Method and apparatus for estimating handover parameter adjustment effect of a wireless cellular network

Abstract

The application discloses a method and device for estimating the adjustment effect of a handover parameter of a wireless cellular network. The application can realize accurate estimation of the KPI value after adjustment of the handover parameter by introducing causal information between the handover parameter and the KPI. In addition, the application can provide a basis for intelligent adjustment of the handover parameter, and improve network operation efficiency.

Description

Technical Field

[0001] This application relates to the field of mobile communication technology, specifically to a method and apparatus for estimating the effect of handover parameter adjustment in wireless cellular networks. Background Technology

[0002] With the development of 5G and next-generation 6G wireless network technologies, wireless cellular networks support an increasing number of services and functions, involving massive connections and ultra-dense coverage. Base stations will face numerous handover-related configuration and optimization parameters. These handover parameters directly affect the network's key performance indicators (KPIs), including the number of connected users, wireless connectivity rate, and wireless packet loss rate. Inappropriate adjustments to handover parameters will lead to performance degradation of some KPIs, negatively impacting user experience. Therefore, to achieve efficient network management and optimization, it is necessary to understand how network performance changes with adjustments to these handover parameters. Accurate estimation of the effects of handover parameter adjustments can provide a basis for intelligent decision-making.

[0003] Current industry practice involves manually adjusting these handover parameters based on domain knowledge. Many solutions require expert engineers to manually analyze data and adjust system parameters to optimize or configure the network. Others require on-site expert personnel to address detected issues promptly. For mobile operators, these manual adjustment-based solutions are extremely inefficient and costly. Despite operators collecting vast amounts of mobile data daily, their full potential remains untapped. To leverage all the information operators have already collected and provide more adaptive and flexible solutions for the network, machine learning-based solutions are currently being applied to predict the effectiveness of handover parameter adjustments, exploring the different types of data collected by operators. These machine learning methods use adjusted handover parameter values ​​as one of the input features and KPI values ​​as the output, training the model by minimizing the factual error.

[0004] Existing handover parameter adjustment schemes for wireless cellular networks have the following shortcomings:

[0005] (1) The current manual adjustment scheme is time-consuming, inefficient and expensive. At the same time, due to its heavy reliance on the subjective experience of expert engineers, the effect of switching parameters is difficult to estimate.

[0006] (2) Machine learning-based methods use switching parameters as one of the input features, but only extract the correlation of the data. In fact, there is a causal relationship between the adjustment of switching parameters and KPIs, and this causal relationship goes beyond the correlation analysis between the two. Therefore, this method is difficult to accurately estimate the effect of the adjustment of switching parameters. Summary of the Invention

[0007] At least one embodiment of this application provides a method and apparatus for estimating the effect of handover parameter adjustment in a wireless cellular network, which solves the problem that it is difficult to accurately estimate the effect of handover parameter adjustment in existing wireless cellular networks.

[0008] To solve the above-mentioned technical problems, this application is implemented as follows:

[0009] In a first aspect, embodiments of this application provide a method for estimating the effect of handover parameter adjustment in a wireless cellular network, including:

[0010] The network configuration information, handover parameters, and corresponding key performance indicators (KPIs) of the wireless cellular network are collected within a time period to obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurs once within the time period.

[0011] Using the first time series X, the second time series Y, and the third time series T, a training dataset is generated. A causal-stable decomposition neural network model is then trained using multiple training datasets from different time periods. This model includes a causal-stable reconstruction module and a causal-stable estimation module. The causal-stable reconstruction module reconstructs the second time series Y into a first part of features and a second part of features based on the first time series X, the second time series Y, and the third time series T. The causal-stable estimation module estimates a first prediction result for the KPI after switching parameter adjustment based on the first part of features, and estimates a second prediction result for the KPI after switching parameter adjustment based on the second part of features and the adjusted switching parameters. Finally, a final prediction result for the KPI after switching parameter adjustment is generated based on the first and second prediction results.

[0012] The effect of adjusting the switching parameters of the wireless cellular network is predicted using the trained causal-stable decomposition neural network model.

[0013] Optionally, the causal-stable reconstruction module is specifically used for:

[0014] Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, wherein the first feature vector, the second feature vector, and the third feature vector are the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the switching parameters are adjusted, respectively.

[0015] Using a first self-attention layer, the correlation information of the first input feature is learned, and the correlation information of the first input feature is input into a feedforward neural network to obtain the first part feature and the second part feature output by the feedforward neural network. Both the first self-attention layer and the feedforward neural network are multi-layer network structures.

[0016] Optionally, the key, query, and value of the first self-attention layer are all the first input features.

[0017] Optionally, random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, including:

[0018] After performing a random masking operation on the first feature vector, the second feature vector, and the third feature vector, they are concatenated with the randomly initialized fourth feature vector and the fifth feature vector, and added to the trainable positional encoding vector to obtain the first input feature, wherein the fourth feature vector and the fifth feature vector correspond to the first part of the feature and the second part of the feature, respectively.

