A method and system for predicting oil equipment failure based on big data

Abstract

The present invention relates to a method and system for predicting oil equipment failure based on big data, which belongs to the field of data processing, and comprises the following steps: obtaining sensor data of oil equipment; respectively determining abnormal factors of vibration sensor data, current sensor data and temperature sensor data at each moment; judging whether the sensor data is abnormal according to the abnormal factors; if so, entering the next step; otherwise, returning to continue monitoring; performing two-dimensional mapping on the vibration sensor data, the current sensor data and the temperature sensor data to form a two-dimensional vibration signal graph, a current signal graph and a temperature signal graph; performing fusion processing on the vibration signal graph, the current signal graph and the temperature signal graph to obtain a fusion signal graph; performing fault prediction according to the fusion signal graph through a multi-scale convolutional neural network to predict whether the oil equipment has a fault; if so, outputting the fault prediction result; otherwise, returning to continue monitoring.

Description

Technical Field

[0001] The present invention belongs to the field of data processing, and in particular relates to a method and system for predicting oil equipment failures based on big data. Background Art

[0002] Oil equipment (such as pumps, compressors, engines, etc.) often works in high temperature, high pressure, and corrosive environments. The stability and safety of equipment operation are crucial. If the equipment fails during operation, it may cause serious safety accidents such as explosions, fires, oil and gas leaks, etc. Unexpected failures and downtime of oil equipment will cause huge economic losses, especially for large oil platforms and refineries. Through effective fault detection, downtime can be greatly reduced and equipment utilization can be improved.

[0003] At present, the traditional method for fault detection of oil equipment is mainly based on the physical principles of equipment operation. By modeling the mechanics, thermal, and fluid dynamics of the system, equations describing the normal and abnormal operating states of the equipment are established. This method determines the fault by monitoring the deviation of parameters from the expected model. However, for complex oil equipment, the traditional physical model is difficult to accurately reflect the actual operation of the equipment, especially when multiple faults occur at the same time, the model is difficult to handle complex multi-fault modes.

[0004] With the rapid development of science and technology, more and more modern technologies are being applied to oil equipment fault detection. Spectral analysis methods based on Fourier transform, short-time Fourier transform and wavelet transform can extract frequency features from time series data such as vibration signals and current signals of equipment, thereby discovering periodic features and frequency components hidden in time domain signals, thereby helping to detect potential faults and improving the accuracy of oil equipment fault detection.

[0005] However, spectrum analysis methods can often only be used to analyze signals of a single dimension (such as vibration or current signals), and it is difficult to integrate information from multiple sensors for comprehensive fault detection. At the same time, spectrum analysis methods have good effects on simple, periodic fault signals (such as imbalance and bearing failures in rotating machinery), but when complex fault modes occur in the equipment, or multiple faults are superimposed, spectrum analysis methods are difficult to effectively distinguish and identify complex fault modes. Summary of the invention

[0006] In view of the above shortcomings of the prior art, the purpose of the invention is to provide a petroleum equipment fault prediction method and system based on big data, which can integrate vibration sensor data, current sensor data and temperature sensor data to perform comprehensive fault prediction. At the same time, through multi-scale convolutional neural network automatic fault prediction, it can distinguish and identify complex fault modes and improve the accuracy of petroleum equipment fault prediction.

[0007] The first aspect of the present invention provides a method for predicting oil equipment failure based on big data, comprising:

[0008] S1, acquiring sensor data of petroleum equipment, wherein the sensor data includes vibration sensor data, current sensor data and temperature sensor data;

[0009] S2, performing data anomaly detection on the vibration sensor data, the current sensor data, and the temperature sensor data, respectively, to determine abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment;

[0010] S3, combining the abnormal factors of the vibration sensor data, the current sensor data and the temperature sensor data at each moment, and determining whether the sensor data is abnormal; if so, proceed to the next step; otherwise, return to S1 and continue monitoring;

[0011] S4, performing two-dimensional mapping on the vibration sensor data, the current sensor data, and the temperature sensor data to form a two-dimensional vibration signal graph, a current signal graph, and a temperature signal graph;

[0012] S5, fusing the vibration signal graph, the current signal graph, and the temperature signal graph to obtain a fused signal graph;

[0013] S6, through a multi-scale convolutional neural network, according to the fusion signal graph, fault prediction is performed to predict whether there is a fault in the oil equipment; if so, the fault prediction result is output; otherwise, return to S1 and continue monitoring.

[0014] Furthermore, the abnormal factor is specifically a local abnormal factor LOF, and the calculation method of the abnormal factor of the vibration sensor data at each moment in S2 specifically includes:

[0015] S201, for each data point in the vibration sensor data, calculating a k-neighboring distance of each data point, wherein the k-neighboring distance is the distance between the kth nearest neighboring sample point and the current data point among the neighboring sample points closest to the current sample point;

[0016] S202, calculating a reachable distance based on the k-neighbor distance;

[0017] S203, calculating a local reachable density based on the reachable distance;

[0018] S204: Calculate a local anomaly factor based on the local reachable density.

