An operating failure prediction processing method and device for an electric submersible screw pump and a medium

Abstract

The application discloses a kind of electric submersible screw pump operation fault prediction processing method, equipment and medium, belong to electric submersible screw pump fault prediction technical field, for solving in the fault monitoring of existing electric submersible screw pump system, fault response lag, more rely on manual repair, it is difficult to realize the technical problem of dynamic fault prediction of electric submersible screw pump system.Method includes: the multi-source operation characteristics in real-time multi-source operation data are carried out multi-dimensional data fusion processing with the mining working condition characteristics in electric submersible screw pump system;Multi-dimensional feature vector is carried out fault prediction under single fault type and equipment overall operation trend calculation;And the running fault recognition accuracy rate of bottom layer output is adaptively weighted fusion calculation with the comprehensive health index of upper layer output;The fault prediction result corresponding to the multi-fault prediction risk score matrix is carried out logic check between expert logic rule, obtains final operation fault type.

Description

Technical Field

[0001] This application relates to the field of fault prediction for electric submersible screw pumps, and in particular to a method, equipment, and medium for predicting and handling operational faults of electric submersible screw pumps. Background Technology

[0002] In the process of heavy oil extraction, electric submersible screw pumps, as key lifting equipment, operate under complex conditions such as high temperature, high viscosity, and many impurities for a long time, and are prone to various failures such as insufficient protection current, waxing, blockage, abnormal current, power failure and shutdown, and dry pumping and pump burnout.

[0003] Traditional operation and maintenance methods mainly rely on periodic inspections and manual experience-based judgment, which suffers from problems such as delayed fault detection, low diagnostic efficiency, and significant losses due to sudden downtime. Existing monitoring systems are mostly based on threshold alarms, lacking the ability to continuously predict and deeply analyze operational status, making it difficult to achieve early warning and intelligent decision-making for faults.

[0004] In other words, there are still certain limitations in the fault monitoring of existing electric submersible screw pump systems:

[0005] 1. Delayed fault response: Most systems only issue alarms after a fault occurs, making it impossible to predict fault risks in advance, resulting in long unplanned downtime and high maintenance costs.

[0006] 2. Reliance on human experience: Fault diagnosis and handling strategies rely heavily on the experience of on-site personnel and lack systematic and standardized decision support;

[0007] 3. Insufficient data utilization: Real-time monitoring data and historical data are not fully integrated, and there is a lack of dynamic fault prediction models based on multi-dimensional data;

[0008] 4. Limited early warning accuracy: Single threshold or simple statistical analysis is insufficient to accurately distinguish multiple fault types under complex operating conditions, resulting in a high false alarm rate. Summary of the Invention

[0009] This application provides a method, equipment, and medium for predicting and processing operational faults in electric submersible screw pumps, which addresses the following technical problem: In existing fault monitoring of electric submersible screw pump systems, fault response is delayed, relies heavily on manual maintenance, and it is difficult to achieve dynamic fault prediction for electric submersible screw pump systems.

[0010] The embodiments of this application adopt the following technical solutions:

[0011] On one hand, this application provides a method for predicting and processing operational faults in an electric submersible screw pump, comprising: collecting real-time multi-source operational data from the electric submersible screw pump system; performing multi-dimensional data fusion processing on the multi-source operational features in the real-time multi-source operational data and the mining condition features in the electric submersible screw pump system to obtain a multi-dimensional feature vector of the electric submersible screw pump system under real-time operation; using a preset two-layer prediction model, performing fault prediction and overall equipment operation trend calculation on the multi-dimensional feature vector under a single fault type; and performing adaptive weighted fusion calculation on the output operational fault identification accuracy of the bottom layer and the comprehensive health index output of the upper layer to generate a multi-fault prediction risk scoring matrix; wherein, the two-layer prediction model includes: a lightweight machine learning model at the bottom layer and a time-series prediction model at the upper layer; and performing logical verification between the fault prediction results corresponding to the multi-fault prediction risk scoring matrix and expert logic rules to obtain the final operational fault type.

[0012] This application embodiment, by collecting real-time multi-source operational data and performing multi-dimensional data fusion, can more comprehensively reflect the operating status of the electric submersible screw pump, thereby improving the accuracy of fault prediction. Utilizing a dual-layer prediction model can adapt to the prediction needs under a single fault type, while simultaneously calculating the overall operating trend of the equipment, making the prediction results more adaptable. Furthermore, it can more comprehensively assess the health status of the electric submersible screw pump, providing a basis for maintenance decisions. At the same time, the generated multi-fault prediction risk scoring matrix can effectively identify and assess the risks of multiple faults, avoiding the limitations of single-fault prediction. Moreover, based on the final operational fault type, the system can automatically query and recommend corresponding handling solutions, providing intelligent decision support for maintenance personnel and reducing human error. Ultimately, through early warning and preventative maintenance, downtime caused by faults can be reduced, lowering maintenance costs and operational risks.

[0013] In one feasible implementation, real-time multi-source operating data of the electric submersible screw pump system is collected, specifically including: real-time acquisition of multi-source operating parameter data of the electric submersible screw pump under operating conditions through a sensor network pre-installed in the electric submersible screw pump system; wherein the multi-source operating parameter data includes at least: current, voltage, speed, temperature, pressure, vibration, and flow rate; the multi-source operating parameter data is converted into a time-series data stream and uploaded to an edge computing node; the time-series operating parameter data is cleaned, and the cleaned time-series operating parameter data is integrated into a standardized dataset according to a preset data sampling frequency and format to obtain the real-time multi-source operating data; wherein the data cleaning includes: filtering and denoising, missing value imputation, and outlier removal.

