Base station operation and maintenance method and device, electronic equipment and computer readable storage medium

Abstract

The embodiment of the invention provides a base station operation and maintenance method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: generating an alarm sequence and an initial state embedding representation based on the current historical alarm information of a target base station; determining a final state embedding representation of the target base station based on the alarm sequence; and predicting an alarm at a future target moment based on the final state embedding characterization to obtain a prediction result, and evaluating the health state of the target base station based on the final state embedding characterization and the initial state embedding characterization to obtain an evaluation result. The embodiment of the invention can help maintenance personnel to eliminate hidden dangers in time and reduce service obstacles.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a base station operation and maintenance method, a base station operation and maintenance device, an electronic device and a computer-readable storage medium. Background Art

[0002] The base station is an important link between the terminal and the network, and is also an important facility for realizing communication functions. Ensuring its healthy operation is related to the smooth operation of social production and the happiness index of people's lives. There are some shortcomings in the maintenance of base stations nowadays:

[0003] 1) Maintenance is mainly achieved through fault dispatch and periodic inspections, which has objective factors such as low efficiency and untimely resource allocation. Therefore, this passive maintenance method seems to be somewhat inadequate in the face of the ever-increasing scale of base stations, and naturally cannot guarantee the healthy operation of base stations.

[0004] 2) Base station health level assessment mainly relies on manual assessment, and different deduction weights are set for different alarms based on expert experience. For example, for more serious station disconnection alarms (such as network element connection interruption, RRU (Remote Radio Unit) disconnection, and equipment power failure, the deduction weight is larger. This method of relying solely on manual assessment of base station health obviously does not take into account the changes within the alarm data. Base station alarms are a reflection of internal problems of base stations, so it is necessary not only to pay attention to the alarms that occur in base stations, but also to understand the internal changes of alarms.

[0005] 3) Base stations generate various operation and maintenance data at all times, such as alarm data, business data, perception data, etc. Some internal changes in these data cannot be captured by the naked eye or ordinary data processing methods. The current base station operation and maintenance requires timely and effective maintenance measures based on these regular changes, but most of the current base station operation and maintenance lack this analytical capability.

[0006] In summary, the current healthy operation and maintenance of base stations lacks active early warning capabilities, and there are also deficiencies in the analysis of operation and maintenance data, which results in the inability to effectively guarantee the health of base stations and will also affect user perception. Summary of the invention

[0007] In view of the above problems, embodiments of the present invention are proposed to provide a base station operation and maintenance method, a base station operation and maintenance device, an electronic device and a computer-readable storage medium that overcome the above problems or at least partially solve the above problems.

[0008] An embodiment of the present invention discloses a base station operation and maintenance method, the method comprising:

[0009] Generate an alarm sequence and an initial state embedding representation based on the current historical alarm information of the target base station;

[0010] Determining a final state embedding representation of the target base station based on the alarm sequence;

[0011] Based on the final state embedded representation, an alarm at a future target moment is predicted to obtain a prediction result; and based on the final state embedded representation and the initial state embedded representation, a health state of the target base station is evaluated to obtain an evaluation result.

[0012] In one or more embodiments, the generating of the alarm sequence and the initial state embedded representation based on the current historical alarm information of the target base station includes:

[0013] Obtaining historical alarm information and attribute information of the target base station at the current moment;

[0014] Filtering target alarm information from the historical alarm information;

[0015] Alarm information of a first preset time period is selected from the target alarm information based on the current time to generate an alarm sequence, and an initial state embedding representation is generated based on the attribute information.

[0016] In one or more embodiments, the determining the state embedding representation of the target base station based on the alarm sequence includes:

[0017] Initializing the alarm sequence to obtain an alarm initial embedding matrix;

[0018] The alarm initial embedding matrix is ​​merged with a preset position embedding matrix to obtain an alarm initial position embedding matrix;

[0019] Inputting the alarm initial position embedding matrix into the self-attention network for data processing to obtain the embedding representation matrix output by the self-attention network;

[0020] Inputting the embedding representation matrix into a preset feedforward neural network for data processing to obtain a first candidate matrix output by the feedforward neural network;

[0021] Introducing the embedding representation matrix and the first candidate matrix into a preset residual connection to obtain a second candidate matrix;

[0022] Iterating the second candidate matrix a preset number of times to obtain an iteration result;

[0023] Each vector in the iteration result is summed to obtain a final state embedding representation of the target base station.

[0024] In one or more embodiments, the predicting an alarm at a future target time based on the final state embedded representation, and the evaluating the health state of the target base station based on the final state embedded representation and the initial state embedded representation, include:

[0025] The initial state embedding representation and the final state embedding representation are fused through a preset feedforward neural network to obtain a target embedding representation;

[0026] Predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result;

[0027] Calculating the similarity between the final state embedding representation and the initial state embedding representation;

[0028] The similarity is matched with the preset similarity intervals corresponding to each health state, and the health state corresponding to the matched similarity interval is used as the evaluation result.

