Alarm and early warning management method and system for base station

Abstract

The invention discloses a base station alarm early warning management method and system. The method comprises the following steps: acquiring a first operation state sequence of each of a plurality of first base stations in a target area in a first time period; for each first base station, analyzing the first operation state sequence corresponding to the first base station by using a pre-trained base station alarm type prediction model to obtain a first predicted base station alarm type of the first base station at the first moment; and determining the troubleshooting priority of each first base station according to the first predicted base station alarm type of each of the plurality of first base stations in the target area at the first moment, and managing the plurality of first base stations in the target area according to the troubleshooting priorities. According to the invention, the technical problem that the operation and maintenance efficiency and the fault defense capability of the base station are relatively low because the related base station operation and maintenance technology cannot carry out early warning on the base station alarm in advance is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology. Specifically, it relates to a method and system for managing base station alarm early warning. Background Art

[0002] As the infrastructure of modern mobile communication networks, base stations play an extremely important role. They not only provide people with convenient and efficient communication services but also promote the development of wireless communication technology and the prosperity of related industries. However, the following problems still exist in the operation and maintenance of current base stations:

[0003] First, in the operation and maintenance of base stations, technologies such as alarm location and alarm diagnosis have become mature, but there is still a large gap in alarm early warning. Moreover, with the popularization of 5G technology, the scale of small base stations is increasing day by day, and the resulting large amount of maintenance work also makes maintenance personnel overwhelmed.

[0004] Second, relying solely on the analysis of base station operation and maintenance data by simple data analysis tools is difficult to capture the internal changes in the data, and the analysis is relatively single, making it difficult to analyze the operation status of base stations by combining multiple types of data.

[0005] In response to the above problems, no effective solution has been proposed yet. Summary of the Invention

[0006] Embodiments of this application provide a method and system for managing base station alarm early warning to at least solve the technical problem that the related base station operation and maintenance technology cannot give early warnings to base station alarms, resulting in low operation and maintenance efficiency and fault prevention ability of base stations.

[0007] According to one aspect of the embodiments of this application, a method for managing base station alarm early warning is provided, including: obtaining the first operation status sequences of multiple first base stations in a target area during a first time period; for each first base station, using a pre-trained base station alarm type prediction model to analyze the first operation status sequence corresponding to the first base station to obtain the first predicted base station alarm type of the first base station at a first moment, where the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network using multiple sets of training sample data, and the first moment is a moment after the first time period; determining the troubleshooting priority levels of each first base station according to the first predicted base station alarm types of multiple first base stations in the target area at the first moment, and managing multiple first base stations in the target area according to the troubleshooting priority levels.

[0008] Optionally, obtaining the first operation status sequences of multiple first base stations within the target area in the first time period includes: determining multiple first base stations within the target area; for each first base station, obtaining the initial operation status sequence of the first base station in the first time period, where the operation status data corresponding to multiple monitoring dimensions is included in the initial operation status sequence, and at least one of the following is included in the multiple monitoring dimensions: the operation and maintenance center dimension, the performance dimension; performing a preprocessing operation on the initial operation status sequence to obtain the corresponding first operation status sequence, where the preprocessing operation includes at least one of the following: data cleaning, data format conversion, normalization processing, standardization processing.

[0009] Optionally, the training process of the base station alarm type prediction model includes: obtaining multiple groups of training sample data, where each group of training sample data includes: the second operation status sequence of the second base station in the second time period and the second base station alarm type at the second moment after the second time period, and the second time period is a historical time period before the first time period, and the second moment is a historical moment before the first moment; constructing a neural network model, where at least the following is included in the neural network model: a self-attention network, a long short-term memory network; using multiple groups of training sample data to perform iterative training on the neural network model to obtain the base station alarm type prediction model.

[0010] Optionally, using multiple groups of training sample data to perform iterative training on the neural network model to obtain the base station alarm type prediction model includes: for each group of training sample data, using a sliding window to divide the second operation status sequence of the second base station in the second time period in the training sample data into second operation status data sets corresponding to n equal-length time steps, and using the self-attention network and the long short-term memory network in the neural network model to sequentially process the second operation status data sets at n time steps to obtain the second predicted base station alarm type of the second base station in the second time period; constructing a target loss function using the second base station alarm type and the corresponding second predicted base station alarm type in each group of training sample data; adjusting the model parameters of the neural network model according to the target loss function until the model parameters converge to obtain the trained base station alarm type prediction model.

[0011] Optionally, the long short-term memory network includes a plurality of memory units corresponding to a plurality of time steps recursively arranged according to a time series, and each memory unit includes: a forget gate, an input gate, and an output gate. Among them, the self-attention network and the long short-term memory network in the neural network model are used to process the second operating state data set at n time steps in sequence, and the second predicted base station alarm type of the second base station in the second time period is obtained, including: traversing each time step, and performing the following steps to obtain the memory unit output corresponding to each time step: The first step: input the memory unit output corresponding to the previous time step of the current time step and the second operating state data set in the current time step into the first Sigmoid activation function, and use the obtained output result as the forget gate activation value; The second step: multiply the memory unit state C t-1 corresponding to the previous historical time step of the current time step by the forget gate activation value to update the memory unit state C t-1 corresponding to the previous historical time step of the current time step, and obtain the memory unit state C' t-1 , and use the memory unit state C' t-1 as the output of the forget gate; The third step: determine the input gate activation value using the forget gate activation value; The fourth step: use the tanh activation function to integrate the second operating state data set in the current time step and the memory unit output corresponding to the previous time step of the current time step to obtain the first candidate state value; The fifth step: multiply the input gate activation value by the first candidate state value to determine the important information to be retained in the first candidate state value, and add the important information to the memory unit state C' t-1 to obtain the memory unit state C t corresponding to the current time step; The sixth step: use the tanh activation function to process the memory unit state C t corresponding to the current time step to obtain the second candidate state value; The seventh step: input the second operating state data set in the current time step and the memory unit output corresponding to the previous historical time step of the current time step into the second Sigmoid activation function to obtain the output gate activation value; The eighth step: multiply the second candidate state value by the output gate activation value to obtain the memory unit output corresponding to the current time step; use the memory unit outputs corresponding to the previous n-1 time steps as key vectors, and use the memory unit output corresponding to the last time step as the query vector, and analyze using the self-attention network to obtain the second predicted base station alarm type of the second base station in the second time period.

[0012] Optionally, after obtaining the first predicted base station alarm type of the first base station at the first moment, the method further includes: determining the number of first base stations whose first predicted base station alarm type is the batch outage alarm type according to the first predicted base station alarm types of the respective first base stations at the first moment, where the batch outage alarm type set includes at least one of the following: RRU disconnection, input power supply disconnection, radio frequency unit link anomaly; determining the base station batch outage probability of the target area according to the number of first base stations and the total number of base stations in the target area; when the base station batch outage probability is not lower than the preset batch outage probability threshold, determining the troubleshooting priority of each first base station according to the first predicted base station alarm types of the multiple first base stations in the target area at the first moment, and managing the multiple first base stations in the target area according to the troubleshooting priority.

