Intelligent supervision method, system and equipment for business outsourcing and medium

Abstract

The invention discloses a service outsourcing intelligent supervision method, system and device and a medium, and relates to the technical field of intelligent supervision, and the method comprises the steps: obtaining the target service time sequence data of each supervision node, and determining the prediction service time sequence data of each supervision node based on a service IMF component prediction model and according to the current service time sequence data; for each supervision node, determining a transverse feature matrix according to the target service time sequence data and the predicted service time sequence data; for each supervision node, based on a data relation model between the supervision node and the next supervision node, determining associated service time sequence data of the next supervision node according to the service time sequence prediction data, and determining a longitudinal feature matrix according to the associated service time sequence data and the prediction service time sequence data; and based on a supervision model, generating a supervision result according to the transverse feature matrix and the longitudinal feature matrix. The method has the effect of high supervision accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent supervision, in particular to a business outsourcing intelligent supervision method, system, device and medium. BACKGROUND

[0002] The current business supervision mainly relies on manual experience and rule-based detection means, and uses pre-defined threshold and judgment rules to monitor the state of each supervision node. Although this method can achieve certain effect in early application, with the expansion of business scale and data growth, the monitoring method relying on static rules alone gradually exposes defects such as low accuracy, low efficiency and slow response, and it is difficult to reflect the business running state from all aspects and multiple angles, and it is also difficult to effectively trace the problem source, which limits the further development of business supervision technology in the aspects of intelligence and automation. SUMMARY

[0003] In order to improve the accuracy of supervision, the present application provides a business outsourcing intelligent supervision method, system, device and medium.

[0004] In a first aspect, the present application provides a business outsourcing intelligent supervision method, which adopts the following technical solution: The business outsourcing intelligent supervision method comprises: Obtaining target business time series data of each supervision node, based on a business IMF component prediction model, and determining predicted business time series data of the supervision node according to the current business time series data; For each supervision node, determining a horizontal feature matrix according to the target business time series data and the predicted business time series data, wherein the horizontal feature matrix is used to represent the data change characteristics of a single supervision node at the current time node and the future time node; For each supervision node, determining the associated business time series data of the next supervision node based on a data relationship model between the supervision node and the next supervision node, and according to the business time series prediction data, and determining a vertical feature matrix according to the associated business time series data and the predicted business time series data, wherein the vertical feature matrix is used to represent the data difference characteristics of adjacent supervision nodes at the future time node; Based on a supervision model, and according to the horizontal feature matrix and the vertical feature matrix, a supervision result is generated, which is used to represent the abnormality of a single supervision node and the abnormal association of adjacent supervision nodes.

[0005] By adopting the technical solution, the target business time sequence data of each supervision node is acquired first, the predicted business time sequence data of each supervision node is determined based on a business IMF component prediction model and according to the current business time sequence data, then for each supervision node, the transverse feature matrix is determined according to the target business time sequence data and the predicted business time sequence data, wherein the transverse feature matrix is used to represent the data change feature of a single supervision node at the current time node and the future time node, then for each supervision node, the associated business time sequence data of the next supervision node is determined based on a data relationship model between the supervision node and the next supervision node and according to the business time sequence prediction data, and the longitudinal feature matrix is determined according to the associated business time sequence data and the predicted business time sequence data, wherein the longitudinal feature matrix is used to represent the data difference feature of adjacent supervision nodes at the future time node, then the supervision result is generated based on a supervision model and according to the transverse feature matrix and the longitudinal feature matrix, and the supervision result is used to represent the abnormality of a single supervision node and the abnormal association of adjacent supervision nodes; in the above method, the dynamic supervision system is constructed by fusing the node self-state monitoring in the time dimension (i.e. the transverse feature) and the inter-node influence analysis in the topology dimension (i.e. the longitudinal feature), thereby improving the accuracy of the abnormal detection of the supervision node, enhancing the correlation analysis capability between nodes and inside the node, improving the effect of the multi-node and multi-level risk assessment through multi-angle analysis, and thereby improving the intelligent level of the supervision system.

[0006] Optionally, the step of determining the predicted business time sequence data of each supervision node based on the business IMF component prediction model and according to the current business time sequence data comprises: For each supervision node, the business IMF component prediction is performed based on the business IMF component prediction model and according to the current business time sequence data, to obtain the business IMF prediction component of the supervision node. The predefined residual component is acquired, and the linear superposition of the business IMF prediction component and the residual component is performed to obtain the predicted business time sequence data of the supervision node.

[0007] By adopting the technical solution, in order to obtain the predicted business time sequence data, for each supervision node, the business IMF component prediction is performed based on the business IMF component prediction model and according to the current business time sequence data, to obtain the business IMF prediction component of the supervision node, then the predefined residual component is acquired, and the linear superposition of the business IMF prediction component and the residual component is performed to obtain the predicted business time sequence data of the supervision node.