[0014] In one feasible implementation, before performing multi-dimensional data fusion processing on the multi-source operating features in the real-time multi-source operating data and the mining condition features in the electric submersible screw pump system to obtain the multi-dimensional feature vector of the electric submersible screw pump system under real-time operation, the method further includes: extracting time-domain feature data related to screw pump operation from the real-time multi-source operating data to obtain operating frequency-domain features; wherein, the operating frequency-domain features are the mean difference feature, variance feature, and peak value feature of each operating parameter; extracting frequency-domain feature data related to screw pump operation from the real-time multi-source operating data to obtain operating frequency-domain features; wherein, the operating frequency-domain features are the spectral energy and dominant frequency component of each operating parameter; extracting time-frequency domain feature data related to screw pump operation from the real-time multi-source operating data to obtain operating frequency-domain features; wherein, the operating frequency-domain features are the wavelet packet energy entropy of each operating parameter; wherein, the multi-source operating features include: the operating frequency-domain features, the operating frequency-domain features, and the operating frequency-domain features.

[0015] In one feasible implementation, the multi-source operating features in the real-time multi-source operating data are fused with the mining condition features in the electric submersible screw pump system to obtain a multi-dimensional feature vector of the electric submersible screw pump system under real-time operation. Specifically, this includes: collecting real-time operating condition data under different operating conditions in the electric submersible screw pump system and performing spatiotemporal data alignment processing between the real-time operating condition data and the real-time multi-source operating data; extracting the mining condition features from the real-time operating data that are related upstream and downstream to each operating parameter; wherein the mining condition features include at least: heavy oil viscosity features, mining depth features, and ambient temperature features; performing feature mapping processing on the mining condition features and the multi-source operating features under multiple operating conditions; and, based on the feature mapping results, associating the mining condition features and the multi-source operating features with features related to the same operating condition dimension, and fusing the correlated features to obtain a feature fusion vector; and, based on the equipment fault history maintenance record tags under different operating conditions, performing data labeling processing on the feature fusion vector under historical fault point trend features to generate the multi-dimensional feature vector.

[0016] In one feasible implementation, the multi-dimensional feature vectors are used for fault prediction and overall equipment operation trend calculation under a single fault type. Specifically, this includes: obtaining a set of historical fault multi-dimensional feature vectors under historical equipment fault maintenance record tags; using the set of historical fault multi-dimensional feature vectors as input data for training models; performing prediction training on single fault type features in the set of historical fault multi-dimensional feature vectors using multiple pre-set lightweight machine learning models to obtain a trained lightweight machine learning model; and using the trained lightweight machine learning models to perform single fault type prediction on the current multi-dimensional feature vectors. Fault prediction calculations are performed on each attribute object under the fault type characteristics to obtain the predicted operational fault type and the corresponding operational fault identification accuracy. Using a pre-set temporal prediction model at the upper layer, and based on the operational fault type and the operational fault identification accuracy, the mining operation module corresponding to each fault type in the electric submersible screw pump system is trained using temporal learning under the overall operational trend to obtain the trained temporal prediction model. Using the trained temporal prediction model, the mining operation module corresponding to each fault type is processed for equipment operational health scoring under the overall operational trend to obtain the comprehensive health index output at the upper layer.

[0017] In one feasible implementation, the operational fault identification accuracy output from the lower layer and the comprehensive health index output from the upper layer are adaptively weighted and fused to generate a multi-fault prediction risk scoring matrix. Specifically, this includes: adaptively adjusting the weights of the lightweight machine learning model's prediction calculation based on the operational fault types output from the lower layer and the operational fault identification accuracy corresponding to each fault type, thus obtaining a weight adjustment strategy; adaptively weighting and fusing the weight adjustment strategy based on the comprehensive health index output from the upper layer to obtain fault prediction results jointly predicted by the lightweight machine learning model and the time-series prediction model; and calculating the prediction risk score for each fault type based on the fault prediction results to generate the multi-fault prediction risk scoring matrix.

[0018] In one feasible implementation, the fault prediction results corresponding to the multi-fault prediction risk scoring matrix are logically verified with expert logic rules to obtain the final operational fault type. Specifically, this includes: constructing an expert logic rule base; wherein the expert logic rule base is a set of fault judgment logic rules based on experience in the field of electric submersible screw pump mining; obtaining the multi-fault prediction risk scoring matrix and the corresponding model fault prediction results, and determining them as model prediction results; according to the expert logic rule base, performing fault logic query processing on the multi-fault prediction risk scoring matrix to obtain rule base fault prediction results, and generating expert prediction results; if the model... If the model prediction result is consistent with the expert prediction result, then the model prediction result is determined as the final operational failure type. If the model prediction result is inconsistent with the expert prediction result, then the reinforcement learning module of the electric submersible screw pump is activated, and the model prediction result is optimized by rule weighting based on the actual on-site status feedback information collected at the mining site to obtain an optimized model prediction result. If the optimized model prediction result is consistent with the expert prediction result, then the optimized model prediction result is determined as the final operational failure type. Otherwise, the expert prediction result is determined as the final operational failure type.

[0019] In one feasible implementation, after logically verifying the fault prediction results corresponding to the multi-fault prediction risk scoring matrix with expert logic rules to obtain the final operational fault type, the method further includes: obtaining the corresponding multi-fault prediction risk scoring level and real-time operating condition information of the mining equipment based on the final operational fault type; extracting semantic features from the final operational fault type, the multi-fault prediction risk scoring level, and the real-time operating condition information of the mining equipment to obtain fault problem term information; inputting the fault problem term information into the strategy library; if the strategy library finds a corresponding fully matching solution, obtaining a fault resolution strategy; if the strategy library does not find a corresponding fully matching solution, retrieving historical similar fault handling records based on case reasoning technology and recommending an adaptive fault resolution strategy; generating and outputting the fault early warning diagnosis report based on the fault resolution strategy and / or the adaptive fault resolution strategy.

[0020] Secondly, embodiments of this application also provide an operational fault prediction and processing device for an electric submersible screw pump, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to execute an operational fault prediction and processing method for an electric submersible screw pump as described in any of the above embodiments.