[0019] Optionally, the causal-stable estimation module includes:

[0020] A stable decoder is used to input the first feature vector and the first part of the features into a second self-attention layer, and use the second self-attention layer to learn the first prediction result;

[0021] A causal decoder is used to input a first sub-feature vector and a second part of the features into a third self-attention layer, and learn a sixth feature vector using the third self-attention layer; and to input the sixth feature vector and a seventh feature vector into a fourth self-attention layer, and learn a second prediction result using the fourth self-attention layer, wherein the first sub-feature vector is the feature vector corresponding to the first sub-time series before the switching parameter adjustment, and the first sub-time series is a time series composed of part of the configuration information in the network configuration information; the seventh feature vector is the feature vector corresponding to the third time series T after the switching parameter adjustment.

[0022] Optionally, the key and value of the second self-attention layer are both the first feature vector, and the query of the second self-attention layer is the first part of the features;

[0023] The key and value of the third self-attention layer are both the first sub-feature vector, and the query of the third self-attention layer is the second part of the feature.

[0024] The key and value of the fourth self-attention layer are both the seventh feature vector, and the query of the fourth self-attention layer is the sixth feature vector.

[0025] Optionally, the total loss function during the training process of the causal-stable decomposition neural network model includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module;

[0026] The first loss function includes reconstruction loss, periodic matching loss, and consistency constraint loss;

[0027] The reconstruction loss is characterized by the average absolute error between the eighth feature vector and the second feature vector, wherein the eighth feature vector is obtained by adding the first part of the features and the second part of the features;

[0028] The periodic matching loss is characterized by the sum of the first cosine similarity and the second cosine similarity, where the first cosine similarity is the cosine similarity between the first part features and the second feature vector, and the second cosine similarity is the cosine similarity between the second part features and the second feature vector.

[0029] The consistency constraint loss is characterized by the degree of difference between the first part of the features and the second part of the features;

[0030] The second loss function is characterized by the mean absolute error between the final prediction result and the actual result of the KPI after the switching parameters are adjusted.

[0031] Optionally, the network configuration information is predetermined configuration information that affects the relationship between the switching parameters and KPIs.

[0032] Secondly, embodiments of this application provide an estimation device for the effect of handover parameter adjustment in a wireless cellular network, comprising:

[0033] The acquisition module is used to acquire network configuration information, handover parameters and corresponding key performance indicators (KPIs) of a wireless cellular network within a time period, and obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurred once within the time period.

[0034] A training module is used to generate training data using the first time series X, the second time series Y, and the third time series T, and to train a causal-stable decomposition neural network model using multiple training data from different time periods. The causal-stable decomposition neural network model includes a causal-stable reconstruction module and a causal-stable estimation module. The causal-stable reconstruction module is used to reconstruct the second time series Y into a first part of features and a second part of features based on the first part of features. The causal-stable estimation module is used to estimate a first prediction result of the KPI after switching parameter adjustment based on the first part of features, and to estimate a second prediction result of the KPI after switching parameter adjustment based on the second part of features and the adjusted switching parameters. Finally, a final prediction result of the KPI after switching parameter adjustment is generated based on the first and second prediction results.

[0035] The prediction module is used to predict the effect of the switching parameter adjustment of the wireless cellular network using a trained causal-stable decomposition neural network model.

[0036] Optionally, the causal-stable reconstruction module is specifically used for:

[0037] Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, wherein the first feature vector, the second feature vector, and the third feature vector are the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the switching parameters are adjusted, respectively.

[0038] Using a first self-attention layer, the correlation information of the first input feature is learned, and the correlation information of the first input feature is input into a feedforward neural network to obtain the first part feature and the second part feature output by the feedforward neural network. Both the first self-attention layer and the feedforward neural network are multi-layer network structures.

[0039] Optionally, the key, query, and value of the first self-attention layer are all the first input features.

[0040] Optionally, the causal-stable reconstruction module is further configured to: perform a random masking operation on the first feature vector, the second feature vector, and the third feature vector, concatenate them with the randomly initialized fourth feature vector and the fifth feature vector, and add them with the trainable position encoding vector to obtain the first input feature, wherein the fourth feature vector and the fifth feature vector correspond to the first part of the feature and the second part of the feature, respectively.

[0041] Optionally, the causal-stable estimation module includes:

[0042] A stable decoder is used to input the first feature vector and the first part of the features into a second self-attention layer, and use the second self-attention layer to learn the first prediction result;

[0043] A causal decoder is used to input a first sub-feature vector and a second part of the features into a third self-attention layer, and learn a sixth feature vector using the third self-attention layer; and to input the sixth feature vector and a seventh feature vector into a fourth self-attention layer, and learn a second prediction result using the fourth self-attention layer, wherein the first sub-feature vector is the feature vector corresponding to the first sub-time series before the switching parameter adjustment, and the first sub-time series is a time series composed of part of the configuration information in the network configuration information; the seventh feature vector is the feature vector corresponding to the third time series T after the switching parameter adjustment.

[0044] Optionally, the total loss function during the training process of the causal-stable decomposition neural network model includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module;

[0045] The first loss function includes reconstruction loss, periodic matching loss, and consistency constraint loss;

[0046] The reconstruction loss is characterized by the average absolute error between the eighth feature vector and the second feature vector, wherein the eighth feature vector is obtained by adding the first part of the features and the second part of the features;

[0047] The periodic matching loss is characterized by the sum of the first cosine similarity and the second cosine similarity, where the first cosine similarity is the cosine similarity between the first part features and the second feature vector, and the second cosine similarity is the cosine similarity between the second part features and the second feature vector.