[0008] Optionally, the step of generating the business IMF component prediction model comprises: Obtaining historical business time series data, dividing the historical business time series data through time series cross-validation to obtain an initial training set and an initial test set; Based on the genetic algorithm, the data in the initial training set and the initial test set are decomposed by EMD to obtain a decomposed training set and a decomposed test set; The decomposed training set and the decomposed test set are constructed for features and labels to obtain a first supervised learning set and a second supervised learning set, wherein the first supervised learning set includes first input feature data and first label data, the first input feature data includes first IMF components, and the label data is first business time series data in the future D days, the second supervised learning set includes second input feature data and second label data, the second input feature data includes second IMF components, and the label data is second business time series data in the future D days. Based on the RF model, and according to the first supervised learning set and the second supervised learning set, a business IMF component prediction model is constructed.

[0009] By adopting the above technical solution, in order to generate a business IMF component prediction model, historical business time series data is obtained, the historical business time series data is divided through time series cross-validation to obtain an initial training set and an initial test set, then based on the genetic algorithm, the data in the initial training set and the initial test set are decomposed by EMD to obtain a decomposed training set and a decomposed test set, then the decomposed training set and the decomposed test set are constructed for features and labels to obtain a first supervised learning set and a second supervised learning set, wherein the first supervised learning set includes first input feature data and first label data, the first input feature data includes first IMF components, and the label data is first business time series data in the future D days, the second supervised learning set includes second input feature data and second label data, the second input feature data includes second IMF components, and the label data is second business time series data in the future D days, then based on the RF model, and according to the first supervised learning set and the second supervised learning set, a business IMF component prediction model is constructed.

[0010] Optionally, the step of constructing features and labels for the decomposed training set and the decomposed test set to obtain a first supervised learning set and a second supervised learning set comprises: The IMF components in the decomposed training set and the decomposed test set are subjected to feature engineering processing to obtain IMF signal features, wherein the IMF signal features include time domain features and frequency domain features; Based on a preset sliding step, the IMF components in the decomposed training set and the decomposed test set are translated to construct corresponding IMF sequences and label vectors Y; perform feature fusion on the IMF sequence and the IMF signal feature to obtain a corresponding feature matrix X; generate a first supervised learning set and a second supervised learning set according to the feature matrix X and a label vector Y, wherein the first supervised learning set corresponds to the decomposition training set, and the second supervised learning set corresponds to the decomposition test set.

[0011] By adopting the technical solution, in order to obtain the first supervised learning set and the second supervised learning set, the IMF components in the decomposition training set and the decomposition test set are subjected to feature engineering processing to obtain IMF signal features, wherein the IMF signal features include time domain features and frequency domain features, then the IMF components in the decomposition training set and the decomposition test set are shifted based on a preset sliding step to construct corresponding IMF sequences and a label vector Y, then the IMF sequences and the IMF signal features are subjected to feature fusion to obtain a corresponding feature matrix X, and then the first supervised learning set and the second supervised learning set are generated according to the feature matrix X and the label vector Y, wherein the first supervised learning set corresponds to the decomposition training set, and the second supervised learning set corresponds to the decomposition test set.

[0012] Optionally, the step of determining a horizontal feature matrix according to the target business time series data and the predicted business time series data comprises: perform matrix processing on the target business time series data to obtain a target business time series matrix, wherein a row of the target business time series matrix is used to represent a time point, and a column of the target business time series matrix is used to represent a first data dimension; perform matrix processing on the predicted business time series data to obtain a predicted business time series matrix, wherein a row of the predicted business time series matrix is used to represent the time point, and a column of the predicted business time series matrix is used to represent the first data dimension; for a same supervision node, perform matrix subtraction operation on the target business time series matrix corresponding to the supervision node and the predicted business time series matrix corresponding to the supervision node to obtain a horizontal difference matrix; perform matrix normalization processing on the horizontal difference matrix to obtain a first normalized matrix; obtain a first weight of the first data dimension, and perform weighted difference processing on the first normalized matrix according to the weight to obtain a horizontal feature matrix.

[0013] By adopting the technical scheme, in order to determine the horizontal feature matrix, the target business time series data is subjected to matrix processing to obtain a target business time series matrix, wherein the rows of the target business time series matrix are used to represent time points, and the columns of the target business time series matrix are used to represent the first data dimension; then the predicted business time series data is subjected to matrix processing to obtain a predicted business time series matrix, wherein the rows of the predicted business time series matrix are used to represent time points, and the columns of the predicted business time series matrix are used to represent the first data dimension; then, for the same supervision node, the target business time series matrix corresponding to the supervision node and the corresponding predicted business time series matrix are subjected to matrix subtraction operation to obtain a horizontal difference matrix; then the horizontal difference matrix is subjected to matrix normalization processing to obtain a first normalized matrix; then a first weight of the first data dimension is obtained, and the first normalized matrix is subjected to weighted difference processing according to the weight to obtain the horizontal feature matrix.

