Prediction

Abstract

To provide a computer-implemented method, a computer program, and an information programming apparatus for predicting network usage data.SOLUTION: A method comprises the steps of: generating at least one of first to third 5G usage data predictions; generating, on the basis of predicted non-network data of a first interim prediction and predicted non-5G network usage data of a second interim prediction, a first 5G usage data prediction; generating, on the basis of the predicted non-network data of the first interim prediction and predicted non-5G network usage data of a third interim prediction, a second 5G usage data prediction; and generating, on the basis of the predicted non-network data of the first interim prediction and predicted non-5G network usage data of a sixth interim prediction, a third 5G usage data prediction.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to forecasting network usage data, and in particular to a computer implemented method, a computer program product and an information programming apparatus. [Background technology]

[0002] Mobile technologies and services are forecast to contribute €757 billion to Europe's GDP in 2021 and account for almost half (44%) of Europe's mobile connections by 2025, with key European markets lagging behind their global peers (https: / / www.telefonica.com / en / communication-room / blog / the-future-of-connectivity-in-europe-perspectives-and-policies / ).

[0003] Setting up 5G connections and infrastructure requires more hardware resources compared to other networks such as LTE, 4G and 3G. Achieving maximum resource utilization can reduce hardware costs, thus potentially resulting in cheaper Internet and more profits for network service providers. Thus, projecting 5G usage data in areas that currently do not have 5G connections is useful, for example, to ascertain what infrastructure will be required and / or whether the cost / effort of setting up the infrastructure is justified. Summary of the Invention [Problem to be solved by the invention]

[0004] In view of the above, improved methods for predicting 5G usage data in areas that do not have 5G connectivity are desired. [Means for solving the problem]

[0005] According to an embodiment of a first aspect, a computer-implemented method is disclosed, the method comprising: executing a prediction process to predict 5G usage data for a target geographic area (having no 5G connection / no historical 5G usage data exists or is available), the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data of the target geographic region for a future time period based on non-network data of the target geographic region for a past time period using a first model trained using data of the target geographic region (to predict non-network data of the second time period based on non-network data of the first time period prior to the second time period); generating a second intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a second model trained using data of at least one reference geographic region (which has or had a 5G connection) (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the first 5G usage data prediction by predicting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the second interim prediction using a third model trained using data of the at least one reference geographic region (to predict 5G usage data for the second time period based on non-network data and non-5G network usage data for the second time period); Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using data for the target geographic region (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating the fourth interim prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data of the at least one reference geographic region (to predict combined network usage data for the second time period based on non-network data for the first time period, the combined network usage data including usage data related to 5G and non-5G networks); generating the fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region (to predict 5G usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, (geographical data), demographic data, weather data, infrastructure data, and (vehicle) traffic data.

[0006] Features associated with any aspect / embodiment may be applied to any other aspect / embodiment. [Brief description of the drawings]

[0007] By way of example, reference is made to the accompanying drawings in which: [Figure 1] FIG. 1 shows exemplary data. [Diagram 2] FIG. 1 shows exemplary data. [Diagram 3] FIG. 2 illustrates the training process. [Figure 4] FIG. 1 illustrates a prediction process. [Diagram 5] FIG. 1 illustrates a prediction process. [Figure 6] FIG. 1 illustrates a prediction process. [Figure 7] 1 is a diagram useful for understanding an embodiment. [Figure 8] FIG. 1 illustrates a prediction process. [Figure 9] FIG. 1 illustrates an exemplary architecture. [Figure 10] FIG. 1 illustrates an exemplary architecture. [Figure 11] FIG. 1 illustrates an exemplary architecture. [Figure 12] FIG. 1 illustrates an exemplary architecture. [Figure 13] FIG. 1 illustrates an exemplary architecture. [Figure 14] FIG. 1 illustrates an exemplary architecture. [Figure 15] FIG. 1 illustrates an exemplary architecture. [Figure 16] FIG. 1 illustrates an exemplary architecture. [Figure 17] FIG. 1 illustrates an exemplary architecture. [Figure 18] FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] The following definitions may be used in the description but are not exhaustive.

