Wireless performance and load prediction method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a wireless performance and load prediction method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring a current time period data matrix of a cell to be predicted; the current time period data matrix comprises a current time period wireless performance data matrix and a current time period wireless load data matrix; obtaining a current time period coding matrix of the to-be-predicted cell according to a preset holiday mask and the current time period data matrix; and inputting the current time period coding matrix and the wireless performance and load short-term characteristics and long-term characteristics of the to-be-predicted cell into a decoder in a wireless performance and load prediction model to obtain a wireless performance and load prediction result of the to-be-predicted cell. By adopting the method, the prediction accuracy of the wireless performance and the load can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and particularly to a method, device, computer device, computer-readable storage medium, and computer program product for predicting wireless performance and load. Background Art

[0002] Wireless performance and load are key indicators for measuring the operating status and carrying capacity of a wireless network, and are of great significance for network planning, optimization, and ensuring user experience.

[0003] In traditional technologies, the method for predicting the wireless performance and load of a wireless network cell generally processes the time series data of the wireless performance and load of the cell based on an RNN (Recurrent Neural Network). However, the time series data of wireless performance and load is both long-term time series data and complex time series data, and the processing ability of the RNN for long-term time series data and complex time series data is limited, resulting in a low prediction accuracy for wireless performance and load. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, computer device, computer-readable storage medium, and computer program product for predicting wireless performance and load that can improve the prediction accuracy of wireless performance and load in view of the above technical problems.

[0005] In a first aspect, the present application provides a method for predicting wireless performance and load, including:

[0006] Obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes the current time period wireless performance data matrix and the current time period wireless load data matrix;

[0007] According to a preset holiday mask and the current time period data matrix, obtain the current time period encoding matrix of the cell to be predicted;

[0008] Input the current time period encoding matrix, as well as the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted, into the decoder in the wireless performance and load prediction model to obtain the prediction result of the wireless performance and load of the cell to be predicted.

[0009] In one of the embodiments, the step of obtaining the current time period encoding matrix of the cell to be predicted according to a preset holiday mask and the current time period data matrix includes:

[0010] Perform positional encoding processing on each data in the current time period data matrix of the to-be-predicted cell to obtain the initial current time period encoding matrix of the to-be-predicted cell;

[0011] Update the initial current time period encoding matrix according to the preset holiday mask to obtain the current time period encoding matrix of the to-be-predicted cell.

[0012] In one embodiment, the preset holiday mask includes holiday masks corresponding to multiple holiday types and time step information;

[0013] The step of updating the initial current time period encoding matrix according to the preset holiday mask to obtain the current time period encoding matrix of the to-be-predicted cell includes:

[0014] Obtain the holiday type and time step information corresponding to each positional encoding in the initial current time period encoding matrix;

[0015] Query the preset holiday mask according to the holiday type and time step information corresponding to each positional encoding to obtain the holiday mask corresponding to each positional encoding;

[0016] Fuse each positional encoding in the initial current time period encoding matrix with the holiday mask corresponding to each positional encoding respectively to obtain the current time period encoding matrix of the to-be-predicted cell.

[0017] In one embodiment, the step of obtaining the current time period data matrix of the to-be-predicted cell includes:

[0018] Obtain the current data of the to-be-predicted cell; the current data includes current wireless performance data and current wireless load data;

[0019] Arrange the current data in chronological order to obtain the arranged current data;

[0020] Slice the arranged current data according to a preset time period to obtain multiple time period data;

[0021] Construct the current time period data matrix of the to-be-predicted cell according to the multiple time period data.

[0022] In one embodiment, before inputting the current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the to-be-predicted cell into the decoder of the wireless performance and load prediction model to obtain the wireless performance and load prediction results of the to-be-predicted cell, it further includes:

[0023] Obtain the historical time period data matrix of the cell to be predicted; the historical time period data matrix includes a historical time period radio performance data matrix and a historical time period radio load data matrix;

[0024] According to the preset holiday mask and the historical time period data matrix, obtain the historical time period encoding matrix of the cell to be predicted;

[0025] Input the historical time period encoding matrix into the encoder in the radio performance and load prediction model to obtain the short-term features and long-term features of the radio performance and load of the cell to be predicted.

[0026] In one embodiment, the obtaining the historical time period encoding matrix of the cell to be predicted according to the preset holiday mask and the historical time period data matrix includes:

[0027] Perform position encoding processing on each data in the historical time period data matrix to obtain the initial historical time period encoding matrix of the cell to be predicted;

[0028] According to the preset holiday mask, perform update processing on the initial historical time period encoding matrix to obtain the historical time period encoding matrix of the cell to be predicted.

[0029] In one embodiment, after inputting the current time period encoding matrix and the short-term features and long-term features of the radio performance and load of the cell to be predicted into the decoder in the radio performance and load prediction model to obtain the radio performance and load prediction result of the cell to be predicted, it further includes:

[0030] When the radio performance and load prediction result is a complex number, confirm that the radio performance and load prediction result is a holiday prediction result;

[0031] When the radio performance and load prediction result is not a complex number, confirm that the radio performance and load prediction result is a non-holiday prediction result.

[0032] In a second aspect, the present application further provides a radio performance and load prediction device, including:

[0033] A matrix acquisition module, configured to acquire a current time period data matrix of a cell to be predicted; the current time period data matrix includes a current time period radio performance data matrix and a current time period radio load data matrix;

[0034] A matrix processing module, configured to obtain the current time period encoding matrix of the cell to be predicted according to a preset holiday mask and the current time period data matrix;

[0035] A result prediction module, configured to input the current time period coding matrix, the short-term and long-term characteristics of the radio performance and load of the cell to be predicted into a decoder in a radio performance and load prediction model, so as to obtain the radio performance and load prediction results of the cell to be predicted.

[0036] In a third aspect, the present application further provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0037] Obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes a current time period radio performance data matrix and a current time period radio load data matrix;

[0038] According to a preset holiday mask and the current time period data matrix, obtain the current time period coding matrix of the cell to be predicted;

[0039] Input the current time period coding matrix, the short-term and long-term characteristics of the radio performance and load of the cell to be predicted into a decoder in a radio performance and load prediction model, so as to obtain the radio performance and load prediction results of the cell to be predicted.