[0014] Optionally, the step of determining the longitudinal feature matrix according to the associated business time series data and the predicted business time series data comprises: the associated business time series data is subjected to matrix processing to obtain an associated business time series matrix, wherein the rows of the associated business time series data are used to represent the time points, and the columns of the associated business time series data are used to represent a second data dimension; for each supervision node, the target business time series matrix corresponding to the supervision node and the associated business time series matrix of the next supervision node of the supervision node are subjected to matrix subtraction operation to obtain a longitudinal difference matrix; the longitudinal difference matrix is subjected to matrix normalization processing to obtain a second normalized matrix; a second weight of the second data dimension is obtained, and the second normalized matrix is subjected to weighted difference processing according to the second weight to obtain the longitudinal feature matrix.

[0015] By adopting the technical scheme, in order to determine the longitudinal feature matrix, the associated business time series data is subjected to matrix processing to obtain an associated business time series matrix, wherein the rows of the associated business time series data are used to represent time points, and the columns of the associated business time series data are used to represent a second data dimension; then, for each supervision node, the target business time series matrix corresponding to the supervision node and the associated business time series matrix of the next supervision node of the supervision node are subjected to matrix subtraction operation to obtain a longitudinal difference matrix; then the longitudinal difference matrix is subjected to matrix normalization processing to obtain a second normalized matrix; finally, a second weight of the second data dimension is obtained, and the second normalized matrix is subjected to weighted difference processing according to the second weight to obtain the longitudinal feature matrix.

[0016] Optionally, the supervision model comprises an input layer, a hidden layer and an output layer, and the step of generating the supervision result based on the supervision model and according to the transverse feature matrix and the longitudinal feature matrix comprises: generating, by the input layer, an input feature vector according to the transverse feature matrix and the longitudinal feature matrix; generating, by the hidden layer, a feature extraction vector according to the input feature vector; generating, by the output layer, an output matrix according to the feature extraction vector, wherein the output matrix has a dimension of 2x(2n-1), n is consistent with the number of supervision nodes, the first row of the output matrix is used to represent a node anomaly index, and the second row of the output matrix is used to represent an anomaly correlation index; generating the supervision result based on the output matrix.

[0017] By adopting the above technical solution, in order to generate the supervision result, the input layer is used to generate the input feature vector according to the transverse feature matrix and the longitudinal feature matrix, then the hidden layer is used to generate the feature extraction vector according to the input feature vector, then the output layer is used to generate the output matrix according to the feature extraction vector, wherein the output matrix has a dimension of 2x(2n-1), n is consistent with the number of supervision nodes, the first row of the output matrix is used to represent the node anomaly index, and the second row of the output matrix is used to represent the anomaly correlation index, and finally the supervision result is generated based on the output matrix.

[0018] In a second aspect, the application further provides a business outsourcing intelligent supervision system, which adopts the following technical solution: The business outsourcing intelligent supervision system comprises: a business time sequence prediction module, configured to acquire target business time sequence data of each supervision node, determine predicted business time sequence data of the supervision nodes based on a business IMF component prediction model and according to the current business time sequence data; a transverse feature matrix generation module, configured to determine, for each supervision node, a transverse feature matrix according to the target business time sequence data and the predicted business time sequence data, wherein the transverse feature matrix is used to represent data change characteristics of a single supervision node at a current time node and a future time node; a longitudinal feature matrix generation module, configured to determine, for each supervision node, associated business time sequence data of a next supervision node based on a data relationship model between the supervision node and the next supervision node and according to the business time sequence prediction data, and determine a longitudinal feature matrix according to the associated business time sequence data and the predicted business time sequence data, wherein the longitudinal feature matrix is used to represent data difference characteristics of adjacent supervision nodes at the future time node; The supervision result generation module is configured to generate a supervision result based on the supervision model and according to the transverse feature matrix and the longitudinal feature matrix, the supervision result being used to represent abnormality of a single supervision node and abnormal correlation of adjacent supervision nodes.

[0019] In a third aspect, the present application further provides a computer device, which adopts the technical scheme as follows: The computer device comprises a memory and a processor, the memory storing a computer program capable of running on the processor, and the processor implements the method in the first aspect when executing the computer program.

[0020] In a fourth aspect, the present application further provides a computer readable storage medium, which adopts the technical scheme as follows: The computer readable storage medium stores a computer program capable of being loaded and executed by the processor to implement the method in the first aspect.

