Cloud-computing-based intelligent management and control method and system for 5g base station node failure

Abstract

The application relates to the technical field of wireless communication, and discloses a 5G base station node fault intelligent management and control method and system based on cloud computing, which comprises the following steps: constructing a state response matrix of a 5G base station node to be analyzed when the 5G base station node is faulty; according to the network level to which each row vector in the state response matrix belongs, performing aggregation processing on the row vectors of the same level, performing differential transformation on the aggregation results between different levels, and generating an inter-level disturbance propagation matrix; performing misplacement cross-correlation scanning on each row vector in the inter-level disturbance propagation matrix; according to the scanning result, determining the propagation direction and propagation time delay of the fault disturbance between adjacent levels; and performing backtracking positioning on the network element node in the propagation starting level in the state response matrix, and determining the root cause network element node of the current fault; the application can automatically determine the propagation path of the 5G base station cross-layer fault and backtrack and position the root cause network element node.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a cloud-based intelligent management and control method and system for 5G base station node faults. Background Technology

[0002] In existing technologies, root cause analysis methods for 5G base station nodes often rely on pre-built knowledge graphs for fault propagation path reasoning. These methods pre-construct a knowledge graph containing network element topology relationships and alarm propagation dependencies, then inject real-time alarm data into the graph for path retrieval and matching. The effectiveness of these methods is limited by the completeness of fault modes and propagation paths in the knowledge graph. When the fault type is not found in historical data or the propagation path shifts due to network configuration changes, the knowledge graph lacks corresponding nodes and edges, making reasoning difficult. Furthermore, the construction of the knowledge graph relies on expert experience and historical alarm annotation data, resulting in long construction cycles, high update and maintenance costs, and difficulty in adapting to the dynamic changes in 5G network topology and configuration.

[0003] In existing technologies, another type of fault location method employs a correlation analysis approach based on alarm timing, inferring the direction of fault propagation by comparing the occurrence sequence of alarm events at various network element nodes. 5G networks adopt a cloud-native architecture, functionally decoupled into multiple network layers, each generating independent operation and maintenance data with different formats and collection frequencies. Existing monitoring tools are mostly deployed layer by layer, lacking a mechanism for unified correlation analysis of cross-layer data. When faults exhibit symptoms simultaneously at multiple layers, relying solely on the timing of alarm events is insufficient to accurately determine the true propagation direction of disturbances between layers. Fault disturbances propagate across layers with time-delay correlations of state signals, but existing methods lack the technical means to utilize these time-delay correlations to determine the propagation direction and propagation delay. Therefore, there is an urgent need to develop a method and system that does not rely on pre-built knowledge graphs and can reconstruct cross-layer disturbance propagation paths and backtrack to locate the root cause network element node based on time-delay correlation analysis of measured state data at the time of the fault, thereby improving the comprehensiveness and accuracy of 5G base station node fault root cause location. Summary of the Invention

[0004] This invention provides a cloud computing-based intelligent management and control method and system for 5G base station node faults to solve the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides a cloud computing-based intelligent fault management method for 5G base station nodes, comprising: S1. Construct a state response matrix when a 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. S2. Based on the network level to which each row vector in the state response matrix belongs, aggregate the row vectors at the same level, perform differential transformation on the aggregation results between different levels, and generate an inter-level perturbation propagation matrix. S3. Perform a staggered cross-correlation scan on each row vector in the inter-level disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent levels based on the scan results. S4. Based on the propagation direction, determine the propagation start level of the fault disturbance. Based on the propagation delay, backtrack and locate the network element nodes in the state response matrix that are at the propagation start level to determine the root cause network element node of the current fault.

[0006] To address the aforementioned problems, this invention also provides a cloud-based intelligent management and control system for 5G base station node faults, the system comprising: The state response matrix construction module is used to construct the state response matrix when the 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. The inter-level perturbation differential extraction module is used to aggregate the row vectors of the same level according to the network level to which each row vector in the state response matrix belongs, perform differential transformation on the aggregation results between different levels, and generate an inter-level perturbation propagation matrix. The cross-layer propagation parameter scanning and determination module is used to perform misaligned cross-correlation scanning on each row vector in the inter-layer disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent layers based on the scanning results. The root cause network element node backtracking and positioning module is used to determine the propagation start level of the fault disturbance according to the propagation direction, and to perform backtracking and positioning of the network element nodes in the state response matrix that are at the propagation start level according to the propagation delay, so as to determine the root cause network element node of the current fault.

[0007] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention enables root cause localization of faults without the need for a pre-built knowledge graph. Existing technologies rely on pre-built knowledge graphs for fault propagation path reasoning. Building such knowledge graphs requires manual analysis of network element topology relationships and alarm propagation dependencies, resulting in long construction cycles, high update and maintenance costs, and an inability to cover unknown fault types. This invention directly starts from the measured state response matrix at the time of the fault, generating an inter-layer perturbation propagation matrix through intra-layer aggregation and inter-layer differentiation. It eliminates the need for pre-entering any fault modes or propagation paths, and can still achieve root cause localization for fault types not found in historical data, adapting to the dynamic changes in 5G network topology and configuration.

[0008] 2. This invention determines the propagation direction and propagation delay of cross-layer disturbances based on the time-delay correlation analysis of state signals. Existing technologies deploy monitoring tools layer by layer, lacking a unified cross-layer correlation analysis mechanism. Relying solely on the order of alarms makes it difficult to accurately determine the true propagation direction of disturbances between layers. This invention performs staggered cross-correlation scanning on the row vectors of the inter-layer disturbance propagation matrix, determines the propagation delay and direction based on the cross-correlation peak value, and projects the state change times of nodes in non-propagation initiation layers onto the time axis of the propagation initiation layer for backtracking and positioning based on the propagation delay. This achieves the reconstruction of cross-layer disturbance propagation paths and the precise positioning of root cause network elements. Attached Figure Description

[0009] Figure 1 A flowchart illustrating a cloud computing-based intelligent fault management method for 5G base station nodes provided in an embodiment of the present invention; Figure 2 This is a functional block diagram of a cloud-based intelligent management and control system for 5G base station node faults, provided in an embodiment of the present invention.

[0010] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0011] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0012] This application provides a cloud-based intelligent fault management method for 5G base station nodes. The executing entity of this cloud-based intelligent fault management method for 5G base station nodes includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the cloud-based intelligent fault management method for 5G base station nodes can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0013] Reference Figure 1 The diagram shown is a flowchart illustrating a cloud-based intelligent fault management method for 5G base station nodes according to an embodiment of the present invention. In this embodiment, the cloud-based intelligent fault management method for 5G base station nodes includes: S1. Construct a state response matrix when a 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. In this embodiment of the invention, when the cloud platform monitoring system detects that the 5G base station node to be analyzed has issued a fault alarm signal, the process of obtaining the status response matrix is ​​initiated.

