Wireless network load prediction method, device, equipment, medium and program product

Abstract

The invention provides a wireless network load prediction method, device and equipment, a medium and a program product, and belongs to the technical field of artificial intelligence, and the method comprises the steps: determining a covariant input vector based on wireless network load data and covariant load data; inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset moment output by the load prediction model; the load prediction model is obtained by training on the basis of historical covariant input vectors and load data labels of the historical covariant input vectors. According to the wireless network load prediction method and device, the equipment, the medium and the program product provided by the invention, the index data related to wireless network load data covariant is considered during wireless network load prediction, so that predicted load data with higher precision is obtained; and the network performance and the resource allocation are better optimized by depending on the predicted load data with higher precision.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a wireless network load prediction method, device, equipment, medium and program product. Background Art

[0002] Wireless network load prediction plays an important role in wireless network communication by predicting wireless network load data (such as data traffic and load) in the wireless network, which can optimize network performance and resource allocation.

[0003] At present, the commonly used wireless network load prediction method is to use historical wireless network load data to train the load prediction model. After the load prediction model is trained, the wireless network load data is used to construct an input vector, and the input vector is input into the trained load prediction model. The load prediction model outputs the predicted load data.

[0004] However, in actual wireless network communications, the current or future wireless network load data is not only related to its own past time series characteristics. If the current or future wireless network load data is only predicted based on historical wireless network load data, the load data prediction accuracy will be low, and it will be difficult to optimize network performance and resource allocation based on the predicted load data. Summary of the invention

[0005] The present invention provides a wireless network load prediction method, device, equipment, medium and program product, which are used to solve the defect of low load data prediction accuracy in the prior art.

[0006] The present invention provides a wireless network load prediction method, comprising: Determining a covariant input vector based on the wireless network load data and the covariant load data; Inputting the covariant input vector into a load forecasting model to obtain forecasted load data at a first preset time output by the load forecasting model; The load forecasting model is trained based on historical covariant input vectors and their load data labels.

[0007] According to a wireless network load prediction method provided by the present invention, the wireless network load data includes downlink physical resource block utilization data, and the covariant load data includes uplink physical resource block utilization data and uplink interference data; the covariant input vector is determined based on the wireless network load data and the covariant load data, including: Determining a first sequence of data based on downlink physical resource block utilization data of a historical preset period; Determining a second sequence of data based on the uplink physical resource block utilization rate data of the historical preset time period; Determining a third sequence of data based on the uplink interference data of the historical preset time period; Determining the covariant multidimensional data matrix based on the first sequence data, the second sequence data, and the third sequence data; Based on the covariant multidimensional data matrix, the covariant input vector is determined.

[0008] According to a wireless network load prediction method provided by the present invention, the determining of the covariant input vector based on the covariant multidimensional data matrix includes: Determining an interoperability parameter vector based on a threshold range of an interoperability parameter affecting a wireless network load; The covariant input vector is determined based on the covariant multidimensional data matrix and the interoperation parameter vector.

[0009] According to a wireless network load prediction method provided by the present invention, the determining of the covariant input vector based on the covariant multidimensional data matrix and the interoperability parameter vector includes: Inputting the covariant multidimensional data matrix into a time domain feature extraction layer to obtain a time domain feature vector output by the time domain feature extraction layer; the time domain feature extraction layer comprises a first long short-term memory network layer and a second long short-term memory network layer connected in sequence; The covariant input vector is determined based on the time domain feature vector and the interoperation parameter vector.

[0010] According to a wireless network load prediction method provided by the present invention, the load prediction model includes a first Kolmogorov-Arnold network layer, a second Kolmogorov-Arnold network layer and a third Kolmogorov-Arnold network layer which are connected in sequence.

[0011] According to a wireless network load prediction method provided by the present invention, after inputting the covariant input vector data into a load prediction model to obtain the predicted load data at a first preset time output by the load prediction model, the method further includes: Determining a new covariant input vector based on the predicted load data at the first preset time and the covariant load data; Inputting the new covariant input vector into the load forecasting model to obtain the forecasted load data at a second preset time output by the load forecasting model; The second preset time is later than the first preset time.