[0021] Thirdly, embodiments of this application also provide a non-volatile computer storage medium, which is a non-volatile computer-readable storage medium storing at least one program. Each program includes instructions, which, when executed by a terminal, cause the terminal to execute a method for predicting and processing operational faults in an electric submersible screw pump as described in any of the above embodiments.

[0022] This application provides a method, equipment, and medium for predicting and handling operational failures of an electric submersible screw pump. Compared with the prior art, the embodiments of this application have the following beneficial technical effects:

[0023] 1. By collecting real-time multi-source operating data and performing multi-dimensional data fusion, the operating status of the electric submersible screw pump can be reflected more comprehensively, thereby improving the accuracy of fault prediction.

[0024] 2. The pre-defined two-layer prediction model can adapt to the prediction needs under a single fault type, while calculating the overall operating trend of the equipment, making the prediction results more adaptable.

[0025] 3. By adaptively weighted fusion calculation of operational fault identification accuracy and comprehensive health index, the health status of electric submersible screw pumps can be more comprehensively assessed, providing a basis for maintenance decisions.

[0026] 4. The generated multi-fault prediction risk score matrix can effectively identify and assess the risks of multiple faults, avoiding the limitations of single fault prediction.

[0027] 5. Logically verifying the fault prediction results against expert logic rules ensures the reliability and practicality of the prediction results. At the same time, combining expert experience improves the accuracy of fault diagnosis.

[0028] 6. Based on the final type of operational failure, the system can automatically query and recommend corresponding solutions, providing intelligent decision support for maintenance personnel and reducing human error.

[0029] 7. By providing early warnings and implementing preventative maintenance, downtime caused by malfunctions can be reduced, thereby lowering maintenance costs and operational risks. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0031] Figure 1A flowchart illustrating a method for predicting and handling operational faults in an electric submersible screw pump, as provided in this application embodiment;

[0032] Figure 2 This is a schematic diagram of the structure of a fault prediction and processing device for an electric submersible screw pump provided in an embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0034] This application provides a method for predicting and handling operational faults in an electric submersible screw pump, such as... Figure 1 As shown, the method for predicting and handling operational faults in an electric submersible screw pump specifically includes steps S101-S105:

[0035] S101. Collect real-time multi-source operating data from the electric submersible screw pump system.

[0036] Specifically, it is necessary to first collect multi-source operating parameter data of the electric submersible screw pump in real time under operating conditions through a sensor network pre-installed in the electric submersible screw pump system. The multi-source operating parameter data includes at least: current, voltage, speed, temperature, pressure, mechanical vibration of the electric submersible screw pump, and flow rate.

[0037] Furthermore, the multi-source operating parameter data is converted into a time-series data stream and uploaded to the edge computing node.

[0038] Furthermore, the time-series runtime parameter data undergoes data cleaning, and based on a preset data sampling frequency and format, the cleaned time-series runtime parameter data is then integrated into a standardized dataset to obtain real-time multi-source runtime data. Data cleaning includes: filtering and denoising, missing value imputation, and outlier removal.

[0039] S102. Perform multi-dimensional data fusion processing on the multi-source operation characteristics in the real-time multi-source operation data and the mining condition characteristics in the electric submersible screw pump system to obtain the multi-dimensional feature vector of the electric submersible screw pump system under real-time operation.

[0040] Specifically, it is necessary to first extract the time-domain feature data related to the operation of the screw pump from the real-time multi-source operating data to obtain the runtime features. Among them, the runtime features are the mean difference feature, variance feature, and peak value feature of each operating parameter.

[0041] Furthermore, frequency domain feature data related to screw pump operation is extracted from the real-time multi-source operating data to obtain the operating frequency domain features. These features include the spectral energy and dominant frequency component of each operating parameter.

[0042] Furthermore, the real-time multi-source operational data is further processed to extract time-frequency domain features related to screw pump operation, yielding operational frequency domain features. These operational frequency domain features are the wavelet packet energy entropy for each operational parameter. In other words, the multi-source operational features include: operational frequency domain features, operational frequency domain features, and operational frequency domain features.

[0043] In one embodiment, the mean deviation (average value), variance (variance), and peak value (maximum or minimum value) of each operating parameter need to be calculated first. Then, methods such as Fourier transform (FFT) or wavelet transform are used to extract the spectral energy and dominant frequency component from the time-domain data. Next, wavelet packet decomposition (WPD) or short-time Fourier transform (STFT) is applied to extract the wavelet packet energy entropy.

[0044] Furthermore, it is necessary to collect real-time operating data under different operating conditions in the electric submersible screw pump system, and perform spatiotemporal data alignment processing between the real-time operating data and the real-time multi-source operation data.

[0045] Furthermore, extraction condition features that are interconnected with each operating parameter in the real-time operating data are extracted. These extraction condition features include at least: heavy oil viscosity characteristics, extraction depth characteristics, and ambient temperature characteristics.

[0046] Furthermore, the mining condition features and the multi-source operation features are subjected to feature mapping processing under multiple operation conditions. Based on the feature mapping results, the mining condition features and the multi-source operation features are associated with features related to the same operation condition dimension, and the associated features are fused to obtain a feature fusion vector.

[0047] In one embodiment, the extracted time-domain, frequency-domain, and time-frequency-domain features need to be integrated to form a multi-source operational feature set. Then, extraction condition data from the electric submersible screw pump system, such as heavy oil viscosity, extraction depth, and ambient temperature, are collected. Next, spatiotemporal data alignment processing is performed on the real-time operational condition data and the real-time multi-source operational data to ensure data comparability on the same time scale. Then, the real-time operational condition data is analyzed to extract extraction condition features related to each operational parameter. A feature mapping is established to associate the extraction condition features with the multi-source operational features; that is, under the same operational condition dimension, the mapped features are fused to generate a feature fusion vector.

[0048] Furthermore, based on the historical maintenance record tags of equipment failures under different operating conditions, the feature fusion vector is processed by data labeling under the historical failure point trend features to generate a multi-dimensional feature vector.