[0040] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0041] Obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes a current time period radio performance data matrix and a current time period radio load data matrix;

[0042] According to a preset holiday mask and the current time period data matrix, obtain the current time period coding matrix of the cell to be predicted;

[0043] Input the current time period coding matrix, the short-term and long-term characteristics of the radio performance and load of the cell to be predicted into a decoder in a radio performance and load prediction model, so as to obtain the radio performance and load prediction results of the cell to be predicted.

[0044] In a fifth aspect, the present application further provides a computer program product, including a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0045] Obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes a current time period radio performance data matrix and a current time period radio load data matrix;

[0046] Obtaining a current time period coding matrix of the cell to be predicted according to a preset holiday mask and the current time period data matrix;

[0047] The current time period coding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted are input into a decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted.

[0048] The above-mentioned wireless performance and load prediction method, device, computer equipment, computer-readable storage medium and computer program product obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes the current time period wireless performance data matrix and the current time period wireless load data matrix; then, according to the preset holiday mask and the current time period data matrix, the current time period coding matrix of the cell to be predicted is obtained; finally, the current time period coding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted are input into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction results of the cell to be predicted. In this way, when predicting the wireless performance and load of the cell to be predicted, the current time period data matrix of the cell to be predicted can be used to deeply explore the deep-level features in the wireless performance and load data, effectively extract the long-term dependence and periodic characteristics of the wireless performance and load data, and significantly improve the processing ability and accuracy of the wireless performance and load prediction model for the complex time series data of the wireless performance and load, thereby improving the prediction accuracy of the wireless performance and load; at the same time, the impact of holiday factors on the wireless performance and load prediction results is comprehensively considered, a preset holiday mask is introduced, and combined with the current time period data matrix, the current time period coding matrix of the cell to be predicted is obtained, so that the wireless performance and load prediction model can better capture the changing rules caused by holiday factors in the complex time series data when processing the current time period coding matrix, enhance the ability of the wireless performance and load prediction model to capture the periodicity and particularity of the time series, and improve the accuracy of the wireless performance and load prediction model in related performance and load prediction tasks; in addition, the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted are also considered, which is conducive to further improving the prediction accuracy of the wireless performance and load. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 It is a schematic flowchart of a wireless performance and load prediction method in an embodiment;

[0051] Figure 2 It is a schematic flowchart of the steps for updating and processing an initial current time period coding matrix in an embodiment;

[0052] Figure 3 It is a schematic flowchart of the steps for determining short-term and long-term characteristics of the wireless performance and load of a cell to be predicted in an embodiment;

[0053] Figure 4 It is a schematic flowchart of a wireless performance and load prediction method in another embodiment;

[0054] Figure 5 It is a schematic flowchart of a wireless performance and load prediction method based on an improved Transformer in an embodiment;

[0055] Figure 6 It is a structural block diagram of a wireless performance and load prediction device in an embodiment;

[0056] Figure 7 It is an internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0057] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0058] The existing prediction technologies for the performance and load of wireless network cells have the following problems: (1) Limited processing ability of traditional RNN: The wireless communication network environment is complex and changeable, and network traffic, user behavior, etc. have the characteristics of dynamic changes. When traditional RNN processes long sequences and multi-modal time series data such as load, performance, and device status, complex structure designs are usually required to integrate information of different modalities, but its processing ability is still limited; (2) Difficulty of linear regression model in capturing non-linear characteristics: The performance and load of wireless network cells are affected by factors such as seasonality and holidays, and there are complex non-linear relationships between these factors. It is difficult for traditional linear regression models to capture these non-linear characteristics, resulting in prediction results deviating from the actual situation; (3) Insufficient ability to handle emergencies: Emergencies (such as natural disasters, accidents, etc.) may have a greater impact on the performance and load of wireless network cells. Existing methods such as traditional recursive structure models tend to focus on time series data close to the prediction point, resulting in inaccurate prediction results when facing emergencies. Based on this, the present application proposes a wireless performance and load prediction method, which can improve the prediction accuracy of wireless performance and load.

[0059] To more clearly illustrate the wireless performance and load prediction method provided by the embodiments of the present application, the following are explanations of some terms:

[0060] Transformer: It is a deep learning model architecture based on the attention mechanism. The core of the Transformer is the attention mechanism, which enables the model to dynamically focus on different parts of the input sequence when processing sequence data. To enable the model to learn information in different subspaces, the Transformer adopts a multi-head attention mechanism. The Transformer itself does not have the ability to capture the position information of elements in the sequence. It encodes the position information as a vector and adds it to the input feature vector to enable the model to perceive the position of elements. The encoder is responsible for encoding the input sequence and extracting features; the decoder then generates a complete output sequence based on the output of the encoder and the partially generated output.

[0061] Wireless performance and load: Key metrics for measuring the operating status and carrying capacity of a wireless network, which are of great significance for network planning, optimization, and ensuring user experience. Wireless performance is used to evaluate the performance of a wireless network in aspects such as data transmission and signal quality. It covers multiple key metrics such as data transmission rate, signal strength, and signal quality, reflecting the quality and efficiency of the network from different dimensions. Wireless load represents the workload carried by the wireless network during operation, reflecting the degree of occupancy of network resources.

[0062] In an exemplary embodiment, as Figure 1 shown, a wireless performance and load prediction method is provided. In this embodiment, an example is given where this method is applied to a server; it can be understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. In this embodiment, the following steps S101 to step S103 are included. Among them:

[0063] Step S101, obtain the current time period data matrix of the cell to be predicted; the current time period data matrix includes the current time period wireless performance data matrix and the current time period wireless load data matrix.

[0064] Among them, a cell refers to a wireless network cell.

[0065] Among them, the current wireless performance data of the cell to be predicted is arranged in chronological order to obtain the arranged current wireless performance data; according to a preset time period (such as a week), the arranged current wireless performance data is segmented to obtain the current wireless performance data of multiple time periods; according to the current wireless performance data of multiple time periods, a current time period wireless performance data matrix of the cell to be predicted is constructed, such as a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix; where N represents the number of days; the number of rows represents the number of weeks, such as the first week, the second week, etc.; the number of columns represents the number of days in a week, such as Monday, Tuesday... Sunday.