[0048] The consistency constraint loss is characterized by the degree of difference between the first part of the features and the second part of the features.

[0049] The second loss function is characterized by the mean absolute error between the final prediction result and the actual result of the KPI after the switching parameters are adjusted.

[0050] Thirdly, embodiments of this application provide an estimation device for the effect of handover parameter adjustment in a wireless cellular network, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method described in the first aspect.

[0051] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that, when executed by a processor, implements the steps of the method described above.

[0052] Compared with existing technologies, the present application provides a method and apparatus for estimating the effect of handover parameter adjustment in wireless cellular networks. By introducing causal information between handover parameters and KPIs, it can achieve accurate estimation of KPI values ​​after handover parameter adjustment. Furthermore, the present application also provides a basis for intelligent adjustment of handover parameters, improving network operation and maintenance efficiency. Attached Figure Description

[0053] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0054] Figure 1 This is a flowchart illustrating a method for estimating the effect of handover parameter adjustment in a wireless cellular network according to an embodiment of this application.

[0055] Figure 2 A schematic diagram of a device for estimating the effect of handover parameter adjustment in a wireless cellular network according to an embodiment of this application;

[0056] Figure 3 This is another schematic diagram of a device for estimating the effect of switching parameters adjustment in a wireless cellular network according to an embodiment of this application. Detailed Implementation

[0057] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0058] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The terms “and / or” in the specification and claims indicate at least one of the connected objects.

[0059] The technologies described in this document are not limited to NR systems and Long Time Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in various wireless communication systems such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" are often used interchangeably. CDMA systems can implement radio technologies such as CDMA2000 and Universal Terrestrial Radio Access (UTRA). UTRA includes Wideband Code Division Multiple Access (WCDMA) and other CDMA variants. TDMA systems can implement radio technologies such as the Global System for Mobile Communication (GSM). OFDMA systems can implement radio technologies such as Ultra Mobile Broadband (UMB), Evolution-UTRA (E-UTRA), IEEE 802.21 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, and Flash-OFDM. UTRA and E-UTRA are part of the Universal Mobile Telecommunications System (UMTS). LTE and more advanced versions of LTE (such as LTE-A) are newer UMTS versions using E-UTRA. UTRA, E-UTRA, UMTS, LTE, LTE-A, and GSM are described in documents from an organization called the 3rd Generation Partnership Project (3GPP). CDMA2000 and UMB are described in documents from an organization called 3rd Generation Partnership Project 2 (3GPP2).The techniques described herein can be used with the systems and radio technologies mentioned above, as well as with other systems and radio technologies. However, the following description describes NR systems for illustrative purposes, and NR terminology is used in most of the following description, although these techniques can also be applied to applications beyond NR systems.

[0060] The following description provides examples and is not intended to limit the scope, applicability, or configuration set forth in the claims. Changes may be made to the function and arrangement of the elements discussed without departing from the spirit and scope of this disclosure. Various procedures or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with reference to certain examples may be combined in other examples.

[0061] As described in the background section, existing technologies struggle to accurately estimate the effects of handover parameter adjustments, making it difficult to provide a reliable basis for intelligent decision-making regarding handover parameters. To address at least one of these problems, embodiments of this application provide a method for estimating the effects of handover parameter adjustments in wireless cellular networks. This method can accurately estimate the adjustment effects (KPI values) after handover parameter adjustments, thereby providing a reliable basis for intelligent decision-making regarding handover parameters.

[0062] Please refer to Figure 1 The method for estimating the effect of handover parameter adjustment in wireless cellular networks provided in this application includes:

[0063] Step 101: Collect network configuration information, handover parameters, and corresponding KPIs of the wireless cellular network within a time period to obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurred once within the time period.

[0064] Here, in this embodiment of the application, network configuration information, handover parameters, and corresponding KPIs of the wireless cellular network are collected. Based on the collected network configuration information, handover parameters, and corresponding KPIs in chronological order, a time series composed of the network configuration information, a time series composed of the KPIs, and a time series composed of the handover parameters can be obtained. Each time series includes a series of points (also called elements), each point corresponding to the network configuration information, handover parameters, or KPIs collected at a certain moment. For example, the current network configuration information, handover parameters, and corresponding KPIs of the wireless cellular network are collected every 5 minutes. This yields a time series composed of a series of network configuration information collected at different times, a time series composed of a series of KPIs collected at different times, and a time series composed of a series of handover parameters collected at different times. Furthermore, this embodiment of the application may collect the above data periodically or non-periodically; this application does not specifically limit this.

[0065] For ease of analysis, this embodiment uses the time series before and after a single switching parameter adjustment as training data, and trains the causal-stable decomposition neural network model using multiple training data sets. Specifically, this embodiment divides the time series into time periods, with a switching parameter adjustment occurring within each time period. This yields a first time series X composed of network configuration information collected within the time period, a second time series Y composed of KPIs collected within the time period, and a third time series T composed of switching parameters collected within the time period. Each time series within a time period serves as training data. After collecting time series from multiple time periods, multiple training data sets are obtained, each corresponding to the first, second, and third time series within a specific time period. These multiple training data sets can then be used for model training in subsequent steps.