[0014] The fault alarm signal includes the node identification of the faulty base station node and the time when the fault occurred.

[0015] Based on the node identity of the faulty base station node, the cloud platform monitoring system queries the network resource management system of the cloud platform to obtain the set of node identity of the superior network element node and the set of node identity of the subordinate network element node that have a direct topological connection with the faulty base station node.

[0016] The cloud platform monitoring system merges the node identity identifiers of the faulty base station node, the node identity identifiers of the upper-level network element node, and the node identity identifiers of the lower-level network element node to obtain the identity identifier set of the network element node to be measured.

[0017] The cloud platform monitoring system reads the preset sampling period parameter and the preset sampling quantity parameter from the configuration database. The unit of the sampling period parameter is seconds, and the unit of the sampling quantity parameter is samples.

[0018] The cloud platform monitoring system determines the start and end times of sampling based on the fault occurrence time, preset sampling period parameters, and preset sampling quantity parameters in the fault alarm signal. The start time of sampling is the fault occurrence time minus the product of the preset sampling quantity parameter and the preset sampling period parameter, corresponding to the time length. The end time of sampling is the fault occurrence time plus the product of the preset sampling quantity parameter and the preset sampling period parameter, corresponding to the time length.

[0019] The cloud platform monitoring system sends a status value collection instruction to each network element node in the identity identifier set of the network element node to be measured. The status value collection instruction includes the sampling start time, the sampling end time, and the preset sampling period parameter.

[0020] After receiving the status value acquisition instruction, each network element node acquires its running status value at the current sampling time once every preset sampling period parameter, starting from the start time of sampling, until the end time of sampling is reached. Each network element node obtains a set of status value sequences arranged in chronological order of sampling time. The number of status values ​​contained in the status value sequence is equal to twice the preset sampling quantity parameter plus one.

[0021] The operating status value of a network element node is the physical resource block utilization rate of that network element node at the corresponding sampling time. The unit of physical resource block utilization rate is percentage.

[0022] Each network element node sends its obtained status value sequence and its own node identity identifier back to the cloud platform monitoring system.

[0023] The cloud platform monitoring system receives the status value sequence and node identity identifier sent by each network element node in the identity identifier set of the network element node to be measured.

[0024] The cloud platform monitoring system uses node identity identifiers as row indexes and sampling time as column indexes to construct an initially empty data table. The number of rows in this data table is equal to the number of identity identifiers in the identity identifier set of the network element node to be measured, and the number of columns in this data table is equal to twice the preset sampling quantity parameter plus one.

[0025] The cloud platform monitoring system fills each received state value sequence into the row of the data table containing the node identity identifier corresponding to that state value sequence. Each state value in each state value sequence is then filled into the columns of that row in the order of sampling time to obtain a state response matrix. The rows of this state response matrix correspond to multiple network element nodes, and the columns correspond to multiple consecutive sampling times.

[0026] The element in the i-th row and j-th column of the state response matrix represents the physical resource block utilization rate of the i-th network element node at the j-th sampling time, where i is a positive integer greater than or equal to 1 and less than or equal to the number of identity identifiers in the identity identifier set of the network element node to be measured, and j is a positive integer greater than or equal to 1 and less than or equal to twice the preset sampling number parameter plus one.

[0027] S2. Based on the network level to which each row vector in the state response matrix belongs, aggregate the row vectors at the same level, perform differential transformation on the aggregation results between different levels, and generate an inter-level perturbation propagation matrix. In this embodiment of the invention, the cloud platform monitoring system reads each row vector in the status response matrix and simultaneously reads the network element node identity identifier corresponding to each row vector. Based on the network element node identity identifier, the system queries the network layer category to which the network element node belongs from the network resource management system of the cloud platform. The network layer categories include service layer category, orchestration layer category, and infrastructure layer category. Row vectors belonging to the same network layer category are grouped into the same vector group. The row vectors corresponding to the service layer category are grouped into the first vector group, the row vectors corresponding to the orchestration layer category are grouped into the second vector group, and the row vectors corresponding to the infrastructure layer category are grouped into the third vector group.

[0028] For each vector group, the cloud platform monitoring system extracts the state values ​​of any two row vectors within that vector group at the same sampling time, forming two state value sequences. The first state value sequence contains the state values ​​of the first row vector at all sampling times, and the second state value sequence contains the state values ​​of the second row vector at all sampling times. The Pearson correlation coefficient between these two state value sequences is calculated, and this Pearson correlation coefficient is used as the state similarity between the two row vectors. The state similarity value ranges from negative one to positive one.

[0029] For each vector group, the cloud platform monitoring system traverses all possible pairwise row vector combinations within the vector group to obtain the state similarity of all row vector pairs within the vector group.

[0030] The cloud platform monitoring system obtains a preset first similarity threshold and a preset second similarity threshold, wherein the first similarity threshold is greater than the second similarity threshold.

[0031] The cloud platform monitoring system compares the state similarity of all row vector pairs in each vector group with the first similarity threshold.

[0032] If the state similarity of all row vector pairs in the vector group is higher than the first similarity threshold, then the vector group meets the applicable conditions for weighted aggregation. For each row vector in the vector group, the cloud platform monitoring system calculates the average state similarity between the row vector and all other row vectors in the vector group, and uses the average value as the relevance weight corresponding to the row vector.

[0033] For a vector group that meets the applicable conditions for weighted aggregation, the cloud platform monitoring system multiplies the state values ​​of all row vectors in the vector group at the same sampling time with the correlation weights corresponding to the row vectors where each state value is located, and then sums them up. The summation result is divided by the sum of the correlation weights of all row vectors in the vector group to obtain the aggregated state value of the level corresponding to the vector group at the sampling time.

[0034] If any row vector pair in the vector group has a state similarity lower than the second similarity threshold, then the vector group does not meet the applicable conditions for weighted aggregation. The cloud platform monitoring system performs an arithmetic mean on the state values ​​of all row vectors in the vector group at the same sampling time, and uses the arithmetic mean as the aggregated state value of the level corresponding to the vector group at the sampling time.

[0035] After the cloud platform monitoring system completes the calculation of the aggregated status value for all sampling times for each vector group, it obtains the aggregated status value of the service layer category, the orchestration layer category, and the infrastructure layer category at each sampling time.

[0036] The following is a specific example of weighted merging.

[0037] Suppose that a certain vector group contains three row vectors, denoted as row vector A, row vector B and row vector C, and their state values ​​at sampling time t1 are 80, 90 and 70 respectively.

[0038] The average state similarity between row vector A and other row vectors in the same vector group is 0.9, the average state similarity between row vector B and other row vectors in the same vector group is 0.85, and the average state similarity between row vector C and other row vectors in the same vector group is 0.95.

[0039] The cloud platform monitoring system sets the relevance weight of row vector A to 0.9, the relevance weight of row vector B to 0.85, and the relevance weight of row vector C to 0.95.