[0066] Among them, the current wireless load data of the cell to be predicted is arranged in chronological order to obtain the arranged current wireless load data; according to a preset time period (such as a week), the arranged current wireless load data is segmented to obtain the current wireless load data of multiple time periods; according to the current wireless load data of multiple time periods, a current time period wireless load data matrix of the cell to be predicted is constructed, such as a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix.

[0067] Among them, the current time period data matrix refers to the data matrix obtained by segmenting the arranged current time period wireless performance and load data of the cell to be predicted according to a preset time period, such as a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix.

[0068] Among them, the current time period data matrix can contain both the current time period wireless performance data matrix and the current time period wireless load data matrix at the same time, and then the final output is the wireless performance and load prediction result. The current time period data matrix can only contain the current time period wireless performance data matrix, and then the final prediction result is the wireless performance prediction result. The current time period data matrix can only contain the current time period wireless load data matrix, and then the final prediction result is the wireless load prediction result.

[0069] Exemplarily, the server obtains the arranged current wireless performance and load data of the cell to be predicted, such as the wireless performance and load data of each day in the most recent week, the wireless performance and load data of each hour in the morning of the current day, etc., and segments the arranged current wireless performance and load data of the cell to be predicted according to a preset time period to obtain multiple time period data, and constructs a data matrix based on the multiple time period data as the current time period data matrix of the cell to be predicted.

[0070] Step S102, according to the preset holiday mask and the current time period data matrix, obtain the current time period encoding matrix of the cell to be predicted.

[0071] Among them, the holiday mask is in the plural form; the holiday masks corresponding to holidays (such as Labor Day, Mid-Autumn Festival, etc.) are in the plural, and the holiday mask corresponding to non-holidays is 1.

[0072] Among them, the holiday masks corresponding to different holiday types are different.

[0073] Among them, the current time period encoding matrix of the cell to be predicted refers to the encoding matrix obtained by multiplying the position encoding corresponding to each data in the current time period data matrix by the corresponding holiday mask.

[0074] Exemplarily, the server performs position encoding processing on each data in the current time period data matrix through a position encoding algorithm to obtain the position encoding corresponding to each data in the current time period data matrix; then queries the preset holiday mask to obtain the holiday mask corresponding to each data in the current time period data matrix; finally, multiplies the position encoding corresponding to each data in the current time period data matrix by the corresponding holiday mask to obtain the encoding matrix corresponding to the current time period data matrix, which is used as the current time period encoding matrix of the cell to be predicted.

[0075] Step S103: Input the current time period encoding matrix, as well as the short-term and long-term characteristics of the radio performance and load of the cell to be predicted, into the decoder in the radio performance and load prediction model to obtain the radio performance and load prediction results of the cell to be predicted.

[0076] Among them, the short-term and long-term characteristics of the radio performance and load refer to the short-term characteristics of the radio performance and load, as well as the long-term characteristics of the radio performance and load. The short-term characteristics of the radio performance and load are used to characterize the short-term evolution trend of the radio performance and load; the long-term characteristics of the radio performance and load are used to characterize the long-term evolution trend of the radio performance and load.

[0077] Among them, the radio performance and load prediction model refers to the trained Transformer model, specifically refer to Figure 5 . The radio performance and load prediction model includes an encoder and a decoder; both the encoder and the decoder are composed of a multi-head attention module, a residual connection and a normalization module, and a fully connected network module. It should be noted that both the encoder and the decoder include N layers.

[0078] Among them, the encoder in the radio performance and load prediction model can output the short-term and long-term characteristics of the radio performance and load of the cell to be predicted by processing the historical time period encoding matrix of the cell to be predicted.

[0079] Among them, the wireless performance and load prediction results refer to the wireless performance and load data in a future time period, such as the wireless performance and load data in the next hour, the wireless performance and load data in the next day, the wireless performance and load data in the next three days, the wireless performance and load data in the next five days, etc.

[0080] Exemplarily, the server obtains the historical time period encoding matrix of the cell to be predicted, inputs the historical time period encoding matrix of the cell to be predicted into the encoder in the wireless performance and load prediction model, and obtains the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted; then inputs the current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model, and the decoder performs a series of processes on the current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted, such as multi-head attention processing, residual connection and normalization processing, and fully connected processing, to obtain the wireless performance and load prediction results of the cell to be predicted.

[0081] Further, the server can also combine the wireless performance and load prediction results of the cell to be predicted (such as the wireless performance and load prediction results tomorrow) and the current wireless performance and load data (such as the wireless performance and load data of each of the last three days) to obtain a new current time period data matrix of the cell to be predicted; then, according to the preset holiday mask and the new current time period data matrix, obtain a new current time period encoding matrix of the cell to be predicted; finally, input the new current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model to obtain a new wireless performance and load prediction result of the cell to be predicted, such as the wireless performance and load prediction results the day after tomorrow; and so on, the server can obtain the wireless performance and load prediction results of each day in the next three to five days.

[0082] In the above wireless performance and load prediction method, the current time period data matrix of the cell to be predicted is obtained; the current time period data matrix includes the current time period wireless performance data matrix and the current time period wireless load data matrix; then, according to the preset holiday mask and the current time period data matrix, the current time period encoding matrix of the cell to be predicted is obtained; finally, the current time period encoding matrix, as well as the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted, are input into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction results of the cell to be predicted. In this way, when predicting the wireless performance and load of the cell to be predicted, by using the current time period data matrix of the cell to be predicted, the deep features in the wireless performance and load data can be deeply mined, the long-term dependencies and periodic features of the wireless performance and load data can be effectively extracted, and the processing ability and accuracy of the wireless performance and load prediction model for the complex time series data of wireless performance and load can be significantly improved, thereby improving the prediction accuracy of wireless performance and load; at the same time, considering the influence of holiday factors on the wireless performance and load prediction results, the preset holiday mask is introduced and combined with the current time period data matrix to obtain the current time period encoding matrix of the cell to be predicted, so that when the wireless performance and load prediction model processes the current time period encoding matrix, it can better capture the change rules brought by holiday factors in the complex time series data, enhance the capture ability of the wireless performance and load prediction model for the periodicity and particularity of the time series, and improve the accuracy of the wireless performance and load prediction model in related performance and load prediction tasks; in addition, the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted are also considered, which is beneficial to further improving the prediction accuracy of wireless performance and load.