[0029] In one or more embodiments, the step of predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result includes:

[0030] Calculating the probability of occurrence of each preset candidate alarm at the target time based on the vector inner product and the target embedding representation;

[0031] The candidate alarm with the highest probability of occurrence is taken as the prediction result.

[0032] In one or more embodiments, the method is applied to a processing system of a base station, the processing system comprising an alarm prediction subsystem and a health assessment subsystem, the alarm prediction subsystem is used to predict alarms at future target moments, and the health assessment subsystem is used to assess the health status of the target base station;

[0033] The method further comprises:

[0034] The alarm prediction subsystem and the health assessment subsystem are trained based on the cross entropy loss function to obtain a trained alarm prediction subsystem and a trained health assessment subsystem. The trained alarm prediction subsystem is used to predict the alarm at a future target moment at the next moment, and the trained health assessment subsystem is used to evaluate the health status of the target base station at the next moment.

[0035] In one or more embodiments, further comprising:

[0036] When the target time is reached, detecting whether the prediction result is correct, and obtaining a first detection result;

[0037] Acquire a second detection result of another base station; the other base station and the target base station are in the same area;

[0038] A hit rate of the processing system is calculated based on the first detection result and the second detection result.

[0039] Accordingly, an embodiment of the present invention discloses a base station operation and maintenance device, the device comprising:

[0040] A generation module, used for generating an alarm sequence and an initial state embedding representation based on the current historical alarm information of the target base station;

[0041] A determination module, configured to determine a final state embedding representation of the target base station based on the alarm sequence;

[0042] A prediction module, used to predict the alarm at a future target moment based on the final state embedding representation to obtain a prediction result;

[0043] An evaluation module is used to evaluate the health state of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result.

[0044] In one or more embodiments, the generating module is specifically used to:

[0045] Obtaining historical alarm information and attribute information of the target base station at the current moment;

[0046] Filtering target alarm information from the historical alarm information;

[0047] Alarm information of a first preset time period is selected from the target alarm information based on the current time to generate an alarm sequence, and an initial state embedding representation is generated based on the attribute information.

[0048] In one or more embodiments, the determining module is specifically configured to:

[0049] Initializing the alarm sequence to obtain an alarm initial embedding matrix;

[0050] The alarm initial embedding matrix is ​​merged with a preset position embedding matrix to obtain an alarm initial position embedding matrix;

[0051] Inputting the alarm initial position embedding matrix into the self-attention network for data processing to obtain the embedding representation matrix output by the self-attention network;

[0052] Inputting the embedding representation matrix into a preset feedforward neural network for data processing to obtain a first candidate matrix output by the feedforward neural network;

[0053] Introducing the embedding representation matrix and the first candidate matrix into a preset residual connection to obtain a second candidate matrix;

[0054] Iterating the second candidate matrix a preset number of times to obtain an iteration result;

[0055] Each vector in the iteration result is summed to obtain a final state embedding representation of the target base station.

[0056] In one or more embodiments, the prediction module is specifically used to:

[0057] The initial state embedding representation and the final state embedding representation are fused through a preset feedforward neural network to obtain a target embedding representation;

[0058] Predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result;

[0059] The evaluation module is specifically used for:

[0060] Calculating the similarity between the final state embedding representation and the initial state embedding representation;

[0061] The similarity is matched with the preset similarity intervals corresponding to each health state, and the health state corresponding to the matched similarity interval is used as the evaluation result.

[0062] In one or more embodiments, the prediction module is further configured to:

[0063] Calculating the probability of occurrence of each preset candidate alarm at the target time based on the vector inner product and the target embedding representation;

[0064] The candidate alarm with the highest probability of occurrence is taken as the prediction result.

[0065] In one or more embodiments, the method is applied to a processing system of a base station, the processing system comprising an alarm prediction subsystem and a health assessment subsystem, the alarm prediction subsystem is used to predict alarms at future target moments, and the health assessment subsystem is used to assess the health status of the target base station;

[0066] The device also includes:

[0067] A training module is used to train the alarm prediction subsystem and the health assessment subsystem based on a cross entropy loss function to obtain a trained alarm prediction subsystem and a trained health assessment subsystem, wherein the trained alarm prediction subsystem is used to predict an alarm at a future target moment at the next moment, and the trained health assessment subsystem is used to assess the health status of the target base station at the next moment.

[0068] In one or more embodiments, further comprising:

[0069] A detection module, used for detecting whether the prediction result is correct when the target time is reached, and obtaining a first detection result;

[0070] An acquisition module, used to acquire second detection results of other base stations; the other base stations and the target base station are in the same area;

[0071] A calculation module is used to calculate a hit rate of the processing system based on the first detection result and the second detection result.