[0049] S103. Perform fault prediction and overall equipment operation trend calculation on multi-dimensional feature vectors under a single fault type. Adaptively weightedly fuse the operational fault identification accuracy of the bottom layer output with the comprehensive health index of the upper layer output to generate a multi-fault prediction risk scoring matrix.

[0050] Specifically, firstly, a set of multi-dimensional feature vectors of historical faults under the tag of historical equipment failure and historical maintenance records is obtained. This set of multi-dimensional feature vectors of historical faults is then used as the input data for training the model.

[0051] Furthermore, multiple lightweight machine learning models are pre-defined at the bottom layer to perform prediction training on single fault type features from the multi-dimensional feature vector set of historical faults, resulting in a trained lightweight machine learning model. These lightweight machine learning models include any one or more of the following: gradient boosting tree model and random forest model.

[0052] As a feasible implementation method, the activation criteria for the lightweight model (bottom layer) include: 1) Accuracy threshold: On the validation set, the prediction accuracy and recall for a single fault type must both reach 85% or higher. 2) F1-Score: Due to the potential imbalance in industrial fault data, a weighted F1-Score ≥ 0.82 is typically required. 3) Inference speed: The time for a single prediction must be ≤ 50ms.

[0053] In one embodiment, historical operating data and maintenance records of an oilfield's electric submersible screw pump system over three years were collected, resulting in 12,000 labeled sample data points. These were divided chronologically into three sets: a training set of 70% (8,400 data points), a validation set of 15% (1,800 data points), and a test set of 15% (1,800 data points). The underlying lightweight machine learning model was then trained and validated (in parallel). For example, fault types were first defined, with three single fault types: A: abnormal current, B: wax deposition trend, and C: blockage risk. Then, for feature engineering, multi-dimensional feature vectors (50 dimensions in total, including mean, peak value, wavelet packet energy, etc.) were extracted from the original vibration and current signals. The lightweight machine learning model was then trained: for fault A (abnormal current), a Gradient Boosting Tree (GBDT) model was used, with 200 trees, a maximum depth of 6, and a learning rate of 0.1; for fault B (wax deposition trend), a Random Forest (RF) model was used. The number of trees was set to 150, with a maximum depth of 8. For fault C (congestion risk), a Random Forest (RF) model was also used for training. Finally, validation was performed on the validation set: Fault A (GBDT model): Prediction accuracy was 87.5%, recall was 86.2%, F1-Score was 0.868, and single inference time was 35ms. The validation result was satisfactory and ready for deployment. Fault B (RF model): Prediction accuracy was 89.1%, recall was 84.5%, F1-Score was 0.867, and single inference time was 42ms. The validation result was satisfactory and ready for deployment. Fault C (RF model): Prediction accuracy was 82.3%, recall was 80.1%, and F1-Score was 0.812. The validation result was unsatisfactory (below the 85% threshold), requiring parameter adjustment or retraining with more samples.

[0054] Furthermore, by using multiple trained lightweight machine learning models, fault prediction calculations are performed on the current multi-dimensional feature vector based on attribute objects under single fault type features, to obtain the predicted operational fault types and the corresponding operational fault identification accuracy for each fault type. That is, multiple parallel-trained lightweight machine learning models (such as gradient boosting trees and random forests) are used to make preliminary predictions for single fault types such as abnormal current, wax deposition trend, and blockage risk.

[0055] In one embodiment, the process begins by collecting historical fault maintenance records of the electric submersible screw pump system, including fault type, maintenance time, and pre- and post-maintenance data. Multi-dimensional feature vectors, including time-domain, frequency-domain, and time-frequency-domain features, are extracted from the historical fault data. The historical fault data and multi-dimensional feature vectors are then cleaned, standardized, and normalized. Lightweight machine learning models, such as Gradient Boosting Tree (GBDT) and Random Forest (RF), are then selected. The set of historical fault multi-dimensional feature vectors is used as input to train the GBDT and RF models, and model parameters, such as the number of trees, depth, and learning rate, are adjusted to optimize model performance. The trained GBDT and RF models are then used to predict the current multi-dimensional feature vectors. Predictions are made for different single fault types (such as abnormal current, wax deposition trend, and blockage risk). The prediction results and recognition accuracy for each fault type are recorded.

[0056] Furthermore, based on the pre-set time-series prediction model at the upper level and the operational fault type and operational fault identification accuracy, the mining operation module corresponding to each fault type in the electric submersible screw pump system is trained on the overall operational trend to obtain the trained time-series prediction model.

[0057] As a feasible implementation method, the criteria for enabling the time series model (upper layer) include: 1) Goodness of fit: On the validation set, the coefficient of determination R² between the predicted health index and the actual health index is ≥ 0.85. 2) Error range: The root mean square error (RMSE) must be less than 5% of the full-scale range of the health index (e.g., for a health index of 0-100, the RMSE must be < 5). 3) Trend consistency: A sign test must be passed to ensure that the consistency between the predicted upward / downward trend and the actual trend is not less than 90%.

[0058] In one embodiment, historical operating data and maintenance records of an oilfield's electric submersible screw pump system over three years were collected, resulting in 12,000 labeled sample data entries. These were divided chronologically into three sets: a training set of 70% (8,400 entries), a validation set of 15% (1,800 entries), and a test set of 15% (1,800 entries). The upper-level time-series prediction model was then trained and validated as follows: First, a 30-day historical window of data was selected. Then, the daily fault identification accuracy of the underlying model (taking compliant A and B as examples) was used as a time-series feature sequence (dimension: 30 days × 2 types of fault accuracy). Finally, the comprehensive health score recorded after major equipment overhaul at the corresponding time point was calculated (e.g., the actual health score on day 30 was 75 out of 100). Meanwhile, during the training process of the time series prediction model: an LSTM network needs to be constructed, with an input layer dimension of 2 (corresponding to the accuracy of two faults), 64 hidden layer neurons, and 1 output node (health score); then, the Adam optimizer is used with a learning rate of 0.001, and iterative training is performed for 200 epochs. Finally, the validation criteria and results are as follows: 1) Validation is performed using a test set (time series derived from 1800 data points not used in training). 2) Goodness of fit (R²): The coefficient of determination between the model's predicted overall health score and the actual value is R² = 0.91. 3) Error (RMSE): The root mean square error is 3.2 (health score range is 0-100, 5% of full scale is 5, 3.2 < 5); therefore, R² ≥ 0.85 and RMSE < 5%, indicating that the time series prediction model has excellent performance and can be deployed online.