[0083] In an exemplary embodiment, step S102 above, obtaining the current time period encoding matrix of the cell to be predicted according to the preset holiday mask and the current time period data matrix, specifically includes the following content: performing position encoding processing on each data in the current time period data matrix to obtain the initial current time period encoding matrix of the cell to be predicted; and updating the initial current time period encoding matrix according to the preset holiday mask to obtain the current time period encoding matrix of the cell to be predicted.

[0084] Among them, the initial current time period encoding matrix is constructed based on the position encoding corresponding to each data (i.e., wireless performance and load data) in the current time period data matrix.

[0085] Among them, the current time period encoding matrix refers to the updated initial current time period encoding matrix, which is specifically obtained by updating the initial current time period encoding matrix according to the preset holiday mask.

[0086] Exemplarily, the server performs position encoding processing on each data in the current time period data matrix according to the sine-cosine position encoding method to obtain the position encoding corresponding to each data in the current time period data matrix; then, based on the position encoding corresponding to each data in the current time period data matrix, an initial current time period encoding matrix of the cell to be predicted is constructed; finally, according to the preset holiday mask, each position encoding in the initial current time period encoding matrix is updated to obtain an updated initial current time period encoding matrix, which is used as the current time period encoding matrix of the cell to be predicted.

[0087] In this embodiment, the current time period encoding matrix of the cell to be predicted is obtained according to the preset holiday mask and the current time period data matrix; in this way, when obtaining the current time period encoding matrix, the holiday mask that affects the wireless performance and load prediction results is introduced, and the current time period data matrix in the form of a cycle is used as the processing object, which is beneficial to improving the accuracy of the wireless performance and load prediction results output by the subsequent model, and further improves the prediction accuracy of the wireless performance and load.

[0088] In an exemplary embodiment, as Figure 2 shown, according to the preset holiday mask, the initial current time period encoding matrix is updated to obtain the current time period encoding matrix of the cell to be predicted, which specifically includes the following steps S201 to S203. Among them:

[0089] Step S201, obtain the holiday type and time step information corresponding to each position encoding in the initial current time period encoding matrix.

[0090] Step S202, query the preset holiday mask according to the holiday type and time step information corresponding to each position encoding to obtain the holiday mask corresponding to each position encoding.

[0091] Step S203, respectively fuse each position encoding in the initial current time period encoding matrix with the holiday mask corresponding to each position encoding to obtain the current time period encoding matrix of the cell to be predicted.

[0092] Among them, the preset holiday mask includes holiday masks corresponding to multiple holiday types and time step information. The holiday type refers to Labor Day, Tomb-Sweeping Day, Mid-Autumn Festival, etc. The time step information refers to the number of the time step within each cycle segment, that is, the number of the column of the matrix. For example, assuming that the original data corresponding to the position encoding is in the first column, the time step information corresponding to this position encoding is 1; assuming that the original data corresponding to the position encoding is in the fifth column, the time step information corresponding to this position encoding is 5.

[0093] Among them, under the same type of holiday, the holiday masks corresponding to different time step information are different.

[0094] Among them, the holiday type corresponding to the position encoding refers to the holiday type of the original data corresponding to the position encoding. The time step information corresponding to the position encoding refers to the time step information of the original data corresponding to the position encoding, that is, which column in the data matrix of the current time period the original data corresponding to the position encoding is located in.

[0095] Among them, the holiday mask corresponding to the position encoding refers to the holiday mask corresponding to the holiday type and time step information of the position encoding. For example, assume that the holiday type of the position encoding is A2 and the time step information is M1, and among the preset holiday masks, the holiday mask corresponding to the holiday type A2 and the time step information M1 is Y3, indicating that the holiday mask corresponding to this position encoding is Y3.

[0096] Among them, the fusion processing refers to multiplication. Each element in the current time period encoding matrix of the cell to be predicted refers to the position encoding after being processed by the holiday mask, which is specifically equal to the product of the position encoding and the holiday mask.

[0097] Exemplarily, the server obtains the holiday type and time step information of the original data corresponding to each position encoding in the initial current time period encoding matrix as the holiday type and time step information corresponding to each position encoding in the initial current time period encoding matrix; then, according to the holiday type and time step information corresponding to each position encoding, it queries the preset holiday mask to obtain the holiday mask corresponding to the holiday type and time step information of each position encoding as the holiday mask corresponding to each position encoding; finally, it multiplies each position encoding in the initial current time period encoding matrix by the holiday mask corresponding to each position encoding respectively to obtain the processed initial current time period encoding matrix as the current time period encoding matrix of the cell to be predicted.

[0098] For example, the position encoding after being processed by the holiday mask is calculated through the following formula:

[0099] x (pos,i) =PE (pos,i) ×m (h,t)

[0100] Among them, x (pos,i) is the position encoding after being processed by the holiday mask; PE (pos,i) is the original position encoding (calculated according to the traditional sine-cosine position encoding method); m (h,t)Complex numbers related to different holiday characteristics, with non-holidays being "1", where \(t = 0, 1, \ldots, T\). The holiday type identification variable is \(h\), and different holidays have different masking strategies. Assuming there are a total of \(H\) different holiday types, \(h = 0, 1, \ldots, H\) (\(h = 0\) represents non-holidays). And for each holiday type, there is a corresponding masking vector:

[0101] \(M=\left[m\right.\) h,1 ,m h,2 ,\(\ldots,m\) h,T \( \)

[0102] where \(m\) (h,t) represents the masking value at the \(t\)-th time step under holiday type \(h\) (\(m\) (h,t) is usually 1 in non-holidays, and \(m\) (h,t) can be a complex number related to holiday characteristics). In actual calculation, first, it is necessary to determine the holiday type \(h\) corresponding to the current time step \(t\) according to the time information, and then calculate the position encoding after masking according to the above formula of \(x\) (pps,i) .

[0103] In this embodiment, according to the preset holiday mask, the initial current time period encoding matrix is updated to obtain the current time period encoding matrix of the cell to be predicted, which is beneficial for comprehensively considering holiday factors when predicting wireless performance and load later, and can better capture the variation law brought by holiday factors in complex time series data, enhancing the ability of the wireless performance and load prediction model to capture the periodicity and particularity of time series, and thus improving the accuracy of the wireless performance and load prediction model in related performance and load prediction tasks.