[0021] In summary, the present application at least has the following beneficial technical effects: target business time series data of each supervision node is acquired first, predicted business time series data of each supervision node is determined based on a business IMF component prediction model and according to current business time series data, then for each supervision node, a transverse feature matrix is determined according to the target business time series data and the predicted business time series data, wherein the transverse feature matrix is used to represent data change characteristics of a single supervision node at a current time node and a future time node, then for each supervision node, associated business time series data of a next supervision node is determined based on a data relationship model between the supervision node and the next supervision node and according to the business time series prediction data, and a longitudinal feature matrix is determined according to the associated business time series data and the predicted business time series data, wherein the longitudinal feature matrix is used to represent data difference characteristics of adjacent supervision nodes at the future time node, then a supervision result is generated based on a supervision model and according to the transverse feature matrix and the longitudinal feature matrix, the supervision result being used to represent abnormality of a single supervision node and abnormal correlation of adjacent supervision nodes; in the above method, a dynamic supervision system is constructed by fusing node self-state monitoring in a time dimension (i.e. transverse feature) and inter-node influence analysis in a topology dimension (i.e. longitudinal feature), thereby improving accuracy of abnormal detection of the supervision node, enhancing correlation analysis capability between nodes and inside a node, improving effects of multi-node and multi-level risk assessment through multi-angle analysis, and thus improving intelligent level of the supervision system. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a whole flow schematic diagram of an embodiment of the present application.

[0023] Figure 2 is a structure schematic diagram of a system of the present application.

[0024] Figure 3 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] This application discloses an intelligent supervision method for business outsourcing.

[0027] Reference Figure 1 Intelligent supervision methods for business outsourcing include: Step S11: Obtain the target business time series data for each supervision node, and determine the predicted business time series data for each supervision node based on the business IMF component prediction model and the current business time series data.

[0028] It should be noted that in step S11, historical business time-series data of each supervision node needs to be collected first, and core change features are extracted through IMF decomposition. Then, a prediction model is constructed using the decomposed key components (IMF) to predict future business indicators. When making predictions, EMD (Empirical Mode Decomposition) is performed on the current business time-series data of each node to extract IMF components. Based on the IMF components and the prediction model, the IMF components corresponding to the business time-series data of the next time stage (i.e. the future) are predicted. Then, the predicted IMF components are reconstructed to obtain the predicted business time-series data.

[0029] Step S12: For each supervision node, determine the horizontal feature matrix based on the target business time series data and the predicted business time series data.

[0030] The horizontal feature matrix is ​​used to represent the data change characteristics of a single supervision node at the current time node and future time nodes.

[0031] It should be noted that in step S12, the actual values ​​and predicted values ​​of each supervision node at multiple time points are combined to calculate the difference value of the data and the difference value of data characteristics (such as rate of change, standard deviation, and sliding window statistics) as features, and then a horizontal feature matrix is ​​constructed to reflect the changes of node data at different time points.

[0032] Step S13: For each supervision node, based on the data relationship model between the supervision node and the next supervision node, and according to the business time series prediction data, determine the associated business time series data of the next supervision node, and determine the vertical feature matrix according to the associated business time series data and the predicted business time series data.

[0033] The vertical feature matrix is ​​used to represent the data difference characteristics of adjacent supervision nodes at future time nodes.

[0034] It should be noted that in step S13, a data relationship model (such as a convolutional neural network GNN or a regression model) is established using the relationships between supervision nodes (e.g., spatial adjacency or correlation coefficient) or the data relationships between supervision nodes. Based on the predicted business time series data, the business time series data of the next supervision node is derived, and the data differences and data feature differences between adjacent supervision nodes are calculated based on the business time series data of the next supervision node. In this way, a vertical feature matrix is ​​constructed to record the relationship changes between adjacent supervision nodes.

[0035] Step S14: Based on the supervision model, generate supervision results according to the horizontal feature matrix and the vertical feature matrix.

[0036] The supervision results are used to represent the anomalies of a single supervision node and the abnormal correlations between adjacent supervision nodes.

[0037] It should be noted that in step S13, by combining horizontal and vertical features, the abnormality of the supervision node and the abnormal correlation between supervision nodes can be judged. Statistical thresholds (such as Z-score) or machine learning models (such as IsolationForest, LOF, AE) can be used to score the features for abnormality. Combined with the difference features between nodes, the correlation anomaly can be identified (such as two adjacent supervision nodes being abnormal at the same time, indicating that there may be a correlation problem).