[0013] Optionally, determining the troubleshooting priority of each first base station according to the first predicted base station alarm types of the respective first base stations at the first moment includes: determining the number of second base stations whose first predicted base station alarm type is different batch outage alarm types according to the first predicted base station alarm types of the respective first base stations at the first moment; sorting the first base stations whose first predicted base station alarm type is various batch outage alarm types according to the number of second base stations, and determining the corresponding troubleshooting priority of each first base station according to the obtained sorting result.

[0014] According to another aspect of the embodiments of the present application, there is also provided a base station alarm early warning management system, including: an acquisition module, configured to acquire the first operation state sequences of multiple first base stations in a target area during a first time period; a prediction module, configured to, for each first base station, analyze the first operation state sequence corresponding to the first base station by using a pre-trained base station alarm type prediction model to obtain the first predicted base station alarm type of the first base station at the first moment, where the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network by using multiple sets of training sample data, and the first moment is a moment after the first time period; a management module, configured to determine the troubleshooting priority of each first base station according to the first predicted base station alarm types of the multiple first base stations in the target area at the first moment, and manage the multiple first base stations in the target area according to the troubleshooting priority.

[0015] According to another aspect of the embodiments of the present application, there is also provided a computer program product, including: a computer program, where when the computer program is executed by a processor, it implements the above-mentioned base station alarm early warning management method.

[0016] According to another aspect of the embodiments of the present application, an electronic device is further provided. The electronic device includes: a memory and a processor. Among them, a computer program is stored in the memory, and the processor is configured to execute the above-mentioned base station alarm warning management method through the computer program.

[0017] In the embodiments of the present application, a deep learning model based on a self-attention network and a long short-term memory network is used to analyze the first operation state sequences of multiple base stations in a target area within a first time period, and the first predicted base station alarm types of each first base station at a first moment are obtained, realizing early warning of future alarm types of base stations to actively discover potential network faults and health problems; then, according to the first predicted base station alarm types of each of the multiple first base stations at the first moment, the troubleshooting priorities of each type of first predicted base station alarm type are determined, and each first base station in the target area is managed according to the troubleshooting priorities, changing the traditional passive response base station operation and maintenance method, achieving the purpose of active base station operation and maintenance management, effectively improving the efficiency of base station maintenance, reducing the risk of network interruption caused by lagging fault response, and ensuring the stability and service quality of the communication network. Furthermore, it solves the technical problem that the related base station operation and maintenance technology cannot give early warning of base station alarms, resulting in low operation and maintenance efficiency and fault prevention ability of base stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0019] Figure 1 is a schematic flowchart of an optional base station alarm warning management method according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of the principle of an optional neural network model according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of the principle of an optional sliding window according to an embodiment of the present application;

[0022] Figure 4 is a schematic structural diagram of an optional base station alarm warning management system according to an embodiment of the present application;

[0023] Figure 5 is a schematic structural diagram of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solution in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0025] It should be noted that the terms "first", "second", etc. in the specification, claims and drawings of this application are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] To better understand the embodiments of this application, the following is a translation and explanation of some nouns or terms that appear in the description process of the embodiments of this application:

[0027] LSTM (Long Short Term Memory) network: It is an improved RNN (Recurrent Neural Network), aiming to solve the problems of gradient disappearance and gradient explosion in the training process of RNN for long sequences. LSTM controls the flow of information through a gating mechanism, which includes a forget gate, an input gate, and an output gate, where:

[0028] (1) Input Gate: Responsible for controlling how much information in the current input is updated into the memory cell. It receives the current input vector and the hidden state vector of the previous moment as inputs, and calculates an update ratio between 0 and 1 through the Sigmoid activation function. Therefore, the expression of the input gate can be written as:

[0029] i t = σ(W i × [h t-1 , x t + b i )

[0030] In the formula, σ(·) represents the Sigmoid activation function of the input gate, and W i represents the weight matrix of the input gate; bi represents the bias vector of the input gate, x t represents the input vector at the current time t, h t-1 represents the hidden state vector at the previous time t-1.

[0031] (2) Forget Gate: Responsible for determining which information in the memory cell needs to be retained and which needs to be discarded. It receives the current input vector and the hidden state vector of the previous time as inputs, and maps them to values between 0 and 1 through a Sigmoid activation function. Among them, a value close to 0 indicates that the corresponding information will be forgotten, and a value close to 1 indicates that the corresponding information will be retained. Therefore, the expression of the forget gate can be written as:

[0032] f t =σ(W f ·[h t-1 ,x t +b f )

[0033] In the formula, σ(·) represents the Sigmoid activation function of the forget gate, W f represents the weight matrix of the forget gate, b f represents the bias vector of the forget gate.

[0034] (3) Candidate Cell State: The current input vector and the hidden state vector of the previous time are transformed through the tanh activation function, and then the two are multiplied to obtain the information that needs to be updated to the memory cell. Therefore, the expression of the candidate memory cell can be written as:

[0035] g t =tanh(W c ×[h t-1 ,x t +b c )

[0036] In the formula, W c represents the weight matrix of the candidate memory cell; b c represents the bias vector of the candidate memory cell.

[0037] (4) Update Memory Cell: The memory cell is updated according to the results of the forget gate and the input gate. Therefore, the expression of the updated memory cell can be written as:

[0038] g ′ t =f t ⊙C t-1 +i t ⊙g t

[0039] where C t represents the state of the memory cell at the current time t; C t-1 represents the state of the memory cell at the previous time t - 1; ⊙ represents element-wise multiplication.

[0040] (5) Output Gate: Determines which information in the memory cell will be output as the hidden state at the current time. It receives the current input vector and the hidden state vector at the previous time as inputs, calculates an output ratio between 0 and 1 through the Sigmoid activation function, and then multiplies it by the memory cell processed by the tanh activation function to obtain the hidden state at the current time. Therefore, the expression of the output gate can be written as:

[0041] o t = σ(W o × [h t-1 , x t + b o )

[0042] h t = o t ⊙ tanh(C t )

[0043] where W o represents the weight matrix of the output gate; b o represents the bias vector of the output gate.

[0044] Embodiment 1

[0045] According to an embodiment of the present application, a method for managing base station alarm and early warning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0046] Figure 1 is a flowchart of a method for managing base station alarm and early warning provided according to an embodiment of the present application. As Figure 1 shown, the method includes the following steps:

[0047] Step S102, obtaining the first operation state sequences of multiple first base stations in a target area during a first time period.