[0104] In an exemplary embodiment, step S101, obtaining the current time period data matrix of the cell to be predicted, specifically includes the following content: obtaining the current data of the cell to be predicted; the current data includes current wireless performance data and current wireless load data; arranging the current data in chronological order to obtain the arranged current data; performing segmentation processing on the arranged current data according to a preset time period to obtain multiple time period data; and constructing the current time period data matrix of the cell to be predicted based on the multiple time period data.

[0105] Among them, the current data of the cell to be predicted can refer to the wireless performance and load data of each day in the most recent month, the wireless performance and load data of each day in the most recent two weeks, the wireless performance and load data of each day in the most recent week, the wireless performance and load data of each hour in the morning of the current day, etc.

[0106] Among them, the preset time period can refer to days, weeks, months.

[0107] Among them, the time period data can refer to daily data, weekly data, monthly data, etc.

[0108] Among them, the current time period data matrix refers to the data matrix obtained by combining multiple time period data. For example, the first row is the data of the first week, the second row is the data of the second week, the third row is the data of the third week, etc.

[0109] Exemplarily, the server obtains the current data of the cell to be predicted, such as the current wireless performance data and the current wireless load data. Then, the current data is arranged in chronological order to obtain the arranged current data, and the arranged current data is segmented according to a preset time period to obtain multiple time period data. For example, the data of the most recent 14 days is segmented into two-week data. Finally, based on the multiple time period data, a data matrix containing the multiple time period data is constructed as the current time period data matrix of the cell to be predicted.

[0110] For example, the format of the current time period data matrix of the cell to be predicted is a week (N / 7)×7 matrix: Assume the original weekly time series is X = [x1, x2, …, x N , and the segmented matrix is M. Then, the element M i,j of the matrix M can be expressed as: M i,j = x (i-1)×7+j ; where, i = 1, 2, …, k (representing the i-th period segment, i.e., the i-th row of the matrix), and j = 1, 2, …, 7 (representing the j-th time step within each period segment, i.e., the j-th column of the matrix).

[0111] For example, the format of the current time period data matrix of the cell to be predicted is a week (N - 7 + 1)×7 matrix: Assume the original weekly time series is X = [x1, x2, …, x N , and the segmented matrix is M. Then, the element M i,j of the matrix M can be expressed as:

[0112]

[0113] where, i = 1, 2, …, N - 7 + 1 (representing the i-th row of the matrix), and j = 1, 2, …, 7 (representing the j-th column of the matrix).

[0114] In this embodiment, the arranged current data of the cell to be predicted is obtained, and the arranged current data is segmented according to a preset time period to obtain a plurality of time period data. Finally, based on the plurality of time period data, the current time period data matrix of the cell to be predicted is constructed. In this way, by using the current time period data matrix of the cell to be predicted, the deep features in the wireless performance and load data can be deeply mined, the long-term dependencies and periodic features of the wireless performance and load data can be effectively extracted, and the processing ability and accuracy of the wireless performance and load prediction model for the complex time series data of the wireless performance and load are significantly improved, thereby improving the prediction accuracy of the wireless performance and load.

[0115] In an exemplary embodiment, as Figure 3 shown, before inputting the current time period encoding matrix and the short-term and long-term features of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted, the step of determining the short-term and long-term features of the wireless performance and load of the cell to be predicted is further included, which specifically includes the following steps S301 to S303. Among them:

[0116] Step S301, obtain the historical time period data matrix of the cell to be predicted; the historical time period data matrix includes the historical time period wireless performance data matrix and the historical time period wireless load data matrix.

[0117] Step S302, obtain the historical time period encoding matrix of the cell to be predicted according to the preset holiday mask and the historical time period data matrix.

[0118] Step S303, input the historical time period encoding matrix into the encoder in the wireless performance and load prediction model to obtain the short-term and long-term features of the wireless performance and load of the cell to be predicted.

[0119] Among them, the historical wireless performance data of the cell to be predicted is arranged in chronological order to obtain the arranged historical wireless performance data; according to a preset time period (such as a week), the arranged historical wireless performance data is segmented to obtain the historical wireless performance data of multiple time periods; based on the historical wireless performance data of multiple time periods, the historical time period wireless performance data matrix of the cell to be predicted is constructed, such as a week (N / 7)×7 matrix or a week (N - 7 + 1)×7 matrix.

[0120] Among them, the historical wireless load data of the cell to be predicted is arranged in chronological order to obtain the arranged historical wireless load data; according to a preset time period (such as a week), the arranged historical wireless load data is segmented to obtain the historical wireless load data of multiple time periods; according to the historical wireless load data of multiple time periods, a historical time period wireless load data matrix of the cell to be predicted is constructed, such as a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix.

[0121] Among them, the historical time period data matrix refers to the data matrix obtained by segmenting the arranged historical time period wireless performance and load data of the cell to be predicted according to a preset time period, such as a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix.

[0122] Among them, the historical time period encoding matrix of the cell to be predicted refers to the encoding matrix obtained by multiplying the position encoding corresponding to each data in the historical time period data matrix by the corresponding holiday mask.

[0123] Exemplarily, the server obtains the historical data of the cell to be predicted; the historical data includes historical wireless performance data and historical wireless load data, such as the wireless performance and load data of each day in the past 49 days; the historical data is arranged in chronological order to obtain the arranged historical data; according to a preset time period, the arranged historical data is segmented to obtain multiple time period data; according to the multiple time period data, a historical time period data matrix of the cell to be predicted is constructed; then, through a position encoding algorithm, each data in the historical time period data matrix is subjected to position encoding processing to obtain the position encoding corresponding to each data in the historical time period data matrix; then, the preset holiday mask is queried to obtain the holiday mask corresponding to each data in the historical time period data matrix; the position encoding corresponding to each data in the historical time period data matrix is multiplied by the corresponding holiday mask to obtain the encoding matrix corresponding to the historical time period data matrix, which is used as the historical time period encoding matrix of the cell to be predicted; finally, the historical time period encoding matrix is input into the encoder in the wireless performance and load prediction model, and through a series of processing of the encoder on the historical time period encoding matrix, such as multi-head attention processing, residual connection and normalization processing, and fully connected processing, the short-term and long-term features of the wireless performance and load of the cell to be predicted are obtained.