[0066] In this embodiment, the network configuration information is pre-determined configuration information that affects the relationship between the switching parameters and KPIs. Specifically, the network configuration information includes, but is not limited to, at least one of the following: network coverage scenario, antenna height, antenna azimuth angle, etc.

[0067] The handover parameters include, but are not limited to, at least one of the following: base station transmission power, handover event, threshold corresponding to the handover event, etc.

[0068] The KPIs include, but are not limited to, at least one of the following: maximum number of RRC connected users / user connection rate, user dropout rate, etc.

[0069] Considering the causal relationship between switching parameter adjustments and KPI values, this application embodiment will utilize causal inference to construct a switching parameter adjustment effect estimation model (causal-stable decomposition neural network model). From the perspective of causal inference, adjusting the switching parameter value is equivalent to applying an intervention, and the affected KPI value is the result. Therefore, the focus is on the impact of different switching parameter adjustment strategies on KPI values, i.e., the intervention effect. To estimate the intervention effect, this application embodiment collected a large amount of data on switching parameters and KPI values ​​from the existing network. Furthermore, considering that other indicators (confounding factors) may also affect the relationship between switching parameters and KPIs, such as coverage scenarios, these relevant indicator data are measured and collected from real-world cellular networks as covariates, and then the relationship between switching parameters and KPIs is analyzed from this cellular network data. The KPI value to be estimated is denoted as the result Y, the switching parameter is denoted as the intervention T, and other observed performance indicators are denoted as covariates X. The switching parameter is adjusted once within each time period, and the adjustment time is denoted as t0.

[0070] Step 102: Using the first time series X, the second time series Y, and the third time series T, a causal-stable decomposition neural network model is trained.

[0071] Here, the causal-stable decomposition neural network model includes a causal-stable reconstruction module and a causal-stable estimation module. During model training, the first time series X, the second time series Y, and the third time series T are input into the model, and the model converts the above sequences into corresponding feature vectors.

[0072] The causal-stable reconstruction module is used to reconstruct the second time series Y into a first part feature and a second part feature based on the first time series X, the second time series Y and the third time series T. That is, the result feature vector corresponding to the second time series Y is reconstructed into the sum of the first part feature and the second part feature, wherein the first part feature corresponds to the stable part in the result feature vector and the second part feature corresponds to the causal part in the result feature vector.

[0073] The causal-stability estimation module is used to estimate a first prediction result of the KPI after switching parameter adjustment based on the first part of the features, the first prediction result corresponding to the stable part in the final prediction result; and to estimate a second prediction result of the KPI after switching parameter adjustment based on the second part of the features and the adjusted switching parameters, the second prediction result corresponding to the causal part in the final prediction result. Then, the final prediction result of the KPI after switching parameter adjustment is generated according to the first and second prediction results, that is, the first and second prediction results are added together to obtain the final prediction result.

[0074] Step 103: Use the trained causal-stable decomposition neural network model to predict the effect of adjusting the handover parameters of the wireless cellular network.

[0075] Based on the causal-stable decomposition neural network model trained in Step 103, embodiments of the present application can predict the effect of adjusting the handover parameters of the wireless cellular network. When predicting, the first time series, the second time series, and the third time series collected before the adjustment of the handover parameters within the period to be predicted, as well as the third time series before and after the adjustment of the handover parameters, are input into the model, and the prediction result output by the model, that is, the predicted value of the KPI, is obtained.

[0076] Through the above steps, embodiments of the present application reconstruct the result feature vector into two partial features corresponding to the stable part and the causal part respectively. Then, based on the two partial features, the first prediction result corresponding to the stable part and the second prediction result corresponding to the causal part are predicted respectively, and the final prediction result is obtained. Since the causal relationship between the adjustment of the handover parameters and the KPI is considered in the above prediction process, embodiments of the present application can improve the accuracy of the estimated adjustment effect after the adjustment of the handover parameters, and thus can provide a reliable basis for the intelligent decision-making of the handover parameters.

[0077] Next, an example of the model structure and functions that can be adopted in embodiments of the present application will be given. It should be noted that the following is only an example of a model adopted in embodiments of the present application and is not used to limit the present application.

[0078] Here, the causal-stable reconstruction module includes a first self-attention layer and a feed-forward neural network. The causal-stable reconstruction module is specifically used for:

[0079] Perform random masking and position encoding operations on the first feature vector The second feature vector The third feature vector to obtain the first input feature Input. Among them, the first feature vector The second feature vector The third feature vector are respectively the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the adjustment of the handover parameters (t < t₀). Then, use the first self-attention layer to learn the correlation information of the first input feature Input, and input the correlation information of the first input feature Input into the feed-forward neural network to obtain the first partial feature and the second partial feature Both the first self-attention layer and the feedforward neural network are multi-layer network structures. Specifically, the key, query, and value of the first self-attention layer are all the first input features.

[0080] When performing random masking and positional encoding operations on the first, second, and third feature vectors, this embodiment of the application performs random masking on the first, second, and third feature vectors, then concatenates them with randomly initialized fourth and fifth feature vectors CLS1 and CLS2, and adds them to the trainable positional encoding vector PoE to obtain the first input feature. The fourth and fifth feature vectors respectively correspond to the first part of the features. Second part features

[0081] In other words, the above embodiments utilize the predictability of time series data to design a causal-stable reconstruction module to convert the second feature vector... Reconstructed into a stable part With causal part The sum. Input After positional encoding and random masking, a self-attention mechanism is used to learn the correlation information between input vectors, where m, n, l are the dimensions of X, Y, and T, respectively; w is the time step contained in the time series corresponding to the first, second, or third feature vector. Let each represent the vector space corresponding to a feature vector. The random masking operation is performed in... A variable with a proportion of β is randomly selected from the variable dimensions for masking. Furthermore, a trainable positional encoding is used. Here, β is a preset ratio value.