[0009] Cold-start forecasting: Generating a forecast for a variable for which there is no historical data (or without using any historical data for that variable). Common in the e-commerce domain. For example, a forecast is generated for a new product (such as its sales), but since the product is new, there is no historical data for the product.

[0010] Multivariate forecasting: generating forecasts based on multiple variables taken together and / or together.

[0011] Spatio-temporal forecasting: generating forecasts that include and / or are based on variables that vary in both time and space.

[0012] Distant Supervision is a machine learning / deep learning technique that can be used to learn direct and / or latent relationships between entities or features from differentially distributed data. Distant Supervision in this sense is based on the concept of distant distribution in natural language processing (NLP). In NLP, the central intuition behind distant supervision is that if two entities are related in a knowledge base and / or clean unstructured text data, then all sentences mentioning those two entities are likely to express that relationship. Distant Supervision may involve training a machine learning model with training data that is generated based on existing training data.

[0013] To forecast 5G usage data in areas that currently have no 5G connectivity, a cold-start forecasting approach is necessary as no historical data exists for 5G usage in that area. Cold-start 5G network usage forecasts provide valuable insights that may support informed decision-making, efficient resource allocation, improved service delivery, and enhanced user experience, benefiting various stakeholders in the communications industry.

[0014] In the e-commerce domain, cold-start prediction can be used to predict sales of a new product. For example, past data related to user behavior and related to sales of other similar products (among other data) can be used to predict sales data for a new product. This technique of cold-start prediction in the e-commerce domain is sometimes referred to herein as the first comparison method.

[0015] In the meteorological domain, cold start forecasting can be used to predict weather patterns in an area where no weather data has been previously collected. For example, past data relating to the weather, geographic features, and weather patterns of other areas can be used to predict weather data for a new area. This forecasting can include, for example, using existing overlapping weather patterns of other areas. This technique for cold start weather forecasting is sometimes referred to herein as the second comparison method.

[0016] The approaches used in the first and second comparative methods are not suitable for generating predictions of 5G usage data in areas that do not currently have 5G connectivity, i.e., these approaches do not generate useful predictions when applied to cold start 5G predictions.

[0017] The reason why the approach of the first comparison method is not suitable for 5G forecasting is that the distribution of data in the 5G area is too different from the distribution of data in the non-5G area. That is, the data in the 5G area belongs to a different distribution from the data in the non-5G area. On the other hand, in the first comparison method, there are products with similar past data to the new product and a common user base. The reason for the different distribution in the case of cold start 5G forecasting is thought to be that 5G is too different from other networks (e.g., 4G, 3G, 2G, LTE, etc.).

[0018] The reason why the second comparative method approach is not suitable for 5G forecasting is that the data in the 5G area belongs to a different distribution than the data in the non-5G area, similar to the reason mentioned above. On the other hand, in the second comparative method, the distributions of the data in different geographical areas are more similar to each other (at least in some geographical areas). Another reason is that in the second comparative method, there is not only an overall common weather pattern / background for the whole world (or at least a large area containing several small areas), but also overlapping weather patterns. Such patterns can be learned and used in forecasting. On the other hand, in the cold start 5G forecasting scenario, such overlapping patterns do not exist (or at least there are no useful overlapping patterns).

[0019] The problem of cold start prediction of 5G usage can be explained as follows: D={d 0 , d 1 , d 2 , …, d i , … d n}, where D is a set of time series data distributions from a region with 5G network usage history. d i represents the ith region having spatial information, 5G time series usage history, and non-5G (2G, 3G, 4G, etc. networks) network usage history, where the region is a geographical area having a 5G network infrastructure (e.g., at least 1 Km 2 (area).