[0012] The present invention also provides a wireless network load prediction device, comprising: An input vector determination module, used to determine a covariant input vector based on the wireless network load data and the covariant load data; A future load prediction module, used for inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset time output by the load prediction model; The load forecasting model is trained based on historical covariant input vectors and their load data labels.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the wireless network load prediction methods described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the wireless network load prediction methods described above.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the wireless network load prediction method described above is implemented.

[0016] The wireless network load prediction method, device, equipment, medium and program product provided by the present invention, when using a model to predict the wireless network load, in addition to the wireless network load data itself, also use covariant load data that has a covariant relationship with the wireless network load data to construct a covariant input vector. When predicting the wireless network load, it is possible to consider indicator data related to the covariance of the wireless network load data, thereby obtaining higher-precision predicted load data, and relying on the higher-precision predicted load data to better optimize network performance and resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the wireless network load prediction method provided by the present invention.

[0019] Figure 2 This is the second flow chart of the wireless network load prediction method provided by the present invention.

[0020] Figure 3 It is a structural schematic diagram of a wireless network load prediction device provided by the present invention.

[0021] Figure 4It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] It should be noted that, in the description of the present invention, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0024] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more.

[0025] Combine the following Figure 1-Figure 4 The present invention describes a wireless network load prediction method, device, equipment, medium and program product.

[0026] Wireless network load prediction is an important technical field that involves predicting data traffic and load in wireless networks in order to optimize network performance and resource allocation. This prediction technology can help network operators better manage networks and optimize network resource allocation. By predicting future loads, network resources can be planned and allocated in advance to avoid resource waste or overload; by optimizing network performance, more stable and fast network services can be provided to improve user experience; at the same time, intelligent management and operation of the network can be realized to improve operational efficiency and reduce operating costs.

[0027] Figure 1 FIG. 1 is one of the flow charts of the wireless network load prediction method provided by the present invention, such as Figure 1 As shown, the wireless network load prediction method includes but is not limited to step 110 and step 120.

[0028] It should be noted that the executor of the wireless network load prediction method provided by the present invention is the corresponding wireless network load prediction device, which can specifically be a server, computer equipment, such as a mobile phone, tablet computer, laptop computer, PDA, vehicle-mounted electronic equipment, wearable device, ultra-mobile personal computer (Ultra-Mobile Personal Computer, UMPC), netbook or personal digital assistant (Personal Digital Assistant, PDA), etc.

[0029] Step 101: Determine a covariant input vector based on wireless network load data and covariant load data.

[0030] The covariant load data is indicator data having a covariant relationship with the wireless network load data.

[0031] In one embodiment, the wireless network load data is downlink physical resource block (PRB) utilization data.

[0032] The future downlink PRB utilization is not only related to its own past time series characteristics, but also to the two covariant indicators of uplink PRB utilization and uplink interference. Because uplink and downlink reflect the relationship between the two channels, uplink interference will affect the utilization of the wireless channel, the wireless resources allocated to users, and thus the load. Uplink PRB utilization and uplink interference are indicators that have a covariant relationship with downlink PRB utilization.

[0033] Therefore, when the wireless network load data is downlink PRB utilization data, the corresponding covariant load data includes uplink physical resource block (PRB) utilization data and uplink interference data.

[0034] Specifically, when predicting future wireless network load data, wireless network load data of a preset period in the past is obtained, and covariant load data of several covariant load indicators having a covariant relationship with the wireless network load data is obtained. The obtained wireless network load data and covariant load data are processed to obtain a covariant input vector.

[0035] Step 102: Input the covariant input vector into a load forecasting model to obtain forecasted load data at a first preset time output by the load forecasting model.

[0036] Among them, the load forecasting model is trained based on the historical covariant input vector and its load data labels.

[0037] The predicted load data is wireless network load data at a first preset moment in the present or future obtained by prediction.

[0038] Specifically, before inputting the covariant input vector into the load forecasting model, it is necessary to pre-train the load forecasting model using multiple historical covariant input vectors and their corresponding load data labels to obtain a trained load forecasting model to ensure that the prediction accuracy of the load forecasting model can meet actual application requirements.

[0039] After the load prediction model is trained, the covariant input vector is input into the load prediction model, and the load prediction model outputs the wireless network load data at the first preset moment, that is, the predicted load data.

[0040] It can be understood that the historical covariant input vector used for training the load prediction model is constructed in the same manner as the aforementioned covariant input vector, except that the wireless network load data and covariant load data used to construct the vector are formed at different times.