[0040] At sampling time t1, the cloud platform monitoring system calculates 80 multiplied by 0.9 to get 72, 90 multiplied by 0.85 to get 76.5, 70 multiplied by 0.95 to get 66.5, 72 plus 76.5 plus 66.5 equals 215, 0.9 plus 0.85 plus 0.95 equals 2.7, and 215 divided by 2.7 equals 79.63.

[0041] Therefore, the aggregation state value of the level corresponding to this vector group at sampling time t1 is 79.63.

[0042] The following is a specific example of equal-weighted merging.

[0043] Suppose that a certain vector group contains three row vectors, denoted as row vector D, row vector E and row vector F, and their state values ​​at sampling time t2 are 85, 95 and 75 respectively.

[0044] Within this vector group, the state similarity between row vector D and row vector F is 0.4, which is lower than the second similarity threshold of 0.5. Therefore, this vector group does not meet the applicable conditions for weighted aggregation.

[0045] The cloud platform monitoring system calculates the arithmetic average of 85, 95, and 75. 85 plus 95 plus 75 equals 255, and 255 divided by 3 equals 85.

[0046] Therefore, the aggregation state value of the level corresponding to this vector group at sampling time t2 is 85.

[0047] After the cloud platform monitoring system completes the calculation of aggregated state values ​​for all sampling times, it arranges the aggregated state values ​​of each layer at each sampling time in the order of infrastructure layer category, orchestration layer category, and service layer category, to obtain the infrastructure layer aggregated state sequence, orchestration layer aggregated state sequence, and service layer aggregated state sequence. The aggregated state values ​​in each aggregated state sequence are arranged in the order of sampling times.

[0048] The cloud platform monitoring system subtracts the aggregated state value at the same sampling time from the aggregated state value at the same sampling time in the aggregated state sequence of the orchestration layer from the aggregated state value at the same sampling time in the aggregated state sequence of the infrastructure layer to obtain the first initial difference sequence between the infrastructure layer and the orchestration layer. Each element in the first initial difference sequence is the first initial difference value.

[0049] The cloud platform monitoring system subtracts the aggregated state value at the same sampling time from the aggregated state value at each sampling time in the service layer aggregated state sequence to obtain the second initial difference sequence between the orchestration layer and the service layer. Each element in the second initial difference sequence is the second initial difference value.

[0050] The following is a specific example of calculating the initial difference value.

[0051] Assume that the aggregated state value of the infrastructure layer is 70 at sampling time t3, the aggregated state value of the orchestration layer is 85 at sampling time t3, and the aggregated state value of the service layer is 95 at sampling time t3.

[0052] The cloud platform monitoring system subtracts the aggregated state value of the infrastructure layer from the aggregated state value of the orchestration layer (85) to obtain the first initial difference value of 15 between the infrastructure layer and the orchestration layer.

[0053] The cloud platform monitoring system subtracts the aggregation state value of the orchestration layer from the aggregation state value of the service layer (95) to obtain a second initial difference value of 10 between the orchestration layer and the service layer.

[0054] For the first initial difference sequence, the cloud platform monitoring system calculates the difference between the kth initial difference value and the kth initial difference value in the sequence to obtain the kth first difference change, where k is a positive integer from 1 to N minus 1, and N is the total number of initial difference values ​​in the initial difference sequence.

[0055] For the second initial difference sequence, the cloud platform monitoring system calculates the difference between the kth initial difference value and the kth initial difference value in the sequence to obtain the kth second difference change.

[0056] The cloud platform monitoring system acquires preset threshold values ​​for changes and preset consecutive values.

[0057] For the first differential change sequence, the cloud platform monitoring system checks each differential change one by one starting from the first differential change. When it finds that the differential changes of a consecutive preset number of sampling times all exceed the change threshold, it marks the time interval formed by the consecutive preset number of sampling times as the abrupt change interval between the infrastructure layer and the orchestration layer.

[0058] For the second differential change sequence, the cloud platform monitoring system checks each differential change one by one, starting from the first differential change. When it finds that the differential changes of a consecutive preset number of sampling times all exceed the change threshold, it marks the time interval formed by the consecutive preset number of sampling times as the abrupt change interval between the orchestration layer and the service layer.

[0059] The following is a specific example of a mutation interval labeling.

[0060] Assume the first initial difference sequence has an initial difference value of 10 at sampling time t1, 10.2 at sampling time t2, 12 at sampling time t3, 15 at sampling time t4, 12.5 at sampling time t5, 10.3 at sampling time t6, and 10.1 at sampling time t7.

[0061] In the first differential change sequence, the differential change between sampling times t1 and t2 is 0.2, the differential change between sampling times t2 and t3 is 1.8, the differential change between sampling times t3 and t4 is 3, the differential change between sampling times t4 and t5 is -2.5, the differential change between sampling times t5 and t6 is -2.2, and the differential change between sampling times t6 and t7 is -0.2.

[0062] The preset threshold for change is 0.15, and the preset number of consecutive values ​​is 3.

[0063] The absolute value of the difference change between sampling times t2 and t3 is 1.8, which is greater than 0.15; the absolute value of the difference change between sampling times t3 and t4 is 3, which is greater than 0.15; and the absolute value of the difference change between sampling times t4 and t5 is -2.5, which is greater than 0.15.

[0064] The difference changes at three consecutive sampling times all exceeded the preset change threshold. Therefore, the time interval from sampling time t2 to sampling time t4 was marked as the abrupt change interval.

[0065] For each mutation interval, the cloud platform monitoring system obtains the initial difference values ​​of the consecutive preset number of sampling times immediately preceding the start of the mutation interval, and calculates the average value of the preset number of initial difference values ​​as the average value before the mutation.

[0066] For the mutation interval, the cloud platform monitoring system obtains the initial difference values ​​of a preset number of consecutive sampling times immediately following the end of the mutation interval, and calculates the average value of the preset number of initial difference values ​​as the average value after the mutation.

[0067] The cloud platform monitoring system calculates the arithmetic mean of the average value before and after the mutation, and uses this arithmetic mean as the steady-state mean of the mutation interval. The unit of the steady-state mean is a percentage.

[0068] The following is a specific example of calculating the steady-state mean.

[0069] Assume that the initial difference values ​​of the three sampling times immediately preceding the start of the mutation interval between the infrastructure layer and the orchestration layer are 10, 10.2 and 10.3 respectively, and the average value before the mutation is 10 plus 10.2 plus 10.3 divided by 3, which equals 10.17.

[0070] The initial difference values ​​of the three sampling times immediately following the end of the mutation interval are 10.4, 10.5 and 10.6, respectively. The average value after the mutation is 10.4 plus 10.5 plus 10.6 divided by 3, which equals 10.5.

