Slurry pump operation monitoring and optimizing system and method based on machine learning

Abstract

The invention relates to the technical field of operation monitoring, and discloses a slurry pump operation monitoring and optimizing system and method based on machine learning. According to the method, firstly, slurry pump operation indexes are obtained, original slurry pump operation data are obtained, abnormal data detection and deletion and missing data filling are carried out, and data preprocessing is completed; secondly, calculating a self-adaptive threshold value by using historical slurry pump operation data, verifying the safety of the slurry pump operation data based on the self-adaptive threshold value, and completing slurry pump operation monitoring; training the support vector regression model to obtain a support vector regression prediction model, and outputting a slurry pump operation data prediction value; and finally, optimizing the operation indexes of the slurry pump by using an escape optimization algorithm according to the predicted value of the operation data of the slurry pump and the operation indexes of the slurry pump so as to realize operation optimization of the slurry pump. By processing and analyzing the operation data of the slurry pump, the purposes of monitoring and optimizing the operation of the slurry pump are achieved, and the method is accurate and objective.

Description

Technical Field

[0001] The present invention relates to the technical field of operation monitoring, and specifically to a slurry pump operation monitoring and optimization system and method based on machine learning. Background Art

[0002] The slurry pump is one of the key equipment for material transportation in coal preparation plants. It is widely used in the transportation of materials such as coal slime water, circulating water, coal slurry and heavy medium suspension. Its safe, stable and efficient operation plays a very important role in improving the production efficiency of coal preparation plants.

[0003] The traditional method of monitoring the operation of slurry pumps mainly relies on employee experience or related testing equipment for testing, and uses the experience and equipment results as the final results of the slurry pump operation. There are problems such as low accuracy of slurry pump operation monitoring and slow recognition speed. At the same time, there is a lot of slurry pump operation data, and real-time and accurate detection cannot be achieved. There is no use of high-tech such as neural networks, resulting in the health status of the slurry pump being unknown, posing a safety hazard. Summary of the invention

[0004] In response to the problems in the related technology, the present invention provides a slurry pump operation monitoring and optimization system and method based on machine learning to overcome the above-mentioned technical problems existing in the existing related technology.

[0005] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is a slurry pump operation monitoring and optimization method based on machine learning, comprising the following steps:

[0007] S1. Acquire the slurry pump operation index, obtain the original slurry pump operation data, generate the original slurry pump operation data matrix, perform abnormal data detection and deletion and missing data filling on the original slurry pump operation data matrix, and obtain the processed slurry pump operation data matrix;

[0008] S2. Acquire historical slurry pump operation data, calculate an adaptive threshold value according to the historical slurry pump operation data, verify the security of the processed slurry pump operation data in the processed slurry pump operation data matrix based on the adaptive threshold value, and complete the slurry pump operation monitoring;

[0009] S3, training a support vector regression model to obtain a support vector regression prediction model, performing prediction based on the processed slurry pump operation data matrix, and outputting a predicted value of the slurry pump operation data;

[0010] S4. Establish a fitness function according to the predicted value of the slurry pump operation data and the slurry pump operation index, use the escape optimization algorithm to optimize the slurry pump operation index, obtain the optimized slurry pump operation index by finding the best fitness function value, and complete the slurry pump operation optimization.

[0011] The invention obtains the original slurry pump operation data, uses the improved isolation forest algorithm to calculate the abnormal score of the original slurry pump operation data, and then uses the clustering algorithm to calculate the abnormal threshold, and detects and deletes the abnormal data by comparing the abnormal threshold and the abnormal score; uses the k-nearest neighbor missing data filling method to perform weighted processing on the original slurry pump operation data to achieve missing data filling; the improved isolation forest algorithm can effectively detect abnormal data, has high applicability, and improves the processing speed of slurry pump operation data; the k-nearest neighbor missing data filling method takes into account the difference between each data and effectively improves the accuracy of data filling; secondly, obtains the historical slurry pump operation data, calculates the adaptive threshold, and compares the adaptive threshold to obtain the abnormal data. The value and the slurry pump operation data are combined to obtain whether the slurry pump is in a safe state; this method avoids misjudgment caused by too high thresholds and too low thresholds, relies on historical data to dynamically adjust the threshold, and has a good monitoring effect on the slurry pump operation; then the support vector regression model is trained to obtain the support vector regression prediction model, and the slurry pump operation data prediction value is output; the prediction model has a good effect on the prediction of small sample data and has high prediction accuracy; finally, a fitness function is established, and the slurry pump operation index is optimized based on the escape optimization algorithm, and the optimized slurry pump operation index is obtained by finding the optimal fitness function value; the algorithm solves complex optimization problems by simulating the behavior of the crowd during the evacuation process, and has good optimization performance and strong optimization ability.