[0072] Correspondingly, an embodiment of the present invention discloses an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the various steps of the above-mentioned base station operation and maintenance method embodiment when executed by the processor.

[0073] Correspondingly, an embodiment of the present invention discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the above-mentioned base station operation and maintenance method embodiment are implemented.

[0074] The embodiments of the present invention include the following advantages:

[0075] Generate an alarm sequence and an initial state embedded representation based on the current historical alarm information of the target base station; determine the final state embedded representation of the target base station based on the alarm sequence; predict the alarm at the future target moment based on the final state embedded representation to obtain a prediction result, and evaluate the health status of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result. Through the above method, the required alarm information can be obtained from all the historical alarm information of the target base station, and then an unsupervised deep learning algorithm is used to mine the various alarm information that has occurred in the target base station, capture the internal changes of the alarm information, and explore the alarm occurrence rules of the target base station, so as to learn the final state embedded representation of the target base station. According to the final state embedded representation, the alarms that may occur in the target base station in the near future can be predicted, thereby helping maintenance personnel to eliminate hidden dangers in a timely manner and reduce business obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flowchart of the steps of a base station operation and maintenance method embodiment 1 of the present invention;

[0077] Figure 2 is a schematic diagram of the structure of the processing system of the present invention;

[0078] Figure 3It is a schematic diagram of a process of obtaining a final state embedding representation of a target base station according to the present invention;

[0079] Figure 4 It is a logical schematic diagram of the prediction alarm of the present invention;

[0080] Figure 5 is a flow chart of steps of a second embodiment of a base station operation and maintenance method of the present invention;

[0081] Figure 6 It is a structural block diagram of an embodiment of a base station operation and maintenance device of the present invention. DETAILED DESCRIPTION

[0082] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0083] One of the core concepts of the embodiment of the present invention is to obtain the required alarm information from all the historical alarm information of the target base station, and then use an unsupervised deep learning algorithm to mine the various alarm information that has occurred in the target base station, capture the internal changes of the alarm information, and explore the alarm occurrence rules of the target base station, so as to learn the final state embedded representation of the target base station. According to the final state embedded representation, the alarms that may occur in the target base station in the near future can be predicted, thereby helping maintenance personnel to eliminate hidden dangers in a timely manner and reduce business obstacles.

[0084] Reference Figure 1 , shows a flow chart of the steps of a base station operation and maintenance method embodiment 1 of the present invention, the method is applied to the processing system of the base station, referring to Figure 2 , shows a schematic diagram of the structure of the processing system, wherein the processing system includes a general module and a business module, and the general module serves the business module. The general module includes an alarm acquisition module and an alarm analysis module, and the business module includes an alarm prediction subsystem and a health assessment subsystem. The alarm prediction subsystem is used to predict the alarm at the future target moment, and the health assessment subsystem is used to assess the health status of the target base station.

[0085] Furthermore, the processing system may perform alarm prediction and health assessment on all base stations in a certain area. In the embodiment of the present invention, the processing system is used to illustrate the processing of a base station (ie, a target base station) in a certain area.

[0086] The method may specifically include the following steps:

[0087] Step 101: Generate an alarm sequence and an initial state embedding representation based on current historical alarm information of a target base station.

[0088] For the target base station, a database for storing historical alarm information can be set up, so that the historical alarm information of the target base station can be stored in the database. When it is necessary to make an alarm prediction for a certain moment in the future (referred to as the "target moment"), the historical alarm information of the target base station can be obtained from the database, and then the historical alarm information is analyzed to obtain a sequence of alarm information sorted based on time (referred to as the "alarm sequence"), so as to facilitate subsequent prediction based on the alarm sequence. At the same time, the initial state embedded representation of the target base station can also be generated, so as to facilitate the subsequent evaluation of the health status of the target base station based on the initial state embedded representation.

[0089] In the embodiment of the present invention, the generating of the alarm sequence and the initial state embedding representation based on the current historical alarm information of the target base station includes:

[0090] Obtaining historical alarm information and attribute information of the target base station at the current moment;

[0091] Filtering target alarm information from the historical alarm information;

[0092] Alarm information of a first preset time period is selected from the target alarm information based on the current time to generate an alarm sequence, and an initial state embedding representation is generated based on the attribute information.

[0093] Specifically, the current time may be used as a reference point to obtain historical alarm information of the target base station (recorded as "current historical alarm information"), and to obtain attribute information of the target base station.

[0094] Since some alarm information will not affect the business, after obtaining the current historical alarm information, the alarm information that will not affect the business can be filtered out, and the alarm information that will affect the business (recorded as "target alarm information") can be retained.

[0095] Among them, when screening, the alarm information can be screened according to the title or alarm code. The alarm information can include the actual data of the alarm, the time of occurrence, and the alarm type. The alarm type can include wireless type, transmission type, and dynamic ring type. Of course, in actual applications, the specific screening method and the specific content of the alarm information can be set according to actual needs, and the embodiment of the present invention does not limit this.