[0071] The cloud platform monitoring system adds the average value before the mutation (10.17) and the average value after the mutation (10.5), and then divides the result by 2 to get 10.335, which is taken as the steady-state mean of the mutation interval.

[0072] For each mutation interval, the cloud platform monitoring system calculates the deviation of the initial difference value at each sampling time within that mutation interval from the steady-state mean. The deviation is equal to the absolute value of the initial difference value at that sampling time minus the steady-state mean, and the unit of deviation is percentage.

[0073] The cloud platform monitoring system obtains preset deviation thresholds.

[0074] The cloud platform monitoring system marks abrupt change intervals with deviations greater than the deviation threshold as valid disturbance intervals.

[0075] The following is a specific example of marking an effective perturbation interval.

[0076] Assuming the initial difference value at sampling time t8 within a certain mutation interval is 12.5, the steady-state mean of this mutation interval is 10.335, the deviation is the absolute value of 12.5 minus 10.335 equals 2.165, the preset deviation threshold is 0.2, and 2.165 is greater than 0.2, this mutation interval is marked as an effective perturbation interval.

[0077] Assuming the initial difference value at sampling time t9 in another mutation interval is 10.4, the steady-state mean of this mutation interval is 10.335, and the deviation is the absolute value of 10.4 minus 10.335, which equals 0.065. Since 0.065 is less than 0.2, this mutation interval is not marked as an effective perturbation interval.

[0078] The cloud platform monitoring system obtains a preset first weight coefficient and a preset second weight coefficient, wherein the second weight coefficient is greater than zero and less than the first weight coefficient.

[0079] For each adjacent level, the cloud platform monitoring system retains the initial difference values ​​that belong to the effective disturbance interval by multiplying them by the first weighting coefficient, and retains the initial difference values ​​that do not belong to the effective disturbance interval by multiplying them by the second weighting coefficient.

[0080] The following is a specific example of weighting coefficient processing.

[0081] Assuming the first weighting coefficient is 1 and the second weighting coefficient is 0.1, the initial difference sequence between the infrastructure layer and the orchestration layer contains ten initial difference values ​​from sampling time t1 to sampling time t10, where sampling time t3 to sampling time t6 are marked as the effective perturbation interval.

[0082] The initial difference value at sampling time t3 is 12, which is multiplied by 1 to get 12; the initial difference value at sampling time t4 is 13, which is multiplied by 1 to get 13; the initial difference value at sampling time t5 is 14, which is multiplied by 1 to get 14; the initial difference value at sampling time t6 is 12.5, which is multiplied by 1 to get 12.5.

[0083] The initial difference value at sampling time t1 is 10, multiplied by 0.1 to get 1; the initial difference value at sampling time t2 is 10.2, multiplied by 0.1 to get 1.02; the initial difference value at sampling time t7 is 10.3, multiplied by 0.1 to get 1.03; the initial difference value at sampling time t8 is 10.1, multiplied by 0.1 to get 1.01; the initial difference value at sampling time t9 is 10, multiplied by 0.1 to get 1; the initial difference value at sampling time t10 is 10.2, multiplied by 0.1 to get 1.02.

[0084] The cloud platform monitoring system arranges the initial difference values ​​between the infrastructure layer and the orchestration layer after weighting coefficients according to the order of sampling time to obtain the perturbation difference sequence between the infrastructure layer and the orchestration layer.

[0085] The cloud platform monitoring system arranges the initial difference values ​​between the orchestration layer and the service layer after weighting coefficient processing according to the order of sampling time, thus obtaining the perturbation difference sequence between the orchestration layer and the service layer.

[0086] The cloud platform monitoring system uses the perturbation difference sequence between the infrastructure layer and the orchestration layer as the first row, and the perturbation difference sequence between the orchestration layer and the service layer as the second row. These are stacked in order from low to high level to obtain the inter-level perturbation propagation matrix. The row direction of this inter-level perturbation propagation matrix corresponds to adjacent layer pairs, and the column direction corresponds to continuous sampling time.

[0087] S3. Perform a staggered cross-correlation scan on each row vector in the inter-level disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent levels based on the scan results. In this embodiment of the invention, the cloud platform monitoring system reads the perturbation difference sequence between the infrastructure layer and the orchestration layer from the inter-layer perturbation propagation matrix as the first row vector, and reads the perturbation difference sequence between the orchestration layer and the service layer as the second row vector.

[0088] The cloud platform monitoring system obtains the preset latency search range and preset step length.

[0089] The cloud platform monitoring system uses a preset step length as the increment and traverses each misalignment position within a preset latency search range. Each misalignment position corresponds to a misalignment time length, which is equal to the misalignment position number multiplied by the step length. The misalignment position number is all integers from the negative maximum misalignment position number to the positive maximum misalignment position number. The maximum misalignment position number is equal to the preset latency search range divided by the preset step length and then rounded down.

[0090] For each misalignment position, the cloud platform monitoring system keeps the first row vector unchanged and shifts the second row vector relative to the first row vector on the time axis by the misalignment time length corresponding to the misalignment position. When the shift direction is positive, the second row vector shifts to the right, and when the shift direction is negative, the second row vector shifts to the left.

[0091] The first row vector contains the perturbation difference values ​​at L sampling times, denoted as sequence X, where the nth element of X is denoted as xn. The second row vector contains the perturbation difference values ​​at L sampling times, denoted as sequence Y, where the nth element of Y is denoted as yn, and L is the total number of sampling times.

[0092] After the second row vector is offset by the time length, the first row vector and the second row vector will both have data values ​​only within the overlapping time interval. The number of sampling moments within the overlapping time interval is L minus the number of sampling intervals corresponding to the absolute value of the offset.

[0093] For each misaligned position, the cloud platform monitoring system extracts the data values ​​of the first row vector within the overlapping time interval, denoted as sequence Xov, and extracts the data values ​​of the second row vector within the overlapping time interval, denoted as sequence Yov.

[0094] The sequence Xov contains M data values, denoted as xov1 to xovM, and the sequence Yov contains M data values, denoted as yov1 to yovM, where M is the number of sampling times within the overlapping time interval.

[0095] The cloud platform monitoring system calculates the mean of the sequence Xov, denoted as Xmean, which is equal to the sum of xov1 to xovM divided by M. It also calculates the mean of the sequence Yov, denoted as Ymean, which is equal to the sum of yov1 to yovM divided by M.

[0096] The cloud platform monitoring system calculates the difference between each data value in the sequence Xov and Xmean to obtain the X-centered sequence, and calculates the difference between each data value in the sequence Yov and Ymean to obtain the Y-centered sequence.

[0097] The cloud platform monitoring system calculates the sum of the dot products of the X-centered sequence and the Y-centered sequence to obtain the cumulative covariance value.

[0098] The cloud platform monitoring system calculates the sum of squares of each data value in the X-centered sequence to obtain the X energy value, and calculates the sum of squares of each data value in the Y-centered sequence to obtain the Y energy value.