[0020] D'={d 0 ', d 1 ', d 2 ', …, d i ', … d n '}, where D' is a set of time series data distributions from regions with non-5G network usage history. d i ' represents the jth region having spatial information and non-5G (2G, 3G, 4G, etc. networks) network usage history, where the region is a geographical area having a non-5G network infrastructure (e.g., at least 1 Km 2 (area).

[0021] where D and D' are different distributions. There is no overlap or direct correlation between D and D' (at least no useful overlap or correlation. For example, there may be overlap or correlation between weather data, but this is not useful). For the purposes of the embodiments disclosed herein, it is possible to find any given jth region (d j The objective of the present invention is to predict 5G time series usage for a 5G network (represented as 5G_U_S_T_P_S_T_1 , 5G_U_S_T_2 , 5G_U_S_T_3 , 5G_U_S_T_4 , 5G_U_S_T_5 , 5G_U_S_T_6 , 5G_U_S_T_7 , 5G_U_S_T_8 , 5G_U_S_T_9 , 5G_U_S_T_10 , 5G_U_S_T_21 , 5G_U_S_T_32 , 5G_U_S_T_43 , 5G_U_S_T_51 , 5G_U_S_T_11 , 5G_U_S_T_22 , 5G_U_S_

[0022] The embodiments disclosed herein include learning time series correlation patterns between different variables of a distribution D, learning a time series correlation pattern between different variables of a given distribution (e.g., d j The method may involve learning the time series patterns of dj', and predicting the 5G usage time series data of dj' using the (distantly) learned correlation patterns of distinct variables of distribution D and the time series patterns of dj'.

[0023] Aspects disclosed herein may include a cold-start prediction approach to indirectly learn patterns from different distributions (i.e., data from non-5G regions and data from 5G regions) and use the learned patterns (distantly) to predict 5G usage data for a given region or regions. Learning patterns from different distributions may be thought of as employing a Distant Supervision based strategy, as described below.

[0024] The data used in the methods disclosed herein can be broadly divided into three categories: non-network data, non-5G network usage data, and 5G network usage data.

[0025] Non-network data may be referred to as "Geographical & Other Data," non-5G network data may be referred to as "Other Network Data" or "Other Networks (2G, 3G, 4G, etc.) Data," and 5G network usage data may be referred to as "5G Network Data." Figure 1 shows a representation of the three data categories.

[0026] Each data category may include any number of variables. In the example representation shown in FIG. 1, the non-network data includes Location details, Traffic Data, Date, Time, Weather Data, Population Density, and Economic Status. In the example representation shown in FIG. 1, the non-5G network usage data includes number of active users (#activeuse), video stream (which may include the number and / or length of video streams and / or the amount of data / bandwidth used for video streaming), Call (IN, OUT) (which may include the number and / or length of calls and / or the amount of data / bandwidth used for calls), SMS (IN, OUT) (which may include the number and / or length of SMS messages and / or the amount of data / bandwidth used for SMS messages), and Internet (which may include the amount of data / bandwidth used for Internet-related processes and / or the amount of data / bandwidth exchanged over the Internet). In the example representation shown in FIG. 1, the 5G network usage data includes the same variables as the non-5G network usage data.

[0027] Call(IN) means incoming call, Call(OUT) means outgoing call, SMS(IN) means incoming SMS, and SMS(OUT) means outgoing SMS. All of these variables use some network bandwidth. The method disclosed here works for any number of other variables that use network bandwidth, such as IOT-based variables that use internet bandwidth.

[0028] As shown in FIG. 1, the data is time series data. The data can be considered to include time series data. Some variables, such as location data, may not change over time. For example, location data may be data that defines the location and / or range of the area to which the data corresponds / from which the data was collected. The variable "time" may simply include the time step to which other variables of the data correspond. Examples of time units that can be used are, for example, 30 minutes, 1.00 hours, etc. The time units can be used to measure / collect data of other variables (e.g., traffic, weather, 5G and other network usage, etc.). The size of the area can be defined by the location variable. Generally, service provider companies divide the coverage area into cells. The size of the cell can be, in particular, 200 square meters, 500 square meters. The area (e.g., target area or reference area) used here can correspond to a cell or be a cell.

