the pre-processed real-time multi-source oil product data Input a multi-dimensional feature extraction model built based on a deep learning algorithm, where is the time indicator, N is the total number of data points, and M is the feature dimension;

[0074] S3-2: Use the physical feature extraction module, cost feature extraction module, and coupling feature extraction module of the multi-dimensional feature extraction model to extract the real-time physical features, real-time cost features, and real-time coupling features of the pre-processed real-time multi-source oil product data;

[0075] The formula is: ;

[0076] Where, These are real-time physical features, real-time cost features, and real-time coupling features; It is a spatial feature extraction function and a temporal feature extraction function; These are the oil product value feature extraction function, the mining cost feature extraction function, the refining cost feature extraction function, the storage and transportation cost feature extraction function, and the transaction cost feature extraction function; Extract function for coupling features;

[0077] S3-3: Based on the real-time physical characteristics, real-time cost characteristics, and real-time coupling characteristics, a causal discovery algorithm is used to identify the causal chain and obtain the real-time causal chain;

[0078] The formula is: ;

[0079] Where, For real-time cause-effect chain; is the causal discovery algorithm function;

[0080] S3-4: Based on the causal model, the causal feature extraction module of the multidimensional feature extraction model is used to extract the real-time causal features corresponding to the real-time causal chain;

[0081] The formula is: ;

[0082] Where, It is a real-time causal feature; is a causal model function, which may be a graph neural network or other model that can process graph structure input;

[0083] S3-5: Based on the dynamic attention weight value, the multidimensional feature merging module of the multidimensional feature extraction model is used to merge the real-time physical features, real-time cost features, real-time coupling features, and real-time causal features to obtain a real-time merged multidimensional feature, including the following steps:

[0084] S3-5-1: Perform feature projection on real-time physical features, real-time cost features, real-time coupling features, and real-time causal features to obtain projected real-time physical features, projected real-time cost features, projected real-time coupling features, and projected real-time causal features, which are used to project features of different dimensions into a unified dimension.

[0085] The formula is: ;

[0086] Where, They are real-time physical characteristics after projection, real-time cost characteristics after projection, real-time coupling characteristics after projection, and real-time causal characteristics after projection; is the feature projection function; K is the unified dimension;

[0087] S3-5-2: Based on the dynamic attention weight value, the multidimensional feature merging module of the multidimensional feature extraction model is used to merge the post-projection real-time physical features, post-projection real-time cost features, post-projection real-time coupling features, and post-projection real-time causal features to obtain a real-time merged multidimensional feature.

[0088] The formula is: ;

[0089] Where, To merge multi-dimensional features in real time; is the dynamic attention weight value;

[0090] S4: Use the oil product data analysis model built based on the deep learning algorithm to analyze the real-time merged multi-dimensional features and obtain real-time oil product data analysis results;

[0091] The oil product data analysis model is built based on the Random Forest (RF)-Multi-Layer Perceptron (MLP)-Multi-Objective Snow Geese Algorithm (MOSGA) algorithm. The oil product data analysis model includes a key feature screening module based on the RF algorithm, an oil product data analysis module based on the MLP algorithm, and an analysis result optimization module based on the MOSGA algorithm.

[0092] The key feature screening module selects the most critical and informative feature subsets for the oil data analysis task based on the calculated feature importance scores, and eliminates redundant or noisy features, thereby reducing the number of input features and the computational complexity and training time of subsequent analysis. The oil data analysis module can learn the complex nonlinear relationships between features and is used to perform analysis and prediction based on key features to obtain a probability distribution. The analysis result optimization module is usually used to handle multi-objective optimization problems, that is, to find a set of optimal compromise solutions between multiple conflicting objectives. In the context of oil data analysis, these objectives may include: accuracy, computational cost, error value, probability dispersion, and result complexity. It iteratively improves the quality of the solution and seeks optimization results that perform well on multiple objectives. This is used to optimize the probability distribution of oil data analysis and improve the accuracy of oil data analysis.