[0099] The cloud platform monitoring system calculates the product of the X energy value and the Y energy value, and then takes the square root of the product to obtain the normalized denominator.

[0100] The cloud platform monitoring system divides the cumulative covariance value by the normalized denominator to obtain the cross-correlation value at the misalignment position. The cross-correlation value ranges from negative one to positive one.

[0101] The cloud platform monitoring system performs the above calculations for each misalignment position within the preset latency search range to obtain the cross-correlation value corresponding to each misalignment position.

[0102] The cloud platform monitoring system iterates through all the cross-correlation values ​​corresponding to the misaligned positions, obtains the absolute value of the cross-correlation value, and takes the misaligned position corresponding to the position with the largest absolute value of the cross-correlation value as the target misaligned position.

[0103] The cloud platform monitoring system uses the misalignment time length corresponding to the misalignment location of the target as the propagation delay of the fault disturbance between the infrastructure layer and the orchestration layer.

[0104] The cloud platform monitoring system obtains the sign of the cross-correlation value at the target misalignment location. When the sign of the cross-correlation value is positive, it determines that the fault disturbance has propagated from the infrastructure layer to the orchestration layer. When the sign of the cross-correlation value is negative, it determines that the fault disturbance has propagated from the orchestration layer to the infrastructure layer.

[0105] The cloud platform monitoring system performs the same misaligned cross-correlation scan on the disturbance difference sequence between the orchestration layer and the service layer to obtain the propagation delay and direction of the fault disturbance between the orchestration layer and the service layer.

[0106] The following is a specific example of a misaligned cross-correlation scan.

[0107] Suppose that the first row vector sequence X represents the five values ​​from sampling time t1 to t5, which are 1, 2, 3, 4, and 5 respectively, and the second row vector sequence Y represents the five values ​​from sampling time t1 to t5, which are 1, 3, 4, 6, and 8 respectively.

[0108] The preset time delay search range is from two negative sampling intervals to two positive sampling intervals, and the preset step length is one sampling interval.

[0109] When the misalignment position is 0, the overlapping time interval is from sampling time t1 to t5, the sequence Xov is 1, 2, 3, 4, 5, and the sequence Yov is 1, 3, 4, 6, 8.

[0110] Xmean equals 1 plus 2 plus 3 plus 4 plus 5 divided by 5, which equals 3. Ymean equals 1 plus 3 plus 4 plus 6 plus 8 divided by 5, which equals 4.4.

[0111] The centered sequence of X is -2, -1, 0, 1, 2, and the centered sequence of Y is -3.4, -1.4, -0.4, 1.6, 3.6.

[0112] The cumulative covariance is -2 multiplied by -3.4 plus -1 multiplied by -1.4 plus 0 multiplied by -0.4 plus 1 multiplied by 1.6 plus 2 multiplied by 3.6 equals 6.8 plus 1.4 plus 0 plus 1.6 plus 7.2 equals 17.

[0113] The energy value of X is -2 squared 4 plus -1 squared 1 plus 0 plus 1 squared 1 plus 2 squared 4 equals 10. The energy value of Y is -3.4 squared 11.56 plus -1.4 squared 1.96 plus -0.4 squared 0.16 plus 1.6 squared 2.56 plus 3.6 squared 12.96 equals 29.2.

[0114] The normalized denominator is 10 multiplied by 29.2, which equals 292. The square root is approximately 17.09.

[0115] The cross-correlation value is 17 divided by 17.09, which is approximately equal to 0.995.

[0116] When the misalignment position is 1, the second row vector shifts to the right by one sampling interval. The overlapping time interval is from sampling time t1 to t4, the sequence Xov is 1, 2, 3, 4, and the sequence Yov is 3, 4, 6, 8.

[0117] Xmean equals 1 plus 2 plus 3 plus 4 divided by 4, which equals 2.5. Ymean equals 3 plus 4 plus 6 plus 8 divided by 4, which equals 5.25.

[0118] The centered sequence of X is -1.5, -0.5, 0.5, 1.5, and the centered sequence of Y is -2.25, -1.25, 0.75, 2.75.

[0119] The cumulative covariance is -1.5 multiplied by -2.25 plus -0.5 multiplied by -1.25 plus 0.5 multiplied by 0.75 plus 1.5 multiplied by 2.75 equals 3.375 plus 0.625 plus 0.375 plus 4.125 equals 8.5.

[0120] The energy value of X is -1.5 squared = 2.25 + -0.5 squared = 0.25 + 0.5 squared = 0.25 + 1.5 squared = 2.25, which equals 5.0625. The energy value of Y is -2.25 squared = 5.0625 + -1.25 squared = 1.5625 + 0.75 squared = 0.5625 + 2.75 squared = 7.5625, which equals 14.75.

[0121] The normalized denominator is 5 multiplied by 14.75, which equals 73.75. The square root is approximately 8.59.

[0122] The cross-correlation value is 8.5 divided by 8.59, which is approximately equal to 0.989.

[0123] When the misalignment position is negative 1, the second row vector shifts to the left by one sampling interval. The overlapping time interval is from sampling time t2 to t5, the sequence Xov is 2, 3, 4, 5, and the sequence Yov is 1, 3, 4, 6.

[0124] Xmean equals 2 plus 3 plus 4 plus 5 divided by 4, which equals 3.5. Ymean equals 1 plus 3 plus 4 plus 6 divided by 4, which equals 3.5.

[0125] The centered sequence of X is -1.5, -0.5, 0.5, 1.5, and the centered sequence of Y is -2.5, -0.5, 0.5, 2.5.

[0126] The cumulative covariance is -1.5 multiplied by -2.5 plus -0.5 multiplied by -0.5 plus 0.5 multiplied by 0.5 plus 1.5 multiplied by 2.5 equals 3.75 plus 0.25 plus 0.25 plus 3.75 equals 8.

[0127] The energy value of X is -1.5 squared = 2.25 + -0.5 squared = 0.25 + 0.5 squared = 0.25 + 1.5 squared = 2.25, which equals 5. The energy value of Y is -2.5 squared = 6.25 + -0.5 squared = 0.25 + 0.5 squared = 0.25 + 2.5 squared = 6.25, which equals 13.

[0128] The normalized denominator is 5 multiplied by 13, which equals 65. The square root is approximately 8.06.

[0129] The cross-correlation value is approximately 0.993 when 8 is divided by 8.06.

[0130] Comparing the absolute values ​​of the cross-correlation values ​​at each misalignment position, the absolute value of the cross-correlation value at misalignment position 0 is 0.995, the absolute value of the cross-correlation value at misalignment position 1 is 0.989, and the absolute value of the cross-correlation value at misalignment position -1 is 0.993. Among them, the absolute value of the cross-correlation value at misalignment position 0 is the largest. The misalignment time length corresponding to misalignment position 0 is selected as the propagation delay, that is, the propagation delay is zero sampling intervals.