[0096] Moreover, alarms caused by special reasons such as natural disasters are abnormal situations, so the alarm information of such alarms can also be filtered out during screening. Of course, in practical applications, whether the alarm information of such alarms needs to be filtered out can also be set according to actual needs, and the embodiment of the present invention does not limit this.

[0097] Furthermore, after the target alarm information is screened, the current moment can be used as a reference point to select alarm information within a period of time (recorded as the "first preset time period") from the target alarm information to generate an alarm sequence in chronological order. At the same time, the attribute information of the target base station is used to generate an initial state embedded representation.

[0098] The length of the alarm sequence, that is, the length of the first preset time period, can be set according to actual needs, and the embodiment of the present invention does not limit this.

[0099] Step 102: Determine a final state embedding representation of the target base station based on the alarm sequence.

[0100] Since the initial state embedding representation cannot reflect the internal relevance of the alarm information, and the self-attention network in deep learning has shown outstanding advantages in the relevance analysis in the field of machine translation and sequence recommendation. Therefore, the embodiment of the present invention introduces a self-attention network to explore the internal relevance of the alarm information, that is, the alarm sequence is processed by the self-attention network to obtain the final state embedding representation of the target base station.

[0101] In the embodiment of the present invention, the step of determining the state embedding representation of the target base station based on the alarm sequence includes:

[0102] Initializing the alarm sequence to obtain an alarm initial embedding matrix;

[0103] The alarm initial embedding matrix is ​​merged with a preset position embedding matrix to obtain an alarm initial position embedding matrix;

[0104] Inputting the alarm initial position embedding matrix into the self-attention network for data processing to obtain the embedding representation matrix output by the self-attention network;

[0105] Inputting the embedding representation matrix into a preset feedforward neural network for data processing to obtain a first candidate matrix output by the feedforward neural network;

[0106] Introducing the embedding representation matrix and the first candidate matrix into a preset residual connection to obtain a second candidate matrix;

[0107] Iterating the second candidate matrix a preset number of times to obtain an iteration result;

[0108] Each vector in the iteration result is summed to obtain a final state embedding representation of the target base station.

[0109] For the convenience of description, in the embodiment of the present invention, Alarm is used to represent the alarm set, and BS is used to represent the base station set. Then the base station set for a certain area can be represented as: BS = {bs1, bs2...bsn}, the alarm set can be expressed as: Alarm = {alarm1, alarm2...alarm m}. The alarm sequence of the base station set is expressed as: Among them, the alarm sequence of a single base station bs∈BS can be expressed as S bS =[s1, s2...s x ], where s x Indicates an alarm that occurred in the base station x .

[0110] Reference Figure 3 , which shows a schematic diagram of a process for obtaining the final state embedding representation of the target base station according to an embodiment of the present invention. Specifically, the alarm sequence for the target base station, that is, the alarm sequence S for the bs∈BS base station bs =[s1, s2...s x ], it can be initialized to obtain the alarm initial embedding matrix in It indicates an alarm that occurs at a base station bs at a certain moment, that is, the historical alarm information at a certain moment.

[0111] For the initial embedding matrix E of the alarm, a learnable, preset position embedding matrix P = [p1, p2...p x ], and obtain the initial position embedding matrix of the alarm with position enhancement in and The reason for introducing the position embedding matrix is ​​that the self-attention network does not have a positional relationship, that is, exchanging the positions of two elements in the sequence will not affect the final result. This approach does not conform to the rules of some base station alarms, because some alarms are temporal, that is, the alarms occur in a sequential order, and there are many such alarms in practice.

[0112] Then the initial position of the alarm with position enhancement obtained above is embedded in the matrix Input the self-attention network for data processing to obtain a new embedding representation matrix output by the self-attention network The specific implementation is as follows:

[0113]

[0114] In the formula, the input Q, K, and V of the self-attention network are expressed as: Three of the learnable projection matrices are It should be noted that if Figure 4As shown, the embodiment of the present invention predicts the alarm that may occur at the base station at time t (that is, the "target time") based on the alarm that occurred at the target base station at the previous (t-1) time. That is, when learning the state embedding of the base station at time t, it is necessary to pay attention to the alarm that occurred at the previous (t-1) time. Therefore, in the process of self-attention calculation, when j>i, the q of the query vector matrix Q i k with the key vector matrix K j Perform a masking operation, namely q i .(k j ) T =0.

[0115] Next, in order to explore the potential nonlinear features between different dimensions, B is fed into a two-layer feedforward neural network to obtain (denoted as the “first candidate matrix”), where The specific implementation is as follows:

[0116] z * =FNN(b * )=(W2σ(W1b * +t1)+t2)

[0117] In the formula, are all trainable parameter matrices, and σ is the nonlinear activation function Relu.