[0019] Furthermore, the determining whether the sensor data is abnormal in S3 specifically includes:

[0020] S301, calculating a comprehensive abnormality factor according to abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment:

[0021]

[0022] Among them, τ c represents the comprehensive abnormality factor, τ z represents the abnormal factor of the vibration sensor data, τ d Represents the abnormal factor of the current sensor data, τ w represents the abnormal factor of the temperature sensor data, λ z Represents the weight coefficient of vibration sensor data, λ d Represents the weight coefficient of the current sensor data, λ w Represents the weight coefficient of temperature sensor data;

[0023] S302: When any of the following conditions is met, it is determined that the sensor data is abnormal:

[0024]

[0025] Among them, τ1 represents the first abnormal threshold, τ2 represents the second abnormal threshold, τ3 represents the third abnormal threshold, and τ4 represents the fourth abnormal threshold.

[0026] Furthermore, the S4 is specifically:

[0027] The vibration sensor data, the current sensor data and the temperature sensor data are two-dimensionally mapped through a symmetrical dot diagram pattern SDP to form a two-dimensional vibration signal diagram, a current signal diagram and a temperature signal diagram.

[0028] Furthermore, the S5 is specifically:

[0029] The vibration signal graph is used as the R channel, the current signal graph is used as the G channel, and the temperature signal graph is used as the B channel to generate a color image and obtain a fused signal graph.

[0030] Furthermore, the S6 specifically includes:

[0031] S601, using a first convolution kernel, a second convolution kernel, and a third convolution kernel of different sizes to perform feature extraction on the fused signal graph;

[0032] S602, fusing the extracted feature maps of different scales to obtain a fused feature map;

[0033] S603, perform fault prediction based on the fused feature graph to predict whether the oil equipment has a fault; if so, output the fault prediction result; otherwise, return to S1 and continue monitoring.

[0034] Furthermore, the size of the first convolution kernel is 7×1, the size of the second convolution kernel is 5×1, and the size of the third convolution kernel is 3×1.

[0035] Furthermore, the S602 is specifically as follows:

[0036] The extracted feature maps of different scales are fused through the tensor concatenation function to obtain a fused feature map.

[0037] Furthermore, the S602 specifically includes:

[0038] S6021, upsampling the large-scale first feature map extracted by the first convolution kernel so that the first feature map matches the size of the medium-scale second feature map extracted by the second convolution kernel;

[0039] S6022, performing weighted fusion on the upsampled first feature map and the second feature map to obtain an intermediate feature map;

[0040] S6023, upsampling the intermediate feature map so that the intermediate feature map matches the size of the small-scale third feature map extracted by the third convolution kernel;

[0041] S6024, performing weighted fusion on the upsampled intermediate feature map and the third feature map to obtain a fused feature map.

[0042] The second aspect of the present invention proposes a petroleum equipment fault prediction system based on big data, including: a memory and a processor.

[0043] An application is stored in the memory, and the application is suitable for being executed by the processor to implement the petroleum equipment fault prediction method based on big data of the first aspect.

[0044] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which can be loaded and executed by a processor to implement the big data-based petroleum equipment fault prediction method of the first aspect.

[0045] The beneficial effects of the present invention are as follows:

[0046] The method and system of the present invention can integrate vibration sensor data, current sensor data and temperature sensor data to perform comprehensive fault prediction. At the same time, through multi-scale convolutional neural network automatic fault prediction, it can distinguish and identify complex fault modes and improve the accuracy of petroleum equipment fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0048] Figure 1 A schematic diagram of a flow chart of a method for predicting oil equipment failure based on big data provided by an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of the structure of a petroleum equipment fault prediction system based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0051] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0052] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0054] The present invention proposes a method and system for predicting petroleum equipment faults based on big data, which solves the problem that existing spectrum analysis methods can only perform analysis based on single-dimensional signals (such as vibration or current signals) and are difficult to integrate information from multiple sensors for comprehensive fault prediction. At the same time, the spectrum analysis method has good effects on simple and periodic fault signals (such as imbalance and bearing failure of rotating machinery). However, when complex fault modes occur in the equipment, or multiple faults are superimposed, the spectrum analysis method is difficult to effectively distinguish and identify complex fault modes.

[0055] Method Embodiment

[0056] Reference Manual Attached Figure 1 , showing a flow chart of a method for predicting oil equipment failures based on big data provided by an embodiment of the present invention.

[0057] An embodiment of the present invention provides a method for predicting oil equipment failures based on big data, comprising:

[0058] Specifically, the method includes steps S1 to S6.

