Capacity prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a capacity prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: collecting direct feature data associated with capacity data, and carrying out the processing of the direct feature data, and obtaining a reference data set; adding indirect feature data associated with the capacity data into the reference data set to obtain a feature data set; wherein the indirect feature data is feature data which indirectly affects the change of the capacity data; constructing a time sequence data set based on the feature data set; wherein the time sequence data set comprises different types of feature data in continuous time periods; respectively inputting different types of feature data sets in the time sequence data set into a multi-input neural network for prediction to obtain a capacity prediction result; the direct feature data and the indirect feature data are fused, and different types of features are separated by using the multi-input neural network, so that the accuracy of capacity prediction and the generalization ability in different scenes can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and in particular to a capacity prediction method and apparatus, electronic device and storage medium. Background Technology

[0002] In the field of communications, cell capacity prediction is a crucial part of network optimization and resource allocation. It can predict cell capacity load, optimize base station resource scheduling, ensure stable downlink communication rates for users, and support the smooth operation of high-bandwidth services.

[0003] In related technologies, cell capacity prediction uses factors directly related to capacity as input features for prediction, thereby completing the prediction of cell capacity. However, due to the increasing complexity of network environments and the diversification of user behaviors, relying solely on direct factors is difficult to adapt to the changing network scenarios and user behavior patterns, resulting in insufficient generalization ability of capacity prediction models. Summary of the Invention

[0004] This disclosure provides a capacity prediction method, apparatus, electronic device, and storage medium to address problems in related technologies. By fusing direct and indirect feature data and utilizing a multi-input neural network to separate and process different types of features, the accuracy of capacity prediction and its generalization ability in different scenarios can be improved.

[0005] According to a first aspect embodiment of this disclosure, a capacity prediction method is provided, comprising: Collect direct feature data associated with capacity data, and process the direct feature data to obtain a benchmark dataset; Indirect feature data associated with the capacity data is added to the baseline dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the changes in the capacity data; Based on the aforementioned feature dataset, a time series dataset is constructed; wherein, the time series dataset contains feature data of different types for consecutive time periods; Different types of feature datasets from the time series dataset are input into a multi-input neural network for prediction to obtain capacity prediction results.

[0006] In some embodiments of this disclosure, the indirect feature data includes at least one of weather features and holiday features. The step of adding indirect feature data associated with the capacity data to the baseline dataset to obtain a feature dataset includes: Based on the dates corresponding to the benchmark dataset, obtain the weather features and / or holiday features associated with the dates; The weather features and / or the holiday features are added to the baseline dataset to obtain the feature dataset.

[0007] In some embodiments of this disclosure, adding the weather features and / or the holiday features to the baseline dataset to obtain the feature dataset includes: Based on the type of weather characteristics, the weather characteristics are divided into rainy day characteristics and non-rainy day characteristics; Based on the type of holiday characteristics, the holiday characteristics are divided into ordinary holiday characteristics and winter and summer vacation characteristics; The rainy day features, non-rainy day features, ordinary holiday features, and / or summer / winter vacation features are added to the baseline dataset to obtain the feature dataset.

[0008] In some embodiments of this disclosure, constructing a time series dataset based on the feature dataset includes: Select feature data of a preset number of days from the feature dataset, and construct the input time series corresponding to the feature data of the preset number of days; The capacity data of the preset number of days to be predicted in the input time series is determined as the target time series corresponding to the input time series; The time series dataset is constructed based on at least one input time series and a target time series corresponding to the at least one input time series.

[0009] In some embodiments of this disclosure, the step of inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain a capacity prediction result includes: The direct feature data in the input time series is input into at least one first neural network of the multi-input neural network to obtain at least one first feature; The indirect feature data in the input time series is input into at least one second neural network of the multi-input neural network to obtain at least one second feature; The at least one first feature and the at least one second feature are spliced ​​together to obtain a spliced ​​feature; The spliced ​​features are predicted by the third neural network of the multi-input neural network to obtain the capacity prediction result.

[0010] In some embodiments of this disclosure, the collection of direct feature data associated with capacity data and the processing of the direct feature data to obtain a benchmark dataset include: Based on the direct feature data corresponding to each cell, the mean and standard deviation of each cell are calculated, and a statistical interval is determined based on the mean and standard deviation; wherein, the direct feature data includes at least one of downlink average user rate, maximum number of user connections, and cell throughput; Based on the statistical interval, outliers are identified; wherein the outliers are located outside the statistical interval. Based on the proportion of the outlier data to the total data volume in each cell and the proportion of holiday duration to the total statistical duration, it is determined whether the direct feature data of each cell is within a preset range. If the direct feature data of each cell is determined to be within the preset range, the direct feature data of all cells are normalized to obtain the benchmark dataset.