[0029] The weather data may include at least one of temperature, wind speed / flow, rainfall frequency, rainfall (amount). The traffic data may include at least one of number of vehicles, average vehicle speed, and average vehicle travel time. The traffic data may include at least one of number / density of vehicles, average vehicle speed, average vehicle travel time, congestion level, number of traffic jams, utilization level of the road network, and maximum capacity (of the road network).

[0030] Non-network data may include infrastructure data, which may include any of the following: number of rail lines, number of bus lines, number of tram lines, distance to the nearest airport, capacity of the nearest airport, average frequency of buses and / or trains and / or trams, etc. It will be understood that some of these variables do not change over time. Non-network data may include demographic data, which may include any of the following: population or population density, economic status of an area or people therein, etc.

[0031] The variables shown in FIG. 1 are not exhaustive or required. In any example, at least one of the variables in each category can be used / presented. Any prediction disclosed herein can include at least one of the variables. The above data is specific to a given region. For example, the variable "number of active users" is the number of active users in a given region (e.g., target or reference). Although not shown in FIG. 1, another data category "combined network usage data" can also be utilized. The combined network usage data is the sum of the 5G usage data and the non-5G network usage data. For example, the number of active users for a given region in the combined network usage data is the number of active users for the region in the non-5G network usage data plus the number of active users for the region in the 5G usage data.

[0032] Non-network usage data includes data usage on non-5G networks, e.g., any 2G, 3G, 4G, LTE, etc. (e.g., (telecommunications) networks installed / used in the area).

[0033] FIG. 2 shows some example data. Table 1 shows example non-5G network usage data, Table 2 shows example 5G usage data, and Table 3 shows example non-network data. In Tables 1 and 2, the data for the "Internet" variable is the amount of data exchanged over the Internet in the region of interest during each time step, excluding data exchanged for video streaming purposes. In general, the amount of data exchanged over the Internet can include data exchanged for any of video streaming, calls, and messages, or can omit any of these amounts of data.

[0034] FIG. 3 is a diagram showing a representation of the training process. That is, FIG. 3 shows a training process for training models used in the prediction process. The training process according to the implementation example includes training each of models 1-6, and is described below. FIG. 3 shows the training data used to train each model, i.e., the training data input to each model. For each model, the dotted box shows the data the model is trained to predict, and the solid box shows the data on which the model's prediction is based. An example of the model architecture is described below.

[0035] In the following description of the training process by the implementation example, the reference region is a region where 5G usage data is available, and the target region is a region where 5G usage data is not available. The reference region can be considered as a region where there is 5G connection (or at least there was 5G connection in the past). The target region can be considered as a region where there is no 5G connection (or where (sufficient) 5G usage data is not available because there was no 5G connection in the past). Of course, it does not necessarily mean that there is no 5G connection in the target region. For example, 5G usage data for the target region may not have been collected or may not be available in the target region. Alternatively, 5G usage data may be available for the target region, but it is desirable to predict the 5G usage data without using historical 5G usage data for the target region. Having and not having 5G connection can be referred to having and not having 5G network infrastructure.

[0036] The training process according to the implementation includes training Model 1 using non-network data of the reference region before time T, non-5G network usage data of the reference region before time T, and 5G usage data of the reference region after time T. Model 1 is trained to predict 5G usage data of the reference region after time T based on the non-network data of the reference region before time T and the non-5G network usage data of the reference region before time T.

[0037] "Until time T" may be used synonymously herein with "before time T." The period referred to as "before time T" may be referred to as a first period. The period referred to as "after time T" may be referred to as a second period.

[0038] That is, training model 1 includes training model 1 to predict 5G usage data of the reference area for a second period based on non-network data of the reference area for a first period and non-5G network usage data of the reference area for the first period.