[0118] Furthermore, in order to prevent overfitting, a residual connection can be introduced after the feedforward neural network, and B and Z can be introduced into the residual connection to obtain (denoted as the “second candidate matrix”), where s * ∈R d , the specific implementation is as follows:

[0119]

[0120] In order to obtain more complex features, S can be iterated for a preset number of times and the result of the last iteration can be As the final output, The final state embedding representation of the base station bs∈BS is obtained by summing up multiple vectors in, which is specifically implemented as follows:

[0121]

[0122] Step 103: predicting an alarm at a future target time based on the final state embedded representation to obtain a prediction result, and evaluating a health state of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result.

[0123] After obtaining the final state embedded representation of the target base station, the final state embedded representation can be input into the alarm prediction subsystem, so that the alarm prediction subsystem uses the final state embedded representation to predict the alarm at a certain moment in the future (i.e., the "target moment"), thereby obtaining a prediction result, and the final state embedded representation is input into the health assessment subsystem, so that the health assessment subsystem uses the final state embedded representation to evaluate the health status of the target base station, thereby obtaining an evaluation result.

[0124] In an embodiment of the present invention, predicting an alarm at a future target moment based on the final state embedded representation, and evaluating the health state of the target base station based on the final state embedded representation and the initial state embedded representation include:

[0125] The initial state embedding representation and the final state embedding representation are fused through a preset feedforward neural network to obtain a target embedding representation;

[0126] Predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result;

[0127] Calculating the similarity between the final state embedding representation and the initial state embedding representation;

[0128] The similarity is matched with the preset similarity intervals corresponding to each health state, and the health state corresponding to the matched similarity interval is used as the evaluation result.

[0129] Specifically, after obtaining the final state embedding representation of the target base station, a two-layer feedforward neural network can be used to fuse the final state embedding representation and the initial state embedding representation of the target base station to obtain the target embedding representation. The fusion method is as follows:

[0130] e bs =FNN(e bs_i ||e bs_f )=(W4σ(W3(e bs_i ||e bs_f )+t3)+t4)

[0131] In the formula, σ is the non-linear activation function Relu.

[0132] Then, the vector inner product method and target embedding representation are used to predict the alarm at the target moment to obtain the prediction result.

[0133] The step of predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result includes:

[0134] Calculating the probability of occurrence of each preset candidate alarm at the target time based on the vector inner product and the target embedding representation;

[0135] The candidate alarm with the highest probability of occurrence is taken as the prediction result.

[0136] Specifically, each alarm in the alarm set can be used as a preset candidate alarm, and then the vector inner product method and target embedding representation are used to calculate the probability of occurrence of each candidate alarm at the target time. At the same time, each candidate alarm is normalized to obtain a set of occurrence probabilities of each alarm at the target time. The specific implementation is as follows:

[0137]

[0138] In the formula, Where A is the initialization of the alarm set alarm.

[0139] Then, the candidate alarm with the highest probability of occurrence in the set can be used as the prediction result.

[0140] Furthermore, the final state embedding representation e can be calculated bs_f Embedded representation with the initial state bs_i In this way, on the one hand, the interference of human factors can be eliminated to a large extent, and on the other hand, this evaluation method has great interpretability.

[0141] There are many similarity measurement methods in deep learning, such as Euclidean distance, cosine similarity, etc., and the embodiment of the present invention mainly focuses on the types of alarms and does not involve the analysis of the number of alarms. Therefore, in the embodiment of the present invention, cosine similarity, which has obvious advantages in this regard, can be used to measure the two states e of the base station bs∈BS. bs_f With e bs_i The similarity is implemented as follows:

[0142]

[0143] In the formula, cos(e bs_i ,e bs_f ) means calculating the cosine of two vectors.

[0144] After obtaining the similarity of the base station bs∈BS, the health status of the target base station can be evaluated according to the similarity. The greater the similarity, the target base station is in a relatively healthy state. Conversely, the target base station is in a relatively unhealthy state.

[0145] Furthermore, before the evaluation, the health status of the base station can be set to four levels according to the size of the similarity: healthy, sub-healthy, slightly pathological, and pathological. Each level has a one-to-one corresponding similarity interval. For example, the similarity interval corresponding to health is (0.75, 1], the similarity interval corresponding to sub-health is (0.5, 0.75], the similarity interval corresponding to slightly pathological is (0.25, 0.5], and the similarity interval corresponding to pathological is [0, 0.25]. Of course, in actual applications, the specific values ​​of the similarity intervals corresponding to each level can be set according to actual needs, and the embodiments of the present invention are not limited to this.

[0146] In this way, after the similarity is calculated, the similarity can be matched with each similarity interval, and the level corresponding to the matching item is used as the evaluation result of the health status.