[0011] According to a second aspect of this disclosure, a capacity prediction apparatus is provided, comprising: The processing unit is used to collect direct feature data associated with capacity data and process the direct feature data to obtain a benchmark dataset. An adding unit is used to add indirect feature data associated with the capacity data to the benchmark dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the change of the capacity data; A construction unit is used to construct a time series dataset based on the feature dataset; wherein the time series dataset contains feature data of different types for consecutive time periods; The prediction unit is used to input different types of feature datasets from the time series dataset into a multi-input neural network for prediction, and obtain the capacity prediction result.

[0012] In some embodiments of this disclosure, the indirect feature data includes at least one of weather features and holiday features, and the adding unit includes: The acquisition module is used to acquire the weather features and / or holiday features associated with the dates corresponding to the benchmark dataset. An add module is used to add the weather features and / or the holiday features to the baseline dataset to obtain the feature dataset.

[0013] In some embodiments of this disclosure, the adding module includes: The first segmentation submodule is used to segment the weather features into rainy day features and non-rainy day features according to the type of the weather features; The second segmentation submodule is used to segment the holiday features into ordinary holiday features and winter and summer vacation features according to the type of the holiday features; Add a submodule to add the rainy day features, the non-rainy day features, the ordinary holiday features and / or the winter and summer vacation features to the baseline dataset to obtain the feature dataset.

[0014] In some embodiments of this disclosure, the building unit includes: The first construction module is used to select feature data of a preset number of days from the feature dataset and construct the input time series corresponding to the feature data of the preset number of days. The determining module is used to determine the capacity data of the preset number of days to be predicted in the input time series as the target time series corresponding to the input time series; The second construction module is used to construct the time series dataset based on at least one input time series and a target time series corresponding to the at least one input time series.

[0015] In some embodiments of this disclosure, the prediction unit includes: The first input module is used to input the direct feature data in the input time series into at least one first neural network of the multi-input neural network to obtain at least one first feature; The second input module is used to input the indirect feature data in the input time series into at least one second neural network of the multi-input neural network to obtain at least one second feature; A splicing module is used to splice the at least one first feature and the at least one second feature to obtain a spliced ​​feature; The prediction module is used to predict the concatenated features through the third neural network of the multi-input neural network to obtain the capacity prediction result.

[0016] In some embodiments of this disclosure, the processing unit includes: The calculation module is used to calculate the mean and standard deviation of each cell based on the direct feature data corresponding to each cell, and to determine the statistical interval based on the mean and standard deviation; wherein, the direct feature data includes at least one of downlink average user rate, maximum number of user connections and cell throughput; An identification module is used to identify outliers based on the statistical interval; wherein the outliers are located outside the statistical interval. The judgment module is used to determine whether the direct feature data of each cell is within a preset range based on the proportion of the outlier data volume to the total data volume of each cell and the proportion of holiday duration to the total statistical duration. The processing module is used to normalize the direct feature data of all cells to obtain the benchmark dataset when it is determined that the direct feature data of each cell is within the preset range.

[0017] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect embodiment.

[0018] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect of the present disclosure.

[0019] According to a fifth aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described in the first aspect of the preceding embodiments.

[0020] In summary, the capacity prediction method, apparatus, electronic device, and storage medium provided in this disclosure include: collecting direct feature data associated with capacity data and processing the direct feature data to obtain a baseline dataset; adding indirect feature data associated with capacity data to the baseline dataset to obtain a feature dataset; wherein the indirect feature data is feature data that indirectly affects changes in capacity data; constructing a time series dataset based on the feature dataset; wherein the time series dataset contains different types of feature data over consecutive time periods; inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results; by fusing direct and indirect feature data and using a multi-input neural network to separate and process different types of features, the accuracy of capacity prediction and its generalization ability in different scenarios can be improved.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0022] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic flowchart of a capacity prediction method provided in an embodiment of the present disclosure; Figure 2 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 3 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 4 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 5 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 6 This is a schematic diagram of the structure of a multi-input neural network provided in an embodiment of the present disclosure; Figure 7 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 8 This is a schematic flowchart of another capacity prediction method provided in an embodiment of the present disclosure; Figure 9 This is a schematic diagram of the structure of a capacity prediction device provided in an embodiment of the present disclosure; Figure 10 This is a schematic diagram of another capacity prediction device provided in an embodiment of the present disclosure; Figure 11 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0023] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] The capacity prediction method, apparatus, electronic device, and storage medium of this disclosure are described below with reference to the accompanying drawings.