[0059] Furthermore, the trained time-series prediction model is used to score the equipment health of each fault type's corresponding mining operation module under the overall operational trend, resulting in a comprehensive health index output by the upper layer. In other words, the upper layer introduces a time-series prediction model (such as an LSTM network) to learn the overall operational trend of the equipment and output a comprehensive health index.

[0060] As a feasible implementation method, health score processing refers to the process by which an upper-level time-series prediction model (such as LSTM) calculates a scalar value based on the input, time-sorted fault feature sequence through its internal neural network. This scalar value is the comprehensive health index. Taking the wax deposition risk prediction of an electric submersible screw pump as an example, its time-series prediction model uses an LSTM network. The number of nodes in the input layer of this network is set to 30 (representing a 30-day time window), the number of nodes in the hidden layer is 64, and the number of nodes in the output layer is 1 (representing the comprehensive health index). In the real-time prediction stage, the daily "wax deposition fault identification accuracy" output by the underlying lightweight model over the past 30 days is first obtained, forming a sequence of data. This sequence is then input into the trained LSTM network. The LSTM network then extracts trend features from the sequence through its gating structure, such as identifying a deterioration trend of "accuracy continuously declining in the last 5 days." After mapping through the fully connected layer, a floating-point number between 0 and 1 is output. For example, given the wax deposition identification accuracy sequence of a certain well over the past 30 days, the LSTM network calculates an output value of 0.82. The system multiplied this value by 100, converting it into a comprehensive health index of 82% out of 100. This index indicates that although no congestion has occurred at present, trend analysis shows that the module's health reserves have decreased by 18%, posing a potential risk.

[0061] In one embodiment, a suitable model for time-series prediction, such as a Long Short-Term Memory (LSTM) network, needs to be selected. Then, the LSTM model is trained using a multi-dimensional feature vector containing the fault type and identification accuracy as input; and model parameters, such as hidden layer size, learning rate, and batch size, are adjusted to optimize model performance. Next, the trained LSTM model is used to perform time-series learning training on the mining operation module corresponding to each fault type in the ESP system. Then, based on the prediction results for each fault type and the time-series data output by the LSTM model, the overall operational health index of each module is calculated. Finally, an adaptive weighted fusion algorithm is used to weightedly fuse the underlying fault identification accuracy and the upper-level health index.

[0062] Furthermore, based on the operational fault types output from the underlying layer and the operational fault identification accuracy corresponding to each fault type, the weights for model prediction calculation in the lightweight machine learning model are adaptively adjusted to obtain a weight adjustment strategy.

[0063] In one embodiment, multiple parallel lightweight machine learning models (such as GBDT and RF) are used at the underlying level, each independently outputting a prediction result for a fault type (including the fault type and its recognition accuracy). When performing comprehensive fault prediction, these results need to be weighted and combined to form a comprehensive risk score matrix. Here, the "weight" is the contribution coefficient of each individual fault prediction result in the combination. The weight allocation is mainly based on the operational fault recognition accuracy output by each model. Models with higher accuracy have higher reliability in their predictions, therefore, they should be given greater weight during fusion to improve the overall prediction accuracy. Furthermore, by dynamically adjusting the weights, the fused fault prediction result can more accurately reflect the actual operating state of the equipment, avoiding interference from low-accuracy models. For example, optimization algorithms (such as gradient descent and genetic algorithms) can be used to automatically search for the optimal weight allocation scheme, aiming to minimize the error between the fused prediction result and the actual fault. For example, if the recognition accuracies of the three lightweight machine learning models are 90%, 80%, and 70%, respectively, the weights can be adjusted to 0.5, 0.3, and 0.2, maximizing the contribution of the model with the highest accuracy, thus obtaining the final weight adjustment strategy.

[0064] Furthermore, based on the comprehensive health index output from the upper layer, the weight adjustment strategy is adaptively weighted and fused to finally obtain the fault prediction result based on the joint prediction of the lightweight machine learning model and the time series prediction model.

[0065] Furthermore, it is necessary to calculate the predicted risk score for each fault type based on the fault prediction results, generating a multi-fault prediction risk score matrix. This involves adaptively weighting the fault identification accuracy (vector) output from the lower layer with the comprehensive health index (scalar) output from the upper layer. The weights are determined based on the reciprocal of the time-series model prediction error (the smaller the error, the larger the upper-layer weight). Only then can a multi-fault prediction risk score matrix containing fault type, probability, and overall equipment health be finally generated.

[0066] In one embodiment, different fault types can be identified based on the operational fault types output from the lower layer, and the operational fault identification accuracy corresponding to each fault type can be analyzed. The weights of the model prediction calculations are then adaptively adjusted based on the accuracy, increasing the weights of fault types with higher accuracy. An optimization algorithm (such as gradient descent or genetic algorithm) can be used to optimize the weight allocation. Then, based on the comprehensive health index output from the upper layer, the overall operating status of the equipment is evaluated, and the weight adjustment strategy is weighted and fused with the upper-layer health index to obtain the final weight adjustment strategy. Subsequently, a lightweight machine learning model and a time-series prediction model are used to predict faults, and the prediction results of the two models are fused to obtain the final fault prediction result. Finally, a risk score is calculated for the prediction results under each fault type, thereby generating a multi-fault prediction risk score matrix to display the predicted risk of different fault types.