[0124] In this embodiment, by obtaining the historical time period data matrix of the cell to be predicted, and based on the preset holiday mask and the historical time period data matrix, the historical time period encoding matrix of the cell to be predicted is obtained. Finally, the historical time period encoding matrix is input into the encoder in the wireless performance and load prediction model to obtain the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted. In this way, by comprehensively considering the historical time period data matrix and the holiday mask, and combining with the encoder in the wireless performance and load prediction model, it is beneficial to improve the accuracy of the output short-term and long-term characteristics of the wireless performance and load, thereby improving the extraction accuracy of the short-term and long-term characteristics of the wireless performance and load.

[0125] In an exemplary embodiment, step S302 above, obtaining the historical time period encoding matrix of the cell to be predicted according to the preset holiday mask and the historical time period data matrix, specifically includes the following content: performing position encoding processing on each data in the historical time period data matrix to obtain the initial historical time period encoding matrix of the cell to be predicted; according to the preset holiday mask, performing update processing on the initial historical time period encoding matrix to obtain the historical time period encoding matrix of the cell to be predicted.

[0126] Among them, the initial historical time period encoding matrix is constructed based on the position encoding corresponding to each data (i.e., wireless performance and load data) in the historical time period data matrix.

[0127] Among them, the historical time period encoding matrix refers to the updated initial historical time period encoding matrix, which is specifically obtained by performing update processing on the initial historical time period encoding matrix based on the preset holiday mask.

[0128] Exemplarily, the server performs position encoding processing on each data in the historical time period data matrix according to the sine-cosine position encoding method to obtain the position encoding corresponding to each data in the historical time period data matrix; then, based on the position encoding corresponding to each data in the historical time period data matrix, the initial historical time period encoding matrix of the cell to be predicted is constructed; finally, according to the preset holiday mask, each position encoding in the initial historical time period encoding matrix is updated to obtain the updated initial historical time period encoding matrix as the historical time period encoding matrix of the cell to be predicted.

[0129] For example, the server obtains the holiday type and time step information corresponding to the original data of each position code in the initial historical time period coding matrix as the holiday type and time step information corresponding to each position code in the initial historical time period coding matrix. Then, according to the holiday type and time step information corresponding to each position code, the server queries the preset holiday mask to obtain the holiday mask corresponding to the holiday type and time step information of each position code as the holiday mask corresponding to each position code. Finally, the server multiplies each position code in the initial historical time period coding matrix by the corresponding holiday mask of each position code to obtain the processed initial historical time period coding matrix as the historical time period coding matrix of the cell to be predicted.

[0130] In this embodiment, the historical time period coding matrix of the cell to be predicted is obtained according to the preset holiday mask and the historical time period data matrix, which facilitates the subsequent input of the historical time period coding matrix into the encoder in the wireless performance and load prediction model to obtain the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted. This is beneficial to comprehensively considering the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted when predicting the wireless performance and load, and further improves the prediction accuracy of the wireless performance and load.

[0131] In an exemplary embodiment, after the above step S103, after inputting the current time period coding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted, the following content is further included: when the wireless performance and load prediction result is complex, it is confirmed that the wireless performance and load prediction result is a holiday prediction result; when the wireless performance and load prediction result is non-complex, it is confirmed that the wireless performance and load prediction result is a non-holiday prediction result.

[0132] Among them, the wireless performance and load prediction result being complex means that the real part in the wireless performance and load prediction result represents the wireless performance and load prediction value, and the imaginary part represents the corresponding holiday type.

[0133] Among them, the wireless performance and load prediction result being non-complex means that the wireless performance and load prediction result only contains the real part, and the real part represents the wireless performance and load prediction value.

[0134] Exemplarily, the server identifies the format of the wireless performance and load prediction result to determine whether the wireless performance and load prediction result is complex. If the wireless performance and load prediction result is complex, it is confirmed that the wireless performance and load prediction result is a holiday prediction result; if the wireless performance and load prediction result is non-complex, it is confirmed that the wireless performance and load prediction result is a non-holiday prediction result.

[0135] In this embodiment, by determining whether the wireless performance and load prediction results are holiday prediction results, it is possible to intuitively see whether the wireless performance and load prediction results are affected by holidays, which is convenient for taking corresponding resource scheduling measures in advance.

[0136] In an exemplary embodiment, as Figure 4 shown, another wireless performance and load prediction method is provided. Taking the application of this method to a server as an example, it includes the following steps S401 to S410.

[0137] Wherein:

[0138] Step S401, obtain the historical time period data matrix of the cell to be predicted; the historical time period data matrix includes the historical time period wireless performance data matrix and the historical time period wireless load data matrix.

[0139] Step S402, perform position encoding processing on each data in the historical time period data matrix to obtain the initial historical time period encoding matrix of the cell to be predicted; update the initial historical time period encoding matrix according to the preset holiday mask to obtain the historical time period encoding matrix of the cell to be predicted.

[0140] Step S403, input the historical time period encoding matrix into the encoder in the wireless performance and load prediction model to obtain the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted.

[0141] Step S404, obtain the current data of the cell to be predicted; the current data includes the current wireless performance data and the current wireless load data; arrange the current data in chronological order to obtain the arranged current data.

[0142] Step S405, perform segmentation processing on the arranged current data according to the preset time period to obtain multiple time period data; construct the current time period data matrix of the cell to be predicted according to the multiple time period data.

[0143] Step S406, perform position encoding processing on each data in the current time period data matrix to obtain the initial current time period encoding matrix of the cell to be predicted.

[0144] Step S407, obtain the holiday type and time step information corresponding to each position encoding in the initial current time period encoding matrix.

[0145] Step S408, query the preset holiday mask according to the holiday type and time step information corresponding to each position encoding to obtain the holiday mask corresponding to each position encoding; the preset holiday mask includes holiday masks corresponding to multiple holiday types and time step information.

[0146] Step S409: respectively fuse the encodings at each position in the initial current time period encoding matrix with the holiday masks corresponding to the encodings at each position to obtain the current time period encoding matrix of the cell to be predicted.

[0147] Step S410: input the current time period encoding matrix, as well as the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted, into the decoder in the wireless performance and load prediction model to obtain the prediction results of the wireless performance and load of the cell to be predicted.