[0025] Figure 1 This is a schematic flowchart of a capacity prediction method provided in an embodiment of the present disclosure.

[0026] like Figure 1 As shown, the method includes the following steps: Step 101: Collect direct feature data associated with capacity data, and process the direct feature data to obtain a benchmark dataset.

[0027] In some embodiments, direct feature data that is directly related to capacity data is acquired. Direct feature data is relevant data that is directly related to changes in capacity data. After data acquisition is completed, data processing operations are performed on the acquired direct feature data. Data processing may employ conventional data preprocessing methods, including but not limited to data cleaning and standardization. Data cleaning is used to remove invalid and missing values ​​from the data, and standardization is used to unify the magnitude range of the data, transforming the original acquired direct feature data into a standardized and valid benchmark dataset.

[0028] The above methods can ensure the basic data quality of the benchmark dataset, providing a data foundation for feature fusion and prediction model training, and avoiding interference with subsequent processes due to defects in the original data.

[0029] Step 102: Add indirect feature data associated with the capacity data to the benchmark dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the changes in the capacity data.

[0030] In some embodiments, the time dimension information corresponding to each data in the benchmark dataset is first determined, and then the indirect feature data of the corresponding time period is obtained based on the time dimension information. The indirect feature data is feature data that does not directly affect the capacity data, but can affect the capacity data through intermediate factors. The obtained indirect feature data is matched with the benchmark dataset according to the time dimension, and the indirect feature data of the corresponding time period is integrated into the corresponding data entry of the benchmark dataset to form a feature dataset that contains both direct feature data and indirect feature data.

[0031] The above method can effectively integrate direct and indirect features, enrich the data source dimensions for model training, and solve the problem of a single data source caused by relying solely on direct feature data.

[0032] Step 103: Based on the feature dataset, construct a time series dataset; wherein the time series dataset contains feature data of different types for consecutive time periods.

[0033] In some embodiments, based on a preset time span, all feature data within a continuous time period are extracted from the feature dataset, and these discrete single-time period feature data are integrated in chronological order to form a feature data sequence with temporal continuity; by extracting and integrating multiple sets of continuous time periods from the feature dataset, multiple sets of sequence data containing different types of feature data in continuous time periods are formed, and these sequence data together constitute a time series dataset.

[0034] The above methods can transform feature data into time-correlated sequence data, enabling the prediction model to capture the correlation patterns of volume data over time and providing data support for the model to learn time-series features.

[0035] Step 104: Input the different types of feature datasets in the time series dataset into a multi-input neural network for prediction to obtain the capacity prediction result.

[0036] In some embodiments, the feature data in the time series dataset are classified according to data type, and different types of feature data are input to the input ports of the multi-input neural network. The corresponding modules of the multi-input neural network process the different types of feature data respectively, and then the processing results of each module are fused and calculated by the comprehensive processing module of the neural network. Finally, the neural network outputs the capacity prediction result for the corresponding time period.

[0037] The above method enables the separate processing of different types of feature data, avoiding confusion between different types of features during transmission and processing, giving full play to the advantages of multi-input neural networks in processing multiple types of data, and improving the overall effect of capacity prediction.

[0038] In summary, the capacity prediction method provided in this disclosure includes: collecting direct feature data associated with capacity data and processing the direct feature data to obtain a baseline dataset; adding indirect feature data associated with capacity data to the baseline dataset to obtain a feature dataset; wherein the indirect feature data is feature data that indirectly affects changes in capacity data; constructing a time series dataset based on the feature dataset; wherein the time series dataset contains different types of feature data over consecutive time periods; inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results; by fusing direct and indirect feature data and using a multi-input neural network to separate and process different types of features, the accuracy of capacity prediction and its generalization ability in different scenarios can be improved.

[0039] Figure 2 This is a flowchart illustrating a capacity prediction method provided in an embodiment of this disclosure. Further explanation of step 102 is as follows: Figure 2 As shown, it includes: Step 201: Based on the date corresponding to the benchmark dataset, obtain the weather features and / or holiday features associated with the date.