[0012] Preferably, the S1 comprises the following steps:

[0013] S11, obtaining the slurry pump operation index, which includes temperature, current, voltage, flow, pressure, etc., using sensor equipment to obtain the data of the slurry pump operation index, and obtaining the original slurry pump operation data. The original slurry pump operation data corresponding to each slurry pump operation index constitutes an original slurry pump operation data set. The first n original slurry pump operation data in the original slurry pump operation data set are selected to generate the original slurry pump operation data matrix A as follows:

[0014]

[0015] Among them, a mn The nth original slurry pump operation data representing the mth slurry pump operation index;

[0016] S12, using an improved isolation forest algorithm to detect and delete abnormal data on the original slurry pump operation data matrix to obtain a processed slurry pump operation data matrix, the specific steps are as follows:

[0017] S121, select any slurry pump operation index in the original slurry pump operation data matrix to obtain a slurry pump operation data set to be detected, set a dimension set corresponding to the slurry pump operation data to be detected in the slurry pump operation data set to be detected, select any dimension in the dimension set and record it as a cutting point dimension; use the slurry pump operation data to be detected corresponding to the cutting point dimension as a cutting point, add the slurry pump operation data to be detected that is less than or equal to the cutting point in the slurry pump operation data set to be detected to the first left subtree, and add the slurry pump operation data to be detected that is greater than the cutting point in the slurry pump operation data set to be detected to the first right subtree to obtain an isolated tree;

[0018] S122, setting the maximum depth of the isolated tree, repeatedly cutting in the first left subtree and the first right subtree to obtain the second left subtree and the second right subtree, until the isolated tree reaches the maximum depth of the isolated tree or the left subtree and the right subtree contain only one slurry pump operation data to be detected, stop cutting, and obtain the final isolated tree; cutting the original slurry pump operation data corresponding to other slurry pump operation indicators in the original slurry pump operation data matrix to generate an isolated forest;

[0019] S123, selecting a data point to be detected from the slurry pump operation data set to be detected, setting the path length of the data point to be detected in the isolation forest to b, β represents a normalization factor, and the calculation formula for the abnormal score of the data point to be detected is as follows:

[0020]

[0021] Among them, B represents the anomaly score of the data point to be detected;

[0022] sequentially calculating the abnormality scores of all the slurry pump operation data to be detected in the slurry pump operation data set to be detected, to obtain an abnormality score set;

[0023] S124, clustering the abnormal score set, selecting an initial cluster center, calculating the distance between the abnormal score in the abnormal score set and the initial cluster center, dividing the abnormal scores in the abnormal score set into the cluster where the nearest initial cluster center is located, and then recalculating the center of the cluster until the center of the cluster no longer changes, clustering the abnormal score set into a first cluster and a second cluster; selecting the first cluster and the second cluster with fewer abnormal scores as the final cluster, and taking the minimum abnormal score in the final cluster as the abnormal threshold; recording the slurry pump operation data to be detected that is greater than the abnormal threshold in the abnormal score set as abnormal data, and deleting it to obtain a processed slurry pump operation data set, and sequentially deleting the abnormal data from the original slurry pump operation data matrix to obtain a processed slurry pump operation data matrix;

[0024] S13, selecting any processed slurry pump operation data set in the processed slurry pump operation data matrix, recorded as the slurry pump operation data set to be filled, selecting k nearest neighbor data at the missing data in the slurry pump operation data set to be filled, calculating the distance and entropy value of the k nearest neighbor data and the missing data, and then calculating the coefficient of variation of the k nearest neighbor data, and performing weighted processing on the k nearest neighbor data. The calculation formula is as follows:

[0025]

[0026] Among them, B c ′ represents the weight coefficient of the cth nearest neighbor data, b c ′ represents the coefficient of variation of the cth nearest neighbor data, c = 1, 2, 3, ..., k;

[0027] Setting b c ″ represents the cth nearest neighbor data. The missing data is calculated according to the weighted coefficient of the cth nearest neighbor data. The calculation formula is as follows:

[0028]

[0029] in, Indicates missing data in the slurry pump operation data set to be filled;

[0030] All missing data in the slurry pump operation data set to be filled are calculated in sequence to obtain a processed slurry pump operation data set, and generate a processed slurry pump operation data matrix.

[0031] The invention obtains the original slurry pump operation data, uses the improved isolation forest algorithm to calculate the anomaly score of the original slurry pump operation data, and detects and deletes the abnormal data by comparing the threshold and the anomaly score. It can effectively detect the abnormal data, has high applicability, and improves the processing speed of the slurry pump operation data; then uses the k-nearest neighbor missing data filling method to perform weighted processing on the original slurry pump operation data to achieve missing data filling, takes into account the degree of difference of each data, and effectively improves the accuracy of data filling.