[0093] Using an oil product data analysis model built based on a deep learning algorithm, we analyze the real-time merged multi-dimensional features to obtain real-time oil product data analysis results. This includes the following steps:

[0094] S4-1: Real-time merging of multi-dimensional features into the oil product data analysis model built based on deep learning algorithms;

[0095] S4-2: Use the key feature screening module of the oil product data analysis model to extract real-time key features that merge multi-dimensional features in real time;

[0096] S4-3: Based on the real-time key features, the oil product data analysis module of the oil product data analysis model is used to perform analysis and prediction to obtain the real-time oil product data analysis probability distribution;

[0097] S4-4: Using the analysis result optimization module of the oil product data analysis model, optimize the real-time oil product data analysis probability distribution to obtain the real-time oil product data analysis results, including the following steps:

[0098] S4-4-1: Analyze the probability distribution according to the real-time oil product data, set the format of the real-time probability distribution compensation value, and encode the initial real-time probability distribution compensation value into the individual vector of the MOSGA individual in the analysis result optimization module;

[0099] S4-4-2: Based on the individual vectors, a dynamic optimization scheme is used to generate an initialization module of the model to generate several initial solutions; the initial solutions correspond to an initial real-time probability distribution compensation value;

[0100] The formula is: ;

[0101] Where, The initial MOSGA individual generated by the Circle chaotic map sequence, that is, the initial solution; is a randomly generated MOSGA individual; i is the MOSGA individual indicator; To find the remainder function;

[0102] S4-4-3: Based on the fitness function, use the iterative optimization module of the dynamic optimization solution to generate the model, iteratively optimize several initial solutions to obtain the optimal solution, including the following steps:

[0103] S4-4-3-1: Use the fitness function to obtain the initial fitness value of each initial MOSGA individual in the initial MOSGA population, and take the initial MOSGA individual with the lowest fitness value as the leader;

[0104] The formula is: ;

[0105] Where, is the fitness function; To calculate the cost function; is the error value function; is the probability dispersion function; is the result complexity function; For MOSGA individuals; are the first weight value, the second weight value, the third weight value, and the fourth weight value;

[0106] S4-4-3-2: Entering the exploration phase, the leader goose rotation mechanism, the calling guidance mechanism, and the dynamic reverse mechanism are introduced to iteratively update the initial MOSGA population to obtain an updated MOSGA population and retain the best individuals;

[0107] The leader goose rotation mechanism selects a new leader goose in each iteration based on the fitness value of the MOSGA individuals. This mechanism can prevent the leader goose from falling into the local optimum too early and enhance the global search capability of the algorithm.

[0108] The formula is: ;

[0109] Where, Be the leader of an update; For the The initial MOSGA individual with the third lowest fitness value in the initial MOSGA population at the number of iterations; is the fifth-to-last initial MOSGA individual in the initial MOSGA population in the iteration number; is the current iteration number; is the optimal individual; is the first weight factor; is a random number generation function;

[0110] The calling guidance mechanism uses the sound wave propagation attenuation model to adjust the individual position update according to the distance between the MOSGA individual and the leader goose. The position update of the MOSGA individual that is closer is more affected by the leader goose, and it can quickly approach the optimal solution. The position update of the MOSGA individual that is farther away is less affected by the leader goose, and it can maintain a certain exploration ability. This mechanism can avoid excessive aggregation or dispersion of the group and improve the local search accuracy of the algorithm.

[0111] The formula is: ;

[0112] Where, It is an updated MOSGA individual; For the The number of iterations of the initial MOSGA individual; is the sound intensity received by the initial MOSGA individual; is the sound intensity parameter; is the initial sound intensity; is the minimum acceptable sound intensity; is the convergence factor; is the initial MOSGA individual with the farthest distance; is a random parameter; is the Brownian motion function; is the Brownian motion parameter; is the XOR processing symbol; ;

[0113] Where, is the convergence factor; tanh(.) is the hyperbolic tangent function; is the current iteration number; is the maximum number of iterations; a max 、a min are the maximum and minimum values of the convergence factor respectively; λ is the deceleration rate parameter, is the decreasing period parameter, λ=-2π, ;