[0048] Step S104, for each first base station, analyzing the first operation state sequence corresponding to the first base station by using a pre-trained base station alarm type prediction model to obtain the first predicted base station alarm type of the first base station at the first moment.

[0049] Among them, the above base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network using multiple sets of training sample data. Therefore, the base station alarm type prediction model integrates the advantages of the self-attention network and the long short-term memory network. In addition, the above first moment is a future moment after the first time period corresponding to the acquisition of the first operating state sequence.

[0050] Step S106: Determine the troubleshooting priority for each type of first predicted base station alarm type based on the first predicted base station alarm types of multiple first base stations at the first moment, and manage each first base station in the target area according to the troubleshooting priority.

[0051] Based on the solution defined in the above steps S102 to S106, it can be known that in the embodiment of the present application, a deep learning model based on a self-attention network and a long short-term memory network is used to analyze the first operating state sequences of multiple base stations in the target area during the first time period, and the first predicted base station alarm types of each first base station at the first moment are obtained, realizing early warning of future alarm types of base stations to actively discover potential network faults and health problems; then, determine the troubleshooting priority for each type of first predicted base station alarm type based on the first predicted base station alarm types of multiple first base stations at the first moment, and manage each first base station in the target area according to the troubleshooting priority, changing the traditional passive response base station operation and maintenance method, achieving the purpose of active base station operation and maintenance management, effectively improving the efficiency of base station maintenance, reducing the network interruption risk caused by lagging fault response, and ensuring the stability and service quality of the communication network.

[0052] The following describes each step of the base station alarm early warning management method in combination with a specific implementation process.

[0053] As an optional implementation manner, in the technical solution provided in the above step S102, the system may first determine multiple first base stations in the target area. Among them, the target area can be defined according to geographical division, such as a city, district or county, etc., or a specific area according to the network architecture or user distribution; then, for each first base station, determine the corresponding first operating state dataset according to the following steps, including:

[0054] Step 1: Obtain the initial operation status sequence of the first base station within the first time period. Among them, the operation status data corresponding to multiple monitoring dimensions in the initial operation status sequence, and multiple monitoring dimensions include but are not limited to: Operations and Maintenance Center (OMC) dimension, performance dimension, etc. The operation status data corresponding to the OMC dimension includes but is not limited to: hardware status, software operation status, etc.; while the operation status data corresponding to the performance dimension covers the service processing capacity, resource usage, network quality indicators, etc. of the base station, such as radio utilization rate, handover success rate, CPU occupancy rate, etc.

[0055] Step 2: Perform preprocessing operations on the initial operation status sequence to obtain the corresponding first operation status sequence. Among them, the preprocessing operations include but are not limited to:

[0056] (1) Data cleaning: Eliminate invalid, duplicate or abnormal data to ensure the quality and reliability of the data set;

[0057] (2) Data format conversion: Usually, the predefined operation status data is in PIL type (defined using the Python Pillow library), and this kind of data cannot be directly applied to the training of deep learning models. Therefore, the original data format can be converted into a standard format readable by deep learning algorithms, such as PyTorch or Tensor;

[0058] (3) Normalization processing: Perform normalization processing on the data, map all data to the same value range, eliminate the influence of dimension, and improve the efficiency and effect of model training;

[0059] (4) Standardization processing: By subtracting the mean and dividing by the standard deviation, make the data set conform to the standard normal distribution, and further improve the stability and generalization ability of the model.

[0060] Through the above steps, the first operation status sequences of each first base station in the target area within the first time period can be obtained, and these data sets meet the input requirements of the deep learning model.

[0061] As an optional implementation manner, in the technical solution provided in the above step S104, the training process of the above base station alarm type prediction model may include:

[0062] Step S1, obtain multiple groups of training sample data.

[0063] Among them, each group of training sample data includes: the second operation status sequence of the second base station within the second time period and the second base station alarm type at the second moment after the second time period of the second base station, and the second time period is a historical time period before the first time period, and the second moment is a historical moment before the first moment.

[0064] Step S2, construct a neural network model.

[0065] Among them, the neural network model at least includes: a self-attention network for capturing the correlation and importance between state data of different dimensions, and a long short-term memory network. The long short-term memory network is good at processing time series data, can learn long-term dependencies, and understand the trend of the base station operating state changing over time. Among them, the long short-term memory network includes multiple memory units corresponding to multiple time steps recursively arranged according to the time series, and each memory unit includes: a forgetting gate, an input gate, and an output gate.

[0066] Step S3, use multiple sets of training sample data to iteratively train the neural network model to obtain a base station alarm type prediction model.

[0067] In the technical solution provided in the above step S3, the above iterative training process can be Figure 2 implemented according to the schematic diagram shown below and the following steps, including:

[0068] Step S31, for each set of training sample data, use a sliding window to divide the second operating state sequence of the second base station within the second time period in the training sample data into a second operating state data set corresponding to n equal-length time steps, and use the self-attention network and long short-term memory network in the neural network model to process the second operating state data set at n time steps in sequence to obtain the second predicted base station alarm type of the second base station within the second time period.

[0069] In the technical solution provided in the above step S31, considering that the sliding window has good flexibility in time series prediction and is helpful for capturing local features of the time series and other advantages. Therefore, in the embodiments of the present application, a sliding window is used to read the second operating state sequence in each set of training sample data to divide the second operating state sequence into n short sequences of equal length (i.e., the second operating state data set) according to the time series, and the length of each short sequence is determined by the size of the sliding window. In addition, the sequence segmentation is overlapping. For example, for a sliding window of size m, the first short sequence segmented is the short sequence containing the first m data, and the second short sequence segmented is the short sequence containing the data between the m + 1 and 2m, and so on. And each short sequence contains multiple sets of explanatory variables (i.e., the second operating state sequence) state and a response variable (i.e., the corresponding second base station alarm type) lable in the data unit, as Figure 3 shown. Therefore, the mathematical expression of the sliding window can be written as: sw = [state(t b ,t f ),lable(t f+1 )], where tb With t f Respectively representing the start time and end time of the sliding window over the entire second operating state sequence, the model can then use the operating state information from t b to t f to predict the label information at time t f+1 In addition, the size sw width of the sliding window can be adjusted according to the training needs, i.e., it can be adjusted as a hyperparameter.

[0070] Specifically, if BS = {bs1, bs2, …, bs m} is used to represent the set of m base stations in a certain geographical area (such as a city, district, county, etc.), then the long-term operating state of the base station BS in a certain area can be represented using a sliding window as And the second operating state sequence of a single base station bs ∈ BS can be represented as bs = [sw1, sw2, …, sw T , where T represents the number of time instants corresponding to the sliding window.