[0032] Preferably, S2 comprises the following steps:

[0033] S21, acquiring historical slurry pump operation data, forming a historical slurry pump operation data set, generating a historical slurry pump operation data matrix, performing abnormal data detection and deletion and missing data filling on the historical slurry pump operation data matrix, obtaining a processed historical slurry pump operation data matrix, selecting any processed historical slurry pump operation data set from the processed historical slurry pump operation data matrix, calculating the mean and standard deviation of the processed historical slurry pump operation data in the processed historical slurry pump operation data set, and the adaptive threshold calculation formula is as follows:

[0034]

[0035] Where ω represents the adaptive threshold, Represents the mean of the processed historical slurry pump operation data, represents the standard deviation of the processed historical slurry pump operation data, and χ represents the hyperparameter;

[0036] S22. Based on the adaptive threshold and the standard deviation of the processed historical slurry pump operation data, the monitoring threshold calculation formula is as follows:

[0037]

[0038] Among them, ξ represents the monitoring threshold;

[0039] A corresponding processed slurry pump operation data set is selected from the processed slurry pump operation data matrix. When the processed slurry pump operation data in the processed slurry pump operation data set is greater than the monitoring threshold, the slurry pump operation is in a dangerous state and an early warning is issued; otherwise, the slurry pump operation is in a safe state and the slurry pump operation monitoring is completed.

[0040] Preferably, S3 comprises the following steps:

[0041] S31, normalizing the historical slurry pump operation data matrix generated by the historical slurry pump operation data to obtain a normalized historical slurry pump operation data matrix, selecting any normalized historical slurry pump operation data set from the normalized historical slurry pump operation data matrix, and dividing the normalized historical slurry pump operation data set into a sample training set and a sample test set; training a support vector regression model to obtain a support vector regression prediction model, the specific steps are as follows:

[0042] S311, setting the kernel function of the support vector regression model to the radial basis function, the regularization parameter to ε, and the loss function parameter to φ, inputting the sample training set into the support vector regression model, finding the regularization parameter and the loss function parameter when the mean square error is minimized, until the support vector regression model converges, and obtaining a trained support vector regression model;

[0043] S312, input the sample test set into the trained support vector regression model, fit the output result, and then perform linear regression with the actual slurry pump operation data to obtain the linear regression result, set the accuracy threshold to ψ, and when the accuracy of the linear regression result is greater than the accuracy threshold, obtain the support vector regression prediction model, otherwise adjust the regularization parameter and the loss function parameter until the accuracy of the linear regression result is greater than the accuracy threshold;

[0044] S32. Select any processed slurry pump operation data set from the processed slurry pump operation data matrix, input the processed slurry pump operation data set into the support vector regression prediction model, perform inverse normalization processing, and output the slurry pump operation data prediction value.

[0045] The invention obtains a support vector regression prediction model by training a support vector regression model and finding the optimal parameters, and outputs the predicted value of the slurry pump operation data; the prediction model has good prediction effect on small sample data and high prediction accuracy.

[0046] Preferably, the S4 comprises the following steps:

[0047] S41, according to the predicted value of the slurry pump operation data, the predicted values ​​of temperature, current and flow in the slurry pump operation index are obtained, the predicted values ​​of temperature, current and flow are used to establish the slurry pump operation cost function, the slurry pump operation cost function is used as the fitness function, and the escape optimization algorithm is used to optimize the temperature, current and flow in the slurry pump operation index to obtain the optimized slurry pump operation index, and the specific steps are as follows:

[0048] S411. Assume that there is a population of escaped individuals in the search space. There are i escaped individuals in the population and they are randomly distributed. The dimension of each escaped individual is j. In the initialization phase of the population of escaped individuals, the upper and lower bounds of the j-th dimension in the population of escaped individuals are set to l′(j) and l″(j), respectively. 1 represents a random number between 0 and 1, then the initial position C of the i-th escaped individual in the j-th dimension of the escaped individual population 1 The calculation formula for (i,j) is as follows:

[0049] C 1 (i,j)=l″(j)+d 1 (l′(j)-l″(j));

[0050] Set the current number of iterations to f and the maximum number of iterations to F. During each iteration, the escaped individual will experience fear and generate a fear index. The calculation formula is as follows:

[0051]

[0052] in, Indicates the fear index;

[0053] According to the fear index, the escaped individual population is divided into a calm individual population, a gathering individual population and a panic individual population. The fitness function value of the escaped individual in the escaped individual population is calculated according to the updated position in each iteration. By finding the optimal fitness function value, the optimized slurry pump operation index is obtained.

[0054] S412. In the calm individual population, the center position of the jth dimension in the calm individual population is set to C′(j), and the average value of the calm individuals in the calm individual population is The Levy flight simulation step length is introduced, and the Levy weight of the calm individual population is γ 1 , and then introduce Bernoulli distribution to simulate the update of calm individuals, and obtain the update coefficient η of the calm individual population 1 , where the update coefficient takes the value of 0 or 1 with the same probability, and updates the initial position of the i-th escaped individual on the j-th dimension in the calm individual population to obtain the position C of the i-th escaped individual on the j-th dimension in the calm individual population 2 (i,j), the calculation formula is as follows:

[0055]

[0056] In the aggregated individual population, the Levy weight of the aggregated individual population is set to γ 2 , the update coefficient η of the aggregated individual population 2 , select the panic individual on the jth dimension in the panic individual population and record it as C″(j), and the average value of the aggregated individuals in the aggregated individual population is Update the position of the i-th escaped individual on the j-th dimension in the calm individual population. The calculation formula is as follows:

[0057]

[0058] In the panic individual population, the average value of panic individuals in the panic individual population is set to In the current iteration process, the escape individual corresponding to the best fitness function is found, denoted as D(j), and a random panic individual is selected from the panic individual population, denoted as C″′(j). The position of the i-th escape individual on the j-th dimension in the calm individual population is updated again. The calculation formula is as follows:

[0059]

[0060] S413. In the continuous iteration process, when the current number of iterations is greater than half of the maximum number of iterations, the escape individual position is updated according to the fitness function value of each iteration, and the escape individual position corresponding to the current optimal fitness function value is used as the current optimal escape exit. In each iteration, the escape individuals in the escape individual population move closer to the current optimal escape exit; when the current number of iterations reaches the maximum number of iterations, the iteration is stopped to obtain the final optimal escape exit, and the three-dimensional coordinate values ​​of the final optimal escape exit position correspond to the temperature, current and flow in the slurry pump operation index respectively, and the optimized slurry pump operation index is obtained;

[0061] S42, the slurry pump operates according to the optimized slurry pump operation index to achieve the optimal operation cost of the slurry pump, and complete the slurry pump operation optimization.

[0062] The invention optimizes the operating indicators of the slurry pump based on the escape optimization algorithm by establishing a fitness function, solves complex optimization problems by simulating the behavior of the crowd during the evacuation process, and obtains the optimized operating indicators of the slurry pump by finding the best fitness function value, thereby achieving slurry pump operation optimization with good optimization performance and strong optimization capability.

[0063] This embodiment also discloses a system for monitoring and optimizing a slurry pump operation based on machine learning, which specifically includes: a slurry pump operation data cleaning module, a slurry pump operation monitoring module, a slurry pump operation prediction module, and a slurry pump operation optimization module;

[0064] The slurry pump operation data cleaning module is used to delete abnormal data and fill missing data using an improved isolation forest algorithm and a k-nearest neighbor missing data filling method;

[0065] The slurry pump operation monitoring module is used to detect the safety of slurry pump operation data by setting an adaptive threshold;

[0066] The slurry pump operation prediction module is used to predict the slurry pump operation data using a support vector regression prediction model;

[0067] The slurry pump operation optimization module is used to optimize the slurry pump operation index using the escape optimization algorithm to obtain the optimized slurry pump operation index.

[0068] The present invention has the following beneficial effects:

[0069] 1. The invention calculates the anomaly score of the original slurry pump operation data by using an improved isolation forest algorithm, detects and deletes the abnormal data by comparing the threshold and the anomaly score, can effectively detect the abnormal data, has high applicability, and improves the processing speed of the slurry pump operation data.

[0070] 2. The present invention performs weighted processing on the original slurry pump operation data by using the k-nearest neighbor missing data filling method, takes into account the degree of difference of each data, realizes missing data filling, and effectively improves the accuracy of data filling.

[0071] 3. The invention obtains historical slurry pump operation data and calculates an adaptive threshold value, thereby avoiding misjudgment caused by excessively high or low threshold values. By comparing the adaptive threshold value with the slurry pump operation data, it is determined whether the slurry pump is in a safe state, and the slurry pump operation monitoring effect is good.

[0072] 4. The invention trains the support vector regression model, obtains the support vector regression prediction model by finding the optimal parameters, and outputs the predicted value of the slurry pump operation data; the prediction model has good prediction effect on small sample data and high prediction accuracy.

[0073] 5. The invention optimizes the operating indicators of the slurry pump based on the escape optimization algorithm by establishing a fitness function, solves complex optimization problems by simulating the behavior of the crowd during the evacuation process, and obtains the optimized operating indicators of the slurry pump by finding the best fitness function value. It has good optimization performance and strong optimization ability.

[0074] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0076] Figure 1 A schematic diagram of the process of monitoring and optimizing the operation of a slurry pump by a slurry pump operation monitoring and optimization system based on machine learning provided by the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0079] Example 1

[0080] See also Figure 1 This embodiment discloses a slurry pump operation monitoring and optimization method based on machine learning, which specifically includes the following contents:

[0081] S1. Acquire the slurry pump operation index, obtain the original slurry pump operation data, generate the original slurry pump operation data matrix, perform abnormal data detection and deletion and missing data filling on the original slurry pump operation data matrix, and obtain the processed slurry pump operation data matrix;

[0082] The S1 comprises the following steps:

[0083] S11, obtaining the slurry pump operation index, which includes temperature, current, voltage, flow, pressure, etc., using sensor equipment to obtain the data of the slurry pump operation index, and obtaining the original slurry pump operation data. The original slurry pump operation data corresponding to each slurry pump operation index constitutes an original slurry pump operation data set. The first n original slurry pump operation data in the original slurry pump operation data set are selected to generate the original slurry pump operation data matrix A as follows:

[0084]

[0085] Among them, a mn The nth original slurry pump operation data representing the mth slurry pump operation index;

[0086] S12, using an improved isolation forest algorithm to detect and delete abnormal data on the original slurry pump operation data matrix to obtain a processed slurry pump operation data matrix, the specific steps are as follows:

[0087] S121, select any slurry pump operation index in the original slurry pump operation data matrix to obtain a slurry pump operation data set to be detected, set a dimension set corresponding to the slurry pump operation data to be detected in the slurry pump operation data set to be detected, select any dimension in the dimension set and record it as a cutting point dimension; use the slurry pump operation data to be detected corresponding to the cutting point dimension as a cutting point, add the slurry pump operation data to be detected that is less than or equal to the cutting point in the slurry pump operation data set to be detected to the first left subtree, and add the slurry pump operation data to be detected that is greater than the cutting point in the slurry pump operation data set to be detected to the first right subtree to obtain an isolated tree;

[0088] S122, setting the maximum depth of the isolated tree, repeatedly cutting in the first left subtree and the first right subtree to obtain the second left subtree and the second right subtree, until the isolated tree reaches the maximum depth of the isolated tree or the left subtree and the right subtree contain only one slurry pump operation data to be detected, stop cutting, and obtain the final isolated tree; cutting the original slurry pump operation data corresponding to other slurry pump operation indicators in the original slurry pump operation data matrix to generate an isolated forest;

[0089] S123, selecting a data point to be detected from the slurry pump operation data set to be detected, setting the path length of the data point to be detected in the isolation forest to b, β represents a normalization factor, and the calculation formula for the abnormal score of the data point to be detected is as follows:

[0090]

[0091] Among them, B represents the anomaly score of the data point to be detected;

[0092] sequentially calculating the abnormality scores of all the slurry pump operation data to be detected in the slurry pump operation data set to be detected, to obtain an abnormality score set;

[0093] S124, clustering the abnormal score set, selecting an initial cluster center, calculating the distance between the abnormal score in the abnormal score set and the initial cluster center, dividing the abnormal scores in the abnormal score set into the cluster where the nearest initial cluster center is located, and then recalculating the center of the cluster until the center of the cluster no longer changes, clustering the abnormal score set into a first cluster and a second cluster; selecting the first cluster and the second cluster with fewer abnormal scores as the final cluster, and taking the minimum abnormal score in the final cluster as the abnormal threshold; recording the slurry pump operation data to be detected that is greater than the abnormal threshold in the abnormal score set as abnormal data, and deleting it to obtain a processed slurry pump operation data set, and sequentially deleting the abnormal data from the original slurry pump operation data matrix to obtain a processed slurry pump operation data matrix;

[0094] S13, selecting any processed slurry pump operation data set in the processed slurry pump operation data matrix, recorded as the slurry pump operation data set to be filled, selecting k nearest neighbor data at the missing data in the slurry pump operation data set to be filled, calculating the distance and entropy value of the k nearest neighbor data and the missing data, and then calculating the coefficient of variation of the k nearest neighbor data, and performing weighted processing on the k nearest neighbor data. The calculation formula is as follows:

[0095]

[0096] Among them, B c ′ represents the weight coefficient of the cth nearest neighbor data, b c ′ represents the coefficient of variation of the cth nearest neighbor data, c = 1, 2, 3, ..., k;

[0097] Setting b c ″ represents the cth nearest neighbor data. The missing data is calculated according to the weighted coefficient of the cth nearest neighbor data. The calculation formula is as follows:

[0098]

[0099] in, Indicates missing data in the slurry pump operation data set to be filled;

[0100] Calculate all missing data in the slurry pump operation data set to be filled in turn, obtain the processed slurry pump operation data set, and generate a processed slurry pump operation data matrix;

[0101] S2. Acquire historical slurry pump operation data, calculate an adaptive threshold value according to the historical slurry pump operation data, verify the security of the processed slurry pump operation data in the processed slurry pump operation data matrix based on the adaptive threshold value, and complete the slurry pump operation monitoring;

[0102] The S2 comprises the following steps:

[0103] S21, acquiring historical slurry pump operation data, forming a historical slurry pump operation data set, generating a historical slurry pump operation data matrix, performing abnormal data detection and deletion and missing data filling on the historical slurry pump operation data matrix, obtaining a processed historical slurry pump operation data matrix, selecting any processed historical slurry pump operation data set from the processed historical slurry pump operation data matrix, calculating the mean and standard deviation of the processed historical slurry pump operation data in the processed historical slurry pump operation data set, and the adaptive threshold calculation formula is as follows:

[0104]

[0105] Where ω represents the adaptive threshold, Represents the mean of the processed historical slurry pump operation data, represents the standard deviation of the processed historical slurry pump operation data, and χ represents the hyperparameter;

[0106] S22. Based on the adaptive threshold and the standard deviation of the processed historical slurry pump operation data, the monitoring threshold calculation formula is as follows:

[0107]

[0108] Among them, ξ represents the monitoring threshold;

[0109] Selecting a corresponding processed slurry pump operation data set from the processed slurry pump operation data matrix, when the processed slurry pump operation data in the processed slurry pump operation data set is greater than the monitoring threshold, the slurry pump operation is in a dangerous state, and an early warning is issued; otherwise, the slurry pump operation is in a safe state, and the slurry pump operation monitoring is completed;