[0131] When the sign of the cross-correlation value is positive, it is determined that the fault disturbance has propagated from the infrastructure layer to the orchestration layer.

[0132] The cloud platform monitoring system repeats the above steps to read the perturbation difference sequences of other adjacent layer pairs from the inter-layer perturbation propagation matrix, and obtains the propagation direction and propagation delay between all adjacent layer pairs.

[0133] S4. Based on the propagation direction, determine the propagation start level of the fault disturbance. Based on the propagation delay, backtrack and locate the network element nodes in the state response matrix that are at the propagation start level to determine the root cause network element node of the current fault.

[0134] In this embodiment of the invention, after the cloud platform monitoring system obtains the propagation direction between all adjacent layer pairs, it extracts the network layer corresponding to the starting end pointed to by the propagation direction from all propagation directions and determines the network layer as the propagation starting layer.

[0135] When the propagation direction is from the infrastructure layer to the orchestration layer and from the orchestration layer to the service layer, the starting point of the propagation direction is the infrastructure layer, and the cloud platform monitoring system determines the infrastructure layer as the propagation starting layer.

[0136] When the propagation direction is from the service layer to the orchestration layer and from the orchestration layer to the infrastructure layer, the starting point of the propagation direction is the service layer, and the cloud platform monitoring system will determine the service layer as the propagation starting layer.

[0137] The cloud platform monitoring system reads the row vector corresponding to each network element node at the propagation start level from the state response matrix. This row vector contains the state value of the network element node at all sampling times.

[0138] The cloud platform monitoring system obtains the time of the fault occurrence, which is the sampling time corresponding to when the cloud platform monitoring system initially detected the fault in the 5G base station node.

[0139] The cloud platform monitoring system extracts the state values ​​of ten consecutive sampling times before the time of the fault in each row vector. These ten state values ​​are denoted as sequence Z, and the p-th element in sequence Z is denoted as zp, where p is a positive integer from 1 to 10.

[0140] The cloud platform monitoring system calculates the mean μ of the sequence Z, where μ is equal to the sum of z1 to z10 divided by 10.

[0141] The cloud platform monitoring system calculates the standard deviation σ of the sequence Z. σ is equal to the square root of the sum of the squares of the differences between each element in z1 to z10 and μ, divided by 10.

[0142] The cloud platform monitoring system uses μ plus twice σ as the upper boundary of the normal state baseline and μ minus twice σ as the lower boundary of the normal state baseline.

[0143] The following is a specific example of calculating the upper and lower boundaries of a normal baseline.

[0144] Suppose that the state values ​​of a certain network element node at ten consecutive sampling times before the fault occurs are 80, 82, 81, 83, 80, 82, 81, 83, 80, and 82.

[0145] The mean μ of sequence Z is equal to 80 + 82 + 81 + 83 + 80 + 82 + 81 + 83 + 80 + 82 divided by 10, which equals 81.4.

[0146] The difference between each element in sequence Z and μ is -1.4, 0.6, -0.4, 1.6, -1.4, 0.6, -0.4, 1.6, -1.4, and 0.6. The sum of the squares of these differences is 1.96 + 0.36 + 0.16 + 2.56 + 1.96 + 0.36 + 0.16 + 2.56 + 1.96 + 0.36, which equals 12.4.

[0147] The standard deviation σ is equal to 12.4 divided by 10, and the square root is 1.24, which is 1.11.

[0148] The upper boundary of the normal baseline is 81.4 plus 2 multiplied by 1.11, which equals 83.62. The lower boundary of the normal baseline is 81.4 minus 2 multiplied by 1.11, which equals 79.18.

[0149] The cloud platform monitoring system starts from the sampling time corresponding to the detection of a fault in the 5G base station node, and compares the state value of each sampling time in the row vector with the upper and lower boundaries of the normal state baseline one by one according to the order of sampling time.

[0150] When the state value at a certain sampling moment is greater than the upper boundary of the normal state baseline, the cloud platform monitoring system records the sampling moment as a candidate deviation start moment.

[0151] When the state value at a certain sampling moment is less than the lower boundary of the normal state baseline, the cloud platform monitoring system records the sampling moment as a candidate deviation start moment.

[0152] The cloud platform monitoring system obtains the status values ​​of the candidate at three consecutive sampling times after the deviation from the starting time, and compares these three status values ​​one by one with the upper and lower boundaries of the normal status baseline.

[0153] When the state values ​​at three consecutive sampling times are all greater than the upper boundary of the normal state baseline, the cloud platform monitoring system determines the candidate deviation start time as the state change time of the network element node.

[0154] When the state values ​​at three consecutive sampling times are all less than the lower boundary of the normal state baseline, the cloud platform monitoring system determines the candidate deviation start time as the state change time of the network element node.

[0155] When any one of the state values ​​in the three consecutive sampling times is between the upper and lower boundaries of the normal state baseline, the cloud platform monitoring system determines that the candidate deviation start time is a noise misjudgment and does not determine the candidate deviation start time as the state change time of the network element node, and continues to search for the next candidate deviation start time.

[0156] The following is a specific example of determining a candidate deviation from the initial time.

[0157] Suppose that the state values ​​of a certain row vector at five consecutive sampling times after the fault occurs are 84, 85, 86, 87, and 88, respectively, and the upper boundary of the normal state baseline is 83.62.

[0158] The state value 84 at sampling time t1 is greater than 83.62, and the cloud platform monitoring system records sampling time t1 as a candidate deviation from the start time.

[0159] The cloud platform monitoring system obtains the state values ​​85, 86, and 87 for three consecutive sampling times after sampling time t1. All three state values ​​are greater than 83.62. The cloud platform monitoring system determines sampling time t1 as the state change time of this network element node.

[0160] Suppose that the state values ​​of a certain row vector at five consecutive sampling times after the fault occurs are 84, 80, 86, 87, and 88, respectively. The upper boundary of the normal state baseline is 83.62, and the lower boundary of the normal state baseline is 79.18.

[0161] The state value 84 at sampling time t1 is greater than 83.62, and the cloud platform monitoring system records sampling time t1 as a candidate deviation from the start time.

[0162] The cloud platform monitoring system acquired the state values ​​80, 86, and 87 for three consecutive sampling times after sampling time t1. Among them, the state value 80 is between 79.18 and 83.62, which does not meet the condition that all three state values ​​are greater than the upper boundary or less than the lower boundary. Therefore, the cloud platform monitoring system determined that the sampling time t1 was a noise misjudgment.

[0163] The cloud platform monitoring system continues its search. The state value 80 at sampling time t2 is between the boundaries and is not recorded. The state value 86 at sampling time t3 is greater than 83.62, so sampling time t3 is recorded as a candidate deviation from the start time.