[0071] Given that the long short-term memory network is good at processing time series data and can learn the patterns and features of time series data, this gives the long short-term memory network a natural advantage in predicting time series data. For this reason, the embodiments of the present application use the long short-term memory network to analyze the second operating state sequence of the second base station and capture the change trend of its operating state. In addition, the embodiments of the present application also use the self-attention mechanism to evaluate the influence of each past time instant on the future, and then weight the outputs of all time steps to obtain a new prediction as the final prediction result. By combining the local focusing characteristics of the self-attention mechanism and the long-term dependence capture ability of the long short-term memory network, the accuracy of the prediction can be effectively improved, thereby accurately predicting the alarm type that may occur at the future second time instant after the second time period of the second base station.

[0072] Therefore, the embodiments of the present application can use the self-attention network and the long short-term memory network in the neural network model to process the second operating state dataset bs = [sw1, sw2, …, sw T-1 at n time steps in sequence to output the second predicted base station alarm type within the second time period of the second base station, including:

[0073] Traverse each time step and perform the following steps to obtain the memory cell output corresponding to each time step:

[0074] First step: Input the memory cell output corresponding to the previous time step of the current time step and the second operating state dataset within the current time step into the first Sigmoid activation function, and use the obtained output result as the forget gate activation value. Therefore, the expression for the forget gate activation value can be written as:

[0075] f t = σ1(W f ·[l t-1 , s t + b f )

[0076] where σ1(·) represents the first Sigmoid activation function of the first forgetting gate, W f represents the weight matrix of the first Sigmoid activation function, b f represents the bias vector of the first Sigmoid activation function, and W f , b f can continuously update their parameters during the model training process. Additionally, l t-1 represents the output of the memory cell corresponding to the previous time step of the current time step (the label information at time t - 1), and s t represents the second operating state dataset within the current time step.

[0077] Step 2: Multiply the state C of the memory cell corresponding to the previous time step of the current time step by the forgetting gate activation value to update the state C of the memory cell corresponding to the previous historical time step of the current time step t-1 to obtain the memory cell state C′ t-1 , and use the memory cell state C t-1 as the output of the forgetting gate. Therefore, the expression for the output of the forgetting gate can be written as: t ′ -1 C′

[0078] = C t-1 × f t-1 t

[0079] Step 3: Determine the input gate activation value using the forgetting gate activation value. Therefore, the expression for the input gate activation value can be written as:

[0080] i t = 1 - f t

[0081] Step 4: Use the tanh activation function to integrate the second operating state dataset within the current time step and the output of the memory cell corresponding to the previous time step of the current time step to obtain the first candidate state value. Therefore, the expression for the first candidate state value can be written as:

[0082] g t = tanh(W c × [l t-1 , s t + b c

[0083] Among them, tanh(·) represents the tanh activation function, and W c represents the weight matrix of the tanh activation function, and b c represents the bias vector of the tanh activation function, and W c , b c can continuously update their parameters during the model training process.

[0084] Step 5: Multiply the input gate activation value by the first candidate state value to determine the important information to be retained in the first candidate state value, and add the important information to the memory cell state C′ t-1 , to obtain the memory cell state C at the current time step t . Therefore, the expression for the memory cell state corresponding to the current time step can be written as:

[0085] C t = i t × g t + C′ t-1

[0086] Step 6: Process the memory cell state C corresponding to the current time step using the tanh activation function to obtain the second candidate state value. Therefore, the expression for the second candidate state value can be written as: t

[0087] g t ′ t = tanh(C t )

[0088] Step 7: Input the second running state dataset within the current time step and the memory cell output corresponding to the previous historical time step of the current time step into the second Sigmoid activation function to obtain the output gate activation value. Therefore, the expression for the output gate activation value can be written as:

[0089] o t = σ2(W o × [l t-1 , s t + b o

[0090] Among them, σ2(·) represents the second Sigmoid activation function of the output gate, W o represents the weight matrix of the second Sigmoid activation function, and b o represents the bias vector of the second Sigmoid activation function, and W o , b o can continuously update their parameters during the model training process

[0091] Step 8: Multiply the second candidate state value by the output gate activation value to obtain the output of the memory cell corresponding to the current time step. Therefore, the expression for the output of the memory cell corresponding to the current time step can be written as:

[0092] h t = o t × g t ′

[0093] Use the outputs of the memory cells corresponding to the previous n - 1 time steps as key vectors, and use the output of the memory cell corresponding to the last time step as the query vector. Analyze using the self-attention network to obtain the second predicted base station alarm type of the second base station within the second time period.

[0094] That is, use the output h T of the last time step of the long short-term memory network as the query vector Q, and use the outputs of the memory cells corresponding to the previous n - 1 time steps as the key vectors K. Input them into the self-attention network. Then, the output of the self-attention network can be expressed as:

[0095]

[0096] In the formula, Q = W q h T , K = W k h t , V = W v h T , t ∈ [1, T - 1], W q , W k , W v are three learnable projection matrices respectively.

[0097] Step S32: Use the second base station alarm type and the corresponding second predicted base station alarm type in each group of training sample data to construct an objective loss function.

[0098] Therefore, the expression of the above objective loss function can be written as:

[0099]

[0100] Among them, l t represents the true label (i.e., the second base station alarm type) in each group of training sample data, and h t represents the predicted label (i.e., the second predicted base station alarm type) corresponding to each group of training sample data.

[0101] Step S33: Adjust the model parameters of the neural network model according to the objective loss function until the model parameters converge to obtain a trained base station alarm type prediction model.

[0102] Among them, the model parameters not only include the self-attention network parameters (such as the Query, Key, and Value weight matrices), the long short-term memory network parameters (such as the input gate, forget gate, output gate weight matrices, bias vectors, etc.), but also include hyperparameters such as the learning rate, batch size, optimizer type, activation function, etc. Therefore, in the embodiments of the present application, through the gradient descent method and its variants (such as optimization algorithms like Adam, RMSprop, etc.), the above parameters are continuously adjusted during the training process to minimize the objective loss function, making the prediction of the model for the input data more accurate.

[0103] Further, after obtaining the first predicted base station alarm types of multiple base stations in the target area at the first moment, the embodiments of the present application can also identify whether there is a potential risk of batch base station outages in the target area according to the following method, including:

[0104] Based on the first predicted base station alarm types of each first base station at the first moment, determine the number of first base stations whose first predicted base station alarm types are batch outage alarm types. Among them, the batch outage alarm type set includes but is not limited to: RRU disconnection, input power supply disconnection, RF unit link anomaly, etc.

[0105] According to the number of first base stations and the total number of base stations in the target area, determine the batch base station outage probability of the target area. Among them, the expression of the batch base station outage probability can be written as:

[0106]

[0107] Among them, m bs represents the total number of base stations, and hit bs represents the number of first base stations. This probability p reflects the possibility that multiple base stations in the target area go offline simultaneously or successively at the first moment.