[0039] A training process according to the example implementation includes training Model 2 using non-network data of the reference region before T and combined network usage data of the reference region after T. Model 2 is trained to predict combined network usage data of the reference region after T based on the non-network data of the reference region before T. That is, training Model 2 includes training Model 2 to predict combined network usage data of the reference region for a second time period based on the non-network data of the reference region for a first time period.

[0040] The training process according to the implementation includes training a model 3 using non-network data of the reference region before T, non-5G network usage data of the reference region before T, and non-5G network usage data of the reference region after T. The model 3 is trained to predict non-5G network usage data of the reference region after T based on the non-network data of the reference region before T and the non-5G network usage data of the reference region before T. That is, training the model 3 includes training the model 3 to predict non-5G network usage data of the reference region for a second time period based on the non-network data of the reference region for a first time period and the non-5G network usage data of the reference region for the first time period.

[0041] The training process according to the implementation includes training model 4 using non-network data of the target area before T, non-5G network usage data of the target area before T, and non-5G network usage data of the target area after T. Model 4 is trained to predict non-5G network usage data of the target area after T based on the non-network data of the target area before T and the non-5G network usage data of the target area before T. That is, training model 4 includes training model 4 to predict non-5G network usage data of the target area for a second time period based on the non-network data of the target area for a first time period and the non-5G network usage data of the target area for the first time period.

[0042] The training process according to the implementation example includes training a model 5 using non-network data of the target region before T and non-network data of the target region after T. The model 5 is trained to predict non-network data of the target region after T based on the non-network data of the target region before T. That is, training the model 5 includes training the model 5 to predict non-network data of the target region in a second time period based on the non-network data of the target region in a first time period.

[0043] The training process according to the implementation includes training the model 6 using non-network data of the reference region after T, non-5G network usage data of the reference region after T, and 5G usage data of the reference region after T. The model 6 is trained to predict 5G usage data of the reference region after T based on the non-network data of the reference region after T and the non-5G network usage data of the reference region after T. That is, training the model 6 includes training the model 6 to predict 5G usage data of the reference region for the second time period based on the non-network data of the reference region for the second time period and the non-5G network usage data of the reference region for the second time period.

[0044] As disclosed herein, Model 1 can be referred to as the sixth model, Model 2 as the fifth model, Model 3 as the second model, Model 4 as the fourth model, Model 5 as the first model, and Model 6 as the third model, as shown below. Model 1-6th Model Model 2-5th Model Model 3 - 2nd Model Model 4 - 4th model Model 5 - 1st Model Model 6 - 3rd Model

[0045] Training any of Models 1-6 may include adjusting at least one network weight of the model (e.g., to make the prediction output by the model equal to or closer to ground truth data). Training any of Models 1-6 may include utilizing a conventional deep learning-based algorithm, e.g., a conventional supervised deep learning algorithm. That is, training any of Models 1-6 may include deep learning-based training.

[0046] This specification discloses a prediction process for predicting 5G usage data, i.e., a prediction process for estimating / predicting 5G usage data in a target area in a future period. For example, the present specification discloses first to third prediction processes.

[0047] FIG. 4 is a diagram illustrating a first prediction process. The first prediction process generates a first predicted 5G usage data of a target area after time T'. The first prediction process includes using a first model (Model 5) to generate a first interim prediction by predicting non-network data of the target area after T' based on non-network data of the target area before T'. The first prediction process further includes using a second model (Model 3) to generate a second interim prediction by predicting non-5G network usage data of the target area after T' based on non-network data of the target area before T' and non-5G network usage data of the target area before T'. The first prediction process further includes using a third model (Model 6) to generate a first 5G usage data prediction by predicting 5G usage data of the target area after T' based on predicted non-network data of the first interim prediction and predicted non-5G network usage data of the second interim prediction.

[0048] "Until time T'" may be used herein synonymously with "before time T'." A period referred to as "before time T'" may be referred to as a past period. A period referred to as "after time T'" may be referred to as a future period.