[0067] S104. Logically verify the fault prediction results corresponding to the multi-fault prediction risk scoring matrix with the expert logic rules to obtain the final operational fault type.

[0068] Specifically, an expert logic rule base is first constructed. This expert logic rule base is a set of fault judgment logic rules based on experience in the field of electric submersible screw pump mining.

[0069] Next, the multi-fault prediction risk score matrix and the corresponding model fault prediction results are obtained and identified as the model prediction results.

[0070] Furthermore, based on the expert logic rule base, the multi-fault prediction risk scoring matrix is ​​queried under fault logic to obtain the rule base fault prediction results and generate expert prediction results.

[0071] Furthermore, if the model prediction result is consistent with the expert prediction result, then the model prediction result is determined as the final operational failure type. If the model prediction result is inconsistent with the expert prediction result, then the reinforcement learning module of the electric submersible screw pump is activated, and based on the actual on-site status feedback information collected at the mining site, the model prediction result is optimized by applying rule weights to obtain the optimized model prediction result.

[0072] In one embodiment, when the system monitors a multi-fault prediction risk scoring matrix (model prediction result) as follows: gas interference probability: 90%; insufficient liquid supply probability: 20%; comprehensive health index: 65 (sub-healthy), the expert logic rule reasoning (expert prediction result) is: large current fluctuations (triggered), sudden increase in heap pressure (triggered), and continuous decrease in flow pressure (not triggered, flow pressure stable); rule R4 (weight 0.9) is triggered, but R5 is not triggered; therefore, the confidence level of R4 is extremely high, and the expert prediction result is gas interference. Rule R4 includes: rule content: IF large current fluctuations AND sudden increase in heap pressure; conclusion: THEN Fault type = gas interference; initial weight is 0.9. Rule R5 includes: rule content: IF large current fluctuations AND continuous decrease in flow pressure; conclusion: THEN Fault type = insufficient liquid supply; initial weight is 0.8.

[0073] In one embodiment, due to inconsistencies between the model and expert results, the system activates the reinforcement learning module and awaits on-site feedback. When technicians confirm on-site that a fault in the oil-gas separator is causing gas to enter the pump body, the actual fault is "gas interference" (on-site feedback label: gas interference). Upon receiving feedback that "the actual problem is gas interference," the system begins the rule weighting optimization process: 1) Well-performing rules: Rule R4 (large current fluctuation + high casing pressure -> gas interference) is correct in this judgment. 2) Poor-performing rules: Although R5 (insufficient liquid supply) was not triggered this time, if R5 has historically caused misjudgments, it also needs adjustment. Then, a reward optimization algorithm is used: Rule R4 is given a positive reward because it accurately identified the current "gas interference" fault. Then, the formula is adjusted as follows: ,in, For the new rule weights, As the current rule weight, R is the learning rate, R is the true label of the field feedback (0 or 1), P is the prediction confidence of the current rule for the fault type, and F is the rule trigger factor (1 when the rule is triggered, 0 otherwise). When the field feedback verifies that the rule judgment is correct, the weight of the rule increases; conversely, the weight decreases. For example, when the original weight of R4 is 0.9, and the field feedback confirms that it is correct, the weight is fine-tuned to 0.92 (becoming more reliable). For the penalty algorithm: although no rule directly causes misjudgment in this scenario, if there is an incorrect rule R6 (e.g., IF large current fluctuation THEN insufficient liquid supply, weight 0.6), and it interferes with the judgment, it will be penalized. Finally, the rule weights are updated, that is, the weight of rule R4 (gas interference rule) is changed from 0.9 to 0.92. For rule R5 (insufficient liquid supply rule): although it is not used in this case, the system will analyze its feature weights. If it finds that the feature "flow pressure" is not sensitive in the current block, the trigger threshold or weight of R5 will be fine-tuned from 0.8 to 0.78 to avoid future misjudgments.

[0074] Furthermore, if the prediction results of the optimization model are consistent with the prediction results of the experts, then the prediction results of the optimization model are determined as the final operational failure type; otherwise, the prediction results of the experts are determined as the final operational failure type.

[0075] As a feasible implementation method, an updatable expert rule knowledge base can be constructed, containing fault judgment logic based on domain experience. For example, a persistently low current and a sudden temperature rise may indicate a risk of dry pumping. The prediction results are then logically verified against the expert rules. If they match, the fault type is confirmed; if a conflict occurs, a reinforcement learning module is activated to dynamically optimize the rule weights based on on-site feedback, improving the reliability of the judgment. Furthermore, combining expert rules with case-based reasoning can form a closed loop of "prediction-verification-decision," reducing reliance on human experience and providing standardized, traceable handling strategies.

[0076] S105. Based on the final operational fault type, query and reason about the corresponding handling solutions in the strategy library to generate a fault warning and diagnosis report.

[0077] Specifically, based on the final operational failure type, the corresponding multi-failure prediction risk score level and real-time operating condition information of the mining equipment are first obtained. Then, semantic features are extracted from the final operational failure type, multi-failure prediction risk score level, and real-time operating condition information of the mining equipment to obtain failure issue term information.

[0078] Next, the fault-related term information is input into the strategy library. If no matching solution is found in the strategy library, a fault-solving strategy is obtained; if no matching solution is found, historical similar fault handling records are retrieved based on case reasoning technology, and an appropriate fault-solving strategy is recommended.

[0079] In one embodiment, case-based reasoning is an AI problem-solving paradigm. When encountering a new problem, the system does not start from scratch but recalls similar problems (cases) that it has solved in the past and adapts the solution of the old problem to the current new problem. For example, in this application: First, the relevant information of the current fault is converted into a feature vector that can be used for retrieval. Then, the final operational fault type (e.g., "abnormal current"), risk score level (e.g., "high risk, 85 points") and real-time operating conditions (e.g., "well depth 3000m, production rate 50 cubic meters / day") are input, and the corresponding fault problem term information is output. Next, a similarity search (finding old cases) is performed; that is, in the historical case library, the K most similar historical cases to the fault problem term information can be found by calculating Euclidean distance or cosine similarity. After that, the solution is reused and modified (adapted). That is, if the solution of the most similar case is completely applicable, it is directly recommended; if there is a difference (e.g., the ambient temperature of the historical case was 20°C, and now it is 35°C), the system will adjust the solution parameters according to the temperature difference. Finally, an adaptive fault resolution strategy is generated and awaits manual confirmation. Once confirmed, the new case is added to the database and an adaptive fault resolution strategy is recommended.