[0108] When the batch base station outage probability is not lower than the preset batch outage probability threshold, it indicates that there is a relatively high batch outage risk in the target area at the first moment. This will not only cause a large number of users to be unable to communicate normally, but may also cause subsequent chain reactions, further deteriorating the network condition. Therefore, in the face of this situation, the system can determine the troubleshooting priority of each type of first predicted base station alarm type based on the first predicted base station alarm types of multiple first base stations at the first moment, and manage each first base station in the target area according to the troubleshooting priority.

[0109] Specifically, in the embodiments of the present application, the system can determine the troubleshooting priorities of each type of first predicted base station alarm type according to the following method, including: determining the number of second base stations with the first predicted base station alarm type being different bulk out-of-service alarm types based on the first predicted base station alarm type of each first base station at the first moment, where the number of second base stations reflects the base station out-of-service risk level of various bulk out-of-service alarm types in the target area at the first moment; then, sorting the first base stations with the first predicted base station alarm type being various bulk out-of-service alarm types according to the number of second base stations, and determining the corresponding troubleshooting priorities of each first base station according to the obtained sorting result. Among them, the bulk out-of-service alarm type with a larger number of second base stations has a greater potential risk to the network. Therefore, the troubleshooting priority of the first base station with the first predicted base station alarm type being the corresponding bulk out-of-service alarm type is also higher.

[0110] Furthermore, the maintenance personnel or the automated system will, according to the troubleshooting priorities, preferentially process the first base stations corresponding to the alarm types with higher troubleshooting priorities, so as to achieve the proactive management and efficient response to potential network risks, greatly improving the efficiency of base station operation and maintenance and the reliability of the network.

[0111] In summary, the embodiments of the present application use deep learning technology, especially combining the long short-term memory network and the self-attention mechanism, to predict the possible alarm types of multiple base stations in the area in the future, so as to achieve proactive warning, significantly improving the intelligent level of network operation and maintenance and the ability of fault prevention. At the same time, it enables the maintenance personnel to take measures in advance to prevent or reduce the impact of faults, changing the traditional passive response operation and maintenance mode.

[0112] Embodiment 2

[0113] According to the embodiments of the present application, there is also provided a base station alarm warning management system for implementing the base station alarm warning management method in Embodiment 1, as Figure 4 shown. The base station alarm warning management system at least includes: an acquisition module 42, a prediction module 44, and a management module 46, where:

[0114] The acquisition module 42 is used to acquire the first operation status sequences of multiple first base stations in the target area within the first time period;

[0115] The prediction module 44 is used to analyze the first operation status sequence corresponding to each first base station by using a pre-trained base station alarm type prediction model for each first base station, and obtain the first predicted base station alarm type of the first base station at the first moment. Among them, the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network using multiple sets of training sample data, and the first moment is the moment after the first time period;

[0116] A management module 46, which is used to determine the troubleshooting priorities of each first base station according to the first predicted base station alarm types of multiple first base stations at the first moment, and manage each first base station in the target area according to the troubleshooting priorities.

[0117] The functions of each module of the base station alarm early warning management system are described below in combination with the specific implementation process.

[0118] Optionally, the acquisition module 42 may first determine multiple first base stations in the target area. Among them, the definition of the target area may be based on geographical division, such as cities, districts or counties, etc., or may be a specific area according to the network architecture or user distribution. Then, for each first base station, the corresponding first operating state dataset is determined according to the following steps, including:

[0119] The first step: Obtain the initial operating state sequence of the first base station within the first time period. Among them, the operating state data corresponding to multiple monitoring dimensions in the initial operating state sequence, and multiple monitoring dimensions include but are not limited to: Operations and Maintenance Center (OMC) dimension, performance dimension, etc. The operating state data corresponding to the OMC dimension includes but is not limited to: hardware state, software operating state, etc.; while the operating state data corresponding to the performance dimension covers the service processing ability, resource usage, network quality indicators, etc. of the base station, such as radio utilization rate, handover success rate, CPU occupancy rate, etc.

[0120] The second step: Perform preprocessing operations on the initial operating state sequence to obtain the corresponding first operating state sequence. Among them, the preprocessing operations include but are not limited to:

[0121] (1) Data cleaning: Eliminate invalid, duplicate or abnormal data to ensure the quality and reliability of the dataset;

[0122] (2) Data format conversion: Usually, the predefined operating state data is in the PIL type (defined using the Python Pillow library), and this data cannot be directly applied to the training of deep learning models. Therefore, the original data format can be converted into a standard format readable by deep learning algorithms, such as PyTorch or Tensor;

[0123] (3) Normalization processing: Perform normalization processing on the data to map all data to the same value range, eliminate the influence of dimension, and improve the efficiency and effect of model training;

[0124] (4) Standardization processing: By subtracting the mean and dividing by the standard deviation, make the dataset conform to the standard normal distribution, and further improve the stability and generalization ability of the model.

[0125] Through the above steps, the first operating state sequences of each first base station in the target area within the first time period can be obtained, and these data sets meet the input requirements of the deep learning model.

[0126] In addition, the base station alarm warning management system also includes a model training module for training a base station alarm type prediction model according to the following steps, including:

[0127] Step S1, obtain multiple sets of training sample data.

[0128] Among them, each set of training sample data includes: the second operating state sequence of the second base station within the second time period and the second base station alarm type at the second moment after the second time period. The second time period is a historical time period before the first time period, and the second moment is a historical moment before the first moment.

[0129] Step S2, construct a neural network model.

[0130] Among them, the neural network model at least includes: a self-attention network for capturing the correlation and importance between state data of different dimensions, and a long short-term memory network. The long short-term memory network is good at processing time series data, can learn long-term dependencies, and understand the trend of the base station operating state changing over time. Among them, the long short-term memory network includes multiple memory units corresponding to multiple time steps recursively arranged according to the time series, and each memory unit includes: a forget gate, an input gate, and an output gate.

[0131] Step S3, use multiple sets of training sample data to iteratively train the neural network model to obtain a base station alarm type prediction model.

[0132] In the technical solution provided in the above step S3, the above iterative training process can be implemented according to the following steps, including:

[0133] Step S31, for each set of training sample data, use a sliding window to divide the second operating state sequence of the second base station within the second time period in the training sample data into a second operating state data set corresponding to n equal time steps, and use the self-attention network and long short-term memory network in the neural network model to process the second operating state data set at n time steps in sequence to obtain the second predicted base station alarm type of the second base station within the second time period.