[0148] In the above wireless performance and load prediction method, when predicting the wireless performance and load of the cell to be predicted, by using the current time period data matrix of the cell to be predicted, it is possible to deeply mine the deep features in the wireless performance and load data, effectively extract the long-term dependencies and periodic features of the wireless performance and load data, significantly improve the processing ability and accuracy of the wireless performance and load prediction model for the complex time series data of wireless performance and load, thereby improving the prediction accuracy of the wireless performance and load; at the same time, comprehensively considering the influence of holiday factors on the prediction results of wireless performance and load, introducing a preset holiday mask, and combining it with the current time period data matrix to obtain the current time period encoding matrix of the cell to be predicted, so that when the wireless performance and load prediction model processes the current time period encoding matrix, it can better capture the change rules brought by holiday factors in the complex time series data, enhance the capture ability of the wireless performance and load prediction model for the periodicity and particularity of the time series, and improve the accuracy of the wireless performance and load prediction model in related performance and load prediction tasks; in addition, the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted are also considered, which is conducive to further improving the prediction accuracy of the wireless performance and load.

[0149] To more clearly illustrate the wireless performance and load prediction method provided by the embodiments of the present application, the following uses a specific embodiment to specifically describe the wireless performance and load prediction method. In an exemplary embodiment, as Figure 5As shown, aiming at the problems of processing long sequences, holidays and emergencies in the prediction of the performance and load of current wireless network cells, this application also provides a wireless performance and load prediction method based on an improved Transformer. The historical performance and load data of the target cell are sliced into a multi-scale matrix according to the time period, and the time period can be days / weeks / months, forming a multi-scale time period data matrix; the holidays are encoded and the holidays are assigned complex numbers to form the positional encoding of the holiday mask; the time period data matrix is input into the encoder (composed of a multi-head attention module, a residual connection and a normalization module, and a fully connected network module) after passing through the holiday mask to learn the short-term features and long-term evolution trends of the target cell; the current data of the target cell is also sliced into the current time period data matrix; the current time period data matrix is input into the decoder (composed of a multi-head attention module, a residual connection and a normalization module, and a fully connected network module) after passing through the holiday mask, and the performance and load prediction results of the target cell are calculated by combining the learned short-term and long-term information, and it is directly judged whether the prediction time is a holiday prediction result according to the value of the complex number. Specifically, it includes the following steps:

[0150] Step 1: Obtain the performance and load related index data of the target cell at the hourly or daily granularity in the past period of time and arrange them in chronological order;

[0151] Step 2: After slicing the historical performance and load time series data into a multi-scale matrix according to the time period, it is used as the training data input. For example, assuming that N is the data length in days and it is processed weekly, the historical data can be sliced into a weekly (N / 7)×7 matrix or a weekly (N - 7 + 1)×7 matrix;

[0152] Step 3: Encode the dates to form a holiday mask in the form of complex numbers and input it as the positional encoding of the training data. Among them, different holidays have different encoding strategies x (pos,i) =PE (pos,i) ×m (h,t) ; x (pos,i) is the positional encoding in the form of complex numbers for different holidays; PE (pos,i) is the original positional encoding (calculated according to the traditional sine-cosine positional encoding method); m (h,t) is the complex number related to the characteristics of different holidays, and it is "1" for non-holidays;

[0153] Step 4: The training data matrix is input into the encoder after passing through the holiday mask to extract the short-term features and long-term evolution trends of the target cell;

[0154] Step 5: Perform the operations of Step 2 and Step 3 on the current data to form the decoder data matrix after holiday encoding;

[0155] Step 6: Input the decoder data matrix of Step 5 into the decoder. The decoder combines the information extracted in Step 4, predicts the changing trends of the performance and load of the target cell, and outputs the prediction results in complex number form. Then, directly judge whether the prediction time is a holiday according to the value of the complex number.

[0156] The above embodiments can achieve the following technical effects: (1) The method of performing multi-scale matrix segmentation on the performance and load time series data according to time periods and then using the segmented data as training and prediction data to input into the encoder-decoder for prediction results. The formed time period data matrix can extract long-term dependencies and periodic features. This construction method can mine deep features of the data, improve the understanding and processing ability of the Transformer prediction model for complex time series data, and enhance the accuracy of performance and load prediction; (2) The method of encoding the date and forming a holiday mask in complex number form as the positional encoding input of the training data enables the Transformer prediction model to better capture the changing patterns brought by holidays in the time series, enhances the ability of the Transformer prediction model to capture the periodicity and particularity of the time series, and improves the accuracy and adaptability of the Transformer prediction model in related performance and load prediction tasks; (3) Compared with the traditional method, the main advantages are as follows: Segmenting the performance and load data according to time periods to form a time period data matrix, which breaks the limitation of the traditional Transformer prediction model that directly calculates predictions, can deeply mine the deep features in the data, effectively extract long-term dependencies and periodic features, and significantly improve the processing ability and accuracy for complex time series data; Traditional positional encoding can only mark the element order and is difficult to reflect special time information such as holidays, while the holiday mask accurately marks holiday information through specific masking operations, enabling the Transformer prediction model to quickly identify whether the time point is a holiday; The holiday mask greatly enhances the ability of the Transformer prediction model to capture the periodicity and particularity of the time series, improving the accuracy and adaptability of the prediction.

[0157] It should be understood that although the steps in the flowcharts involved in the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments, and the execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0158] Based on the same inventive concept, an embodiment of the present application further provides a wireless performance and load prediction device for implementing the wireless performance and load prediction methods involved above. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the wireless performance and load prediction device provided below can refer to the limitations on the wireless performance and load prediction method in the foregoing, and will not be repeated here.

[0159] In an exemplary embodiment, as Figure 6 shown, a wireless performance and load prediction device is provided, including: a matrix acquisition module 610, a matrix processing module 620, and a result prediction module 630, where:

[0160] The matrix acquisition module 610 is configured to acquire the current time period data matrix of the cell to be predicted; the current time period data matrix includes the current time period wireless performance data matrix and the current time period wireless load data matrix.

[0161] The matrix processing module 620 is configured to obtain the current time period encoding matrix of the cell to be predicted according to the preset holiday mask and the current time period data matrix.

[0162] The result prediction module 630 is configured to input the current time period encoding matrix, as well as the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted, into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted.

[0163] In an exemplary embodiment, the matrix processing module 620 is further configured to perform position encoding processing on each data in the current time period data matrix to obtain the initial current time period encoding matrix of the cell to be predicted; and update the initial current time period encoding matrix according to the preset holiday mask to obtain the current time period encoding matrix of the cell to be predicted.