[0147] Furthermore, the processing system can be a real-time evaluation system. Therefore, after obtaining the prediction results and evaluation results, corresponding report information can be generated and output according to the prediction results and evaluation results, such as the number of abnormalities in a cycle (week, month), the alarm with the greatest impact in the cycle, and the disposal suggestions for the alarm, as shown in Table 1. Maintenance personnel can refer to the report information for early warning and timely eliminate hidden dangers.

[0148] Base station ID / name Health Health status Abnormal cycle days Maximum impact alert Suggestions ***123 0.86 healthy *** alarm1 *** ***124 0.69 Sub-health *** alarm2 *** ***125 0.54 Mild illness *** alarm3 *** ***126 0.41 Pathological *** alarm4 *** … … … … … …

[0149] Table 1

[0150] Of course, in practical applications, the specific form of the report information can be set according to actual needs, and the embodiment of the present invention does not limit this.

[0151] In an embodiment of the present invention, an alarm sequence and an initial state embedded representation are generated based on the current historical alarm information of the target base station; a final state embedded representation of the target base station is determined based on the alarm sequence; an alarm at a future target moment is predicted based on the final state embedded representation to obtain a prediction result, and an evaluation result is obtained by evaluating the health status of the target base station based on the final state embedded representation and the initial state embedded representation. Through the above method, the required alarm information can be obtained from all historical alarm information of the target base station, and then an unsupervised deep learning algorithm is used to mine various alarm information that has occurred in the target base station, capture the internal changes of the alarm information, and explore the alarm occurrence rules of the target base station, so as to learn the final state embedded representation of the target base station. According to the final state embedded representation, the alarms that may occur in the target base station in the near future can be predicted, thereby helping maintenance personnel to eliminate hidden dangers in a timely manner and reduce business obstacles.

[0152] Reference Figure 5, shows a flow chart of the steps of Embodiment 2 of a base station operation and maintenance method of the present invention, which may specifically include the following steps:

[0153] Step 501, generating an alarm sequence and an initial state embedding representation based on current historical alarm information of a target base station;

[0154] Step 502, determining a final state embedding representation of the target base station based on the alarm sequence;

[0155] Step 503: predicting the alarm at the future target time based on the final state embedded representation to obtain a prediction result, and evaluating the health state of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result.

[0156] Among them, steps 501 to 503 are substantially the same as steps 101 to 103, and will not be described here to avoid repetition.

[0157] Step 504: Train the alarm prediction subsystem and the health assessment subsystem based on the cross entropy loss function to obtain a trained alarm prediction subsystem and a trained health assessment subsystem. The trained alarm prediction subsystem is used to predict the alarm at a future target moment at the next moment, and the trained health assessment subsystem is used to evaluate the health status of the target base station at the next moment.

[0158] Specifically, there are many loss functions in the field of deep learning, which can be roughly divided into losses based on distance measurement and losses based on probability distribution measurement. In order to make the probability distribution of the system output close to the true probability distribution (the true probability distribution is the characteristic of the cross entropy loss function itself), after obtaining the prediction results and evaluation results, the alarm prediction subsystem and health assessment subsystem can be trained according to the characteristics of the cross entropy loss function. The specific implementation is as follows:

[0159]

[0160] In the formula, is the system's predicted probability of a certain alarm, y i It is one-hot coding, that is, if the base station bs∈BS has an alarm alarm i , then y i =1.

[0161] Among them, during training, the current moment can be used as a reference, N-1 historical alarm information can be used as a training set, and the Nth historical alarm information can be used as a test set, and then the alarm prediction subsystem and the health assessment subsystem can be trained using the training set, the test set and the loss function. The trained alarm prediction subsystem is used to predict the alarm of a future target moment at the next moment, and the trained health assessment subsystem is used to evaluate the health status of the target base station at the next moment.

[0162] It should be noted that the training set and test set used in training are different at different times. For example, the current time is 2024-01-01, 12:00, and the number of historical alarm information is 1000. Then, the training set can be the 1st to 999th historical alarm information, and the 1000th historical alarm information is used as the test set. The trained alarm prediction subsystem can predict the alarm of a future target time at the next moment (such as 2024-01-02, 12:00), and the trained health assessment subsystem is used to assess the health status of the target base station at the next moment.

[0163] When the current time is 2024-01-02, 12:00, the number of historical alarm information is updated to 1100, then the training set can be the 1st to 1099th historical alarm information, and the 1100th historical alarm information is used as the test set. And so on.

[0164] Step 505, when the target time is reached, detecting whether the prediction result is correct, and obtaining a first detection result;

[0165] Step 506, obtaining second detection results of other base stations; the other base stations and the target base station are in the same area;

[0166] Step 507: Calculate the hit rate of the processing system based on the first detection result and the second detection result.