[0080] As a feasible implementation method, corresponding handling solutions can be matched from the strategy library based on the confirmed fault type, risk level, and real-time operating conditions. For example, wax deposition warning → backflushing is recommended; excessive current → frequency adjustment and load reduction, etc. If there is no completely matching solution in the strategy library, historical similar fault handling records are retrieved based on case reasoning technology to generate adaptive solution strategy recommendations, that is, an adaptive fault handling strategy is recommended.

[0081] Furthermore, based on the fault resolution strategy and / or adaptive fault resolution strategy, a fault early warning and diagnostic report is generated and output. That is, the fault early warning and diagnostic report can ultimately output an early warning report containing fault type, risk score, maintenance suggestions, and handling steps, and push it to the field terminal and remote management center through the human-machine interface.

[0082] In addition, embodiments of this application also provide a fault prediction and processing device for an electric submersible screw pump, such as... Figure 2 As shown, the fault prediction and processing equipment 200 for electric submersible screw pumps specifically includes:

[0083] At least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201 to enable the at least one processor 201 to execute:

[0084] Collect real-time multi-source operating data from the electric submersible screw pump system;

[0085] Multi-source operation features in real-time multi-source operation data are combined with mining condition features in the electric submersible screw pump system to obtain a multi-dimensional feature vector of the electric submersible screw pump system under real-time operation.

[0086] Through a pre-set two-layer prediction model, multi-dimensional feature vectors are used to predict faults under a single fault type and calculate the overall operating trend of the equipment; and the operating fault identification accuracy of the bottom layer output and the comprehensive health index of the upper layer output are adaptively weighted and fused to generate a multi-fault prediction risk score matrix.

[0087] The fault prediction results corresponding to the multi-fault prediction risk scoring matrix are logically verified with the expert logic rules to obtain the final operational fault type.

[0088] Based on the final operational failure type, the strategy library is queried and inferred to recommend corresponding handling solutions, and a failure early warning and diagnosis report is generated.

[0089] This application embodiment, by collecting real-time multi-source operational data and performing multi-dimensional data fusion, can more comprehensively reflect the operating status of the electric submersible screw pump, thereby improving the accuracy of fault prediction. Utilizing a dual-layer prediction model can adapt to the prediction needs under a single fault type, while simultaneously calculating the overall operating trend of the equipment, making the prediction results more adaptable. Furthermore, it can more comprehensively assess the health status of the electric submersible screw pump, providing a basis for maintenance decisions. At the same time, the generated multi-fault prediction risk scoring matrix can effectively identify and assess the risks of multiple faults, avoiding the limitations of single-fault prediction. Moreover, based on the final operational fault type, the system can automatically query and recommend corresponding handling solutions, providing intelligent decision support for maintenance personnel and reducing human error. Ultimately, through early warning and preventative maintenance, downtime caused by faults can be reduced, lowering maintenance costs and operational risks.

[0090] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0091] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0092] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0097] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0098] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0099] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0100] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of this specification.

Claims

1. A method for predicting and handling operational faults in an electric submersible screw pump, characterized in that, The method includes: Collect real-time multi-source operating data from the electric submersible screw pump system; The multi-source operation features in the real-time multi-source operation data are combined with the mining condition features in the electric submersible screw pump system to perform multi-dimensional data fusion processing, thereby obtaining the multi-dimensional feature vector of the electric submersible screw pump system under real-time operation. Using a pre-defined two-layer prediction model, the multi-dimensional feature vectors are used for fault prediction under a single fault type and for calculating the overall operating trend of the equipment. Specifically, this includes: Obtain a set of multi-dimensional feature vectors of historical faults under the tag of historical equipment fault maintenance records; and use the set of multi-dimensional feature vectors of historical faults as the input data for training the model; By using multiple lightweight machine learning models preset at the bottom layer, the single fault type features in the set of multi-dimensional feature vectors of historical faults are subjected to prediction training to obtain the trained lightweight machine learning model. By using multiple trained lightweight machine learning models, fault prediction calculations are performed on the current multi-dimensional feature vector based on single fault type features to obtain the predicted operational fault types and the operational fault identification accuracy corresponding to each fault type. Using the pre-set time-series prediction model at the upper level, and based on the operational fault type and the operational fault identification accuracy, the mining operation module corresponding to each fault type in the electric submersible screw pump system is trained on the overall operational trend to obtain the trained time-series prediction model. The trained time-series prediction model is used to score the equipment health of each mining operation module under the overall operation trend, and the comprehensive health index is obtained from the upper-level output. The underlying output fault identification accuracy and the upper-layer output comprehensive health index are adaptively weighted and fused to generate a multi-fault prediction risk scoring matrix. The dual-layer prediction model comprises a lightweight machine learning model at the bottom layer and a time-series prediction model at the top layer, specifically including: Based on the types of operational faults output from the underlying layer and the operational fault identification accuracy corresponding to each fault type, the weights of the lightweight machine learning model are adaptively adjusted for model prediction calculation, resulting in a weight adjustment strategy. Based on the comprehensive health index output from the upper layer, the weight adjustment strategy is adaptively weighted and fused to obtain the fault prediction result jointly predicted by the lightweight machine learning model and the time series prediction model, including: The overall operating status of the equipment is assessed based on the comprehensive health index output from the upper layer; and the weight adjustment strategy is weighted and fused with the comprehensive health index output from the upper layer to obtain the final weight adjustment strategy. A lightweight machine learning model and a time-series prediction model are used to predict faults, and the prediction results of the two models are fused to obtain the final fault prediction result. The prediction risk score for each fault type is calculated based on the fault prediction results to generate the multi-fault prediction risk score matrix. The fault prediction results corresponding to the multi-fault prediction risk scoring matrix are logically verified with the expert logic rules to obtain the final operational fault type.