[0114] Dynamic reverse mechanism, which dynamically reverses the initial MOSGA individuals to improve the diversity of exploration directions and avoid falling into local optimality;

[0115] The formula is: ;

[0116] Where, is a reverse MOSGA individual updated once; γ is the decreasing inertia coefficient; L max 、Lmin are the maximum and minimum values of the vector space respectively;

[0117] The leader goose of one update, several MOSGA individuals of one update and several reverse MOSGA individuals of one update are integrated to obtain a MOSGA population of one update, and the MOSGA individual with the lowest fitness value is retained as the optimal individual;

[0118] S4-4-3-3: Entering the development stage, introducing the abnormal boundary strategy and Gaussian mutation mechanism, performing a second update on the once-updated MOSGA population, obtaining a second-updated MOSGA population, and retaining the optimal individual;

[0119] Abnormal boundary strategy, calculates the difference between the fitness value of each updated MOSGA individual and the average fitness value of the group. For MOSGA individuals whose fitness value is much higher than the group average, their position update method will be adjusted, such as using Gaussian mutation mechanism, larger step size or smaller step size. This mechanism can help individuals avoid falling into local optimality and improve the convergence speed and accuracy of the algorithm;

[0120] The formula is: ;

[0121] Where, is the second updated MOSGA individual; It is an updated MOSGA individual; is the fitness function; is the average fitness value of the group; is the MOSGA individual with the highest fitness value; are the second weight factor and the third weight factor; is the Gaussian mutation mechanism parameter;

[0122] S4-4-3-4: If the number of iterations is greater than or equal to the iteration threshold or the fitness value of the optimal individual is less than the fitness threshold, the optimal individual is output as the optimal solution;

[0123] S4-4-4: Decode the individual vectors of the optimal solution to obtain the optimal real-time probability distribution compensation value. Based on the optimal real-time probability distribution compensation value, optimize the real-time oil product data analysis probability distribution to obtain the real-time oil product data analysis results.

[0124] S5: Based on the real-time oil product data analysis results, a production, refining, storage and transportation strategy generation model built based on the reinforcement learning algorithm is used to generate a production, refining, storage and transportation strategy to obtain a real-time production, refining, storage and transportation strategy;

[0125] The mining, refining, storage and transportation strategy generation model is built based on the Meta-Policy Optimization (MPO)-Multi-Objective Group Relative Policy Optimization (MOGRPO) algorithm. The model includes a sequentially connected meta-policy optimization module based on the MPO algorithm and a mining, refining, storage and transportation strategy generation module based on the MOGRPO algorithm. The engineering mining, refining, storage and transportation strategy generation module is equipped with an objective function set, an intelligent agent, a policy network, and an experience replay pool.

[0126] The meta-policy optimization module adjusts the network parameters of the policy network in the mining, refining, storage and transportation strategy generation module so that these parameters can quickly adapt to new, unseen performance test prediction results, improving the model's generalization ability. Even under unseen performance test prediction results, the policy network can be updated based on previous learning experience, improving the adaptability of the mining, refining, storage and transportation strategy generation model. The objective function set of the mining, refining, storage and transportation strategy generation module can handle multiple conflicting objectives, such as response time, storage and transportation costs, and transaction impact, and generate mining, refining, storage and transportation strategies that balance these objectives. The intelligent agent learns historical mining, refining, storage and transportation strategies through the experience replay pool, continuously optimizing its own policy generation capabilities. The intelligent agent controls the policy network based on learned experience to generate more effective mining, refining, storage and transportation strategies. The design of the experience replay pool and intelligent agent enables the model to continuously learn and optimize, improving the quality of policy generation. The mining, refining, storage and transportation strategy generation module uses a swarm exploration approach, which can, to a certain extent, avoid falling into local optimal solutions. The policy network outputs the distribution probability of actions under a given state. The mining, refining, storage and transportation strategy generation module directly updates the policy network through gradients, eliminating the value network used in traditional reinforcement learning and making the algorithm structure more concise.