[0167] Specifically, after obtaining the prediction result and when the current moment reaches the target moment, it is possible to detect whether the prediction result is correct, that is, to detect whether an alarm identical to the prediction result occurs at the target moment. If it occurs, it means that the prediction result is correct. If it does not occur, or the alarm that occurs is different from the prediction result, it means that the prediction result is incorrect, thereby obtaining the detection result of the target base station (recorded as the "first detection result").

[0168] Then, the detection results of other base stations in the same area except the target base station are obtained (recorded as "second detection results"). The hit rate of the processing system can be calculated based on the first detection result and the second detection result. HR (HitsRatio) is used to represent the hit rate of the system. The specific implementation is as follows:

[0169]

[0170] In the formula, number represents the number of accurate hits of base stations, and |GT| represents the total number of base stations.

[0171] It should be noted that step 505 can be executed after step 503 or after step 504. In practical applications, the specific execution order can be set according to actual needs, and the embodiment of the present invention does not limit this.

[0172] In an embodiment of the present invention, the required alarm information can be obtained from all historical alarm information of the target base station, and then an unsupervised deep learning algorithm is used to mine various alarm information that has occurred in the target base station, capture the internal changes of the alarm information, and explore the alarm occurrence rules of the target base station, so as to learn the final state embedding representation of the target base station. According to the final state embedding representation, the alarms that may occur in the target base station in the near future can be predicted, and the recent health status of the target base station can also be evaluated. Detailed maintenance report information is output based on the predicted alarm information and health status, and the hit rate of the system prediction is statistically processed, so as to further help maintenance personnel to eliminate hidden dangers in time and reduce business obstacles.

[0173] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0174] Reference Figure 6 , shows a structural block diagram of an embodiment of a base station operation and maintenance device of the present invention, which may specifically include the following modules:

[0175] A generating module 601, configured to generate an alarm sequence and an initial state embedding representation based on the current historical alarm information of the target base station;

[0176] A determination module 602, configured to determine a final state embedding representation of the target base station based on the alarm sequence;

[0177] A prediction module 603 is used to predict an alarm at a future target moment based on the final state embedded representation to obtain a prediction result;

[0178] The evaluation module 604 is used to evaluate the health status of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result.

[0179] In the embodiment of the present invention, the generating module is specifically used for:

[0180] Obtaining historical alarm information and attribute information of the target base station at the current moment;

[0181] Filtering target alarm information from the historical alarm information;

[0182] Alarm information of a first preset time period is selected from the target alarm information based on the current time to generate an alarm sequence, and an initial state embedding representation is generated based on the attribute information.

[0183] In the embodiment of the present invention, the determining module is specifically used to:

[0184] Initializing the alarm sequence to obtain an alarm initial embedding matrix;

[0185] The alarm initial embedding matrix is ​​merged with a preset position embedding matrix to obtain an alarm initial position embedding matrix;

[0186] Inputting the alarm initial position embedding matrix into the self-attention network for data processing to obtain the embedding representation matrix output by the self-attention network;

[0187] Inputting the embedding representation matrix into a preset feedforward neural network for data processing to obtain a first candidate matrix output by the feedforward neural network;

[0188] Introducing the embedding representation matrix and the first candidate matrix into a preset residual connection to obtain a second candidate matrix;

[0189] Iterating the second candidate matrix a preset number of times to obtain an iteration result;

[0190] Each vector in the iteration result is summed to obtain a final state embedding representation of the target base station.

[0191] In the embodiment of the present invention, the prediction module is specifically used for:

[0192] The initial state embedding representation and the final state embedding representation are fused through a preset feedforward neural network to obtain a target embedding representation;

[0193] Predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result;

[0194] The evaluation module is specifically used for:

[0195] Calculating the similarity between the final state embedding representation and the initial state embedding representation;

[0196] The similarity is matched with the preset similarity intervals corresponding to each health state, and the health state corresponding to the matched similarity interval is used as the evaluation result.

[0197] In the embodiment of the present invention, the prediction module is further used for:

[0198] Calculating the probability of occurrence of each preset candidate alarm at the target time based on the vector inner product and the target embedding representation;

[0199] The candidate alarm with the highest probability of occurrence is taken as the prediction result.

[0200] In an embodiment of the present invention, the method is applied to a processing system of a base station, the processing system includes an alarm prediction subsystem and a health assessment subsystem, the alarm prediction subsystem is used to predict alarms at future target moments, and the health assessment subsystem is used to assess the health status of the target base station;

[0201] The device also includes:

[0202] A training module is used to train the alarm prediction subsystem and the health assessment subsystem based on a cross entropy loss function to obtain a trained alarm prediction subsystem and a trained health assessment subsystem, wherein the trained alarm prediction subsystem is used to predict an alarm at a future target moment at the next moment, and the trained health assessment subsystem is used to assess the health status of the target base station at the next moment.