[0041] The wireless network load prediction method provided by the present invention, when using a model to predict the wireless network load, in addition to the wireless network load data itself, also uses covariant load data that has a covariant relationship with the wireless network load data to construct a covariant input vector, which can consider indicator data related to the covariance of the wireless network load data when predicting the wireless network load, thereby obtaining more accurate predicted load data, and relying on the more accurate predicted load data to better optimize network performance and resource allocation.

[0042] In one embodiment, the step of training a load forecasting model based on historical covariant input vectors and their load data labels includes: The historical wireless network load data and historical covariant load data of all 4G or 5G wireless base stations in the analysis area are selected to construct the first data set.

[0043] For example, a residential area and its surrounding 500-meter range are selected as the analysis area, and the uplink PRB utilization data, uplink PRB utilization data and uplink interference data of all 4G or 5G wireless base stations in the analysis area are pulled every day (busy hours) in the past two years to construct the first data set.

[0044] Delete the data in the data set that are missing or seriously deviate from the normal range due to statistics, network element disconnection, etc., obtain the second data set after data cleaning and preprocessing, and divide the second data set into a training set and a test set.

[0045] The second data set is segmented based on a sliding window method, and the preset width of the sliding window is T days, that is, every T days is a group of data segmentation.

[0046] The historical wireless network load data and historical covariant load data of day T-1 in each group of data are used to construct a historical covariant input vector as the input of the original load forecasting model after the model structure and model parameters are initialized, and the historical wireless network load data of day T is used as the load data label corresponding to the historical covariant input vector.

[0047] The original load forecasting model is trained using multiple historical covariant input vectors and their corresponding load data labels until the prediction accuracy meets the preset requirements, thus obtaining a load forecasting model that can be applied in practice.

[0048] Based on the above embodiment, as an optional embodiment, the radio network load data includes downlink physical resource block utilization data, and the covariant load data includes uplink physical resource block utilization data and uplink interference data; and determining the covariant input vector based on the radio network load data and the covariant load data includes: Determining a first sequence of data based on downlink physical resource block utilization data of a historical preset period; Determining a second sequence of data based on the uplink physical resource block utilization rate data of the historical preset time period; Determining a third sequence of data based on the uplink interference data of the historical preset time period; Determining the covariant multidimensional data matrix based on the first sequence data, the second sequence data, and the third sequence data; Based on the covariant multidimensional data matrix, the covariant input vector is determined.

[0049] Specifically, as mentioned above, when the wireless network load data is downlink PRB utilization data, the covariant load data includes uplink PRB utilization data and uplink interference data.

[0050] Accordingly, when determining the covariant input vector based on the wireless network load data and the covariant load data, the downlink PRB utilization data, uplink PRB utilization data and uplink interference data within a historical preset time period are first used to correspondingly determine the first sequence data, the second sequence data and the third sequence data.

[0051] Using historical preset periods (as shown above downlink PRB utilization data of the day) to determine the first sequence data ,in, It is the downlink PRB utilization data of the first day in the historical preset period. The first time in the historical preset period Downlink PRB utilization data for the day.

[0052] Determine the second sequence data using the uplink PRB utilization data of the historical preset period ,in, The uplink PRB utilization data for the first day of the historical preset period. The first time in the historical preset period Uplink PRB utilization data for the day.

[0053] Use the uplink interference data of the historical preset period to determine the third sequence data ,in, It is the uplink interference data of the first day in the historical preset period. The first time in the historical preset period The uplink interference data of the day.

[0054] Further, the first sequence data directly time-related to the downlink PRB utilization data , and a second sequence of data indirectly time-related to the downlink PRB utilization data and the third series data constitute Covariant multidimensional data matrix . Covariant multidimensional data matrix The expression is as follows: (1)

[0055] Finally, the covariant multidimensional data matrix Data processing is performed to obtain a covariant input vector, which is used to input the covariant input vector into the load forecasting model to obtain the first The downlink PRB utilization data of the day is also the predicted load data.

[0056] As an optional embodiment, in the step of determining the covariant multidimensional data matrix based on the first sequence data, the second sequence data and the third sequence data Previously, the first sequence data, the second sequence data and the third sequence data were normalized respectively.