[0127] Based on the real-time oil product data analysis results, a production, refining, storage and transportation strategy generation model based on a reinforcement learning algorithm is used to generate a real-time production, refining, storage and transportation strategy, including the following steps:

[0128] S5-1: Based on the real-time oil product data analysis results, the meta-strategy optimization module of the production, refining, storage and transportation strategy generation model is used to update the strategy network of the production, refining, storage and transportation strategy generation module to obtain an updated strategy network;

[0129] S5-2: Randomly extracting several historical mining, refining, storage and transportation strategy generation experiences from the experience replay pool, generating several possible mining, refining, storage and transportation decision actions based on the several historical mining, refining, storage and transportation strategy generation experiences, and updating the action space of the intelligent agent of the mining, refining, storage and transportation strategy generation module based on the several possible mining, refining, storage and transportation decision actions to obtain an updated action space;

[0130] S5-3: Analyze the real-time oil product data analysis results to obtain several real-time oil product data states, and update the state space of the intelligent agent of the production, refining, storage and transportation strategy generation module based on the real-time oil product data states to obtain an updated state space;

[0131] S5-4: Selecting a real-time objective function from the objective function set of the production, refining, storage and transportation strategy generation module, and based on the real-time objective function, using the updated intelligent agent of the production, refining, storage and transportation strategy generation module to control the updated policy network, generating a probability distribution of all possible production, refining, storage and transportation decision actions in the updated action space corresponding to each real-time oil product data state in the updated state space;

[0132] S5-5: The possible production, refining, storage and transportation decision action with the highest probability distribution in the updated action space is used as the execution production, refining, storage and transportation decision action of the corresponding real-time oil product data state, and the execution production, refining, storage and transportation decision actions of all real-time oil product data states in the updated state space are integrated to obtain the real-time production, refining, storage and transportation strategy.

[0133] Example 2 like Figure 2 As shown, this embodiment provides an oil product data analysis system based on deep learning, which is used to implement the oil product data analysis method. The system includes a cloud data center, several edge computing gateways and several full life cycle data acquisition devices. The engineering cloud data center is connected to several edge computing gateways respectively, and each engineering edge computing gateway is communicated with the full life cycle data acquisition device within the communication range.

[0134] A full-lifecycle data acquisition device, used to collect real-time multi-source oil product data from oil wells and transmit the real-time multi-source oil product data to an edge computing gateway within communication range;

[0135] Edge computing gateway, used to pre-process real-time multi-source oil product data and upload the pre-processed real-time multi-source oil product data to the cloud data center;

[0136] The cloud data center is used to extract the real-time merged multidimensional features of pre-processed real-time multi-source oil product data based on the dynamic multi-dimensional feature space using a multi-dimensional feature extraction model built based on a deep learning algorithm; use the oil product data analysis model built based on a deep learning algorithm to analyze the real-time merged multi-dimensional features to obtain real-time oil product data analysis results; based on the real-time oil product data analysis results, use the production, refining, storage and transportation strategy generation model built based on a reinforcement learning algorithm to generate production, refining, storage and transportation strategies to obtain real-time production, refining, storage and transportation strategies.