[0049] That is, the first prediction process includes: A first intermediate forecast is generated by using a first model (Model 5) trained using data for the target geographic region to forecast non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period. A second interim forecast is generated by using a second model (Model 3) trained using data for the reference geographic region to forecast non-5G network usage data for the target geographic region for a future time period based on the non-network data for the target geographic region for a past time period and the non-5G network usage data for the target geographic region for a past time period. A first 5G usage data forecast is generated by using a third model (Model 6) trained using data for the reference geographic region to forecast 5G usage data for the target geographic region for a future time period based on the predicted non-network data of the first interim forecast and the predicted non-5G network usage data of the second interim forecast.

[0050] FIG. 5 is a diagram illustrating a second prediction process. The second prediction process generates a second predicted 5G usage data of the target area after time T'. The second prediction process includes generating a first interim prediction by predicting the non-network data of the target area after T' based on the non-network data of the target area before T' using a first model (Model 5). The second prediction process further includes generating a third interim prediction by predicting the non-5G network usage data of the target area after T' based on the non-network data of the target area before T' and the non-5G network usage data of the target area before T' using a fourth model (Model 4). The second prediction process further includes generating a second 5G usage data prediction by predicting the 5G usage data of the target area after T' based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the third interim prediction using a third model (Model 6).

[0051] That is, the second prediction process includes: A first intermediate forecast is generated by using a first model (Model 5) trained using data for the target geographic region to forecast non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period. A third interim forecast is generated by using a fourth model (Model 4) trained using data for the reference geographic region to forecast non-5G network usage data for the target geographic region for a future time period based on the non-network data and non-5G network usage data for the target geographic region for a past time period. A third model (Model 6) is used to generate a second 5G usage data forecast by forecasting 5G usage data for the target geographic region for a future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast.

[0052] 6 illustrates a third prediction process, which generates a third predicted 5G usage data of a target area after time T'. The third prediction process includes generating a first intermediate prediction by predicting non-network data of the target area after T' based on non-network data of the target area before T' using a first model (Model 5).

[0053] The third prediction process further includes generating a fourth intermediate prediction by predicting combined network usage data of the target area after T' based on non-network data of the target area before T' using a fifth model (Model 2). The third prediction process further includes generating a fifth intermediate prediction by predicting 5G usage data of the target area after T' based on non-network data of the target area before T' and non-5G network usage data of the target area before T' using a sixth model (Model 1).

[0054] The third prediction process further includes combining the fourth and fifth interim predictions to generate a sixth interim prediction, the combining including subtracting predicted 5G usage data for the target area after T' of the fifth interim prediction from the predicted combined network usage data for the target area after T' of the fourth interim prediction to generate the sixth interim prediction.

[0055] The third prediction process further includes generating a third 5G usage data prediction by using a third model (Model 6) to predict 5G usage data for the target area after T' based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the sixth interim prediction.

[0056] That is, the third prediction process includes: A first intermediate forecast is generated by using a first model (Model 5) trained using data for the target geographic region to forecast non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period. A fourth interim forecast is generated by using a fifth model (Model 2) trained using data for the reference geographic region to forecast combined network usage data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period. A fifth interim forecast is generated by using a sixth model (Model 1) trained using data for the reference geographic region to forecast 5G usage data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period and non-5G network usage data for the target geographic region for a past time period. Subtract the predicted 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes predicted non-5G network usage data for the target area for a future time period; and A third model (Model 6) is used to generate a third 5G usage data forecast by forecasting 5G usage data for the target geographic region for a future period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast.

[0057] The prediction process may include generating any one of a first through a third 5G usage data prediction, i.e., any one of a first through a third prediction process. The first intermediate prediction may be generated only once in the prediction process, rather than being generated multiple times.