[0057] Optionally, an expression for normalizing the first sequence data is as follows: ; (2)

[0058] in, is a certain downlink PRB utilization rate data which is not normalized in the first sequence data; Downlink PRB utilization data Normalized data; is the maximum downlink PRB utilization data in the first sequence data; It is the minimum downlink PRB utilization data in the first sequence data.

[0059] It can be understood that when normalizing the first sequence data, the second sequence data, and the third sequence data respectively, the normalization method used is the same.

[0060] By normalizing the first sequence data, the second sequence data, and the third sequence data respectively and scaling the data to between the maximum and minimum values, the learning process can be accelerated and stabilized to avoid gradient problems.

[0061] In the relevant wireless network load prediction methods, most of them use traffic data as wireless network load data, input the traffic data sequence into the prediction model to obtain predicted traffic data, and use the predicted traffic data to optimize network performance and resource allocation. However, traffic data is not directly related to the load of the wireless network, and the increase and decrease of traffic data or load may also have a strong correlation with other indicators.

[0062] The wireless network load prediction method provided by the present invention utilizes the characteristic that downlink PRB utilization can directly reflect the load situation of a base station or a wireless cell, and constructs a covariant multidimensional data matrix for improving the correlation of wireless network load prediction by using a downlink PRB utilization data sequence that is directly time-related, an uplink PRB utilization data sequence that is indirectly time-related to the downlink PRB utilization data, and an uplink interference data sequence. The method uses multi-feature data as input, fully utilizes the time correlation and covariant relationship between the input data to obtain higher-precision predicted load data, and relies on the higher-precision predicted load data to better optimize network performance and resource allocation.

[0063] Based on the above embodiment, as an optional embodiment, determining the covariant input vector based on the covariant multidimensional data matrix includes: Determining an interoperability parameter vector based on a threshold range of an interoperability parameter affecting a wireless network load; The covariant input vector is determined based on the covariant multidimensional data matrix and the interoperation parameter vector.

[0064] Interoperability parameters are parameter configurations on the network management side that affect the wireless network load. Different settings of interoperability thresholds will lead to load migration.

[0065] Specifically, while constructing a covariant multidimensional data matrix using a sequence of wireless network load data and a sequence of covariant load data, a threshold range of at least one interoperability parameter affecting the wireless network load is divided into several range sets, each range set is encoded, and an interoperability parameter vector is constructed using the encodings of all range sets.

[0066] The covariant multidimensional data matrix and the interoperability parameter vector are further processed to construct the covariant input vector of the input load forecasting model.

[0067] It should be noted that the interoperability parameters include but are not limited to at least one of the interoperability parameters such as the coverage-based inter-frequency A2 RSRP (Reference Signal Receiving Power) trigger threshold, the inter-frequency A1 RSRP trigger threshold, the inter-frequency A4 RSRP trigger threshold and the inter-frequency A5-1 RSRP trigger threshold.

[0068] When the interoperability parameters include two or more types, the threshold range of each interoperability parameter is encoded to obtain a corresponding interoperability parameter sub-vector, and all the interoperability parameter sub-vectors are connected together to obtain an interoperability parameter vector.

[0069] Optionally, a one-hot coding (One-Hot Code) method is used to encode each range set of the threshold range of the interoperability parameter.

[0070] OneHot encoding is an encoding method that converts categorical variables into numerical data. It uses an M-bit state register to encode M states. Each state has its own independent register bit, and at any time, only one of them is valid and set to 1, and the other bits are 0.

[0071] Taking the interoperability parameter as the inter-frequency A2 RSRP trigger threshold as an example, the coverage-based inter-frequency A2 RSRP trigger threshold is the measurement threshold configured on the network management side. When the RSRP measurement value of the serving cell is lower than the interoperability threshold, the A2 measurement report will be reported. The inter-frequency A2 RSRP trigger threshold is closely related to the load of the wireless network base station. The larger the setting of this parameter, the lower the difficulty of triggering the A2 event, that is, the easier it is to start the inter-frequency measurement, and the smaller the load of the wireless cell / base station.

[0072] The threshold range of the inter-frequency A2 RSRP trigger threshold is -156~-31 (unit: milliwatt decibel, dBm). Considering the impact of A2 parameters on wireless network load prediction, the A2 parameters of the analysis area are input into the network in the form of OneHot encoding. Specifically, the inter-frequency A2 RSRP trigger threshold is divided into M range sets, and then the M range sets are encoded into M state codes using OneHot encoding to obtain the interoperability parameter vector .