[0038] In the above implementation, firstly, target business time-series data for each supervision node is acquired. Based on the business IMF component prediction model, and according to the current business time-series data, the predicted business time-series data for each supervision node is determined. Then, for each supervision node, a horizontal feature matrix is ​​determined based on the target business time-series data and the predicted business time-series data. The horizontal feature matrix is ​​used to represent the data change characteristics of a single supervision node at the current time node and future time nodes. Then, for each supervision node, based on the data relationship model between the supervision node and the next supervision node, and according to the business time-series prediction data, the associated business time-series data of the next supervision node is determined. Finally, a vertical feature matrix is ​​determined based on the associated business time-series data and the predicted business time-series data. In this method, the vertical feature matrix represents the data difference characteristics of adjacent supervision nodes at future time nodes. Then, based on the supervision model, and according to the horizontal and vertical feature matrices, the supervision results are generated. The supervision results represent the anomalies of a single supervision node and the anomaly correlations between adjacent supervision nodes. In the above method, by integrating the node self-state monitoring in the time dimension (i.e., horizontal features) and the node inter-influence analysis in the topology dimension (i.e., vertical features), a dynamic supervision system is constructed, thereby improving the accuracy of anomaly detection of supervision nodes, enhancing the correlation analysis capabilities between and within nodes, and improving the effect of multi-node, multi-level risk assessment through multi-angle analysis, thus improving the intelligence level of the supervision system.

[0039] As a further implementation of the method, the step of determining the predicted business time series data for each supervision node based on the business IMF component prediction model and the current business time series data includes: Step S21: For each supervision node, based on the business IMF component prediction model and according to the current business time series data, the business IMF component prediction is performed to obtain the business IMF prediction component of the supervision node.

[0040] Step S22: Obtain the predefined residual components, and linearly superimpose the business IMF prediction components and the residual components to obtain the predicted business time series data of the supervision node.

[0041] In the above implementation, in order to obtain the predicted business time series data, for each supervision node, the business IMF component prediction model is used, and the business IMF component is predicted based on the current business time series data to obtain the business IMF predicted component of the supervision node. Then, the predefined residual component is obtained, and the business IMF predicted component and the residual component are linearly superimposed to obtain the predicted business time series data of the supervision node.

[0042] As a further implementation of the method, the generation step of the business IMF component prediction model includes: Step S31: Obtain historical business time-series data, and divide the historical business time-series data by time series cross-validation to obtain the initial training set and the initial test set.

[0043] It should be noted that in step S31, the historical business time-series data is segmented sequentially using the time-series cross-validation method to ensure the order and obtain multiple training-test splits to enhance the model's generalization ability.

[0044] Step S32: Based on the genetic algorithm, perform EMD decomposition on the data in the initial training set and the initial test set to obtain the decomposed training set and the decomposed test set.

[0045] It should be noted that in step S32, each time series (data in training and testing) is subjected to empirical mode decomposition (EMD) to obtain a limited number of IMF components. The decomposition results are optimized to make the features more representative. Specifically, a genetic algorithm (GA) is used to adjust the EMD parameters (such as stopping criteria, number of modes, etc.) to obtain the optimal EMD decomposition. Then, the optimal EMD decomposition is applied to the initial training set and the initial test set respectively to obtain the IMF components.

[0046] Step S33: Construct features and labels for the decomposed training set and decomposed test set to obtain a first supervised learning set and a second supervised learning set. The first supervised learning set includes first input feature data and first label data. The first input feature data includes a first IMF component, and the label data is the first business time series data for the next D days. The second supervised learning set includes second input feature data and second label data. The second input feature data includes a second IMF component, and the label data is the second business time series data for the next D days.

[0047] Step S34: Based on the RF model, construct a business IMF component prediction model according to the first supervised learning set and the second supervised learning set.

[0048] In the above implementation, in order to generate a business IMF component prediction model, historical business time-series data is obtained. The historical business time-series data is divided using time-series cross-validation to obtain an initial training set and an initial test set. Then, based on a genetic algorithm, EMD decomposition is performed on the data in the initial training set and the initial test set to obtain a decomposed training set and a decomposed test set. Features and labels are then constructed from the decomposed training set and the decomposed test set to obtain a first supervised learning set and a second supervised learning set. The first supervised learning set includes first input feature data and first label data. The first input feature data includes a first IMF component, and the label data is the first business time-series data for the next D days. The second supervised learning set includes second input feature data and second label data. The second input feature data includes a second IMF component, and the label data is the second business time-series data for the next D days. Finally, based on the RF model and the first and second supervised learning sets, a business IMF component prediction model is constructed.

[0049] As a further implementation of the method, the step of constructing features and labels from the decomposed training set and decomposed test set to obtain the first supervised learning set and the second supervised learning set includes: Step S41: Perform feature engineering on the IMF components in the decomposed training set and decomposed test set to obtain IMF signal features, which include time domain features and frequency domain features.

[0050] It should be noted that in step S41, the time-domain features may include statistical features such as the mean, variance, and extrema in the time domain, and the frequency-domain features may include wavelet coefficients and FFT energy spectrum features in the frequency domain.