[0137] The present invention provides a deep learning-based oil product data analysis method and system. Through preprocessing on an edge computing gateway and a multidimensional feature extraction model (including physical, cost, coupling, and causal feature engineering) in a cloud data center, this method can systematically and structuredly integrate various data from oil wells, refineries, storage, and transportation, achieving data fusion and feature extraction from multi-source heterogeneous data, avoiding information silos, and providing comprehensive, high-quality input for subsequent analysis. The "causal feature engineering" and "coupling feature engineering" in the dynamic multidimensional feature space are specifically designed to capture interactions and causal relationships between data. This ensures that analysis results are not only based on correlations but also reflect inherent causal connections. This improves the model's interpretability and the depth of understanding of the system's dynamic behavior, enhancing the systematic, automated, and intelligent nature of oil product data analysis. The multidimensional feature extraction model and oil product data analysis model, constructed using a deep learning algorithm, can automatically learn complex nonlinear relationships and deep patterns in the data, extracting more representative and predictive key features, thereby significantly improving the accuracy and robustness of analysis and prediction, particularly when handling complex operating conditions and abnormal situations. Through the initial processing acceleration of edge computing and the real-time feature extraction, analysis and optimization process of the cloud center, it is possible to quickly respond to real-time data changes in the production, storage and transportation of oil products, generate real-time oil data analysis results and production, refining, storage and transportation strategies, support dynamic adjustment and instant decision-making, improve the flexibility and adaptability of the system, and realize the response mechanism.

[0138] The present invention is not limited to the above optional embodiments. Anyone can derive various other forms of products based on the teachings of the present invention. The above specific embodiments should not be construed as limiting the scope of protection of the present invention. The scope of protection of the present invention shall be based on the scope defined in the claims, and the description can be used to interpret the claims.

Claims

1. A deep learning-based oil product data analysis method, characterized by: The steps include: Use full-lifecycle data acquisition devices to collect real-time multi-source oil product data from oil wells and transmit the real-time multi-source oil product data to the edge computing gateway within the communication range; Use edge computing gateways to pre-process real-time multi-source oil product data and upload the pre-processed real-time multi-source oil product data to the cloud data center; In the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built based on deep learning algorithms is used to extract the real-time merged multidimensional features of pre-processed real-time multi-source oil product data; Use the oil product data analysis model built based on deep learning algorithms to analyze real-time merged multi-dimensional features and obtain real-time oil product data analysis results; According to the real-time oil product data analysis results, the production, refining, storage and transportation strategy generation model built based on the reinforcement learning algorithm is used to generate the production, refining, storage and transportation strategy to obtain the real-time production, refining, storage and transportation strategy.

2. The oil product data analysis method based on deep learning according to claim 1, characterized in that: The full life cycle data acquisition device includes several stage data acquisition devices, including a mining end data acquisition device, a refining end data acquisition device, and a storage and transportation end data acquisition device; The real-time multi-source oil product data includes several real-time single-source oil product data, and the several real-time single-source oil product data correspond one to one to several stage data acquisition devices, and the real-time single-source oil product data includes real-time mining process data, real-time refining process data and real-time inventory logistics data.

3. The oil product data analysis method based on deep learning according to claim 2, characterized in that: Use a full lifecycle data acquisition device to collect real-time multi-source oil product data from oil wells and transmit the real-time multi-source oil product data to an edge computing gateway within communication range, including the following steps: Use the data acquisition device at each stage of the full life cycle data acquisition device to collect real-time single-source oil product data of the oil well and transmit it to the edge computing gateway within the communication range; According to the OPC UA protocol, real-time multi-source oil product data consisting of several real-time single-source oil product data are written to the OPC UA instance in the OPC UA server.

4. The oil product data analysis method based on deep learning according to claim 3, characterized in that: Use the edge computing gateway to preprocess real-time multi-source oil product data and upload the preprocessed real-time multi-source oil product data to the cloud data center, including the following steps: Use the edge computing gateway to perform data cleaning, spatiotemporal alignment, and protocol conversion on real-time multi-source oil product data to obtain pre-processed real-time multi-source oil product data. The pre-processed real-time multi-source oil product data is uploaded to the cloud data center through an encrypted secure channel.