[0073] Table 1 is a OneHot coding table corresponding to the nine range sets of heterofrequency A2 RSRP trigger thresholds provided by the present invention. As shown in Table 1, the interoperability parameter vector obtained is .

[0074]

[0075] The wireless network load prediction method provided by the present invention obtains a corresponding number of codes by dividing the threshold range of the interoperability parameter into a number of range sets, and connects the several codes to obtain an interoperability parameter vector, and considers the interoperability parameter vector and the covariant multidimensional data matrix simultaneously when constructing a covariant input vector. Therefore, by considering the influence of the interoperability parameter on the wireless network load, the parameter configuration of the wireless base station, the downlink PRB utilization data itself, and the influence of the covariant load data on the load prediction can be more comprehensively considered when predicting the network load, so as to predict higher-precision predicted load data, and rely on the higher-precision predicted load data to better optimize the network performance and resource allocation.

[0076] Based on the above embodiment, as an optional embodiment, determining the covariant input vector based on the covariant multidimensional data matrix and the interoperation parameter vector includes: Inputting the covariant multidimensional data matrix into a time domain feature extraction layer to obtain a time domain feature vector output by the time domain feature extraction layer; the time domain feature extraction layer comprises a first long short-term memory network layer and a second long short-term memory network layer connected in sequence; The covariant input vector is determined based on the time domain feature vector and the interoperation parameter vector.

[0077] The Long Short-Term Memory (LSTM) network is a time-recurrent neural network. The LSTM network layer is composed of LSTM neural units. The LSTM neural unit contains a forget gate, an input gate, and an output gate. It controls the transmission state through the gate state and selectively memorizes important information.

[0078] Specifically, a time domain feature extraction layer is constructed in advance using an LSTM network. The time domain feature extraction layer includes a first LSTM network layer and a second LSTM network layer connected in sequence. Meanwhile, the hidden layer size of the time domain feature extraction layer is set to hidden_size1.

[0079] When determining the covariant input vector based on the covariant multidimensional data matrix and the interoperation parameter vector, The covariant multidimensional data matrix is ​​input into the time domain feature extraction layer, and the hidden state of the last time step is determined as the time domain feature vector , where N is hidden_size1.

[0080] The time domain feature vector and the interoperability parameter vector Perform concat connection to obtain covariant input vector , the covariant input vector The data length is .

[0081] The wireless network load prediction method provided by the present invention can extract the common time domain features of the covariant multidimensional data matrix by using LSTM, and can extract the deep features in the wireless network load data sequence without being restricted by a relatively simple hypothesis space; by using a double-layer LSTM network layer to construct a time domain feature extraction layer, while mining the deep features in the wireless network load data sequence as much as possible and improving the feature extraction capability of the time domain feature extraction layer, it also avoids the problems of overfitting, increased complexity, increased training time, etc. caused by too many stacked LSTM network layers; by using a method of multi-feature input and single-step prediction on the whole, the time series features of the downlink PRB utilization rate and its two covariant indicator data are fused with the interoperability parameters, and the future downlink load is fitted and predicted together, so that higher-precision predicted load data can be obtained, and the network performance and resource allocation can be better optimized by relying on the higher-precision predicted load data.

[0082] Based on the above embodiment, as an optional embodiment, the load forecasting model includes a first Kolmogorov-Arnold network layer, a second Kolmogorov-Arnold network layer and a third Kolmogorov-Arnold network layer which are connected in sequence.

[0083] The core idea of ​​Kolmogorov-Arnold Networks (KAN) is to transform the problem of approximating multivariable functions into the problem of learning a set of univariate functions, and replace the weight parameters with learnable univariate functions. KAN is based on the Kolmogorov-Arnold representation theorem, places activation functions on edges (weights) instead of nodes, and activation functions are learnable rather than fixed. KAN has great advantages in accuracy and interpretability.

[0084] Specifically, when predicting the wireless network load, a load prediction model is constructed based on the KAN network, and the load prediction model includes a first KAN network layer, a second KAN network layer, and a third KAN network layer that are connected in sequence.