[0040] In some embodiments, date information corresponding to each data point in the benchmark dataset is extracted. This date information corresponds one-to-one with the collection time of the direct feature data and serves as the time benchmark for feature matching. For weather features, the date is used as the retrieval condition to obtain the complete meteorological record for the corresponding date from the preset historical meteorological data. The meteorological record includes, but is not limited to, specific weather descriptions such as light rain, light rain turning cloudy, and sunny turning partly cloudy. For holiday features, the date is also used as the retrieval condition to obtain information on whether the corresponding date belongs to a national statutory holiday, a regional special holiday (such as "March 3rd"), or a winter / summer vacation from the preset winter / summer vacation schedule file. If the benchmark dataset requires a single type of indirect feature, weather features or holiday features can be obtained separately. If weather features and holiday features need to be combined, they are obtained simultaneously and stored separately.

[0041] Step 202: Add the weather features and / or the holiday features to the baseline dataset to obtain the feature dataset.

[0042] In some embodiments, the acquired weather features and holiday features are first classified and labeled to transform unstructured feature information into structured data that can be fused with the benchmark dataset. Weather features are labeled as binary categories based on "whether it is a rainy day", and holiday features are labeled as combined categories based on "whether it is a regular holiday" and "whether it is a summer or winter vacation". After labeling, the labeled feature data (weather features and / or holiday features) for each benchmark data date are matched to the corresponding field of the benchmark data to achieve the fusion of direct and indirect feature data, and finally form a feature dataset.

[0043] The above method achieves accurate correlation between indirect feature data and benchmark dataset, ensuring that the added feature data and direct feature data are fully matched in the time dimension, providing accurate multi-dimensional feature support for building high-quality time series datasets, while clarifying the specific data source for feature acquisition and improving the reliability of feature data.

[0044] Figure 3 This is a flowchart illustrating a capacity prediction method provided in an embodiment of this disclosure. Further explanation of step 202 is as follows: Figure 3 As shown, it includes: Step 301: Based on the type of the weather feature, classify the weather feature into rainy day features and non-rainy day features.

[0045] In some embodiments, weather classification rules are established for the acquired meteorological records, taking into account local geographical conditions (such as the difference in judgment criteria between rainy and humid areas and arid and dry areas) and meteorological characteristics (such as the local definition of "light rain" rainfall). Weather conditions containing rainfall descriptions, such as "light rain turning cloudy" and "moderate rain turning heavy rain," are uniformly classified as rainy day features. Weather conditions without rainfall descriptions, such as "sunny," "partly cloudy," and "sunny turning partly cloudy," are uniformly classified as non-rainy day features. After classification, the features are quantified using numerical labeling: rainy day features are labeled as -1, and non-rainy day features are labeled as 1, forming structured weather feature data.

[0046] Step 302: Based on the type of the holiday characteristics, the holiday characteristics are divided into ordinary holiday characteristics and winter and summer vacation characteristics.

[0047] In some embodiments, the classification criteria for holiday characteristics are clearly defined: holidays are divided into statutory holidays, regionally specific holidays (such as "March 3rd"), and winter and summer vacations applicable to teachers and students; since winter and summer vacations have a longer duration and the passenger flow changes relatively steadily during these periods, while other holidays have larger passenger flow changes, statutory holidays and regionally specific holidays are uniformly classified as ordinary holiday characteristics, and winter and summer vacations are classified as winter and summer vacation characteristics, with the two categories being classified independently; considering that ordinary holidays and whether the corresponding short period before and after them are ordinary holidays influence each other, while winter and summer vacations are not sensitive to this, therefore, any date included in the training set is considered... , mark , Whether it is a statutory holiday or a holiday with regional characteristics, and Whether it is a winter or summer vacation, the marking rule is: if it is "yes", mark it as 1; if it is "no", mark it as -1.

[0048] in The annotation satisfies the following:

[0049] in, The annotation method and The same applies to both types of holidays: whether they are national statutory holidays or regional holidays. The marking rule is uniform: "yes" is marked as 1, and "no" is marked as -1.

[0050] Step 303: Add the rainy day features, the non-rainy day features, the ordinary holiday features, and / or the winter and summer vacation features to the baseline dataset to obtain the feature dataset.

[0051] In some embodiments, the collection date of each direct feature data in the benchmark dataset is extracted, and the weather feature label value (e.g., -1 for rainy days and 1 for non-rainy days) and holiday feature label result (e.g., 1 for ordinary holidays or winter and summer vacations, and -1 for no holidays) corresponding to that date are matched. These features are then added as independent fields to the corresponding entries in the benchmark dataset.

[0052] The above methods have enabled the standardization, quantification, and precise fusion of indirect feature data, fully preserving the influence and correlation of weather and different types of holidays on capacity data, making the feature data adaptable to model training.