[0051] Step S42: Based on a preset sliding step size, the IMF components in the decomposed training set and decomposed test set are translated to construct the corresponding IMF sequence and label vector Y.

[0052] Step S43: Perform feature fusion on the IMF sequence and IMF signal features to obtain the corresponding feature matrix X.

[0053] Step S44: Generate a first supervised learning set and a second supervised learning set based on the feature matrix X and the label vector Y, wherein the first supervised learning set corresponds to the decomposed training set and the second supervised learning set corresponds to the decomposed test set.

[0054] In the above implementation, in order to obtain the first supervised learning set and the second supervised learning set, feature engineering is performed on the IMF components in the decomposed training set and the decomposed test set to obtain IMF signal features, wherein the IMF signal features include time domain features and frequency domain features. Then, based on a preset sliding step size, the IMF components in the decomposed training set and the decomposed test set are translated to construct the corresponding IMF sequence and label vector Y. Then, the IMF sequence and the IMF signal features are fused to obtain the corresponding feature matrix X. Then, the first supervised learning set and the second supervised learning set are generated according to the feature matrix X and the label vector Y, wherein the first supervised learning set corresponds to the decomposed training set and the second supervised learning set corresponds to the decomposed test set.

[0055] As a further implementation of the method, the step of determining the horizontal feature matrix based on the target business time-series data and the predicted business time-series data includes: Step S51: Perform matrix processing on the target business time series data to obtain the target business time series matrix, wherein the rows of the target business time series matrix are used to represent time points, and the columns of the target business time series matrix are used to represent the first data dimension.

[0056] Step S52: Perform matrix processing on the predicted business time series data to obtain the predicted business time series matrix, wherein the rows of the predicted business time series matrix are used to represent time points, and the columns of the predicted business time series matrix are used to represent the first data dimension.

[0057] Step S53: For the same supervision node, perform matrix subtraction on the target business time series matrix and the corresponding predicted business time series matrix corresponding to the supervision node to obtain the horizontal difference matrix.

[0058] Step S54: Perform matrix normalization on the horizontal difference matrix to obtain the first normalized matrix.

[0059] Step S55: Obtain the first weight of the first data dimension, and perform weighted difference processing on the first normalized matrix according to the weight to obtain the horizontal feature matrix.

[0060] In the above implementation, in order to determine the horizontal feature matrix, the target business time series data is matrix-processed to obtain the target business time series matrix, wherein the rows of the target business time series matrix represent time points and the columns of the target business time series matrix represent the first data dimension. Then, the predicted business time series data is matrix-processed to obtain the predicted business time series matrix, wherein the rows of the predicted business time series matrix represent time points and the columns of the predicted business time series matrix represent the first data dimension. Then, for the same supervision node, matrix subtraction is performed on the target business time series matrix and the corresponding predicted business time series matrix corresponding to the supervision node to obtain the horizontal difference matrix. Then, matrix normalization is performed on the horizontal difference matrix to obtain the first normalized matrix. Then, the first weight of the first data dimension is obtained, and the first normalized matrix is ​​weighted according to the weight to obtain the horizontal feature matrix.

[0061] As a further implementation of the method, the step of determining the vertical feature matrix based on associated business time-series data and predicted business time-series data includes: Step S61: Perform matrix processing on the associated business time series data to obtain the associated business time series matrix, wherein the rows of the associated business time series data are used to represent time points, and the columns of the associated business time series data are used to represent the second data dimension.

[0062] Step S62: For each supervision node, perform matrix subtraction on the target business time sequence matrix corresponding to the supervision node and the associated business time sequence matrix of the next supervision node to obtain the vertical difference matrix.

[0063] Step S63: Perform matrix normalization on the longitudinal difference matrix to obtain the second normalized matrix.

[0064] Step S64: Obtain the second weight of the second data dimension, and perform weighted difference processing on the second normalized matrix according to the second weight to obtain the vertical feature matrix.

[0065] In the above implementation, in order to determine the vertical feature matrix, the associated business time series data is matrix-processed to obtain the associated business time series matrix. The rows of the associated business time series data are used to represent time points, and the columns of the associated business time series data are used to represent the second data dimension. Then, for each supervision node, a matrix subtraction operation is performed on the target business time series matrix corresponding to the supervision node and the associated business time series matrix of the next supervision node to obtain the vertical difference matrix. Then, the vertical difference matrix is ​​normalized to obtain the second normalized matrix. Finally, the second weight of the second data dimension is obtained, and the second normalized matrix is ​​weighted according to the second weight to obtain the vertical feature matrix.

[0066] As a further implementation of the method, the supervision model includes an input layer, a hidden layer, and an output layer. The steps of generating supervision results based on the supervision model and according to the horizontal and vertical feature matrices include: Step S71: Through the input layer, an input feature vector is generated based on the horizontal feature matrix and the vertical feature matrix.