[0059] S1, obtain sensor data of oil equipment.

[0060] The sensor data includes vibration sensor data, current sensor data and temperature sensor data.

[0061] S2, performing data anomaly detection on the vibration sensor data, the current sensor data, and the temperature sensor data, respectively, and determining the abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment.

[0062] Optionally, the outlier factor is specifically a local outlier factor LOF.

[0063] In a possible implementation, the local outlier factor LOF is calculated as follows:

[0064] S201, for each data point in the vibration sensor data, calculate the k-neighboring distance of each data point, where the k-neighboring distance is the distance between the kth nearest neighboring sample point and the current data point among the neighboring sample points closest to the current sample point.

[0065] S202, based on the k-neighbor distance, calculate the reachable distance:

[0066] reachdist k ( p , o ) = max [ kdist ( o ), d ( p , o ) ]

[0067] Among them, reachdist k (p,o) represents the reachable distance from the current data point p to the neighboring sample point o, max represents the maximum value, kdist(o) represents the k-neighboring distance of the neighboring sample point o, and d(p,o) represents the actual distance from the current data point p to the neighboring sample point o.

[0068] S203, based on the reachable distance, calculate the local reachable density:

[0069]

[0070] Among them, LRD k (p) represents the local reachable density of the current data point p, o represents the neighboring sample points, N k (p) represents the set of k nearest neighboring sample points of the current data point p, and k represents the number of neighboring samples.

[0071] S204, based on the local reachable density, calculate the local anomaly factor:

[0072]

[0073] Among them, LOF(p) represents the local outlier factor of the current data point p, LRD k (o) represents the local reachability density of neighboring sample point o.

[0074] It should be noted that the calculation process of the local anomaly factor LOF for the current sensor data and the temperature sensor data may refer to the calculation process of the local anomaly factor LOF for the vibration sensor data above, and to avoid repetition, the present invention will not go into details.

[0075] In the present invention, LOF can effectively detect local anomalies, that is, a data point that is abnormal compared to its neighboring data points. Faults in oil equipment usually manifest as local, sudden anomalies. The local anomaly factor can quickly detect these anomalies by comparing the local density of each data point.

[0076] It should be noted that sensor data has a certain correlation within a period of time. If the correlation between a data point and other data points is abnormal, it often means that the equipment has failed. Therefore, in the process of calculating the abnormality factor, in addition to the local abnormality factor LOF, the correlation factor is also introduced.

[0077] Furthermore, the correlation factors are specifically:

[0078]

[0079] Among them, C(p) represents the correlation factor of the current data point p, N represents the set of data points in the sliding window, n represents the total number of data points in the sliding window, and cor(p,q) represents the correlation between the current data point p and other data points q in the sliding window.

[0080] Optionally, the correlation degree may be specifically a Pearson Correlation Coefficient, a Spearman's Rank Correlation Coefficient, or a cosine similarity.

[0081] Specifically, taking the cosine similarity (value range 0 to 1) as an example, the correlation factor is introduced. Combining the correlation factor and the local anomaly factor, the anomaly factor is specifically:

[0082] t ( p ) = [ 1 − C ( p )] LOF ( p )

[0083] Among them, τ(p) represents the abnormal factor of the current data point p.

[0084] In the present invention, the local anomaly factor mainly measures the local density difference between a data point and its neighboring data points, detects local anomalies in the data, and is suitable for identifying data points that appear inconsistent in the local data distribution. The correlation factor reveals the trend consistency of the data point in the time series by calculating the correlation between the data point and other data points in its time window. By comprehensively considering LOF and correlation factors to determine the anomaly factor, it is possible to detect local anomalies of the data (such as sudden failures) and capture changes in temporal correlation (such as changes in trends), which can more comprehensively and accurately reflect the abnormal conditions of the equipment.

[0085] S3, comprehensively analyze the abnormal factors of vibration sensor data, current sensor data and temperature sensor data at each moment, and judge whether the sensor data is abnormal. If so, proceed to the next step. Otherwise, return to S1 and continue monitoring.

[0086] In a possible implementation manner, determining whether the sensor data is abnormal in S3 specifically includes:

[0087] S301, calculating a comprehensive abnormality factor based on abnormal factors of vibration sensor data, current sensor data, and temperature sensor data at each moment:

[0088]

[0089] Among them, τ c represents the comprehensive abnormality factor, τ z represents the abnormal factor of the vibration sensor data, τ d Represents the abnormal factor of the current sensor data, τ w represents the abnormal factor of the temperature sensor data, λ z Represents the weight coefficient of vibration sensor data, λ d Represents the weight coefficient of the current sensor data, λ w Represents the weight coefficient of the temperature sensor data.