[0053] Figure 4 This is a flowchart illustrating a capacity prediction method provided in an embodiment of this disclosure. Further explanation of step 103 is as follows: Figure 4 As shown, it includes: Step 401: Select feature data of a preset number of days from the feature dataset and construct the input time series corresponding to the feature data of the preset number of days.

[0054] In some embodiments, based on the training requirements of the neural network, feature data for a preset number of days (e.g., three consecutive days) is selected from the feature dataset to construct the input time series; the feature data includes direct feature data and indirect feature data: direct feature data (such as the average user download speed data), denoted as... Indirect feature data includes weather feature data such as whether it is a rainy day, denoted as... The characteristic data of ordinary holidays, including the holiday situation on the same day and the following day, are denoted as... The characteristic data of winter and summer vacations on that day are denoted as follows: For example, using the starting date t as a reference, feature data at t, t+1, and t+2 are extracted and combined in the order of direct and indirect features. The input time series is... ,in, The data consists of direct features from three consecutive days. This is weather characteristic data for three consecutive days. This data represents the characteristics of a typical three-day holiday. This data represents the characteristics of three consecutive days of winter and summer vacation. For prediction Direct feature data for the following three consecutive days; through , , , Represent The data includes the average downlink speed per day, whether it is a rainy day, a holiday, or a summer / winter break. It should be noted that the examples above are for illustrative purposes only, and there is no limitation on the preset number of days.

[0055] Step 402: Determine the capacity data of the preset number of days to be predicted in the input time series as the target time series corresponding to the input time series.

[0056] In some embodiments, the preset number of days to be predicted is consistent with the preset number of days in the input time series, and the target time series is the volume data of the preset number of days after the input time series, which serves as the label for model training. Taking the input time series as an example, the predicted periods are days t+3, t+4, and t+5. Direct feature data (such as the average downlink user rate and volume data indicators) for these three days are extracted as the target time series, specifically in the form of... ,in , , These are direct feature data for three consecutive days following the input time series. It should be noted that the above examples are for illustrative purposes only, and the setting of the preset number of days is not limited.

[0057] Step 403: Construct the time series dataset based on at least one input time series and a target time series corresponding to the at least one input time series.

[0058] In some embodiments, a sliding time window is used to generate multiple sets of associated time series pairs. The sliding step size is a preset number of days. After the input and target time series pairs with the start date t are constructed, a new input time series and a corresponding target time series are constructed with t+1 as the new start date. The sliding continues until all continuous time periods that meet the conditions in the feature dataset are traversed, and at least one set of input time series and corresponding target time series are obtained, which constitutes a time series dataset.

[0059] The above method transforms multi-type feature data into structured samples with temporal correlation, fully preserving the temporal mapping relationship between features and capacity, and providing a dataset that meets the requirements for neural network training.

[0060] Figure 5 This is a flowchart illustrating a capacity prediction method provided in an embodiment of this disclosure. Further explanation of step 104 is as follows: Figure 5 As shown, it includes: Step 501: Input the direct feature data in the input time series into at least one first neural network of the multi-input neural network to obtain at least one first feature.

[0061] In some embodiments, direct feature data is input into a first neural network of a multi-input neural network. The first neural network has a preset number of neurons (e.g., 3, 128, 32, 2) to process the direct feature data. Through the computation of the first neural network, higher-dimensional features of the direct feature data are captured and abstracted. Its training process corresponds to... The final output will contain at least one first feature.

[0062] Step 502: Input the indirect feature data in the input time series into at least one second neural network of the multi-input neural network to obtain at least one second feature.

[0063] In some embodiments, the indirect feature data in the input time series includes at least weather features, summer and winter vacation features, and general holiday features. The second neural network comprises two sub-networks to process different types of indirect features respectively. The first sub-network processes weather features and summer and winter vacation features, with a preset number of neurons (e.g., 6, 64, 32, 2); the second sub-network processes general holiday features, with a preset number of neurons (e.g., 6, 64, 32, 2). After the two types of indirect feature data are input into their respective second neural networks, the high-dimensional abstract features are captured through the second neural network operations. The training process corresponds to... , The final output will have at least two second features.

[0064] Step 503: Perform splicing processing on the at least one first feature and the at least one second feature to obtain spliced ​​features.

[0065] In some embodiments, the first feature and at least two second features are dimensionally concatenated to integrate high-dimensional information of different types of features, forming a dimensionally unified concatenated feature. This concatenated feature serves as input data for the prediction operation of a multi-input neural network, thereby achieving the fusion of different feature information.