[0082] To encode Y simultaneously causal With Y stable Two randomly initialized vectors, CLS1 and CLS2, are set in the input, and z is obtained using a self-attention mechanism. cls1 With z cls2 , respectively corresponding and

[0083]

[0084]

[0085] CLS = [|CLS1|CLS2]

[0086] Where [|] represents the concatenation operation between input variables, self-attention represents the self-attention layer, and FFN represents the feedforward neural network.

[0087] In this embodiment, the causal-stable estimation module includes a stable decoder and a causal decoder, wherein the stable decoder includes a second self-attention layer, and the causal decoder includes a third self-attention layer and a fourth self-attention layer. Specifically,

[0088] The stable decoder is used to convert the first feature vector and the first part of the features The input is fed into a second self-attention layer, and the first prediction result is learned using the second self-attention layer. Right now:

[0089]

[0090] Here, both the key and value of the second self-attention layer are the first feature vector. The query of the second self-attention layer is the first part of the features.

[0091] The causal decoder is used to convert the first sub-feature vector Second part features The input is fed into the third self-attention layer, and the sixth feature vector Z is learned using the third self-attention layer. causal ; and, the sixth feature vector Z causal and the seventh eigenvector The input is a fourth self-attention layer, and the second prediction result is learned using the fourth self-attention layer, wherein the first sub-feature vector... This is the feature vector corresponding to the first sub-time series before the switching parameters are adjusted. The first sub-time series is a time series composed of a portion of the configuration information in the network configuration information. This portion of the configuration information is pre-selected from the network configuration information, for example, configuration information that has a causal relationship with the switching parameters; the seventh feature vector It is the feature vector corresponding to the third time series T after the switching parameters are adjusted.

[0092] This application also uses a self-attention mechanism to construct a causal decoder. Furthermore, to distinguish the influence of T and X on Y, since the dimensionality of X's features is much higher than T's, if X and T are input into the representation network together, the influence of T will be lost in the high-dimensional features. Therefore, X and T are input into the representation network separately, implemented in two steps:

[0093]

[0094]

[0095] Here, the key and value of the third self-attention layer are both the first sub-feature vector. The query of the third self-attention layer is the second part of the feature. The key and value of the fourth self-attention layer are both the seventh feature vector. The query of the fourth self-attention layer is the sixth feature vector Z. causal .

[0096] Finally, the output of the stable decoder will be stabilized. and the output of the causal decoder The final prediction result is obtained by summing the results.

[0097] The training process of the above model is described below, including the following steps:

[0098] The total loss function in the training process of the causal-stable decomposition neural network model in this application embodiment includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module.

[0099] A. Design the training loss for the causal-stable reconstruction module. To enable the causal-stable reconstruction module (Deconstructor) to... Deconstructed into and The system consists of two parts, where an appropriate training objective is designed to constrain the functionality of the causal-stable reconstruction module. The first loss function of the causal-stable reconstruction module includes reconstruction loss, periodic matching loss, and consistency constraint loss.

[0100] 1) Reconstruction loss. To ensure the causal-stable reconstruction module deconstructs the loss... and Capable of reconstructing use The mean absolute error (MAE) between the true value and the actual value is used as the loss function (reconstruction loss), as follows:

[0101]

[0102] in, and These represent the reconstructed value and the true value of the KPI, respectively. That is, the reconstruction loss uses the eighth feature vector and the second feature vector. The average absolute error between them is used to characterize the eighth feature vector, which is the first part of the feature vector. Second part features The results are obtained by adding them together. Here, we assume that the total duration from the start time of the time period to the switching time t0 is t0, that is, we assume that the start time is 0.

[0103] 2) Periodic matching loss. To make... and and Given the similar periodic variations, a periodic matching loss is designed here. This periodic matching loss is characterized by the sum of a first cosine similarity and a second cosine similarity, where the first cosine similarity is the cosine similarity between the first feature subset and the second feature vector, and the second cosine similarity is the cosine similarity between the second feature subset and the second feature vector. Specifically, cosine similarity is used to measure the periodic dynamic changes of two time series. Generally, a larger cosine similarity indicates similar periodicity or predictable dynamic changes between the two sequences; conversely, a smaller cosine similarity indicates diverse, random, or chaotic dynamic phenomena.

[0104] First, use max-min normalization. and get and Then, calculate them with The cosine similarity is used. To amplify the similarity value, its negative logarithm is taken. A proposed periodic matching loss is as follows:

[0105]

[0106]

[0107]

[0108] The ReLU function is introduced to guarantee the numerator term. Non-negative, τ is the damping coefficient to avoid the numerator term being 0.