[0090] It should be noted that by flexibly adjusting the sensor weight coefficient, it is possible to adapt to the needs of different equipment and working conditions. For example, for some equipment, vibration data may be more indicative of faults, and vibration sensors can be given higher weights; while in other scenarios, current or temperature anomalies are more important. This improves the adaptability and flexibility of the detection algorithm.

[0091] S302: When any of the following conditions is met, it is determined that the sensor data is abnormal:

[0092]

[0093] Among them, τ1 represents the first abnormal threshold, τ2 represents the second abnormal threshold, τ3 represents the third abnormal threshold, and τ4 represents the fourth abnormal threshold.

[0094] Among them, those skilled in the art can set the first abnormal threshold, the second abnormal threshold, the third abnormal threshold and the fourth abnormal threshold according to actual conditions, and the present invention does not limit this.

[0095] It should be noted that when the abnormal factor of a certain sensor exceeds the threshold, the sensor data can be judged to be abnormal. This logic is suitable for capturing significant abnormalities in a certain dimension and issuing timely warnings. Even if the abnormal factor of a single sensor does not exceed the threshold, through the calculation of the comprehensive abnormal factor, if the comprehensive abnormal factor exceeds the threshold, it can also be judged that there is an overall fault and an alarm can be issued. This logic is suitable for handling situations where multiple sensors have slight deviations, and improves the sensitivity of detection through comprehensive analysis.

[0096] In the present invention, by combining the vibration, current and temperature sensor data, calculating the comprehensive abnormality factor, and setting a variety of flexible abnormality judgment conditions, it is possible to preliminarily detect the presence of abnormalities in the data and then perform subsequent refined detection. If the preliminary detection data does not show abnormalities, there is no need for subsequent refined detection, which reduces the amount of calculation for equipment fault prediction and is conducive to real-time detection.

[0097] S4, performing two-dimensional mapping on the vibration sensor data, the current sensor data and the temperature sensor data to form a two-dimensional vibration signal graph, a current signal graph and a temperature signal graph.

[0098] In a possible implementation, S4 specifically includes: performing two-dimensional mapping on the vibration sensor data, the current sensor data, and the temperature sensor data through a symmetrical dot diagram pattern SDP to form a two-dimensional vibration signal diagram, a current signal diagram, and a temperature signal diagram.

[0099] Among them, Symmetrized Dot Pattern (SDP) is a technology that converts one-dimensional time series signals into two-dimensional images. It was first used in the field of speech signal processing and later gradually applied to other signal analysis fields. The core idea of ​​SDP is to map time series signals into images through polar coordinates and use symmetrical structures to show the changing characteristics of signals. It can not only show local features in the time series, but also retain global trend information.

[0100] Furthermore, S4 specifically includes sub-steps S401 to S403:

[0101] S401, normalize the vibration sensor data:

[0102]

[0103] Among them, R i represents the normalized data value of the ith data point, x i represents the original data value of the i-th data point, x min Indicates the minimum data value, x max Indicates the maximum data value.

[0104] S402, determining the symmetric mapping angle of each data point:

[0105]

[0106] in, represents the forward mapping angle of the i-th data point, represents the reverse mapping angle of the i-th data point, represents the initial angle, δ represents the rotation range, l represents the time lag coefficient, R i+l Represents the normalized data value of the i+lth data point.

[0107] S403 , using the normalized data value as the radius, and using the forward mapping angle and the reverse mapping angle as mapping angles, to generate a two-dimensional vibration signal graph.

[0108] In the present invention, SDP converts one-dimensional time series data into two-dimensional images, which can more intuitively display the patterns, trends and periodicity of the data. This conversion retains the time series characteristics of the signal and presents more local and global information in the form of images. Compared with directly analyzing time series data, the two-dimensional image form is easier to capture complex fault characteristics, especially for nonlinear and non-stationary signals, and can reflect potential anomalies through changes in symmetry and shape.

[0109] S5, fusing the vibration signal graph, the current signal graph, and the temperature signal graph to obtain a fused signal graph.

[0110] In a possible implementation, S5 specifically includes: using the vibration signal graph as the R channel, the current signal graph as the G channel, and the temperature signal graph as the B channel to generate a color image and obtain a fused signal graph.

[0111] In the present invention, the three types of sensor data are mapped to different channels of the RGB image, so that each type of data can be displayed simultaneously in the same image. This processing method can effectively fuse multi-dimensional data, thereby retaining the characteristics of different data such as vibration, current and temperature. This fused image shows the correlation of the three types of data at the same time, which helps to more comprehensively reflect the overall operating status of the device.

[0112] S6, through the multi-scale convolutional neural network, according to the fusion signal graph, predict whether the oil equipment has a fault. If so, output the fault prediction result. Otherwise, return to S1 and continue monitoring.