[0203] In an embodiment of the present invention, it also includes:

[0204] A detection module, used for detecting whether the prediction result is correct when the target time is reached, and obtaining a first detection result;

[0205] An acquisition module, used to acquire second detection results of other base stations; the other base stations and the target base station are in the same area;

[0206] A calculation module is used to calculate a hit rate of the processing system based on the first detection result and the second detection result.

[0207] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0208] An embodiment of the present invention further provides an electronic device, including:

[0209] It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned base station operation and maintenance method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0210] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned base station operation and maintenance method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0211] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0212] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0213] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0214] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0215] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0216] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0217] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0218] The above is a detailed introduction to a base station operation and maintenance method and a base station operation and maintenance device provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A base station operation and maintenance method, characterized in that: The method comprises: Generate an alarm sequence and an initial state embedding representation based on the current historical alarm information of the target base station; Determining a final state embedding representation of the target base station based on the alarm sequence; Based on the final state embedded representation, an alarm at a future target moment is predicted to obtain a prediction result, and based on the final state embedded representation and the initial state embedded representation, a health state of the target base station is evaluated to obtain an evaluation result.

2. The base station operation and maintenance method according to claim 1, characterized in that: The generating of the alarm sequence and the initial state embedding representation based on the current historical alarm information of the target base station includes: Obtaining historical alarm information and attribute information of the target base station at the current moment; Filtering target alarm information from the historical alarm information; Alarm information of a first preset time period is selected from the target alarm information based on the current time to generate an alarm sequence, and an initial state embedding representation is generated based on the attribute information.

3. The base station operation and maintenance method according to claim 1, characterized in that: The step of determining the state embedding representation of the target base station based on the alarm sequence includes: Initializing the alarm sequence to obtain an alarm initial embedding matrix; The alarm initial embedding matrix is ​​merged with a preset position embedding matrix to obtain an alarm initial position embedding matrix; Inputting the alarm initial position embedding matrix into the self-attention network for data processing to obtain the embedding representation matrix output by the self-attention network; Inputting the embedding representation matrix into a preset feedforward neural network for data processing to obtain a first candidate matrix output by the feedforward neural network; Introducing the embedding representation matrix and the first candidate matrix into a preset residual connection to obtain a second candidate matrix; Iterating the second candidate matrix a preset number of times to obtain an iteration result; Each vector in the iteration result is summed to obtain a final state embedding representation of the target base station.

4. The base station operation and maintenance method according to claim 1, characterized in that: The predicting of an alarm at a future target moment based on the final state embedded representation, and the evaluating of a health state of the target base station based on the final state embedded representation and the initial state embedded representation, include: The initial state embedding representation and the final state embedding representation are fused through a preset feedforward neural network to obtain a target embedding representation; Predicting an alarm at a future target moment based on the vector inner product and the target embedding representation to obtain a prediction result; Calculating the similarity between the final state embedding representation and the initial state embedding representation; The similarity is matched with the preset similarity intervals corresponding to each health state, and the health state corresponding to the matched similarity interval is used as the evaluation result.

5. The base station operation and maintenance method according to claim 4, characterized in that: The predicting of the alarm at the future target moment based on the vector inner product and the target embedding representation to obtain the prediction result includes: Calculating the probability of occurrence of each preset candidate alarm at the target time based on the vector inner product and the target embedding representation; The candidate alarm with the highest probability of occurrence is taken as the prediction result.

6. The base station operation and maintenance method according to claim 1, characterized in that: The method is applied to a processing system of a base station, the processing system comprising an alarm prediction subsystem and a health assessment subsystem, the alarm prediction subsystem is used to predict alarms at a future target moment, and the health assessment subsystem is used to assess the health status of the target base station; The method further comprises: The alarm prediction subsystem and the health assessment subsystem are trained based on the cross entropy loss function to obtain a trained alarm prediction subsystem and a trained health assessment subsystem. The trained alarm prediction subsystem is used to predict the alarm at a future target moment at the next moment, and the trained health assessment subsystem is used to evaluate the health status of the target base station at the next moment.

7. The base station operation and maintenance method according to claim 1, characterized in that: Also includes: When the target time is reached, detecting whether the prediction result is correct, and obtaining a first detection result; Obtaining second detection results of other base stations; The other base stations and the target base station are in the same area; A hit rate of the processing system is calculated based on the first detection result and the second detection result.

8. A base station operation and maintenance device, characterized in that: The device comprises: A generation module, used for generating an alarm sequence and an initial state embedding representation based on the current historical alarm information of the target base station; A determination module, configured to determine a final state embedding representation of the target base station based on the alarm sequence; A prediction module, used to predict the alarm at a future target moment based on the final state embedding representation to obtain a prediction result; An evaluation module is used to evaluate the health state of the target base station based on the final state embedded representation and the initial state embedded representation to obtain an evaluation result.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the base station operation and maintenance method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the base station operation and maintenance method according to any one of claims 1 to 7 are implemented.

Images