[0067] Step S72: Generate a feature extraction vector based on the input feature vector through the hidden layer.

[0068] Step S73: Through the output layer, an output matrix is ​​generated based on the feature extraction vector. The dimension of the output matrix is ​​2×(2n-1), where n is the same as the number of supervision nodes. The first row of the output matrix is ​​used to represent the node anomaly index, and the second row of the output matrix is ​​used to represent the anomaly correlation index.

[0069] Step S74: Generate the supervision results based on the output matrix.

[0070] In the above implementation, in order to generate the supervision result, an input feature vector is generated based on the horizontal and vertical feature matrices through the input layer. Then, a feature extraction vector is generated based on the input feature vector through the hidden layer. Finally, an output matrix is ​​generated based on the feature extraction vector through the output layer. The dimension of the output matrix is ​​2×(2n-1), where n is the same as the number of supervision nodes. The first row of the output matrix is ​​used to represent the node anomaly index, and the second row of the output matrix is ​​used to represent the anomaly correlation index. Finally, the supervision result is generated based on the output matrix.

[0071] This application also discloses an intelligent supervision system for business outsourcing.

[0072] refer to Figure 2 The business outsourcing intelligent supervision system includes: The business time series prediction module is used to acquire the target business time series data of each supervision node, based on the business IMF component prediction model, and determine the predicted business time series data of each supervision node according to the current business time series data. The horizontal feature matrix generation module is used to determine the horizontal feature matrix for each supervision node based on the target business time series data and the predicted business time series data. The horizontal feature matrix is ​​used to represent the data change characteristics of a single supervision node at the current time node and future time nodes. The vertical feature matrix generation module is used to determine the associated business time series data of the next supervision node for each supervision node based on the data relationship model between the supervision node and the next supervision node and the business time series prediction data, and to determine the vertical feature matrix based on the associated business time series data and the predicted business time series data. The vertical feature matrix is ​​used to represent the data difference characteristics of adjacent supervision nodes at future time nodes. The supervision result generation module is used to generate supervision results based on the supervision model and according to the horizontal and vertical feature matrices. The supervision results are used to represent the anomalies of a single supervision node and the abnormal correlations between adjacent supervision nodes.

[0073] The intelligent supervision system for business outsourcing of the present invention can implement any of the methods in the intelligent supervision method for business outsourcing, and the specific working process of the intelligent supervision system for business outsourcing of the present invention can refer to the corresponding process in the above-mentioned intelligent supervision method for business outsourcing.

[0074] This application also discloses a computer device.

[0075] refer to Figure 3 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described intelligent supervision methods for business outsourcing.

[0076] This application also discloses a computer-readable storage medium.

[0077] A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing any of the above-described intelligent supervision methods for business outsourcing.

[0078] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0079] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A business outsourcing intelligent supervision method, characterized in that, include: Obtain the target business time series data of each supervision node, and determine the predicted business time series data of each supervision node based on the business IMF component prediction model and the current business time series data. For each supervision node, a horizontal feature matrix is ​​determined based on the target business time series data and the predicted business time series data, wherein the horizontal feature matrix is ​​used to represent the data change characteristics of a single supervision node at the current time node and future time nodes; For each supervision node, based on the data relationship model between the supervision node and the next supervision node, and according to the business time-series prediction data, the associated business time-series data of the next supervision node is determined, and a vertical feature matrix is ​​determined according to the associated business time-series data and the predicted business time-series data, wherein the vertical feature matrix is ​​used to represent the data difference characteristics of adjacent supervision nodes at future time nodes. Based on the supervision model, and according to the horizontal feature matrix and the vertical feature matrix, the supervision results are generated. The supervision results are used to represent the anomalies of a single supervision node and the abnormal correlations between adjacent supervision nodes.

2. The intelligent supervision method for business outsourcing according to claim 1, characterized in that, The step of determining the predicted business time series data for each supervision node based on the business IMF component prediction model and the current business time series data includes: For each supervision node, the business IMF component prediction component is obtained by using the business IMF component prediction model and the current business time series data. Obtain the predefined residual components, and linearly superimpose the business IMF prediction components and the residual components to obtain the predicted business time series data of the supervision node.

3. The intelligent supervision method for business outsourcing according to claim 1, characterized in that, The steps for generating the business IMF component prediction model include: Historical business time-series data is acquired, and the historical business time-series data is divided using time series cross-validation to obtain an initial training set and an initial test set. Based on the genetic algorithm, EMD decomposition is performed on the data in the initial training set and the initial test set to obtain the decomposed training set and the decomposed test set. Features and labels are constructed on the decomposed training set and the decomposed test set to obtain a first supervised learning set and a second supervised learning set. The first supervised learning set includes first input feature data and first label data. The first input feature data includes a first IMF component. The label data is first business time series data for the next D days. The second supervised learning set includes second input feature data and second label data. The second input feature data includes a second IMF component. The label data is second business time series data for the next D days. Based on the RF model, a business IMF component prediction model is constructed according to the first and second supervised learning sets.