[0058] The prediction process may include performing at least two of the first through third prediction processes and combining the 5G usage data predictions to generate a final 5G prediction. The final 5G prediction may include an average of the at least two 5G usage data predictions or a prediction range for the 5G usage data.

[0059] For example, considering a prediction process in which two of the first through third prediction processes are performed, the prediction range for the 5G usage data may include, for each variable in the prediction, for each time step, a range of the variable that includes as endpoints the value of that variable at that time step of the two 5G usage data predictions.

[0060] The prediction process can include first through third prediction processes. The 5G usage data prediction range can include, for each variable in the prediction, for each time step, a range of the variable that includes as endpoints the highest and lowest values ​​of that variable at that time step among the first through third 5G usage data predictions.

[0061] For a prediction process including the first to third prediction processes, the prediction range of 5G usage data can be calculated as follows. Figure 7 is a diagram that helps understand how the prediction range of 5G usage data is calculated. For each variable, an average value is calculated at each time step (i.e., the average of the values ​​of that variable at that time step of the first to third prediction processes). Then, among the first to third values ​​of the variable at that time step (corresponding to the first to third prediction processes), the two values ​​closest to the average value are selected as the end points of the prediction range. This process can be performed for each time step and for each variable to generate a prediction range of 5G usage data for each variable as time series data. Figure 7 shows Tables 1 to 3, which respectively include the first to third predicted 5G usage data. Specifically, the values ​​of the variable "internet usage" are shown. Table 4 includes a row for each time step. For a given time step, Table 4 contains in the left column the average of the first three values, in the middle column the smaller of the two values ​​closest to the average among the first three values, and in the right column the larger of the two values ​​closest to the average among the first three values. The forecast range for each time step is the range defined by the "minimum" and "maximum" values, i.e., the values ​​in the middle and right columns. Not all of Table 4 is complete.

[0062] In other words, in an example implementation, combining the first to third predictions includes, for at least one (or each) variable, calculating an average of the predicted values ​​of the variable (at each time step) in the first to third 5G usage data predictions, and selecting two values ​​among the predicted values ​​of the variable (at each time step) that are closest to the corresponding average as the endpoints of the prediction range for the variable.

[0063] FIG. 8 is a diagram showing a prediction process including the first to third prediction processes. According to the above first to third prediction processes, as shown in FIG. 8, this prediction process uses models 1 and 2 to generate a sixth interim prediction (predicted non-5G network usage data of the target area after T'), uses model 3 to generate a second interim prediction (predicted non-5G network usage data of the target area after T'), uses model 4 to generate a third interim prediction (predicted non-5G network usage data of the target area after T'), uses model 5 to generate a first interim prediction (predicted non-network data of the target area after T'), uses model 6 to generate first to third 5G usage data predictions based on the sixth and first interim predictions, the second and first interim predictions, and the third and first interim predictions, respectively, and generates a range of predicted 5G usage data based on the first to third 5G usage data predictions.

[0064] Aspects disclosed herein include methods that include a training process and / or a prediction process, where the training process in such methods may include training only the model that is used in a subsequent prediction process (if included).

[0065] In tests, the prediction process including the first to third prediction processes described with reference to FIG. 7 and combinations thereof achieved 20% higher accuracy than the first prediction process alone, 35% higher accuracy than the second prediction process alone, and 40% higher accuracy than the third prediction process alone.

[0066] The training and / or prediction process disclosed herein may include collecting at least a portion of the training and / or input data (either non-network or network data). The traffic data may include data obtained from sensors in the relevant geographic region. The sensors may include any of the following: at least one vehicle mounted sensor, at least one user equipment, at least one camera, and at least one speed sensor. The weather data may include data obtained from sensors in the relevant geographic region.

[0067] The first period may be, for example, several months (e.g., 1, 6, etc.) or several weeks or years, etc. The second period may be, for example, several months (e.g., 1, 6, etc.) or several weeks or years, etc. The past period may include the first and second periods, for example in a method including a training process followed by a prediction process using the trained model. The time T' may be present or close to the present, or may be in the past. Although the first and second periods are described above as directly following each other, this is not necessary, i.e. instead of "before time T" and "after time T", periods "before time T1" and "after time T2" ​​may be used, such that there is a period between the first and second periods. Corresponding considerations may apply to past and future periods.