[0066] Step 504: The spliced ​​features are predicted by the third neural network of the multi-input neural network to obtain the capacity prediction result.

[0067] In some embodiments, the time series dataset is processed as follows: after shuffling all data, it is divided into a training set with a pre-preset proportion (e.g., 70%) and a test set with a pre-preset proportion (e.g., 30%). The third neural network has a pre-preset number of neurons (e.g., 6, 128, 32, 8, 3). After inputting the concatenated features into the third neural network, the model is trained using Adam as the optimizer (e.g., with a learning rate set to 0.001) and Mean Absolute Error (MAE) as the loss function. The formula for calculating MAE is: The MAE model was chosen because it is insensitive to outliers, which can reduce the interference of outliers on training during network management data capture; the third neural network further generalizes the spliced ​​features through calculations, approximating the training target, and its training process corresponds to... Finally, the predicted values ​​corresponding to the three neurons are output, which are the predicted values ​​of the normalized direct feature data of the last three days (such as the average rate of downstream users). These predicted values ​​are the capacity prediction results.

[0068] Using the above method, multi-input neural networks achieve separate input and independent processing of various unrelated information types, avoiding the mixing of different types of features and solving the problem of poor training performance caused by a single network input; at the same time With clear training features and a concise structure, the entire network is easy to train, has strong universality, and good interpretability. Combined with the advantages of the MAE loss function, the reliability of the capacity prediction results is further improved.

[0069] Figure 6 This is a schematic diagram of the structure of a multi-input neural network provided in an embodiment of the present disclosure, such as... Figure 6 As shown, as one possible implementation, this structure corresponds to the composition of a multi-input neural network, including a first neural network, at least one second neural network, and a third neural network. The first neural network is a sub-network that processes direct feature data, taking the direct feature data in the input time series as input. At least one second neural network is a sub-network that processes indirect feature data, corresponding to the processing of weather features, summer / winter vacation features, and ordinary holiday features, respectively. The plus sign in the figure represents the concatenation of the first feature output by the first neural network and the second feature output by the second neural network to obtain a concatenated feature with unified dimensions. This concatenated feature is input to the third neural network, and through further fusion and generalization of the concatenated feature, the capacity prediction result for the corresponding time period is finally output.

[0070] Figure 7 This is a flowchart illustrating a capacity prediction method provided in an embodiment of the present disclosure. Further explanation of step 101 is as follows: Figure 7 As shown, it includes: Step 601: Calculate the mean and standard deviation of each cell based on the direct feature data corresponding to each cell, and determine the statistical interval based on the mean and standard deviation; wherein, the direct feature data includes at least one of downlink average user rate, maximum number of user connections, and cell throughput.

[0071] In some embodiments, direct feature data is collected from network management data. The direct feature data includes at least one of downlink average user rate, maximum number of user connections, and cell throughput. Taking downlink average user rate as an example, for different cells, the maximum value of the downlink average user rate is taken as the corresponding data for each day, with hourly granularity. Simultaneously, this type of data from all cells within a specified area is aggregated to form a dataset. For the direct feature data of each cell, its mean is calculated. with standard deviation Based on the mean with standard deviation Determine the statistical interval as .

[0072] Step 602: Identify outliers based on the statistical interval; wherein the outliers are located outside the statistical interval.

[0073] In some embodiments, statistical intervals determined by the direct feature data of each cell are compared, and those not included in the data are excluded. Data within a given interval is labeled as outliers for that cell.

[0074] Step 603: Based on the proportion of the outlier data volume to the total data volume of each cell and the proportion of holiday duration to the total statistical duration, determine whether the direct feature data of each cell is within a preset range.

[0075] In some embodiments, the amount of outlier data in each cell is counted. Total data volume The proportion of outliers is calculated as follows: At the same time, the total length of holidays within the statistical period is calculated. With total date length The percentage of holiday time is calculated as follows: ; through formula Determine whether the data is within a preset range, where, (e.g., 0.01) (e.g., 0.05) is a preset threshold. If the data falls within this preset range, it is considered to conform to the actual situation; if it is greater than or equal to... The data distribution is too discrete; if it is less than or equal to If outliers are discontinuous, determine whether to delete the cell based on the actual situation.

[0076] Step 604: If the direct feature data of each cell is determined to be within the preset range, the direct feature data of all cells are normalized to obtain the benchmark dataset.

[0077] In some embodiments, cell data within a preset range are selected as the training set, and the maximum value of the direct feature data in the training set is taken. and minimum value For all direct feature data in the training set Perform normalization operation: The processed data is then integrated to form a benchmark dataset.