2. The method for predicting and handling operational faults in an electric submersible screw pump according to claim 1, characterized in that, Real-time multi-source operating data is collected from the electric submersible screw pump system, specifically including: The sensor network pre-installed in the electric submersible screw pump system collects multi-source operating parameter data of the electric submersible screw pump in real time under operating conditions; wherein the multi-source operating parameter data includes at least: current, voltage, speed, temperature, pressure, vibration and flow rate; The multi-source operating parameter data is converted into a time-series data stream and uploaded to the edge computing node; The time-series running parameter data is cleaned, and the cleaned time-series running parameter data is integrated into a standardized dataset according to a preset data sampling frequency and format to obtain the real-time multi-source running data; wherein, the data cleaning includes: filtering and denoising, missing value imputation, and outlier removal.

3. The method for predicting and handling operational faults in an electric submersible screw pump according to claim 1, characterized in that, Before performing multi-dimensional data fusion processing on the multi-source operation features in the real-time multi-source operation data and the mining condition features of the electric submersible screw pump system to obtain the multi-dimensional feature vector of the electric submersible screw pump system under real-time operation, the method further includes: The real-time multi-source operating data is used to extract time-domain feature data related to screw pump operation to obtain runtime features; wherein, the runtime features are the mean difference feature, variance feature and peak value feature of each operating parameter; The real-time multi-source operating data is used to extract frequency domain feature data related to screw pump operation to obtain operating frequency domain features; wherein, the operating frequency domain features are the spectral energy and main frequency component of each operating parameter; The real-time multi-source operating data is subjected to time-frequency domain feature data extraction related to screw pump operation to obtain runtime frequency domain features; wherein, the runtime frequency domain features are wavelet packet energy entropy for each operating parameter; The multi-source operating characteristics include: the runtime domain characteristics, the operating frequency domain characteristics, and the operating frequency domain characteristics.

4. The method for predicting and handling operational faults in an electric submersible screw pump according to claim 3, characterized in that, The multi-source operational features in the real-time multi-source operational data are fused with the mining condition features of the electric submersible screw pump system to obtain a multi-dimensional feature vector of the electric submersible screw pump system under real-time operation, specifically including: Real-time operating condition data of the electric submersible screw pump system under different operating conditions are collected, and the real-time operating condition data is aligned with the real-time multi-source operation data in a time-space data alignment process. Extract the mining condition features from the real-time operating data that are related upstream and downstream to each operating parameter; wherein, the mining condition features include at least: heavy oil viscosity features, mining depth features, and ambient temperature features; The mining condition features and the multi-source operation features are subjected to feature mapping processing under multiple operation conditions; and based on the feature mapping results, the mining condition features and the multi-source operation features are associated with features under the same operation condition dimension, and the features that are associated with each other are fused to obtain a feature fusion vector. Based on the equipment failure history maintenance record tags under different operating conditions, the feature fusion vector is processed by data labeling under the historical failure point trend features to generate the multi-dimensional feature vector.

5. The method for predicting and handling operational faults in an electric submersible screw pump according to claim 1, characterized in that, The fault prediction results corresponding to the multi-fault prediction risk scoring matrix are logically verified with expert logic rules to obtain the final operational fault type, which specifically includes: An expert logic rule base is constructed; wherein, the expert logic rule base is a set of fault judgment logic rules based on experience in the field of electric submersible screw pump mining; Obtain the multi-fault prediction risk score matrix and the corresponding model fault prediction results, and determine them as model prediction results; Based on the expert logic rule base, the multi-fault prediction risk scoring matrix is ​​queried under fault logic to obtain the rule base fault prediction result and generate the expert prediction result. If the model prediction result is consistent with the expert prediction result, then the model prediction result is determined as the final operational failure type; If the model prediction result is inconsistent with the expert prediction result, the reinforcement learning module of the electric submersible screw pump is activated, and the model prediction result is optimized by feedback of rule weights based on the actual on-site status feedback information collected at the mining site, so as to obtain the optimized model prediction result. If the prediction result of the optimization model is consistent with the prediction result of the expert, then the prediction result of the optimization model is determined as the final operational failure type; otherwise, the prediction result of the expert is determined as the final operational failure type.

6. The method for predicting and handling operational faults in an electric submersible screw pump according to claim 1, characterized in that, After performing logical verification between the fault prediction results corresponding to the multi-fault prediction risk scoring matrix and the expert logic rules to obtain the final operational fault type, the method further includes: Based on the final operational failure type, obtain the corresponding multi-failure prediction risk score level and real-time operating information of the mining equipment. Semantic features are extracted from the final operational fault type, the multi-fault prediction risk score level, and the real-time operating information of the mining equipment to obtain fault problem term information. Input the fault-related term information into the strategy library; If the strategy library finds a matching solution, then a fault resolution strategy is obtained. If no matching solution is found in the strategy library, then based on case reasoning technology, historical similar fault handling records are retrieved, and an adaptive fault resolution strategy is recommended. Based on the fault resolution strategy and / or the adaptive fault resolution strategy, generate and output a fault early warning and diagnostic report.

7. A fault prediction and processing device for an electric submersible screw pump, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, enabling the at least one processor to execute a method for predicting and processing operational faults in an electric submersible screw pump according to any one of claims 1-6.

8. A non-volatile computer storage medium, characterized in that, The storage medium is a non-volatile computer-readable storage medium that stores at least one program, each program including instructions that, when executed by a terminal, cause the terminal to perform a method for predicting and handling operational faults in an electric submersible screw pump according to any one of claims 1-6.

Images