[0134] In the technical solution provided in the above step S31, considering the advantages that the sliding window has good flexibility in time series data prediction and helps to capture the local features of the time series. Therefore, in the embodiments of the present application, a sliding window is used to read the second operating state sequence in each group of training sample data, and the second operating state sequence is segmented into n short sequences of equal length according to the time series (i.e., the second operating state data set), and the length of each short sequence is determined by the size of the sliding window. In addition, the sequence segmentation is overlapping. For example, for a sliding window of size m, the first short sequence segmented is the short sequence containing the first m data, and the second short sequence segmented is the short sequence containing the data between the (m + 1)th and the 2mth data, and so on. And each short sequence contains multiple sets of explanatory variables (i.e., the second operating state sequence) state and a response variable (i.e., the corresponding second base station alarm type) lable in the data unit. Therefore, the mathematical expression of the sliding window can be written as: sw = [state(t b ,t f ),lable(t f+1 )], where t b and t f respectively represent the start time and the end time of the sliding window on the entire second operating state sequence. Then the model can predict the label information at time t b to t f based on the operating state information from t f+1 . In addition, the size sw width of the sliding window can be adjusted according to the training needs, that is, it can be adjusted as a hyperparameter.

[0135] Specifically, if BS = {bs1, bs2,..., bs m} is used to represent the set of m base stations in a certain geographical area (such as a city, district, county, etc.), then the long-term operating state of the base stations BS in a certain area can be represented by a sliding window as And the second operating state sequence of a single base station bs ∈ BS can be represented as bs = [sw1, sw2,..., sw T , where T represents the number of time instances corresponding to the sliding window.

[0136] Since the long short-term memory network is good at processing time series data and can learn the patterns and features of time series data, this makes the long short-term memory network have a natural advantage in predicting time series data. For this reason, the embodiments of the present application use the long short-term memory network to analyze the second operating state sequence of the second base station and capture the change trend of its operating state. In addition, the embodiments of the present application also use the self-attention mechanism to evaluate the influence of each past moment on the future, and then weight the outputs of all time steps to obtain a new prediction as the final prediction result. By combining the local focusing characteristics of the self-attention mechanism and the long-term dependence capture ability of the long short-term memory network, the accuracy of the prediction can be effectively improved, so as to accurately predict the alarm type that may occur at the future second moment after the second time period of the second base station.

[0137] Therefore, the embodiments of the present application can use the self-attention network and the long short-term memory network in the neural network model to process the second operating state data set bs = [sw1, sw2,..., sw T-1 in the following steps in sequence to output the second predicted base station alarm type within the second time period of the second base station, including:

[0138] Traverse each time step and perform the following steps to obtain the memory cell output corresponding to each time step:

[0139] The first step: Input the memory cell output corresponding to the previous time step of the current time step and the second operating state data set within the current time step into the first Sigmoid activation function, and use the obtained output result as the forgetting gate activation value. Therefore, the expression of the forgetting gate activation value can be written as:

[0140] f t =σ1(W f ·[l t-1 ,s t +b f )

[0141] where σ1(·) represents the first Sigmoid activation function of the first forgetting gate, W f represents the weight matrix of the first Sigmoid activation function, b f represents the bias vector of the first Sigmoid activation function, and W f ,b f can continuously update their parameters during the model training process. In addition, l t-1 represents the memory cell output corresponding to the previous time step of the current time step (the label information at time t-1), and s t represents the second operating state data set within the current time step.

[0142] Step 2: Multiply the memory cell state C corresponding to the previous time step of the current time step by the forget gate activation value to update the memory cell state C corresponding to the previous historical time step of the current time step, obtaining the memory cell state C′, and use the memory cell state C′ as the output of the forget gate. Therefore, the expression for the output of the forget gate can be written as: t-1 C′ = C × f t-1 t-1 t ′ -1 Step 3: Determine the input gate activation value using the forget gate activation value. Therefore, the expression for the input gate activation value can be written as:

[0143] C′ t-1 = C t-1 × f t

[0144] i = 1 - f

[0145] t t

[0146] Step 4: Use the tanh activation function to integrate the second running state dataset within the current time step and the memory cell output corresponding to the previous time step of the current time step to obtain the first candidate state value. Therefore, the expression for the first candidate state value can be written as:

[0147] g t = tanh(W c × [l t-1 , s t + b c

[0148] where tanh(·) represents the tanh activation function, W c represents the weight matrix of the tanh activation function, b c represents the bias vector of the tanh activation function, and W c , b c can continuously update their parameters during the model training process.

[0149] Step 5: Multiply the input gate activation value by the first candidate state value to determine the important information to be retained in the first candidate state value, and add the important information to the memory cell state C′ to obtain the memory cell state C of the current time step. Therefore, the expression for the memory cell state corresponding to the current time step can be written as: t-1 C = i × g + C′ t

[0150] C t = i t × g t + C′ t-1 ​​​​​

[0151] Step 6: Process the memory cell state C corresponding to the current time step using the tanh activation function to obtain the second candidate state value. Therefore, the expression for the second candidate state value can be written as: t Perform processing to obtain the second candidate state value. Thus, the expression for the second candidate state value can be written as:

[0152] g t ′ = tanh(C t )

[0153] Step 7: Input the second operating state data set within the current time step and the memory cell output corresponding to the previous historical time step of the current time step into the second Sigmoid activation function to obtain the output gate activation value. Therefore, the expression for the output gate activation value can be written as:

[0154] o t = σ2(W o × [l t-1 , s t + b o

[0155] where σ2(·) represents the second Sigmoid activation function of the output gate, W o represents the weight matrix of the second Sigmoid activation function, b o represents the bias vector of the second Sigmoid activation function, and W o , b o can continuously update their parameters during the model training process

[0156] Step 8: Multiply the second candidate state value by the output gate activation value to obtain the memory cell output corresponding to the current time step. Therefore, the expression for the memory cell output corresponding to the current time step can be written as:

[0157] h t = o t × g t ′

[0158] Use the memory cell outputs corresponding to the first n - 1 time steps as key vectors, and use the memory cell output corresponding to the last time step as the query vector, and perform analysis using the self-attention network to obtain the second predicted base station alarm type of the second base station within the second time period.

[0159] That is to say, use the output h T of the last time step of the long short-term memory network as the query vector Q, and use the memory cell outputs corresponding to the first n - 1 time steps as the key vectors K, and input them into the self-attention network. Then, the output of the self-attention network can be expressed as:

[0160]

[0161] where Q = W q h T , K = W k h t , V = W v h T , t ∈ [1, T - 1], W q , W k , W v are three learnable projection matrices respectively, and d represents the dimension of the vector.

[0162] Step S32: Construct a target loss function by using the second base station alarm type and the corresponding second predicted base station alarm type in each group of training sample data.

[0163] Therefore, the expression of the above target loss function can be written as:

[0164]

[0165] where l t represents the true label (i.e., the second base station alarm type) in each group of training sample data, and h t represents the predicted label (i.e., the second predicted base station alarm type) corresponding to each group of training sample data.