[0085] Accordingly, the expression of the multi-layer cascade function of the load forecasting model is as follows: ; (3)

[0086] in, is a multi-layer cascade function of the load forecasting model; is the covariant input vector; is the transfer matrix of the first KAN network layer; is the transfer matrix of the second KAN network layer; is the transfer matrix of the third KAN network layer.

[0087] Furthermore, the expression of the output of each KAN network layer in the load forecasting model is as follows: ; (4)

[0088] in, It is l The output of the KAN network layer, It is l The input of the KAN network layer, It is l The transfer matrix of the KAN network layer, It is l The activation function on each edge of the KAN network layer, that is, the nonlinear transformation.

[0089] by For example, input the covariant input vector of the first KAN network layer The length is , the output length is the number of hidden layers of the first KAN network layer hidden_size2, yes The transfer matrix of , each activation function is learned in back-propagation.

[0090] Furthermore, the activation function of each edge on the KAN network layer is determined based on the sum of the basis function and the spline function; the expression of the activation function of each edge is as follows: ; (5)

[0091] ; (6)

[0092] ; (7)

[0093] in, is the activation function of each edge; is the basis function; is a spline function; is a factor that controls the overall size of the activation function; is the B-spline function; is a trainable coefficient.

[0094] In one embodiment, the loss function of the load forecasting model is set to root mean square error (RMSE) or mean square error (MSE).

[0095] For some prediction methods that directly output load forecast results from traditional LSTM networks or add a fully connected layer at the end of the network to splice and output load forecast results, no matter how they are transformed, LSTM is still based on feedforward neural networks, resulting in the prediction accuracy of these methods being not high enough and there is still much room for improvement.

[0096] Experiments have shown that if the traditional LSTM network is used to predict the wireless network load, the RMSE is 0.0185 and the MAE is 0.0148, while using the load prediction model based on LSTM-KAN provided by the present invention, the RMSE is 0.0163 and the MAE is 0.0139.

[0097] The wireless network load forecasting method provided by the present invention constructs a load forecasting model by utilizing the KAN network, and utilizes the higher accuracy of KAN in solving the task of data fitting and prediction. The internal structure of the model is visualized and the model behavior can be understood in the form of mathematical formulas, thereby having stronger interpretability. With a small number of parameters, it can achieve performance comparable to that of a fully connected layer and even have advantages such as higher parameter efficiency. It can obtain higher-precision predicted load data, and rely on the higher-precision predicted load data to better optimize network performance and resource allocation.

[0098] Based on the above embodiment, as an optional embodiment, after the covariant input vector data is input into the load forecasting model to obtain the predicted load data at the first preset time output by the load forecasting model, the method further includes: Determining a new covariant input vector based on the predicted load data at the first preset time and the covariant load data; Inputting the new covariant input vector into the load forecasting model to obtain the forecasted load data at a second preset time output by the load forecasting model; The second preset time is later than the first preset time.

[0099] Specifically, in one embodiment, after constructing a covariant input vector using the actual wireless network load data and covariant load data of the past T-1 days and predicting the predicted load data for the first preset moment, that is, the Tth day, a new covariant input vector is further predicted using the predicted load data of the Tth day and the actual wireless network load data and covariant load data of the past T-2 days, and then the new covariant input vector is input into the load prediction model to predict the predicted load data for the second preset moment, that is, the T+1th day.

[0100] The wireless network load prediction method provided by the present invention continuously constructs a new input vector using the predicted load data obtained by prediction and inputs it into the load prediction model for prediction, thereby predicting the wireless network load data for a period of time in the future in a sequence prediction manner, thereby being able to formulate an optimization plan for network performance and resource allocation in advance, and perform personalized configuration and operation and maintenance of the network.

[0101] In addition, by forecasting the load for the next few days, the load trend in a certain area can be seen from the forecast data, and personalized adjustments to the network can be achieved through threshold setting. For example, capacity guarantee parameter configuration can be sent to base stations in advance during the time period when the forecast load exceeds the threshold.

[0102] By performing statistical analysis on the errors between the predicted downlink PRB utilization data and the actual load, the traffic model and equipment status in the area can be evaluated. For example, when the deviation between the predicted load and the actual load exceeds the threshold, an early warning is activated to assist in determining whether the network equipment on site has failed or whether there is a sudden gathering of people on site.

[0103] In order to better illustrate the wireless network load prediction method provided by the present invention, an embodiment is provided below to illustrate the complete process of the wireless network load prediction method.