[0113] Among them, the convolutional neural network (CNN) is a deep learning model specifically designed to process grid-structured data (such as images, audio, etc.), and is particularly suitable for processing two-dimensional data and images. Through a series of convolution operations, pooling operations, and fully connected layers, it can automatically extract local and global features of data, thereby achieving outstanding performance in tasks such as classification and detection.

[0114] In a possible implementation, S6 specifically includes sub-steps S601 to S603:

[0115] S601, using the first convolution kernel, the second convolution kernel and the third convolution kernel of different sizes to extract features from the fused signal graph.

[0116] Optionally, the size of the first convolution kernel is 7×1, the size of the second convolution kernel is 5×1, and the size of the third convolution kernel is 3×1.

[0117] In one possible implementation, an attention mechanism may be introduced into a convolutional neural network.

[0118] First, compress the feature map extracted by the convolution kernel:

[0119]

[0120] Among them, Y represents the feature map after compression processing, S represents the compression function, X represents the input feature map, including the state feature map, frequency domain feature map and fault density feature map, (h, w) represents the coordinates of the similarity point, H represents the height of the feature map, and W represents the width of the feature map.

[0121] After that, the excitation is processed through two fully connected layers:

[0122] E ( Y ) = S i g m o i d [ W s ( R e L U ( W r ( Y ) ) ) ]

[0123] Among them, E represents the attention weight function, Y represents the feature map after compression processing, and W s represents the reconstruction of the fully connected layer, W r Represents a compressed fully connected layer, Sigmoid represents the Sigmoid activation function, and ReLU represents the ReLU activation function.

[0124] Furthermore, the attention weight function is multiplied with the feature map to highlight the target area that contributes to fault prediction and suppress irrelevant areas that do not contribute to fault prediction.

[0125] In the present invention, the dynamic and nonlinear interdependence between feature channels is determined by squeezing and excitation, the weight distribution of each feature channel is determined, and then the important features are enhanced and the useless and redundant features are suppressed according to the channel weights, thereby simplifying the learning process and improving the quality of the output features and the expressive power of the network.

[0126] S602, fusing the extracted feature maps of different scales to obtain a fused feature map.

[0127] In a possible implementation manner, S602 specifically includes:

[0128] Through the tensor concatenation function, the extracted feature maps of different scales are fused to obtain a fused feature map:

[0129]

[0130] Among them, F c Represents a fused feature map, torch.cat represents a tensor concatenation function, f1 represents the first feature map extracted by the first convolution kernel, f2 represents the second feature map extracted by the second convolution kernel, and f3 represents the third feature map extracted by the third convolution kernel.

[0131] Among them, the concatenation tensor function is a function used to concatenate tensors in PyTorch. It can concatenate multiple tensors into one tensor along the specified dimension. The concatenation tensor function is often used to fuse features in neural networks, especially to concatenate the outputs of different convolutional layers as the input of subsequent networks.

[0132] In the present invention, feature maps are spliced ​​through tensor splicing functions, and feature fusion is not forced. Instead, the independence of each scale feature is retained, which can ensure that multi-scale features are effectively utilized, avoid information loss during feature fusion, and process fault features in complex scenarios more carefully.

[0133] It should be noted that tensor concatenation simply connects different feature maps by dimension without any feature weighting mechanism. Therefore, each feature map has the same influence in the fusion result after concatenation, and cannot reflect the difference in feature importance. However, in the fault prediction scenario, the importance of feature maps of different scales is different. Large-scale feature maps are often used to capture the overall change trend of the equipment and are suitable for identifying long-term progressive faults. Small-scale feature maps are more sensitive to local anomalies or sudden fault signals and are suitable for detecting short-term spikes or shocks. For different detection requirements, large-scale feature maps and small-scale feature maps differ in importance.

[0134] In a possible implementation, S602 specifically includes steps S6021 to S6024:

[0135] S6021, up-sample the large-scale first feature map extracted by the first convolution kernel so that the first feature map matches the size of the medium-scale second feature map extracted by the second convolution kernel.

[0136] S6022, perform weighted fusion on the upsampled first feature map and the second feature map to obtain an intermediate feature map.

[0137] S6023, up-sample the intermediate feature map so that the intermediate feature map matches the size of the small-scale third feature map extracted by the third convolution kernel.

[0138] S6024, performing weighted fusion on the upsampled intermediate feature map and the third feature map to obtain a fused feature map.

[0139] In the present invention, through the multi-scale feature fusion strategy of layer-by-layer upsampling and weighted fusion, weights can be allocated according to the importance of feature maps, while effectively retaining feature information of different scales, and enhancing the comprehensive analysis capability of global and local features. It can flexibly adapt to multiple fault modes, ensure that information is not lost and models are not confused, and improve the accuracy and sensitivity of fault prediction.

[0140] S603, based on the fusion feature graph, perform fault prediction to predict whether the oil equipment has a fault. If so, output the fault prediction result. Otherwise, return to S1 and continue monitoring.