[0166] Step S33: Adjust the model parameters of the neural network model according to the target loss function until the model parameters converge, and obtain the trained base station alarm type prediction model.

[0167] Among them, the model parameters include not only the self-attention network parameters (such as Query, Key, Value weight matrices), long short-term memory network parameters (such as input gate, forget gate, output gate weight matrices, bias vectors, etc.), but also hyperparameters such as learning rate, batch size, optimizer type, activation function, etc. Therefore, in the embodiments of the present application, through the gradient descent method and its variants (such as optimization algorithms like Adam, RMSprop, etc.), the above parameters are continuously adjusted during the training process to minimize the target loss function and make the prediction of the model for the input data more accurate.

[0168] Furthermore, after obtaining the first predicted base station alarm types of multiple base stations in the target area at the first moment, the management module 46 can identify whether there is a potential risk of batch base station outages in the target area according to the following method, including:

[0169] Determine the number of first base stations whose first predicted base station alarm type is the batch outage alarm type according to the first predicted base station alarm types of each first base station at the first moment. Among them, the batch outage alarm type set includes but is not limited to: remote control radio frequency unit (RRU) disconnection, input power supply disconnection, radio frequency unit link anomaly, etc.

[0170] Determine the base station batch outage probability of the target area according to the number of first base stations and the total number of base stations of the first base stations in the target area. Among them, the expression of the base station batch outage probability can be written as:

[0171]

[0172] Among them, m bs represents the total number of base stations, and hit bs represents the number of first base stations. This probability p reflects the possibility that multiple base stations in the target area go offline simultaneously or successively at the first moment.

[0173] When the base station batch outage probability is not lower than the preset batch outage probability threshold, it indicates that there is a high batch outage risk in the target area at the first moment. This will not only cause a large number of users to be unable to communicate normally, but may also cause subsequent chain reactions, further deteriorating the network condition. Therefore, in the face of this situation, the management module 46 can determine the troubleshooting priority of each type of first predicted base station alarm type according to the first predicted base station alarm types of multiple first base stations at the first moment, and manage each first base station in the target area according to the troubleshooting priority.

[0174] Specifically, in the embodiment of the present application, the management module 46 can determine the troubleshooting priority of each type of first predicted base station alarm type according to the following method, including: determining the number of second base stations with the first predicted base station alarm type being different batch outage alarm types according to the first predicted base station alarm types of each first base station at the first moment, where the number of second base stations reflects the base station outage risk level of each type of batch outage alarm type in the target area at the first moment; then, sorting the first base stations with the first predicted base station alarm type being various batch outage alarm types according to the number of second base stations, and determining the troubleshooting priority corresponding to each first base station according to the obtained sorting result. Among them, the greater the number of second base stations of a batch outage alarm type, the greater the potential risk of its impact on the network. Therefore, the troubleshooting priority of the first base station with the first predicted base station alarm type being the corresponding batch outage alarm type is also higher.

[0175] Furthermore, maintenance personnel or the automation system will preferentially process the first base stations corresponding to the alarm types with higher troubleshooting priorities according to the troubleshooting priorities, so as to realize the proactive management and efficient response to potential network risks, greatly improving the efficiency of base station operation and maintenance and the reliability of the network.

[0176] It should be noted that each module in the base station alarm warning management system in the embodiments of the present application corresponds one by one to each implementation step of the base station alarm warning management method in Embodiment 1. Since detailed descriptions have been made in Embodiment 1, some details not shown in this embodiment can be referred to Embodiment 1 and will not be elaborated here.

[0177] Embodiment 3

[0178] According to an embodiment of the present application, there is also provided a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the base station alarm warning management method in Embodiment 1.

[0179] According to an embodiment of the present application, there is also provided a non-volatile storage medium, which includes a stored computer program. When the device where the non-volatile storage medium is located runs this computer program, it executes the base station alarm warning management method in Embodiment 1.

[0180] According to an embodiment of the present application, there is also provided a processor, which is used to run a computer program. When the computer program runs, it executes the base station alarm warning management method in Embodiment 1.

[0181] According to an embodiment of the present application, there is also provided an electronic device, which includes: a memory and a processor. Among them, a computer program is stored in the memory, and the processor is configured to execute the base station alarm warning management method in Embodiment 1 through the computer program.

[0182] Specifically, when the computer program runs, it executes the following steps: obtaining the first operation status sequence of each of multiple first base stations in a target area within a first time period; for each first base station, using a pre-trained base station alarm type prediction model to analyze the first operation status sequence corresponding to the first base station, and obtaining the first predicted base station alarm type of the first base station at a first moment, where the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network using multiple sets of training sample data, and the first moment is a moment after the first time period; determining the investigation priority of each first base station according to the first predicted base station alarm type of each of multiple first base stations in the target area at the first moment, and managing multiple first base stations in the target area according to the investigation priority.

[0183] As an optional implementation manner, the above-mentioned electronic device may exist in the form of a mobile terminal, a computer terminal, or a similar computing device. Figure 5 Shows a hardware structure block diagram of an electronic device for implementing the base station alarm warning management method. As Figure 5As shown, the electronic device 50 may include one or more processors 502 (illustrated as 502a, 502b, ……, 502n in the figure) (the processor 502 may include, but is not limited to, processing devices such as a microprocessor MCU or a programmable logic device FPGA), a memory 504 for storing data, and a transmission device 506 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 5 the structure shown is only schematic and does not limit the structure of the above-mentioned electronic device. For example, the electronic device 50 may further include more or fewer components than Figure 5 shown in, or have a different configuration from Figure 5 that shown.

[0184] It should be noted that the above one or more processors 502 and / or other data processing circuits can generally be referred to as "data processing circuits" in this article. The data processing circuit may be embodied in software, hardware, firmware, or any other combination, in whole or in part. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the electronic device 50. As involved in the embodiments of the present application, the data processing circuit is a kind of processor control (such as the selection of a variable resistance terminal path connected to an interface).

[0185] The memory 504 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the base station alarm and early warning management method in the embodiments of the present application. The processor 502 executes various functional applications and data processing by running the software programs and modules stored in the memory 504, that is, implements the vulnerability detection method of the above application program. The memory 504 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 504 may further include a memory remotely set relative to the processor 502, and these remote memories can be connected to the electronic device 50 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and their combinations.

[0186] The transmission device 506 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the electronic device 50. In one example, the transmission device 506 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 506 can be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0187] The display can be, for example, a touch-screen Liquid Crystal Display (LCD), which enables a user to interact with the user interface of the electronic device 50.

[0188] The above-mentioned serial numbers of the embodiments are only for description and do not represent the advantages or disadvantages of the embodiments.

[0189] In the above embodiments of the present application, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0190] In several embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0191] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0192] In addition, the functional units in the various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0193] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs.