[0078] The above method not only ensures the quality of direct feature data (by removing cell data with excessive missing data or abnormal distribution), but also expands the coverage of training samples by aggregating data from multiple cells within the region. At the same time, normalization processing unifies the data volume, providing reliable and suitable benchmark data for subsequent feature fusion and model training.

[0079] Figure 8 This is a flowchart illustrating a capacity prediction method provided in an embodiment of the present disclosure, as shown below. Figure 8 As shown, the process performs the following operations in sequence: collecting direct feature data associated with capacity data and processing the direct feature data to obtain a baseline dataset; adding indirect feature data associated with capacity data to the baseline dataset to obtain a feature dataset that contains both direct and indirect feature data; constructing a time series dataset containing different types of feature data over consecutive time periods based on the feature dataset; and inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results.

[0080] Corresponding to the capacity prediction method described above, this invention also proposes a capacity prediction device. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0081] Figure 9 This is a schematic diagram of the structure of a capacity prediction device provided in an embodiment of this disclosure, as shown below. Figure 9 As shown, it includes: a processing unit 91, an adding unit 92, a building unit 93, and a prediction unit 94.

[0082] Processing unit 91 is used to collect direct feature data associated with capacity data and process the direct feature data to obtain a benchmark dataset; Adding unit 92 is used to add indirect feature data associated with the capacity data to the benchmark dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the change of the capacity data; Construction unit 93 is used to construct a time series dataset based on the feature dataset; wherein, the time series dataset contains feature data of different types for consecutive time periods; The prediction unit 94 is used to input different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results.

[0083] In summary, the capacity prediction device provided in this disclosure includes: collecting direct feature data associated with capacity data and processing the direct feature data to obtain a baseline dataset; adding indirect feature data associated with capacity data to the baseline dataset to obtain a feature dataset; wherein the indirect feature data is feature data that indirectly affects changes in capacity data; constructing a time series dataset based on the feature dataset; wherein the time series dataset contains different types of feature data over consecutive time periods; inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results; by fusing direct feature data and indirect feature data and using a multi-input neural network to separate and process different types of features, the accuracy of capacity prediction and its generalization ability in different scenarios can be improved.

[0084] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 10 As shown, the indirect feature data includes at least one of weather features and holiday features, and the adding unit 92 includes: The acquisition module 921 is used to acquire the weather features and / or holiday features associated with the date based on the date corresponding to the benchmark dataset; Add module 922, used to add the weather features and / or the holiday features to the baseline dataset to obtain the feature dataset.

[0085] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 10 As shown, the adding module 922 includes: The first segmentation submodule 9221 is used to segment the weather features into rainy day features and non-rainy day features according to the type of the weather features; The second division submodule 9222 is used to divide the holiday features into ordinary holiday features and winter and summer vacation features according to the type of the holiday features; Add submodule 9223 to add the rainy day features, the non-rainy day features, the ordinary holiday features and / or the winter and summer vacation features to the baseline dataset to obtain the feature dataset.

[0086] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 10 As shown, the building unit 93 includes: The first construction module 931 is used to select feature data of a preset number of days from the feature dataset and construct the input time series corresponding to the feature data of the preset number of days. The determining module 932 is used to determine the capacity data of the preset number of days to be predicted in the input time series as the target time series corresponding to the input time series; The second construction module 933 is used to construct the time series dataset based on at least one input time series and a target time series corresponding to the at least one input time series.

[0087] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 10 As shown, the prediction unit 94 includes: The first input module 941 is used to input the direct feature data in the input time series into at least one first neural network of the multi-input neural network to obtain at least one first feature; The second input module 942 is used to input the indirect feature data in the input time series into at least one second neural network of the multi-input neural network to obtain at least one second feature; The splicing module 943 is used to splice the at least one first feature and the at least one second feature to obtain a spliced ​​feature; The prediction module 944 is used to predict the spliced ​​features through the third neural network of the multi-input neural network to obtain the capacity prediction result.

[0088] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 10 As shown, the processing unit 91 includes: The calculation module 911 is used to calculate the mean and standard deviation of each cell based on the direct feature data corresponding to each cell, and to determine the statistical interval based on the mean and the standard deviation; wherein, the direct feature data includes at least one of downlink average user rate, maximum number of user connections and cell throughput; The identification module 912 is used to identify outliers based on the statistical interval; wherein the outliers are located outside the statistical interval. The judgment module 913 is used to determine whether the direct feature data of each cell is within a preset range based on the proportion of the outlier data volume to the total data volume of each cell and the proportion of holiday duration to the total statistical duration. The processing module 914 is used to normalize the direct feature data of all cells when it is determined that the direct feature data of each cell is within the preset range, so as to obtain the benchmark dataset.