5. The oil product data analysis method based on deep learning according to claim 4, characterized in that: The dynamic multidimensional feature space includes physical feature engineering, cost feature engineering, coupling feature engineering and causal feature engineering; The physical feature engineering includes parallel spatial feature engineering and temporal feature engineering; The cost feature engineering includes parallel oil product value feature engineering, mining cost feature engineering, refining cost feature engineering, storage and transportation cost feature engineering, and transaction cost feature engineering; The coupling feature engineering is physical-cost feature engineering; The causal feature engineering is causal chain-causal structure feature engineering; The multidimensional feature extraction model is constructed based on the MPFEFN algorithm, and the multidimensional feature extraction model includes a physical feature extraction module, a cost feature extraction module, a coupling feature extraction module, a causal feature extraction module and a multidimensional feature merging module. The physical feature extraction module, the cost feature extraction module, the coupling feature extraction module and the causal feature extraction module correspond to the physical feature engineering, cost feature engineering, coupling feature engineering and causal feature engineering of the dynamic multidimensional feature space respectively, and the physical feature extraction module, the cost feature extraction module, the coupling feature extraction module and the causal feature extraction module are all connected to the multidimensional feature merging module, and the multidimensional feature merging module is provided with an attention mechanism.

6. The oil product data analysis method based on deep learning according to claim 5, characterized in that: The oil product data analysis model is constructed based on the RF-MLP-MOSGA algorithm, and the oil product data analysis model includes a key feature screening module constructed based on the RF algorithm, an oil product data analysis module constructed based on the MLP algorithm, and an analysis result optimization module constructed based on the MOSGA algorithm, which are connected in sequence.

7. The oil product data analysis method based on deep learning according to claim 6, characterized in that: The mining, refining, storage and transportation strategy generation model is constructed based on the MPO-MOGRPO algorithm, and the mining, refining, storage and transportation strategy generation model includes a meta-strategy optimization module constructed based on the MPO algorithm and a mining, refining, storage and transportation strategy generation module constructed based on the MOGRPO algorithm, which are connected in sequence. The mining, refining, storage and transportation strategy generation module is provided with an objective function set, an intelligent agent, a strategy network and an experience replay pool.

8. The oil product data analysis method based on deep learning according to claim 7, characterized in that: In the cloud data center, based on the dynamic multidimensional feature space, a multidimensional feature extraction model built with a deep learning algorithm is used to extract the real-time merged multidimensional features of pre-processed real-time multi-source oil product data. The steps include: According to the dynamic multi-dimensional feature space, the pre-processed real-time multi-source oil product data is input into the multi-dimensional feature extraction model built based on the deep learning algorithm; The physical feature extraction module, cost feature extraction module and coupling feature extraction module of the multi-dimensional feature extraction model are used to extract the real-time physical features, real-time cost features and real-time coupling features of the pre-processed real-time multi-source oil product data; Based on real-time physical characteristics, real-time cost characteristics, and real-time coupling characteristics, a causal discovery algorithm is used to identify causal chains and obtain real-time causal chains. Based on the causal model, the causal feature extraction module of the multidimensional feature extraction model is used to extract the real-time causal features corresponding to the real-time causal chain; According to the dynamic attention weight value, the multidimensional feature merging module of the multidimensional feature extraction model is used to merge the real-time physical features, real-time cost features, real-time coupling features and real-time causal features to obtain real-time merged multidimensional features.

9. The oil product data analysis method based on deep learning according to claim 8, characterized in that: Using an oil product data analysis model built based on a deep learning algorithm, we analyze the real-time merged multi-dimensional features to obtain real-time oil product data analysis results. This includes the following steps: The real-time merging of multi-dimensional features is input into the oil product data analysis model built based on the deep learning algorithm; Use the key feature screening module of the oil product data analysis model to extract real-time key features that merge multi-dimensional features in real time; Based on the real-time key features, the oil product data analysis module of the oil product data analysis model is used to perform analysis and prediction to obtain the real-time oil product data analysis probability distribution; The analysis result optimization module of the oil product data analysis model is used to optimize the probability distribution of real-time oil product data analysis to obtain real-time oil product data analysis results.

10. A deep learning-based oil product data analysis system, used to implement the oil product data analysis method according to any one of claims 1 to 9, characterized in that: The system includes a cloud data center, several edge computing gateways and several full life cycle data acquisition devices. The cloud data center is connected to several edge computing gateways respectively, and each edge computing gateway is communicatively connected to a full life cycle data acquisition device within the communication range.

Images