[0194] The above are only the preferred embodiments of this application. It should be noted that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. A base station alarm warning management method, characterized in that, Including: Obtain the first operation status sequences of multiple first base stations in a target area during a first time period; For each first base station, use a pre-trained base station alarm type prediction model to analyze the first operation status sequence corresponding to the first base station, and obtain the first predicted base station alarm type of the first base station at a first moment, where the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network using multiple sets of training sample data, and the first moment is a moment after the first time period; According to the first predicted base station alarm types of multiple first base stations in the target area at the first moment, determine the troubleshooting priorities of each of the first base stations, and manage the multiple first base stations in the target area according to the troubleshooting priorities.

2. The method according to claim 1, wherein Obtaining the first operation status sequences of multiple first base stations in a target area during a first time period includes: Determine multiple first base stations in the target area; For each of the first base stations, obtain the initial operation status sequence of the first base station during the first time period, where the initial operation status sequence contains operation status data corresponding to multiple monitoring dimensions, and at least one of the following is included in the multiple monitoring dimensions: operation and maintenance center dimension, performance dimension; perform preprocessing operations on the initial operation status sequence to obtain the corresponding first operation status sequence, where the preprocessing operations include at least one of the following: data cleaning, data format conversion, normalization processing, standardization processing.

3. The method according to claim 1, characterized in that, The training process of the base station alarm type prediction model includes: Obtain the multiple sets of training sample data, where each set of training sample data includes: the second operation status sequence of a second base station during a second time period and the second base station alarm type of the second base station at a second moment after the second time period, and the second time period is a historical time period before the first time period, and the second moment is a historical moment before the first moment; Construct the neural network model, where the neural network model at least includes: a self-attention network, a long short-term memory network; Use the multiple sets of training sample data to iteratively train the neural network model to obtain the base station alarm type prediction model.

4. The method according to claim 3, wherein Using the multiple sets of training sample data to iteratively train the neural network model to obtain the base station alarm type prediction model includes: For each set of training sample data, use a sliding window to divide the second operation status sequence of the second base station in the training sample data during the second time period into n second operation status data sets corresponding to equal time steps, and use the self-attention network and the long short-term memory network in the neural network model to process the second operation status data sets at n time steps in sequence to obtain the second predicted base station alarm type of the second base station during the second time period; Construct a target loss function using the second base station alarm type and the corresponding second predicted base station alarm type in each set of training sample data; Adjust the model parameters of the neural network model according to the target loss function until the model parameters converge, and obtain the trained base station alarm type prediction model.

5. The method according to claim 4, wherein The long short-term memory network includes a plurality of memory units corresponding to a plurality of time steps recursively arranged according to a time series, and each memory unit includes: a forget gate, an input gate, and an output gate. Among them, the self-attention network and the long short-term memory network in the neural network model are used to process the second operating state data set at n time steps in sequence, and the second predicted base station alarm type of the second base station in the second time period is obtained, including: Traverse each time step, and execute the following steps to obtain the memory unit output corresponding to each time step: The first step: Input the memory unit output corresponding to the previous time step of the current time step and the second operating state data set within the current time step into the first Sigmoid activation function, and use the obtained output result as the forget gate activation value; Step 2: Multiply the memory cell state C corresponding to the previous time step of the current time step t-1 by the forget gate activation value to update the memory cell state C corresponding to the previous historical time step of the current time step t-1 , obtaining a memory cell state C' t-1 , and use the memory cell state C' t-1 as the output of the forget gate; The third step: Use the forget gate activation value to determine the input gate activation value; The fourth step: Use the tanh activation function to integrate the second operating state data set within the current time step and the memory unit output corresponding to the previous time step of the current time step to obtain the first candidate state value; Step 5: Multiply the input gate activation value by the first candidate state value to determine the important information to be retained in the first candidate state value, and add the important information to the memory cell state C' t-1 , obtaining the memory cell state C corresponding to the current time step t ; Step 6: Process the memory cell state C corresponding to the current time step by using the tanh activation function t to obtain a second candidate state value; The seventh step: Input the second operating state data set within the current time step and the memory unit output corresponding to the previous historical time step of the current time step into the second Sigmoid activation function to obtain the output gate activation value; The eighth step: Multiply the second candidate state value and the output gate activation value to obtain the memory unit output corresponding to the current time step; Use the memory unit outputs corresponding to the previous n - 1 time steps as key vectors, and use the memory unit output corresponding to the last time step as the query vector, and analyze using the self-attention network to obtain the second predicted base station alarm type of the second base station in the second time period.

6. The method according to claim 1, wherein After obtaining the first predicted base station alarm type of the first base station at the first moment, the method further includes: Determine the number of first base stations whose first predicted base station alarm type is the batch outage alarm type according to the first predicted base station alarm types of each first base station at the first moment, where the batch outage alarm type set includes at least one of the following: remote radio unit (RRU) disconnection, input power supply disconnection, and radio unit link anomaly; Determine the base station batch outage probability of the target area according to the number of first base stations and the total number of base stations in the target area; When the base station batch outage probability is not lower than the preset batch outage probability threshold, determine the troubleshooting priority of each first base station according to the first predicted base station alarm types of multiple first base stations in the target area at the first moment, and manage multiple first base stations in the target area according to the troubleshooting priority.

7. The method according to claim 6, wherein Determine the troubleshooting priority of each first base station according to the first predicted base station alarm types of the multiple first base stations, including: Determine the number of second base stations with the first predicted base station alarm type being different batch outage alarm types according to the first predicted base station alarm type of each of the first base stations at the first moment; Sort the first base stations with the first predicted base station alarm type being various batch outage alarm types according to the number of the second base stations, and determine the troubleshooting priority levels corresponding to each of the first base stations according to the obtained sorting result.

8. A base station alarm and early warning management system, characterized in that, Comprising: An acquisition module, configured to acquire first operation state sequences of multiple first base stations in a target area respectively within a first time period; A prediction module, configured to, for each first base station, analyze the first operation state sequence corresponding to the first base station by using a pre-trained base station alarm type prediction model, so as to obtain a first predicted base station alarm type of the first base station at a first moment, wherein the base station alarm type prediction model is obtained by iteratively training a neural network model based on a self-attention network and a long short-term memory network by using multiple groups of training sample data, and the first moment is a moment after the first time period; A management module, configured to determine the troubleshooting priority levels of each of the first base stations according to the first predicted base station alarm type of each of the multiple first base stations in the target area at the first moment, and manage the multiple first base stations in the target area according to the troubleshooting priority levels.

9. A computer program product, characterized in that, Comprising: A computer program, wherein when the computer program is executed by a processor, the base station alarm early warning management method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that, Comprising: A memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the base station alarm early warning management method according to any one of claims 1 to 7 through the computer program.

Images