[0109] 3) Consistency Constraint Loss. The consistency constraint loss is characterized by the degree of difference between the first and second part features. To avoid excessive differences between the two decomposition variables output by the causal-stable reconstruction module, a consistency constraint loss is designed here. Specifically, a margin parameter α is introduced to ensure the consistency of the two decomposition variables. The proposed consistency constraint loss limits... and The difference is less than the margin, which can be defined as follows:

[0110]

[0111] in, and They represent and The L1 norm is d(a,b)=|ab|.

[0112] In summary, by combining the proposed reconstruction loss, periodic matching loss, and consistency constraint loss, the overall training loss L for the causal-stable reconstruction module is obtained. R :

[0113] L R =L MAE +L PM +L CC

[0114] B. Design the training loss for the causal-stable estimation module. For the causal-stable estimation module, the mean absolute error between the final prediction result and the actual result of the KPI after parameter adjustment is used to characterize the loss; that is, the predicted value is used. Compared with the true value The MAE loss is used as a constraint:

[0115]

[0116] C. Joint Optimization Model. The training losses of the causal-stable reconstruction module and the causal-stable estimation module are summed to obtain the total training loss of the model. During training, the entire causal-stable decomposition network model is optimized based on the total training loss.

[0117] As can be seen from the above, this application addresses the challenges of existing handover parameter adjustment methods for wireless cellular networks by proposing a handover parameter adjustment effect method that combines causal inference and deep learning. By accurately predicting the handover parameter adjustment effect, it provides guidance for achieving intelligent handover parameter adjustment. Specifically:

[0118] (1) Causal analysis is introduced into the estimation of the effect of switching parameter adjustment. A causal-stable decomposition network model is proposed. The decomposition idea is used to distinguish between stable components (insensitive to parameter adjustment) and causal components (sensitive to parameter adjustment) when modeling the relationship between switching parameters and KPIs. Multiple self-attention modules are used to capture the temporal and dynamic patterns of the observed cellular network data.

[0119] (2) A causal-stable reconstruction module based on a self-attention mechanism was designed to characterize the complex causal stability of KPIs caused by COP adjustment. In addition, two new loss functions were proposed, including gradient matching loss and consistency constraint loss, to ensure the training of the causal-stable reconstruction module.

[0120] (3) Secondly, a causal decoder and a stable decoder were designed to accurately estimate the effect of switching parameter adjustment by introducing causal information.

[0121] Compared with the prior art, the embodiments of this application have at least the following advantages:

[0122] 1. The causal stability model proposed in this application can achieve accurate estimation of KPI values ​​after the switching parameters are adjusted by introducing causal information between the switching parameters and KPIs.

[0123] 2. The handover parameter adjustment effect estimation model based on the causal stable decomposition model proposed in this application provides a basis for intelligent adjustment of handover parameters and improves network operation and maintenance efficiency.

[0124] The various methods of the embodiments of this application have been described above. Apparatus for implementing the above methods will now be provided.

[0125] Please refer to Figure 2 This application embodiment also provides an estimation device for the effect of handover parameter adjustment in a wireless cellular network, comprising:

[0126] The acquisition module 201 is used to acquire network configuration information, handover parameters and corresponding key performance indicators (KPIs) of a wireless cellular network within a time period, and obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurred once within the time period.

[0127] Training module 202 is used to generate training data using the first time series X, the second time series Y, and the third time series T, and to train a causal-stable decomposition neural network model using multiple training data from different time periods. The causal-stable decomposition neural network model includes a causal-stable reconstruction module and a causal-stable estimation module. The causal-stable reconstruction module is used to reconstruct the second time series Y into a first part of features and a second part of features based on the first part of features. The causal-stable estimation module is used to estimate a first prediction result of the KPI after switching parameter adjustment based on the first part of features, and to estimate a second prediction result of the KPI after switching parameter adjustment based on the second part of features and the adjusted switching parameters. Finally, a final prediction result of the KPI after switching parameter adjustment is generated based on the first and second prediction results.

[0128] The prediction module 203 is used to predict the effect of the switching parameter adjustment of the wireless cellular network using the trained causal-stable decomposition neural network model.

[0129] Through the above modules, the embodiments of this application can achieve accurate estimation of KPI values ​​after the switching parameters are adjusted.

[0130] Optionally, the causal-stable reconstruction module is specifically used for:

[0131] Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, wherein the first feature vector, the second feature vector, and the third feature vector are the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the switching parameters are adjusted, respectively.

[0132] Using a first self-attention layer, the correlation information of the first input feature is learned, and the correlation information of the first input feature is input into a feedforward neural network to obtain the first part feature and the second part feature output by the feedforward neural network. Both the first self-attention layer and the feedforward neural network are multi-layer network structures.

[0133] Optionally, the key, query, and value of the first self-attention layer are all the first input features.

[0134] Optionally, the causal-stable reconstruction module is further configured to: perform a random masking operation on the first feature vector, the second feature vector, and the third feature vector, concatenate them with the randomly initialized fourth feature vector and the fifth feature vector, and add them with the trainable position encoding vector to obtain the first input feature, wherein the fourth feature vector and the fifth feature vector correspond to the first part of the feature and the second part of the feature, respectively.