[0089] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0090] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0091] Figure 11 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0092] like Figure 11 As shown, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 702 or loaded from storage unit 708 into RAM (Random Access Memory) 703. The RAM 703 can also store various programs and data required for the operation of the electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An I / O (Input / Output) interface 705 is also connected to the bus 704.

[0093] Multiple components in electronic device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of displays, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0094] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as capacity prediction methods. For example, in some embodiments, the capacity prediction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the aforementioned capacity prediction method by any other suitable means (e.g., by means of firmware).

[0095] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0096] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0097] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0098] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0099] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0100] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0101] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0102] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0103] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A capacity prediction method, characterized in that, The method includes: Collect direct feature data associated with capacity data, and process the direct feature data to obtain a benchmark dataset; Indirect feature data associated with the capacity data is added to the baseline dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the changes in the capacity data; Based on the aforementioned feature dataset, a time series dataset is constructed; wherein, the time series dataset contains feature data of different types for consecutive time periods; Different types of feature datasets from the time series dataset are input into a multi-input neural network for prediction to obtain capacity prediction results.

2. The method according to claim 1, characterized in that, The indirect feature data includes at least one of weather features and holiday features. The addition of indirect feature data associated with the capacity data to the baseline dataset to obtain a feature dataset includes: Based on the dates corresponding to the benchmark dataset, obtain the weather features and / or holiday features associated with the dates; The weather features and / or the holiday features are added to the baseline dataset to obtain the feature dataset.

3. The method according to claim 2, characterized in that, The step of adding the weather features and / or the holiday features to the baseline dataset to obtain the feature dataset includes: Based on the type of weather characteristics, the weather characteristics are divided into rainy day characteristics and non-rainy day characteristics; Based on the type of holiday characteristics, the holiday characteristics are divided into ordinary holiday characteristics and winter and summer vacation characteristics; The rainy day features, non-rainy day features, ordinary holiday features, and / or summer / winter vacation features are added to the baseline dataset to obtain the feature dataset.

4. The method according to claim 1, characterized in that, The construction of a time series dataset based on the feature dataset includes: Select feature data of a preset number of days from the feature dataset, and construct the input time series corresponding to the feature data of the preset number of days; The capacity data of the preset number of days to be predicted in the input time series is determined as the target time series corresponding to the input time series; The time series dataset is constructed based on at least one input time series and a target time series corresponding to the at least one input time series.

5. The method according to claim 4, characterized in that, The step of inputting different types of feature datasets from the time series dataset into a multi-input neural network for prediction to obtain capacity prediction results includes: The direct feature data in the input time series is input into at least one first neural network of the multi-input neural network to obtain at least one first feature; The indirect feature data in the input time series is input into at least one second neural network of the multi-input neural network to obtain at least one second feature; The at least one first feature and the at least one second feature are spliced ​​together to obtain a spliced ​​feature; The spliced ​​features are predicted by the third neural network of the multi-input neural network to obtain the capacity prediction result.

6. The method according to claim 1, characterized in that, The acquisition of direct feature data associated with capacity data, and the processing of the direct feature data to obtain a benchmark dataset, includes: Based on the direct feature data corresponding to each cell, the mean and standard deviation of each cell are calculated, and a statistical interval is determined based on the mean and standard deviation; wherein, the direct feature data includes at least one of downlink average user rate, maximum number of user connections, and cell throughput; Based on the statistical interval, outliers are identified; wherein the outliers are located outside the statistical interval. Based on the proportion of the outlier data to the total data volume in each cell and the proportion of holiday duration to the total statistical duration, it is determined whether the direct feature data of each cell is within a preset range. If the direct feature data of each cell is determined to be within the preset range, the direct feature data of all cells are normalized to obtain the benchmark dataset.

7. A capacity prediction device, characterized in that, include: The processing unit is used to collect direct feature data associated with capacity data and process the direct feature data to obtain a benchmark dataset. An adding unit is used to add indirect feature data associated with the capacity data to the benchmark dataset to obtain a feature dataset; wherein, the indirect feature data is feature data that indirectly affects the change of the capacity data; A construction unit is used to construct a time series dataset based on the feature dataset; wherein the time series dataset contains feature data of different types for consecutive time periods; The prediction unit is used to input different types of feature datasets from the time series dataset into a multi-input neural network for prediction, and obtain the capacity prediction result.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images