[0110] S3, training a support vector regression model to obtain a support vector regression prediction model, performing prediction based on the processed slurry pump operation data matrix, and outputting a predicted value of the slurry pump operation data;

[0111] The S3 comprises the following steps:

[0112] S31, normalizing the historical slurry pump operation data matrix generated by the historical slurry pump operation data to obtain a normalized historical slurry pump operation data matrix, selecting any normalized historical slurry pump operation data set from the normalized historical slurry pump operation data matrix, and dividing the normalized historical slurry pump operation data set into a sample training set and a sample test set; training a support vector regression model to obtain a support vector regression prediction model, the specific steps are as follows:

[0113] S311, setting the kernel function of the support vector regression model to the radial basis function, the regularization parameter to ε, and the loss function parameter to φ, inputting the sample training set into the support vector regression model, finding the regularization parameter and the loss function parameter when the mean square error is minimized, until the support vector regression model converges, and obtaining a trained support vector regression model;

[0114] S312, input the sample test set into the trained support vector regression model, fit the output result, and then perform linear regression with the actual slurry pump operation data to obtain the linear regression result, set the accuracy threshold to ψ, and when the accuracy of the linear regression result is greater than the accuracy threshold, obtain the support vector regression prediction model, otherwise adjust the regularization parameter and the loss function parameter until the accuracy of the linear regression result is greater than the accuracy threshold;

[0115] S32, selecting any processed slurry pump operation data set from the processed slurry pump operation data matrix, inputting the processed slurry pump operation data set into the support vector regression prediction model, performing a denormalization process, and outputting a predicted value of the slurry pump operation data;

[0116] S4, establishing a fitness function according to the predicted value of the slurry pump operation data and the slurry pump operation index, optimizing the slurry pump operation index using the escape optimization algorithm, obtaining the optimized slurry pump operation index by finding the best fitness function value, and completing the slurry pump operation optimization;

[0117] The S4 comprises the following steps:

[0118] S41, according to the predicted value of the slurry pump operation data, the predicted values ​​of temperature, current and flow in the slurry pump operation index are obtained, the predicted values ​​of temperature, current and flow are used to establish the slurry pump operation cost function, the slurry pump operation cost function is used as the fitness function, and the escape optimization algorithm is used to optimize the temperature, current and flow in the slurry pump operation index to obtain the optimized slurry pump operation index, and the specific steps are as follows:

[0119] S411. Assume that there is a population of escaped individuals in the search space. There are i escaped individuals in the population and they are randomly distributed. The dimension of each escaped individual is j. In the initialization phase of the population of escaped individuals, the upper and lower bounds of the j-th dimension in the population of escaped individuals are set to l′(j) and l″(j), respectively. 1 represents a random number between 0 and 1, then the initial position C of the i-th escaped individual in the j-th dimension of the escaped individual population 1 The calculation formula for (i,j) is as follows:

[0120] C 1 (i,j)=l″(j)+d 1 (l′(j)-l″(j));

[0121] Set the current number of iterations to f and the maximum number of iterations to F. During each iteration, the escaped individual will experience fear and generate a fear index. The calculation formula is as follows:

[0122]

[0123] in, Indicates the fear index;

[0124] According to the fear index, the escaped individual population is divided into a calm individual population, a gathering individual population and a panic individual population. The fitness function value of the escaped individual in the escaped individual population is calculated according to the updated position in each iteration. By finding the optimal fitness function value, the optimized slurry pump operation index is obtained.

[0125] S412. In the calm individual population, the center position of the jth dimension in the calm individual population is set to C′(j), and the average value of the calm individuals in the calm individual population is The Levy flight simulation step length is introduced, and the Levy weight of the calm individual population is γ 1 , and then introduce Bernoulli distribution to simulate the update of calm individuals, and obtain the update coefficient η of the calm individual population 1 , where the update coefficient takes the value of 0 or 1 with the same probability, and updates the initial position of the i-th escaped individual on the j-th dimension in the calm individual population to obtain the position C of the i-th escaped individual on the j-th dimension in the calm individual population 2 (i,j), the calculation formula is as follows:

[0126]

[0127] In the aggregated individual population, the Levy weight of the aggregated individual population is set to γ 2 , the update coefficient η of the aggregated individual population 2 , select the panic individual on the jth dimension in the panic individual population and record it as C″(j), and the average value of the aggregated individuals in the aggregated individual population is Update the position of the i-th escaped individual on the j-th dimension in the calm individual population. The calculation formula is as follows:

[0128]

[0129] In the panic individual population, the average value of panic individuals in the panic individual population is set to In the current iteration process, the escape individual corresponding to the best fitness function is found, denoted as D(j), and a random panic individual is selected from the panic individual population, denoted as C″′(j). The position of the i-th escape individual on the j-th dimension in the calm individual population is updated again. The calculation formula is as follows:

[0130]

[0131] S413. In the continuous iteration process, when the current number of iterations is greater than half of the maximum number of iterations, the escape individual position is updated according to the fitness function value of each iteration, and the escape individual position corresponding to the current optimal fitness function value is used as the current optimal escape exit. In each iteration, the escape individuals in the escape individual population move closer to the current optimal escape exit; when the current number of iterations reaches the maximum number of iterations, the iteration is stopped to obtain the final optimal escape exit, and the three-dimensional coordinate values ​​of the final optimal escape exit position correspond to the temperature, current and flow in the slurry pump operation index respectively, and the optimized slurry pump operation index is obtained;

[0132] S42, the slurry pump operates according to the optimized slurry pump operation index to achieve the optimal operation cost of the slurry pump, and complete the slurry pump operation optimization.