[0135] Optionally, the causal-stable estimation module includes:

[0136] A stable decoder is used to input the first feature vector and the first part of the features into a second self-attention layer, and use the second self-attention layer to learn the first prediction result;

[0137] A causal decoder is used to input a first sub-feature vector and a second part of the features into a third self-attention layer, and learn a sixth feature vector using the third self-attention layer; and to input the sixth feature vector and a seventh feature vector into a fourth self-attention layer, and learn a second prediction result using the fourth self-attention layer, wherein the first sub-feature vector is the feature vector corresponding to the first sub-time series before the switching parameter adjustment, and the first sub-time series is a time series composed of part of the configuration information in the network configuration information; the seventh feature vector is the feature vector corresponding to the third time series T after the switching parameter adjustment.

[0138] Optionally, the key and value of the second self-attention layer are both the first feature vector, and the query of the second self-attention layer is the first part of the feature; the key and value of the third self-attention layer are both the first sub-feature vector, and the query of the third self-attention layer is the second part of the feature; the key and value of the fourth self-attention layer are both the seventh feature vector, and the query of the fourth self-attention layer is the sixth feature vector.

[0139] Optionally, the total loss function during the training process of the causal-stable decomposition neural network model includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module;

[0140] The first loss function includes reconstruction loss, periodic matching loss, and consistency constraint loss;

[0141] The reconstruction loss is characterized by the average absolute error between the eighth feature vector and the second feature vector, wherein the eighth feature vector is obtained by adding the first part of the features and the second part of the features;

[0142] The periodic matching loss is characterized by the sum of the first cosine similarity and the second cosine similarity, where the first cosine similarity is the cosine similarity between the first part features and the second feature vector, and the second cosine similarity is the cosine similarity between the second part features and the second feature vector.

[0143] The consistency constraint loss is characterized by the degree of difference between the first part of the features and the second part of the features.

[0144] The second loss function is characterized by the mean absolute error between the final prediction result and the actual result of the KPI after the switching parameters are adjusted.

[0145] Optionally, the network configuration information is predetermined configuration information that affects the relationship between the switching parameters and KPIs.

[0146] It should be noted that the device in this embodiment is a device corresponding to the above method, and the implementation methods in each of the above embodiments are applicable to the embodiments of this device, and can also achieve the same technical effect. The device provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0147] Please refer to Figure 3This application embodiment also provides an estimation device 300 for the effect of handover parameter adjustment of wireless cellular network, including a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 1101. When the computer program is executed by the processor 301, it implements the various processes of the above-described method embodiment for estimating the effect of handover parameter adjustment of wireless cellular network and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0148] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described method embodiment for estimating the effect of handover parameter adjustment in a wireless cellular network, and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0149] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0151] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for estimating the effect of handover parameter adjustment in a wireless cellular network, characterized in that, include: The network configuration information, handover parameters, and corresponding key performance indicators (KPIs) of the wireless cellular network are collected within a time period to obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurs once within the time period. Using the first time series X, the second time series Y, and the third time series T, a training dataset is generated. A causal-stable decomposition neural network model is then trained using multiple training datasets from different time periods. This model includes a causal-stable reconstruction module and a causal-stable estimation module. The causal-stable reconstruction module reconstructs the second time series Y into a first part of features and a second part of features based on the first time series X, the second time series Y, and the third time series T. The causal-stable estimation module estimates a first prediction result for the KPI after switching parameter adjustment based on the first part of features, and estimates a second prediction result for the KPI after switching parameter adjustment based on the second part of features and the adjusted switching parameters. Finally, a final prediction result for the KPI after switching parameter adjustment is generated based on the first and second prediction results. The effect of adjusting the switching parameters of the wireless cellular network is predicted using the trained causal-stable decomposition neural network model.

2. The method according to claim 1, characterized in that, The causal-stable reconstruction module is specifically used for: Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, wherein the first feature vector, the second feature vector, and the third feature vector are the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the switching parameters are adjusted, respectively. Using a first self-attention layer, the correlation information of the first input feature is learned, and the correlation information of the first input feature is input into a feedforward neural network to obtain the first part feature and the second part feature output by the feedforward neural network. Both the first self-attention layer and the feedforward neural network are multi-layer network structures.

3. The method according to claim 2, characterized in that, The key, query, and value of the first self-attention layer are all the first input features.

4. The method according to claim 2, characterized in that, Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, including: After performing a random masking operation on the first feature vector, the second feature vector, and the third feature vector, they are concatenated with the randomly initialized fourth feature vector and the fifth feature vector, and added to the trainable positional encoding vector to obtain the first input feature, wherein the fourth feature vector and the fifth feature vector correspond to the first part of the feature and the second part of the feature, respectively.

5. The method according to claim 2, characterized in that, The causal-stable estimation module includes: A stable decoder is used to input the first feature vector and the first part of the features into a second self-attention layer, and use the second self-attention layer to learn the first prediction result; A causal decoder is used to input a first sub-feature vector and a second part of the features into a third self-attention layer, and learn a sixth feature vector using the third self-attention layer; and to input the sixth feature vector and a seventh feature vector into a fourth self-attention layer, and learn a second prediction result using the fourth self-attention layer, wherein the first sub-feature vector is the feature vector corresponding to the first sub-time series before the switching parameter adjustment, and the first sub-time series is a time series composed of part of the configuration information in the network configuration information; the seventh feature vector is the feature vector corresponding to the third time series T after the switching parameter adjustment.