[0141] Specifically, the fused feature map can be used to predict faults through a detection head with a Softmax activation function to determine the specific fault type. The Softmax activation function can convert the multi-class prediction results output by the network into a probability distribution, ensuring that the sum of the probabilities of each category is 1. In this way, the model can clearly give the probability of each fault type, thereby helping to identify the type of the current fault of the device.

[0142] In the present invention, fault prediction is performed on the fused signal graph through a multi-scale convolutional neural network, multi-scale feature extraction and automated feature learning can be performed, the accuracy of fault prediction can be improved, and it can also adapt to complex equipment operating environments and various fault types, effectively improving the health monitoring level of the equipment and ensuring the safe and stable operation of the equipment.

[0143] The traditional training method for convolutional neural networks is often to use the gradient descent method. However, the gradient descent method relies on gradient information. In complex high-dimensional space, the surface of the loss function may contain multiple local optimal solutions. Since the gradient descent method is a local optimization algorithm, the model may be stuck at the local optimal point, causing the network to be unable to find the global optimal solution, thereby affecting the final performance of the model. The present invention proposes to use the firefly algorithm to determine the network parameters of the multi-scale convolutional neural network.

[0144] Among them, the Firefly Algorithm (FA) is a meta-heuristic optimization algorithm based on swarm intelligence. The algorithm finds the optimal solution by simulating the attraction behavior between fireflies in nature based on light intensity. The brightness of each firefly individual represents its fitness (the quality of the solution), and they attract each other based on the brightness difference between individuals. The brighter the firefly, the stronger its attraction, and other fireflies will approach it, and eventually the entire population gradually converges to the global optimal solution.

[0145] Specifically, the inverse of the mean square error loss function or the cross entropy loss function can be used as the fitness function of the firefly algorithm.

[0146] After that, initialize the firefly individuals, each of which represents a feasible set of network parameters, including the weights, sizes, biases, number of convolutional layers, number of pooling layers, and pooling types of each convolution kernel. Each firefly individual consists of multiple dimensional components, each of which represents a certain network parameter.

[0147] Compare each firefly individual with other individuals in the population, generate a random number u, and move according to the following adaptive movement rules:

[0148]

[0149] in, represents the position of the a-th firefly individual at the t+1-th iteration, ω represents the inertia weight factor, represents the position of the a-th firefly individual at the t-th iteration, represents the position of the cth firefly individual at the tth iteration, β represents the attraction, α represents the step factor, sign represents the sign function, r represents a random number in the range of 0 to 1, Levy represents the Levy flight random step, exp represents the exponential function with the natural constant e as the base, b represents the logarithmic spiral shape parameter, g represents a uniform random number between -1 and 1, R t represents the adaptive switching factor at the tth iteration, and u represents a uniform random number between 0 and 1.

[0150] In the present invention, through the adaptive switching factor, the algorithm can dynamically adjust the search strategy according to the change of the global optimal fitness during the iteration process. When the fitness value changes greatly, the algorithm tends to use Levy flight for global search, and when the fitness change is small, it tends to use logarithmic spiral motion for local search. This adaptive adjustment mechanism enables the algorithm to have better adaptive ability and can flexibly respond to the optimization needs of different stages. In the early stage, the algorithm can focus on global exploration, and gradually turn to local fine-tuning in the later stage, which improves the overall efficiency and convergence of the algorithm.

[0151] Optionally, the adaptive switching factor is specifically:

[0152]

[0153] in, represents the fitness value of the global best position at the t-1th iteration, Represents the fitness value of the global best position at the t-2th iteration.

[0154] In the present invention, the adaptive switching factor dynamically adjusts the search strategy through the change of the global optimal fitness value of the first two iterations. When the global optimal fitness values ​​of the first two iterations differ greatly, the value of the adaptive switching factor is small, and the algorithm is more inclined to global search. When these two values ​​are close, it indicates that the change of the global optimum slows down, the value of the adaptive switching factor increases, and the algorithm turns to more refined local optimization. By adaptively adjusting according to the change of fitness, the algorithm can be balanced at different stages. This makes the algorithm tend to perform global search in the early stage, and gradually focus on local optimization in the later stage, which improves the flexibility and efficiency of the algorithm.

[0155] Optionally, the attraction is specifically:

[0156]

[0157] Among them, β0 represents the attraction between two fireflies when the distance between them is 0, exp represents the exponential function with the natural constant e as the base, γ represents the light intensity attraction coefficient, and r ac represents the Euclidean distance between the a-th firefly individual and the c-th firefly individual.

[0158] In the present invention, by introducing an exponential decay function, the attraction between individual fireflies decreases exponentially with the increase of the Euclidean distance. That is, when the distance between two individuals is small, the attraction is large and the individuals will quickly approach each other. When the distance is large, the attraction is weakened, thereby maintaining a larger exploration range.