[0164] In an exemplary embodiment, the preset holiday mask includes holiday masks corresponding to multiple holiday types and time step information;

[0165] The matrix processing module 620 is further configured to obtain the holiday type and time step information corresponding to each position encoding in the initial current time period encoding matrix; query the preset holiday mask according to the holiday type and time step information corresponding to each position encoding to obtain the holiday mask corresponding to each position encoding; and respectively perform fusion processing on each position encoding in the initial current time period encoding matrix and the holiday mask corresponding to each position encoding to obtain the current time period encoding matrix of the cell to be predicted.

[0166] In an exemplary embodiment, the matrix acquisition module 610 is further configured to acquire the current data of the cell to be predicted; the current data includes current radio performance data and current radio load data; arrange the current data in chronological order to obtain the arranged current data; perform segmentation processing on the arranged current data according to a preset time period to obtain a plurality of time period data; and construct a current time period data matrix of the cell to be predicted based on the plurality of time period data.

[0167] In an exemplary embodiment, the radio performance and load prediction device further includes a feature determination module, configured to acquire a historical time period data matrix of the cell to be predicted; the historical time period data matrix includes a historical time period radio performance data matrix and a historical time period radio load data matrix; obtain a historical time period coding matrix of the cell to be predicted according to a preset holiday mask and the historical time period data matrix; and input the historical time period coding matrix into an encoder in the radio performance and load prediction model to obtain short-term features and long-term features of the radio performance and load of the cell to be predicted.

[0168] In an exemplary embodiment, the feature determination module is further configured to perform position coding processing on each data in the historical time period data matrix to obtain an initial historical time period coding matrix of the cell to be predicted; and update the initial historical time period coding matrix according to a preset holiday mask to obtain a historical time period coding matrix of the cell to be predicted.

[0169] In an exemplary embodiment, the radio performance and load prediction device further includes a result analysis module, configured to confirm that the radio performance and load prediction result is a holiday prediction result when the radio performance and load prediction result is a complex number; and confirm that the radio performance and load prediction result is a non-holiday prediction result when the radio performance and load prediction result is a non-complex number.

[0170] Each module in the above radio performance and load prediction device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0171] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as Figure 7As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as a preset holiday mask. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for wireless performance and load prediction.

[0172] Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0173] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0174] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0175] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0177] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0178] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0179] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A wireless performance and load prediction method, characterized in that, The method includes: Obtaining a current time period data matrix of the cell to be predicted; the current time period data matrix includes a current time period radio performance data matrix and a current time period radio load data matrix; Obtaining a current time period encoding matrix of the cell to be predicted according to a preset holiday mask and the current time period data matrix; Inputting the current time period encoding matrix, as well as the short-term and long-term characteristics of the radio performance and load of the cell to be predicted, into a decoder in a radio performance and load prediction model to obtain a radio performance and load prediction result of the cell to be predicted.

2. The method according to claim 1, wherein The obtaining a current time period encoding matrix of the cell to be predicted according to a preset holiday mask and the current time period data matrix includes: Performing position encoding processing on each data in the current time period data matrix to obtain an initial current time period encoding matrix of the cell to be predicted; Updating the initial current time period encoding matrix according to the preset holiday mask to obtain a current time period encoding matrix of the cell to be predicted.

3. The method according to claim 2, wherein The preset holiday mask includes a plurality of holiday types and holiday masks corresponding to time step information; The updating the initial current time period encoding matrix according to the preset holiday mask to obtain a current time period encoding matrix of the cell to be predicted includes: Obtaining the holiday type and time step information corresponding to each position encoding in the initial current time period encoding matrix; Querying the preset holiday mask according to the holiday type and time step information corresponding to each position encoding to obtain the holiday mask corresponding to each position encoding; Fusing each position encoding in the initial current time period encoding matrix with the holiday mask corresponding to each position encoding respectively to obtain a current time period encoding matrix of the cell to be predicted.

4. The method according to claim 1, wherein The obtaining a current time period data matrix of the cell to be predicted includes: Obtaining the current data of the cell to be predicted; the current data includes current radio performance data and current radio load data; Arranging the current data in chronological order to obtain the arranged current data; Performing slicing processing on the arranged current data according to a preset time period to obtain a plurality of time period data; Constructing a current time period data matrix of the cell to be predicted according to the plurality of time period data.

5. The method according to claim 1, wherein Before inputting the current time period encoding matrix, as well as the short-term and long-term characteristics of the radio performance and load of the cell to be predicted, into a decoder in a radio performance and load prediction model to obtain a radio performance and load prediction result of the cell to be predicted, it further includes: Obtaining a historical time period data matrix of the cell to be predicted; the historical time period data matrix includes a historical time period radio performance data matrix and a historical time period radio load data matrix; Obtaining a historical time period encoding matrix of the cell to be predicted according to the preset holiday mask and the historical time period data matrix; Input the historical time period encoding matrix into the encoder in the wireless performance and load prediction model to obtain the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted.

6. The method according to claim 5, characterized in that The obtaining of the historical time period encoding matrix of the cell to be predicted according to the preset holiday mask and the historical time period data matrix includes: Perform position encoding processing on each data in the historical time period data matrix to obtain the initial historical time period encoding matrix of the cell to be predicted; Update the initial historical time period encoding matrix according to the preset holiday mask to obtain the historical time period encoding matrix of the cell to be predicted.

7. The method according to any one of claims 1 to 6, characterized in that, After inputting the current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted, it further includes: In the case where the wireless performance and load prediction result is a complex number, confirm that the wireless performance and load prediction result is a holiday prediction result; In the case where the wireless performance and load prediction result is not a complex number, confirm that the wireless performance and load prediction result is a non-holiday prediction result.

8. A wireless performance and load prediction device, characterized in that The device includes: A matrix acquisition module, configured to acquire the current time period data matrix of the cell to be predicted; the current time period data matrix includes a current time period wireless performance data matrix and a current time period wireless load data matrix; A matrix processing module, configured to obtain the current time period encoding matrix of the cell to be predicted according to the preset holiday mask and the current time period data matrix; A result prediction module, configured to input the current time period encoding matrix and the short-term and long-term characteristics of the wireless performance and load of the cell to be predicted into the decoder in the wireless performance and load prediction model to obtain the wireless performance and load prediction result of the cell to be predicted.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images