6. The method according to claim 5, characterized in that, The key and value of the second self-attention layer are both the first feature vector, and the query of the second self-attention layer is the first part of the features; The key and value of the third self-attention layer are both the first sub-feature vector, and the query of the third self-attention layer is the second part of the feature; The key and value of the fourth self-attention layer are both the seventh feature vector, and the query of the fourth self-attention layer is the sixth feature vector.

7. The method according to claim 4, characterized in that, The total loss function during the training process of the causal-stable decomposition neural network model includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module; The first loss function includes reconstruction loss, periodic matching loss, and consistency constraint loss; The reconstruction loss is characterized by the average absolute error between the eighth feature vector and the second feature vector, wherein the eighth feature vector is obtained by adding the first part of the features and the second part of the features; The periodic matching loss is characterized by the sum of the first cosine similarity and the second cosine similarity, where the first cosine similarity is the cosine similarity between the first part features and the second feature vector, and the second cosine similarity is the cosine similarity between the second part features and the second feature vector. The consistency constraint loss is characterized by the degree of difference between the first part of the features and the second part of the features; The second loss function is characterized by the mean absolute error between the final prediction result and the actual result of the KPI after the switching parameters are adjusted.

8. The method according to claim 1, characterized in that, The network configuration information is predetermined configuration information that affects the relationship between the switching parameters and KPIs.

9. A device for estimating the effect of handover parameter adjustment in a wireless cellular network, characterized in that, include: The acquisition module is used to acquire network configuration information, handover parameters and corresponding key performance indicators (KPIs) of a wireless cellular network within a time period, and obtain a first time series X composed of the network configuration information, a second time series Y composed of the KPIs, and a third time series T composed of the handover parameters, wherein an adjustment of the handover parameters occurred once within the time period. A training module is used to generate training data using the first time series X, the second time series Y, and the third time series T, and to train a causal-stable decomposition neural network model using multiple training data from different time periods. The causal-stable decomposition neural network model includes a causal-stable reconstruction module and a causal-stable estimation module. The causal-stable reconstruction module is used to reconstruct the second time series Y into a first part of features and a second part of features based on the first part of features. The causal-stable estimation module is used to estimate a first prediction result of the KPI after switching parameter adjustment based on the first part of features, and to estimate a second prediction result of the KPI after switching parameter adjustment based on the second part of features and the adjusted switching parameters. Finally, a final prediction result of the KPI after switching parameter adjustment is generated based on the first and second prediction results. The prediction module is used to predict the effect of the switching parameter adjustment of the wireless cellular network using a trained causal-stable decomposition neural network model.

10. The apparatus according to claim 9, characterized in that, The causal-stable reconstruction module is specifically used for: Random masking and positional encoding operations are performed on the first feature vector, the second feature vector, and the third feature vector to obtain the first input feature, wherein the first feature vector, the second feature vector, and the third feature vector are the feature vectors corresponding to the first time series X, the second time series Y, and the third time series T before the switching parameters are adjusted, respectively. Using a first self-attention layer, the correlation information of the first input feature is learned, and the correlation information of the first input feature is input into a feedforward neural network to obtain the first part feature and the second part feature output by the feedforward neural network. Both the first self-attention layer and the feedforward neural network are multi-layer network structures.

11. The apparatus according to claim 10, characterized in that, The causal-stable reconstruction module is further configured to: perform a random masking operation on the first feature vector, the second feature vector, and the third feature vector, concatenate them with the randomly initialized fourth feature vector and the fifth feature vector, and add them with the trainable position encoding vector to obtain the first input feature, wherein the fourth feature vector and the fifth feature vector correspond to the first part of the feature and the second part of the feature, respectively.

12. The apparatus according to claim 10, characterized in that, The causal-stable estimation module includes: A stable decoder is used to input the first feature vector and the first part of the features into a second self-attention layer, and use the second self-attention layer to learn the first prediction result; A causal decoder is used to input a first sub-feature vector and a second part of the features into a third self-attention layer, and learn a sixth feature vector using the third self-attention layer; and to input the sixth feature vector and a seventh feature vector into a fourth self-attention layer, and learn a second prediction result using the fourth self-attention layer, wherein the first sub-feature vector is the feature vector corresponding to the first sub-time series before the switching parameter adjustment, and the first sub-time series is a time series composed of part of the configuration information in the network configuration information; the seventh feature vector is the feature vector corresponding to the third time series T after the switching parameter adjustment.

13. The apparatus according to claim 11, characterized in that, The total loss function during the training process of the causal-stable decomposition neural network model includes: a first loss function for training the causal-stable reconstruction module, and a second loss function for training the causal-stable estimation module; The first loss function includes reconstruction loss, periodic matching loss, and consistency constraint loss; The reconstruction loss is characterized by the average absolute error between the eighth feature vector and the second feature vector, wherein the eighth feature vector is obtained by adding the first part of the features and the second part of the features; The periodic matching loss is characterized by the sum of the first cosine similarity and the second cosine similarity, where the first cosine similarity is the cosine similarity between the first part features and the second feature vector, and the second cosine similarity is the cosine similarity between the second part features and the second feature vector. The consistency constraint loss is characterized by the degree of difference between the first part of the features and the second part of the features; The second loss function is characterized by the mean absolute error between the final prediction result and the actual result of the KPI after the switching parameters are adjusted.

14. An estimation device for the effect of handover parameter adjustment in a wireless cellular network, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 8.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 8.

Images