[0159] Optionally, the Levy flight random step size is specified as:

[0160]

[0161] Among them, u means obedience Normal distribution, v means obey The normal distribution of , λ represents the exponential parameter.

[0162] s u = { t ( 1 + l ) s i n ( π l / 2 ) t [ ( 1 + l ) / 2 ] l 2 ( l − 1 ) / 2 } 1 / l , s v = 1

[0163] Among them, τ is the standard Gamma function.

[0164] In the present invention, Levy flight is a random step distribution pattern with a long-tail distribution. Unlike normal distribution, Levy distribution allows for larger jump steps, which means that individuals can jump over a large range in the search space. Since the normal distribution of parameter v allows the random step to take a larger value, individuals will conduct long-distance exploration in some cases. This mechanism enables the algorithm to jump out of the local optimal solution trap and conduct a large-scale global search, increasing the possibility of finding the global optimal solution. This is particularly important in complex nonlinear optimization problems, as it can avoid falling into the local optimum.

[0165] After the firefly individual moves to a new location, update the individual fitness;

[0166] Determine whether the current number of iterations has reached the maximum number of iterations; if so, output the network parameter set represented by the firefly individual with the highest current fitness; otherwise, return to continue iterating.

[0167] In the present invention, the firefly algorithm realizes the effective combination of global search and local fine-tuning through multiple mechanisms such as attraction between individuals, Levy flight step, inertia weight and adaptive switching factor. It can better deal with the high-dimensional complexity, local optimality and search efficiency problems in the optimization of convolutional neural network parameters. The set of convolutional neural network parameters optimized by this method can significantly improve the performance of multi-scale convolutional neural networks, especially in complex tasks such as fault prediction.

[0168] The beneficial effects of the present invention are as follows:

[0169] The method and system of the present invention can integrate vibration sensor data, current sensor data and temperature sensor data to perform comprehensive fault prediction. At the same time, through multi-scale convolutional neural network automatic fault prediction, it can distinguish and identify complex fault modes and improve the accuracy of petroleum equipment fault prediction.

[0170] System Example

[0171] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a petroleum equipment fault prediction system based on big data provided by an embodiment of the present invention.

[0172] The present invention proposes a petroleum equipment fault prediction system 30 based on big data, including: a memory 303 and a processor 301.

[0173] The memory 303 stores an application program, and the application program is suitable for being executed by the processor 301 to implement the petroleum equipment fault prediction method based on big data of the method embodiment.

[0174] The oil equipment fault prediction system 30 based on big data includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, through a bus 302.

[0175] The structure of the petroleum equipment failure prediction system 30 based on big data does not constitute a limitation on the embodiment of the present invention.

[0176] The beneficial effects of the present invention are as follows:

[0177] The method and system of the present invention can integrate vibration sensor data, current sensor data and temperature sensor data to perform comprehensive fault prediction. At the same time, through multi-scale convolutional neural network automatic fault prediction, it can distinguish and identify complex fault modes and improve the accuracy of petroleum equipment fault prediction.

[0178] Processor 301 may be a CPU, a general purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. Processor 301 may also be a combination that implements computing functions, such as a combination of microprocessors, a combination of DSPs and microprocessors, etc.

[0179] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI bus or an EISA bus, etc. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0180] The memory 303 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0181] Computer Readable Storage Medium Embodiments

[0182] The present invention proposes a computer-readable storage medium having a computer program stored thereon, which can be loaded and executed by a processor to implement a petroleum equipment fault prediction method based on big data according to the first aspect.