[0133] Example 2

[0134] This embodiment also discloses a system for monitoring and optimizing a slurry pump operation based on machine learning, which specifically includes: a slurry pump operation data cleaning module, a slurry pump operation monitoring module, a slurry pump operation prediction module, and a slurry pump operation optimization module;

[0135] The slurry pump operation data cleaning module is used to delete abnormal data and fill missing data using an improved isolation forest algorithm and a k-nearest neighbor missing data filling method;

[0136] The slurry pump operation monitoring module is used to detect the safety of slurry pump operation data by setting an adaptive threshold;

[0137] The slurry pump operation prediction module is used to predict the slurry pump operation data using a support vector regression prediction model;

[0138] The slurry pump operation optimization module is used to optimize the slurry pump operation index using the escape optimization algorithm to obtain the optimized slurry pump operation index.

[0139] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0140] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A slurry pump operation monitoring and optimization method based on machine learning, characterized in that: The steps include: S1. Acquire the slurry pump operation index, obtain the original slurry pump operation data, generate the original slurry pump operation data matrix, perform abnormal data detection and deletion and missing data filling on the original slurry pump operation data matrix, and obtain the processed slurry pump operation data matrix; S2. Acquire historical slurry pump operation data, calculate an adaptive threshold value according to the historical slurry pump operation data, verify the security of the processed slurry pump operation data in the processed slurry pump operation data matrix based on the adaptive threshold value, and complete the slurry pump operation monitoring; S3, training a support vector regression model to obtain a support vector regression prediction model, performing prediction based on the processed slurry pump operation data matrix, and outputting a predicted value of the slurry pump operation data; S4. Establish a fitness function according to the predicted value of the slurry pump operation data and the slurry pump operation index, optimize the slurry pump operation index, obtain the optimized slurry pump operation index by finding the best fitness function value, and complete the slurry pump operation optimization.

2. A slurry pump operation monitoring and optimization method based on machine learning according to claim 1, characterized in that: The S1 comprises the following steps: S11, obtaining a slurry pump operation index to obtain original slurry pump operation data, the original slurry pump operation data corresponding to each slurry pump operation index constitutes an original slurry pump operation data set, and generates an original slurry pump operation data matrix; S12, using an improved isolation forest algorithm to detect and delete abnormal data on the original slurry pump operation data matrix to obtain a processed slurry pump operation data matrix; S13, selecting k nearest neighboring data of the missing data in the processed slurry pump operation data matrix, calculating the weighted coefficients of the k nearest neighboring data, calculating the missing data according to the weighted coefficients of the k nearest neighboring data, completing the missing data filling, and obtaining the processed slurry pump operation data matrix.

3. A slurry pump operation monitoring and optimization method based on machine learning according to claim 2, characterized in that: The S12 comprises the following steps: S121, selecting a cutting point in the original slurry pump operation data matrix to establish an isolated tree; S122, setting the maximum depth of the isolated tree, until the isolated tree reaches the maximum depth of the isolated tree, stopping cutting, and generating an isolated forest; S123, calculating the anomaly scores of the data points to be detected in the isolation forest to obtain an anomaly score set; S124, performing clustering processing on the abnormal score set, clustering the abnormal score set into a first cluster and a second cluster; Selecting the cluster with the least number of anomaly scores in the first and second clusters as the final cluster, and taking the minimum anomaly score in the final cluster as the anomaly threshold; The abnormal threshold is compared with the abnormal score set, abnormal data is screened out, and deleted to obtain a processed slurry pump operation data matrix.

4. The method for monitoring and optimizing slurry pump operation based on machine learning according to claim 3, characterized in that: The S2 comprises the following steps: S21, obtaining historical slurry pump operation data, calculating the mean and standard deviation of the historical slurry pump operation data, and then calculating the adaptive threshold; S22, calculating the monitoring threshold according to the adaptive threshold and the mean and standard deviation of the historical slurry pump operation data; The monitoring threshold is compared with the processed slurry pump operation data in the processed slurry pump operation data matrix to obtain the slurry pump operation safety and complete the slurry pump operation monitoring.