[0164] The cloud platform monitoring system obtains the state values ​​87, 88, and 89 for three consecutive sampling times after sampling time t3. All three state values ​​are greater than 83.62. The cloud platform monitoring system determines sampling time t3 as the state change time of this network element node.

[0165] The cloud platform monitoring system performs the above-mentioned normal state baseline calculation and candidate deviation start time determination steps on each network element node at the propagation start level to obtain the state change time of each network element node at the propagation start level.

[0166] The cloud platform monitoring system will treat network layers other than the propagation initiation layer as non-propagation initiation layers.

[0167] For each network element node that is not at the propagation initiation level, the cloud platform monitoring system obtains the state change time of each network element node using the same method.

[0168] The cloud platform monitoring system obtains the propagation latency, which is the time required for a fault disturbance to propagate from the initial propagation level to the adjacent level.

[0169] The time interval between adjacent sampling moments in the state response matrix is ​​the sampling period, which is a preset fixed duration.

[0170] The cloud platform monitoring system calculates the ratio of the propagation delay to the sampling period, and decomposes the ratio into an integer part and a fractional part. The integer part is the number of sampling intervals that are integer multiples of the propagation delay, and the fractional part is the margin of sampling intervals that are fractional multiples of the propagation delay.

[0171] The following is a specific example of calculating the number of sampling intervals as integer multiples and the margin of sampling intervals as fractional multiples.

[0172] Assuming a propagation delay of 7.5 seconds and a sampling period of 1 second, the propagation delay divided by the sampling period equals 7.5.

[0173] The integer part is 7, the decimal part is 0.5, the number of integer multiple sampling intervals is 7 sampling intervals, and the decimal multiple sampling interval margin is 0.5 sampling intervals.

[0174] The cloud platform monitoring system subtracts the integer multiple of the sampling interval from the state change time of each network element node in the non-propagation initiation layer. The integer multiple of the sampling interval is equal to the integer multiple of the sampling interval multiplied by the sampling period to obtain the first projection time.

[0175] The following is a specific example of calculating the first projection time.

[0176] Assume that the state change time of a certain network element node in the non-propagation initiation layer is the sampling time t10, the number of integer multiples of the sampling interval is 7, the sampling period is 1 second, and the time length is 7 seconds.

[0177] The first projection time is the sampling time t10 minus 7 seconds.

[0178] When sampling time t10 is the 10th sampling time, the first projection time is the 3rd sampling time.

[0179] The cloud platform monitoring system performs a fractional offset compensation on the first projection time based on a fractional multiple of the sampling interval margin to obtain the second projection time.

[0180] Fractional offset compensation is achieved through linear interpolation. The cloud platform monitoring system obtains the first state value at the sampling time of the first projection time, and obtains the second state value at the sampling time after the first projection time. The difference between the first state value and the second state value is multiplied by the fractional sampling interval margin, and then added to the first state value to obtain the projection time after fractional offset compensation.

[0181] The following is a specific example of calculating the second projection time.

[0182] Assume the first projection time is the third sampling time, the state value at the third sampling time is 80, the state value at the fourth sampling time is 84, and the fractional sampling interval margin is 0.5.

[0183] The state value difference is 84 minus 80 equals 4, 4 multiplied by 0.5 equals 2, and the state value corresponding to the projected time after fractional offset compensation is 80 plus 2 equals 82.

[0184] The cloud platform monitoring system uses the second projection moment as the projection moment of the non-propagation start-level network element node on the propagation start-level time axis.

[0185] The cloud platform monitoring system performs the above projection calculation on each network element node in the non-propagation initiation layer to obtain the projection time of each network element node in the non-propagation initiation layer.

[0186] The cloud platform monitoring system merges the state change times of each network element node at the propagation initiation level with the projection times of each network element node at the non-propagation initiation level onto the same time axis for arrangement.

[0187] The cloud platform monitoring system selects the earliest time from all the time intervals as the root cause network element node of the current fault.

[0188] The following is a specific example of root cause network element node selection.

[0189] Assume that the state change time of network element node A at the propagation initiation level is sampling time t3, the state change time of network element node B is sampling time t5, the projection time of network element node C at the non-propagation initiation level is sampling time t4, and the projection time of network element node D is sampling time t6.

[0190] The cloud platform monitoring system combines and arranges the sampling times t3, t4, t5, and t6, with the earliest sampling time being t3. The corresponding network element node is network element node A. The cloud platform monitoring system determines network element node A as the root cause network element node of the current fault.

[0191] like Figure 2 The diagram shown is a functional block diagram of a cloud computing-based intelligent management and control system for 5G base station node faults, provided in an embodiment of the present invention.

[0192] The cloud-based intelligent management and control system for 5G base station node faults described in this invention can be installed in electronic devices. Depending on the functions implemented, the cloud-based intelligent management and control system for 5G base station node faults may include a state response matrix construction module, an inter-level disturbance differential extraction module, a cross-layer propagation parameter scanning and determination module, and a root cause network element node backtracking and location module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0193] In this embodiment, the functions of each module / unit are as follows: The state response matrix construction module is used to construct a state response matrix when a 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. The inter-level perturbation differential extraction module is used to aggregate the row vectors of the same level according to the network level to which each row vector in the state response matrix belongs, and to perform differential transformation on the aggregation results between different levels to generate an inter-level perturbation propagation matrix. The cross-layer propagation parameter scanning and determination module is used to perform misaligned cross-correlation scanning on each row vector in the inter-layer disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent layers based on the scanning results. The root cause network element node backtracking and positioning module is used to determine the propagation start level of the fault disturbance according to the propagation direction, and to perform backtracking and positioning of the network element nodes in the state response matrix that are at the propagation start level according to the propagation delay, so as to determine the root cause network element node of the current fault.

[0194] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0195] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0197] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0198] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0199] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-based intelligent fault management method for 5G base station nodes, characterized in that, The method includes: S1. Construct a state response matrix when a 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. S2. Based on the network level to which each row vector in the state response matrix belongs, aggregate the row vectors at the same level, perform differential transformation on the aggregation results between different levels, and generate an inter-level perturbation propagation matrix. S3. Perform a staggered cross-correlation scan on each row vector in the inter-level disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent levels based on the scan results. S4. Based on the propagation direction, determine the propagation start level of the fault disturbance. Based on the propagation delay, backtrack and locate the network element nodes in the state response matrix that are at the propagation start level to determine the root cause network element node of the current fault.

2. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 1, characterized in that, The construction of the state response matrix when the 5G base station node to be analyzed fails includes: When a fault is detected in the 5G base station node to be analyzed, the state values ​​of the faulty base station node and its adjacent network element nodes in the communication link are measured to obtain the state values ​​of each network element node at multiple consecutive sampling times. The state values ​​of each network element node are arranged according to the network element node identity, sampling time order, and hierarchical affiliation and connection relationship between network element nodes to obtain the state response matrix.

3. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 1, characterized in that, The aggregation process for row vectors at the same level includes: Based on the network element node identity identifiers corresponding to each row vector in the state response matrix, the network level to which each row vector belongs is determined, and row vectors belonging to the same network level are grouped into the same vector group. Based on the state similarity between any two row vectors within each vector group, determine whether the vector group meets the applicable conditions for weighted aggregation.

4. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 3, characterized in that, The step of determining whether a vector group meets the applicable conditions for weighted aggregation based on the state similarity between any two row vectors within each vector group includes: If the state similarity between each row vector in the vector group is higher than the first preset similarity threshold, then the relevance weights corresponding to each row vector in the vector group are determined based on the state similarity. Based on the correlation weight, the state values ​​of all row vectors in the vector group at the same sampling time are weighted and merged. If there are row vector pairs in the vector group whose state similarity is lower than the second preset similarity threshold, then the state values ​​of all row vectors in the vector group at the same sampling time are merged with equal weight. The result of weighted merging or equal-weighted merging is used as the aggregation state value of the level corresponding to the vector group at each sampling time.

5. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 1, characterized in that, The step of performing a differential transformation on the aggregation results between different levels to generate an inter-level perturbation propagation matrix includes: For adjacent levels, the aggregated state values ​​of the lower level and the aggregated state values ​​of the higher level are numerically differ at the same sampling time to obtain the initial difference value; Calculate the change in difference between the initial difference values ​​at adjacent sampling times; When the differential change of a consecutive preset number of sampling times all exceed a preset change threshold, the time interval formed by the consecutive preset number of sampling times is marked as a sudden change interval.

6. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 5, characterized in that, The step of performing a difference transformation on the aggregation results between different levels to generate an inter-level perturbation propagation matrix further includes: Calculate the mean of the initial difference value before the start of the mutation interval and the initial difference value after the end of the mutation interval, and use the mean as the steady-state mean of the mutation interval; Calculate the deviation of the initial difference value at each sampling time within the mutation interval from the steady-state mean, and mark the mutation interval with a deviation greater than a preset deviation threshold as an effective disturbance interval; The initial difference value within the effective disturbance interval is retained with a first weighting coefficient, and the initial difference value outside the effective disturbance interval is retained with a second weighting coefficient, wherein the second weighting coefficient is greater than zero and less than the first weighting coefficient. The initial difference values ​​of all adjacent levels after weighting are arranged in ascending order of level to obtain the inter-level perturbation propagation matrix.

7. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 1, characterized in that, The step of performing a staggered cross-correlation scan on each row vector in the inter-level disturbance propagation matrix, and determining the propagation direction and propagation delay of the fault disturbance between adjacent levels based on the scan results, includes: Using a preset time delay search range and a preset step length, the two row vectors belonging to adjacent levels in the inter-level perturbation propagation matrix are relatively misaligned on the time axis; At each misalignment position, calculate the cross-correlation value of the two row vectors within the overlapping time interval; Traverse all misalignment positions within the time delay search range to obtain the target misalignment position corresponding to the maximum absolute value of the cross-correlation value; The misalignment time length corresponding to the target misalignment position is used as the propagation delay of the fault disturbance between adjacent levels; The propagation direction of the fault disturbance between adjacent levels is determined based on the sign of the cross-correlation value at the target misalignment location.

8. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 1, characterized in that, The step of determining the propagation initiation level of the fault disturbance based on the propagation direction, and backtracking the network element nodes at the propagation initiation level in the state response matrix based on the propagation delay to determine the root cause network element node of the current fault, includes: The network layer corresponding to the starting point to which the propagation direction points is determined as the propagation starting layer; The deviation time of the state value of each network element node in the state response matrix from the normal state baseline in the state response matrix at the propagation start level is taken as the state change time of each network element node. Based on the ratio of the time length value corresponding to the propagation delay to the time interval between adjacent sampling moments in the state response matrix, determine the number of integer multiple sampling intervals and the fractional multiple sampling interval margin included in the propagation delay; Other network layers besides the propagation initiation layer are taken as non-propagation initiation layers. The state change time of each network element node in the non-propagation initiation layer is subtracted from the integer multiple of the sampling interval number to obtain the first projection time. Based on the fractional sampling interval margin, the first projection time is compensated by a fractional offset to obtain the second projection time, and the second projection time is used as the projection time of each network element node on the time axis of the propagation start level. The state change time is merged with the projection time, and the network element node with the earliest time is selected as the root cause network element node of the current fault.

9. The intelligent fault management method for 5G base station nodes based on cloud computing as described in claim 8, characterized in that, The step of taking the deviation of the state value of each network element node at the propagation initiation level from the normal state baseline in the corresponding row vector of the state response matrix as the state change time of each network element node includes: Extract the state values ​​of consecutive sampling moments in each row vector before the fault occurs, calculate the mean and standard deviation of the state values ​​of the consecutive sampling moments, use the weighted sum of the mean and the standard deviation as the upper boundary of the normal state baseline, and use the weighted difference of the mean and the standard deviation as the lower boundary of the normal state baseline. Starting from the sampling time corresponding to when the 5G base station node is detected to have a fault, the state values ​​of each row vector are compared with the upper and lower boundaries of the normal state baseline one by one according to the sampling time order; When the state value at a certain sampling moment exceeds the upper boundary of the normal state baseline or falls below the lower boundary of the normal state baseline, the sampling moment is recorded as a candidate deviation start moment. The state values ​​of a preset number of consecutive sampling times after the candidate deviation start time are obtained. If the state values ​​of the preset number of consecutive sampling times all exceed the upper boundary of the normal state baseline or are all lower than the lower boundary of the normal state baseline, then the candidate deviation start time is determined as the state change time of each network element node.

10. A cloud-based intelligent fault management and control system for 5G base station nodes, characterized in that, The system for implementing the cloud computing-based intelligent fault management method for 5G base station nodes as described in claim 1 includes: The state response matrix construction module is used to construct the state response matrix when the 5G base station node to be analyzed fails. The row direction of the state response matrix corresponds to multiple network element nodes, and the column direction corresponds to multiple continuous sampling times. The inter-level perturbation differential extraction module is used to aggregate the row vectors of the same level according to the network level to which each row vector in the state response matrix belongs, perform differential transformation on the aggregation results between different levels, and generate an inter-level perturbation propagation matrix. The cross-layer propagation parameter scanning and determination module is used to perform misaligned cross-correlation scanning on each row vector in the inter-layer disturbance propagation matrix, and determine the propagation direction and propagation delay of the fault disturbance between adjacent layers based on the scanning results. The root cause network element node backtracking and positioning module is used to determine the propagation start level of the fault disturbance according to the propagation direction, and to perform backtracking and positioning of the network element nodes in the state response matrix that are at the propagation start level according to the propagation delay, so as to determine the root cause network element node of the current fault.

Images