[0183] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting oil equipment failure based on big data, characterized in that: include: S1, acquiring sensor data of petroleum equipment, wherein the sensor data includes vibration sensor data, current sensor data and temperature sensor data; S2, performing data anomaly detection on the vibration sensor data, the current sensor data, and the temperature sensor data, respectively, to determine abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment; S3, combining the abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment to determine whether the sensor data is abnormal; If yes, go to the next step; Otherwise, return to S1 and continue monitoring; S4, performing two-dimensional mapping on the vibration sensor data, the current sensor data, and the temperature sensor data to form a two-dimensional vibration signal graph, a current signal graph, and a temperature signal graph; S5, fusing the vibration signal graph, the current signal graph, and the temperature signal graph to obtain a fused signal graph; S6, using a multi-scale convolutional neural network, based on the fused signal graph, predicting whether the oil equipment has a fault; if so, outputting the fault prediction result; otherwise, returning to S1 and continuing monitoring; The abnormal factor is determined by comprehensively considering the local abnormal factor LOF and the correlation factor; the calculation method of the abnormal factor of the vibration sensor data at each moment in S2 specifically includes: S201, for each data point in the vibration sensor data, calculating a k-neighboring distance of each data point, wherein the k-neighboring distance is the distance between the kth nearest neighboring sample point and the current data point among the neighboring sample points closest to the current sample point; S202, based on the k-neighbor distance, calculate the reachable distance: ； Among them, reachdist k (p,o) represents the reachable distance from the current data point p to the neighboring sample point o, max represents the maximum value, kdist(o) represents the k-neighboring distance of the neighboring sample point o, and d(p,o) represents the actual distance from the current data point p to the neighboring sample point o; S203: Calculate the local reachable density based on the reachable distance: ； Among them, LRD k (p) represents the local reachable density of the current data point p, o represents the neighboring sample points, N k (p) represents the set of k nearest sample points of the current data point p, where k represents the number of neighboring samples; S204: Calculate a local anomaly factor based on the local reachable density: ； Among them, LOF(p) represents the local outlier factor of the current data point p, LRD k (o) represents the local reachability density of neighboring sample point o; S205, in the process of calculating the anomaly factor, in addition to the local anomaly factor LOF, a correlation factor is also introduced to calculate the correlation factor: ； Where C(p) represents the correlation factor of the current data point p, N represents the set of data points in the sliding window, n represents the total number of data points in the sliding window, and cor(p,q) represents the correlation between the current data point p and other data points q in the sliding window; S206, the correlation degree uses cosine similarity, introduces the correlation factor, combines the correlation factor and the local anomaly factor, and calculates the anomaly factor: ； Among them, τ(p) represents the abnormal factor of the current data point p; The step of determining whether the sensor data is abnormal in S3 specifically includes: S301, calculating a comprehensive abnormality factor according to abnormal factors of the vibration sensor data, the current sensor data, and the temperature sensor data at each moment: ； Among them, τ c represents the comprehensive abnormality factor, τ z represents the abnormal factor of the vibration sensor data, τ d Represents the abnormal factor of the current sensor data, τ w represents the abnormal factor of the temperature sensor data, λ z Represents the weight coefficient of vibration sensor data, λ d Represents the weight coefficient of the current sensor data, λ w Represents the weight coefficient of temperature sensor data; S302: When any of the following conditions is met, it is determined that the sensor data is abnormal: ； Among them, τ1 represents the first abnormal threshold, τ2 represents the second abnormal threshold, τ3 represents the third abnormal threshold, and τ4 represents the fourth abnormal threshold.

2. The method for predicting oil equipment failure based on big data according to claim 1 is characterized in that: The S4 is specifically: The vibration sensor data, the current sensor data and the temperature sensor data are two-dimensionally mapped through a symmetrical dot diagram pattern SDP to form a two-dimensional vibration signal diagram, a current signal diagram and a temperature signal diagram.

3. The method for predicting oil equipment failure based on big data according to claim 1 is characterized in that: The S5 is specifically: The vibration signal graph is used as the R channel, the current signal graph is used as the G channel, and the temperature signal graph is used as the B channel to generate a color image and obtain a fused signal graph.

4. The method for predicting oil equipment failure based on big data according to claim 1 is characterized in that: The S6 specifically includes: S601, using a first convolution kernel, a second convolution kernel, and a third convolution kernel of different sizes to perform feature extraction on the fused signal graph; S602, fusing the extracted feature maps of different scales to obtain a fused feature map; S603, perform fault prediction based on the fused feature graph to predict whether the oil equipment has a fault; if so, output the fault prediction result; otherwise, return to S1 and continue monitoring.

5. The method for predicting oil equipment failure based on big data according to claim 4 is characterized in that: The size of the first convolution kernel is 7×1, the size of the second convolution kernel is 5×1, and the size of the third convolution kernel is 3×1.

6. The method for predicting oil equipment failure based on big data according to claim 4 is characterized in that: The S602 is specifically as follows: The extracted feature maps of different scales are fused through the tensor concatenation function to obtain a fused feature map.

7. The method for predicting oil equipment failure based on big data according to claim 4 is characterized in that: The S602 specifically includes: S6021, upsampling the large-scale first feature map extracted by the first convolution kernel so that the first feature map matches the size of the medium-scale second feature map extracted by the second convolution kernel; S6022, performing weighted fusion on the upsampled first feature map and the second feature map to obtain an intermediate feature map; S6023, upsampling the intermediate feature map so that the intermediate feature map matches the size of the small-scale third feature map extracted by the third convolution kernel; S6024, performing weighted fusion on the upsampled intermediate feature map and the third feature map to obtain a fused feature map.

8. A petroleum equipment failure prediction system based on big data, characterized in that: include: Memory and processor; An application is stored in the memory, and the application is suitable for being executed by the processor to implement the petroleum equipment fault prediction method based on big data as described in any one of claims 1 to 7.

Images