5. A method for monitoring and optimizing slurry pump operation based on machine learning according to claim 4, characterized in that: The S3 comprises the following steps: S31, normalizing the historical slurry pump operation data matrix generated by the historical slurry pump operation data to obtain a normalized historical slurry pump operation data matrix, selecting any normalized historical slurry pump operation data set from the normalized historical slurry pump operation data matrix, and dividing the normalized historical slurry pump operation data set into a sample training set and a sample test set; Train the support vector regression model to obtain a support vector regression prediction model; S32. Select any processed slurry pump operation data set from the processed slurry pump operation data matrix, input the processed slurry pump operation data set into the support vector regression prediction model, perform inverse normalization processing, and output the slurry pump operation data prediction value.

6. The method for monitoring and optimizing slurry pump operation based on machine learning according to claim 5, characterized in that: The training of the support vector regression model to obtain the support vector regression prediction model comprises the following steps: The kernel function of the support vector regression model is set to a radial basis function, the regularization parameter is ε, and the loss function parameter is φ, the sample training set is input into the support vector regression model, and the regularization parameter and the loss function parameter are found when the mean square error is minimized, until the support vector regression model converges, thereby obtaining a trained support vector regression model; The sample test set is input into the trained support vector regression model, the output result is fitted, and then linear regression is performed with the actual slurry pump operation data to obtain the linear regression result. The accuracy threshold is set to ψ. When the accuracy of the linear regression result is greater than the accuracy threshold, the support vector regression prediction model is obtained. Otherwise, the regularization parameter and the loss function parameter are adjusted until the accuracy of the linear regression result is greater than the accuracy threshold.

7. A method for monitoring and optimizing slurry pump operation based on machine learning according to claim 6, characterized in that: The S4 comprises the following steps: S41, according to the predicted value of the slurry pump operation data, the predicted values ​​of temperature, current and flow in the slurry pump operation index are obtained, the predicted values ​​of temperature, current and flow are used to establish the slurry pump operation cost function, the slurry pump operation cost function is used as the fitness function, and the escape optimization algorithm is used to optimize the temperature, current and flow in the slurry pump operation index to obtain the optimized slurry pump operation index; S42, the slurry pump operates according to the optimized slurry pump operation index to achieve the optimal operation cost of the slurry pump, and complete the slurry pump operation optimization.

8. The method for monitoring and optimizing slurry pump operation based on machine learning according to claim 7, characterized in that: The method of using the escape optimization algorithm to optimize the temperature, current and flow rate in the slurry pump operation index to obtain the optimized slurry pump operation index comprises the following steps: Assume that there is a population of escaped individuals in the search space, and there are i escaped individuals in the population of escaped individuals and they are randomly distributed. The dimension of each escaped individual is j. In the initialization phase of the population of escaped individuals, the initial position of the i-th escaped individual on the j-th dimension in the population of escaped individuals is obtained; Set the current number of iterations to f and the maximum number of iterations to F. During each iteration, the escaped individual will experience fear and generate a fear index. According to the fear index, the escaped individual population is divided into a calm individual population, a gathering individual population and a panic individual population. The fitness function value of the escaped individual in the escaped individual population is calculated according to the updated position in each iteration. By finding the optimal fitness function value, the optimized slurry pump operation index is obtained. In the calm individual population, the Levy flight simulation step length is introduced, the Levy weight of the calm individual population is γ1, and then the Bernoulli distribution is introduced to simulate the calm individual update situation, and the update coefficient η1 of the calm individual population is obtained, and the position of the i-th escaped individual in the j-th dimension in the calm individual population is obtained; In the clustered individual population, the Levy weight of the clustered individual population is set to γ2, and the update coefficient of the clustered individual population is η2, and the position of the i-th escaped individual on the j-th dimension in the calm individual population is updated; in the panic individual population, the position of the i-th escaped individual on the j-th dimension in the calm individual population is updated again; In the continuous iteration process, when the current number of iterations is greater than half of the maximum number of iterations, the position of the escape individual is updated according to the fitness function value of each iteration, and the escape individual position corresponding to the current best fitness function value is used as the current best escape exit. In each iteration, the escape individuals in the escape individual population move closer to the current best escape exit; when the current number of iterations reaches the maximum number of iterations, the iteration is stopped to obtain the final best escape exit. The three-dimensional coordinate values ​​of the final best escape exit position correspond to the temperature, current and flow in the slurry pump operation indicators, respectively, to obtain the optimized slurry pump operation indicators.

9. A system for implementing the method for monitoring and optimizing the operation of a slurry pump based on machine learning as claimed in any one of claims 1 to 8, characterized in that: Specifically include: Slurry pump operation data cleaning module, slurry pump operation monitoring module, slurry pump operation prediction module and slurry pump operation optimization module; The slurry pump operation data cleaning module is used to delete abnormal data and fill missing data using an improved isolation forest algorithm and a k-nearest neighbor missing data filling method; The slurry pump operation monitoring module is used to detect the safety of slurry pump operation data by setting an adaptive threshold; The slurry pump operation prediction module is used to predict the slurry pump operation data using a support vector regression prediction model; The slurry pump operation optimization module is used to optimize the slurry pump operation index using the escape optimization algorithm to obtain the optimized slurry pump operation index.

Images