[0104] Figure 2 FIG. 2 is a flow chart of the wireless network load prediction method provided by the present invention. Figure 2 As shown, the downlink PRB utilization data, uplink PRB utilization data and uplink interference data within a preset historical period are obtained, and after preprocessing operations such as data cleaning and normalization, the first sequence data, the second sequence data and the third sequence data are obtained respectively.

[0105] A covariant multidimensional data matrix is ​​constructed using the first sequence data, the second sequence data, and the third sequence data, and the covariant multidimensional data matrix is ​​input into a time series feature extraction layer composed of two LSTM network layers connected in sequence to obtain a time series feature vector output by the time series feature extraction layer.

[0106] At the same time, the threshold ranges of several interoperability parameters are divided and one-hot encoded to obtain the interoperability parameter vector, the time series feature vector and the interoperability parameter vector are concat-connected as a covariant input vector, and then the covariant input vector is input into the load forecasting model composed of three KAN network layers connected in sequence, and finally the predicted load data at the preset time output by the load forecasting model is obtained.

[0107] Figure 3 is a schematic diagram of the structure of the wireless network load prediction device provided by the present invention, such as Figure 3As shown, the wireless network load prediction device includes but is not limited to an input vector determination module 301 and a future load prediction module 302 .

[0108] The input vector determination module 301 is used to determine the covariant input vector based on the wireless network load data and the covariant load data.

[0109] The future load prediction module 302 is used to input the covariant input vector into the load prediction model to obtain the predicted load data at the first preset time output by the load prediction model.

[0110] The load forecasting model is trained based on historical covariant input vectors and their load data labels.

[0111] It should be noted that the wireless network load prediction device provided by the present invention can execute the wireless network load prediction method described in any of the above embodiments during specific operation, which will not be described in detail in this embodiment.

[0112] The wireless network load prediction device provided by the present invention, when using a model to predict the wireless network load, in addition to the wireless network load data itself, also uses covariant load data that has a covariant relationship with the wireless network load data to construct a covariant input vector, which can consider indicator data related to the covariance of the wireless network load data when predicting the wireless network load, thereby obtaining more accurate predicted load data, and relying on the more accurate predicted load data to better optimize network performance and resource allocation.

[0113] As an optional embodiment, the input vector determination module also includes a first sequence determination module, a second sequence determination module, a third sequence determination module, a covariant multi-dimensional matrix determination module and an input vector determination submodule.

[0114] The radio network load data includes downlink physical resource block utilization data, and the covariant load data includes uplink physical resource block utilization data and uplink interference data.

[0115] The first sequence determination module is used to determine the first sequence data based on downlink physical resource block utilization data in a historical preset time period.

[0116] The second sequence determination module is used to determine the second sequence data based on the uplink physical resource block utilization data of the historical preset time period.

[0117] The third sequence determination module is used to determine the third sequence data based on the uplink interference data of the historical preset time period.

[0118] The covariant multidimensional matrix determination module is used to determine the covariant multidimensional data matrix based on the first sequence data, the second sequence data and the third sequence data.

[0119] The input vector determination submodule is used to determine the covariant input vector based on the covariant multi-dimensional data matrix.

[0120] As an optional embodiment, the input vector determination submodule further includes: an interoperability parameter vector determination module and a vector fusion module.

[0121] The interoperability parameter vector determination module is used to determine the interoperability parameter vector based on a threshold range of the interoperability parameter affecting the wireless network load.

[0122] The vector fusion module is used to determine the covariant input vector based on the covariant multidimensional data matrix and the interoperability parameter vector.

[0123] As an optional embodiment, the vector fusion module further includes: a time domain feature extraction module and a vector connection module.

[0124] The time domain feature extraction module is used to input the covariant multidimensional data matrix into the time domain feature extraction layer to obtain the time domain feature vector output by the time domain feature extraction layer; the time domain feature extraction layer includes a first long short-term memory network layer and a second long short-term memory network layer connected in sequence.

[0125] The vector connection module is used to determine the covariant input vector based on the time domain feature vector and the interoperability parameter vector.

[0126] As an optional embodiment, the wireless network load prediction device further includes a sequence load prediction module.