4. The intelligent supervision method for business outsourcing according to claim 1, characterized in that, The step of constructing features and labels from the decomposed training set and the decomposed test set to obtain a first supervised learning set and a second supervised learning set includes: Feature engineering is performed on the IMF components in the decomposed training set and the decomposed test set to obtain IMF signal features, wherein the IMF signal features include time-domain features and frequency-domain features. Based on a preset sliding step size, the IMF components in the decomposed training set and the decomposed test set are translated to construct the corresponding IMF sequence and label vector Y. The IMF sequence and the IMF signal features are fused to obtain the corresponding feature matrix X; A first supervised learning set and a second supervised learning set are generated based on the feature matrix X and the label vector Y, wherein the first supervised learning set corresponds to the decomposed training set and the second supervised learning set corresponds to the decomposed test set.

5. The intelligent supervision method for business outsourcing according to claim 1, characterized in that, The step of determining the horizontal feature matrix based on the target service time-series data and the predicted service time-series data includes: The target business time series data is matrixed to obtain a target business time series matrix, wherein the rows of the target business time series matrix represent time points and the columns of the target business time series matrix represent the first data dimension. The predicted business time series data is matrixed to obtain a predicted business time series matrix, wherein the rows of the predicted business time series matrix represent the time points, and the columns of the predicted business time series matrix represent the first data dimension. For the same supervision node, matrix subtraction is performed on the target business time series matrix and the corresponding predicted business time series matrix corresponding to the supervision node to obtain the horizontal difference matrix; The horizontal difference matrix is ​​normalized to obtain the first normalized matrix; Obtain the first weight of the first data dimension, and perform weighted difference processing on the first normalized matrix according to the weight to obtain the horizontal feature matrix.

6. The intelligent supervision method for business outsourcing according to claim 5, characterized in that, The step of determining the vertical feature matrix based on the associated business time-series data and the predicted business time-series data includes: The associated business time series data is matrixed to obtain an associated business time series matrix, wherein the rows of the associated business time series data are used to represent the time points, and the columns of the associated business time series data are used to represent the second data dimension; For each supervision node, a matrix subtraction operation is performed on the target business time sequence matrix corresponding to the supervision node and the associated business time sequence matrix of the next supervision node of the supervision node to obtain a vertical difference matrix; The longitudinal difference matrix is ​​normalized to obtain a second normalized matrix; Obtain the second weight of the second data dimension, and perform weighted difference processing on the second normalized matrix according to the second weight to obtain the vertical feature matrix.

7. The intelligent supervision method for business outsourcing according to claim 1, characterized in that, The supervision model includes an input layer, a hidden layer, and an output layer. The step of generating supervision results based on the supervision model and according to the horizontal feature matrix and the vertical feature matrix includes: Through the input layer, an input feature vector is generated based on the horizontal feature matrix and the vertical feature matrix; Through the hidden layer, a feature extraction vector is generated based on the input feature vector; Through the output layer, an output matrix is ​​generated based on the feature extraction vector, wherein the dimension of the output matrix is ​​2×(2n-1), where n is the same as the number of supervision nodes, the first row of the output matrix is ​​used to represent the node anomaly index, and the second row of the output matrix is ​​used to represent the anomaly correlation index. Based on the output matrix, the supervision results are generated.

8. A business outsourcing intelligent supervision system, characterized in that: include: The business time series prediction module is used to acquire the target business time series data of each supervision node, based on the business IMF component prediction model, and determine the predicted business time series data of each supervision node according to the current business time series data. The horizontal feature matrix generation module is used to determine the horizontal feature matrix for each supervision node based on the target business time series data and the predicted business time series data, wherein the horizontal feature matrix is ​​used to represent the data change characteristics of a single supervision node at the current time node and future time nodes; The vertical feature matrix generation module is used to determine the associated business time series data of the next supervision node for each supervision node based on the data relationship model between the supervision node and the next supervision node and the business time series prediction data, and to determine the vertical feature matrix based on the associated business time series data and the predicted business time series data, wherein the vertical feature matrix is ​​used to represent the data difference characteristics of adjacent supervision nodes at future time nodes. The supervision result generation module is used to generate supervision results based on the supervision model and according to the horizontal feature matrix and the vertical feature matrix. The supervision results are used to represent the anomalies of a single supervision node and the abnormal correlations between adjacent supervision nodes.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and execute the method of any one of claims 1 to 7.

Images