[0127] The sequence load prediction module is used to determine a new covariant input vector based on the predicted load data at the first preset moment and the covariant load data after inputting the covariant input vector data into the load prediction model to obtain the predicted load data at the first preset moment output by the load prediction model; inputting the new covariant input vector into the load prediction model to obtain the predicted load data at the second preset moment output by the load prediction model; the second preset moment is later than the first preset moment.

[0128] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor (Processor) 410, a communication interface (Communications Interface) 420, a memory (Memory) 430 and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the wireless network load prediction method provided in any of the above embodiments, and the wireless network load prediction method includes but is not limited to the following steps: determining a covariant input vector based on wireless network load data and covariant load data; inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset moment output by the load prediction model; the load prediction model is obtained by training based on historical covariant input vectors and their load data labels.

[0129] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0130] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the wireless network load prediction method provided in any of the above embodiments, and the wireless network load prediction method includes but is not limited to the following steps: determining a covariant input vector based on wireless network load data and covariant load data; inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset moment output by the load prediction model; the load prediction model is trained based on historical covariant input vectors and their load data labels.

[0131] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the wireless network load prediction method provided in any of the above embodiments is implemented. The wireless network load prediction method includes but is not limited to the following steps: determining a covariant input vector based on wireless network load data and covariant load data; inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset moment output by the load prediction model; the load prediction model is trained based on historical covariant input vectors and their load data labels.

[0132] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wireless network load prediction method, characterized in that: include: Determining a covariant input vector based on the wireless network load data and the covariant load data; Inputting the covariant input vector into a load forecasting model to obtain forecasted load data at a first preset time output by the load forecasting model; The load forecasting model is trained based on historical covariant input vectors and their load data labels.

2. The wireless network load prediction method according to claim 1, characterized in that: The wireless network load data includes downlink physical resource block utilization data, and the covariant load data includes uplink physical resource block utilization data and uplink interference data; and determining the covariant input vector based on the wireless network load data and the covariant load data includes: Determining a first sequence of data based on downlink physical resource block utilization data of a historical preset period; Determining a second sequence of data based on the uplink physical resource block utilization rate data of the historical preset time period; Determining a third sequence of data based on the uplink interference data of the historical preset time period; Determining the covariant multidimensional data matrix based on the first sequence data, the second sequence data, and the third sequence data; Based on the covariant multidimensional data matrix, the covariant input vector is determined.

3. The wireless network load prediction method according to claim 2, characterized in that: The step of determining the covariant input vector based on the covariant multidimensional data matrix comprises: Determining an interoperability parameter vector based on a threshold range of an interoperability parameter affecting a wireless network load; The covariant input vector is determined based on the covariant multidimensional data matrix and the interoperation parameter vector.

4. The wireless network load prediction method according to claim 3, characterized in that: The step of determining the covariant input vector based on the covariant multidimensional data matrix and the interoperation parameter vector comprises: Inputting the covariant multidimensional data matrix into a time domain feature extraction layer to obtain a time domain feature vector output by the time domain feature extraction layer; the time domain feature extraction layer comprises a first long short-term memory network layer and a second long short-term memory network layer connected in sequence; The covariant input vector is determined based on the time domain feature vector and the interoperation parameter vector.

5. The wireless network load prediction method according to claim 1, characterized in that: The load forecasting model includes a first Kolmogorov-Arnold network layer, a second Kolmogorov-Arnold network layer and a third Kolmogorov-Arnold network layer which are connected in sequence.

6. The wireless network load prediction method according to claim 1, characterized in that: After inputting the covariant input vector data into the load forecasting model to obtain the forecasted load data at the first preset time output by the load forecasting model, the method further includes: Determining a new covariant input vector based on the predicted load data at the first preset time and the covariant load data; Inputting the new covariant input vector into the load forecasting model to obtain the forecasted load data at a second preset time output by the load forecasting model; The second preset time is later than the first preset time.

7. A wireless network load prediction device, characterized in that: include: An input vector determination module, used to determine a covariant input vector based on the wireless network load data and the covariant load data; A future load prediction module, used for inputting the covariant input vector into a load prediction model to obtain predicted load data at a first preset time output by the load prediction model; The load forecasting model is trained based on historical covariant input vectors and their load data labels.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the wireless network load prediction method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the wireless network load prediction method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the wireless network load prediction method according to any one of claims 1 to 6 is implemented.

Images