[0068] The training process can include using data from multiple target regions and / or multiple reference regions. The prediction process can include predicting 5G data usage for multiple target regions. The prediction process can include predicting 5G data usage for one or more or all of the target regions used in the training process that are used to train a model used in the prediction process.

[0069] FIG. 9 illustrates the architecture of models 1, 3, 4, and 6 in an implementation example. The architecture includes a pair of first bidirectional long short-term memory (BiLSTM) layers (for receiving data on which the associated predictions are based), a concatenation layer, at least one deep neural network (DNN), in this case two (DNN1 and DNN2), a self-attention network, a second BiLSTM layer, and a second at least one DNN (for outputting prediction data). The pair of first BiLSTM layers receives input data. There are pairs of BiLSTM layers because models 1, 3, 4, and 6 are each configured to operate based on two input datasets (e.g., model 1 is configured to receive non-network data of a reference region before T as the first dataset and non-5G network usage data of a reference region before T as the second dataset). Inputs "C-1" and "C-2" represent the two input datasets in FIG. 9.

[0070] The first BiLSTM layer can be thought of as learning the (long-term) dependencies between time steps of each set of input data, which is a time series. Each layer performs additive interactions, which helps to improve the gradient flow over long sequences during training. Each layer contains a number of BiLSTM blocks. The function of a BiLSTM layer is defined as follows: h_t = LSTM_forward(x_t,h_{t-1}) g_t=LSTM_backward(x_t,g_{t+1}) y_t = f(concatenate(h_t,g_t)) Where: Let x_t represent the input at time step t. h_t represents the hidden state of the forward LSTM at time step t. g_t represents the hidden state of the backward LSTM at time step t. h_{t-1} represents the hidden state of the forward LSTM at the previous time step t-1. g_{t+1} represents the hidden state of the backward LSTM at the next time step t+1.

[0071] LSTM_forward and LSTM_backward represent the forward and backward LSTM functions, respectively. "concatenate" is the operation of concatenating the outputs of the forward and backward LSTMs. "f'" is the activation function that transforms the concatenated output into the final output y_t.

[0072] Bidirectional LSTM processes the input sequence x_t from left to right in the forward LSTM and from right to left in the backward LSTM. The final output y_t is passed to the next layer.

[0073] The concatenation layer takes the inputs (the outputs from the first BiLSTM layer) and concatenates them along a specified dimension. That is, the concatenation layer concatenates the outputs of the first BiLSTM layer (based on the two input datasets) to generate a concatenation with the appropriate dimensionality for the next stage.

[0074] The first DNN learns relationships and correlations in the data and adds nonlinearity to the data. The function of the DNN is defined as:

[0075] DNN (Deep Neural Network) layer refers to a layer in a deep neural network architecture. A typical DNN layer can be mathematically represented as follows: y=f(Wx+b) Where: x represents the input to the layer, which can be the output of a previous layer or the input data itself. W represents the weight matrix, which contains the learnable parameters of the layer that determine the strength of the connections between neurons. b represents the bias vector, which provides an offset to the weighted sum of the inputs. f is an activation function that introduces nonlinearity into the output of the layer. Common activation functions include ReLU (Rectified Linear Unit), sigmoid, or tanh.

[0076] A DNN layer takes an input x, performs a linear transformation by multiplying it by a weight matrix W, and adds a bias vector b. It then passes the resulting weighted sum through an activation function f to compute the layer's output y, which is passed to the next layer.

[0077] The self-attention layer focuses on "important" parts of the data based on correlations calculated between parts of the input data. The self-attention mechanism allows the model to focus on different parts of the input sequence based on their relevance to each other. By attending to relevant context, the model can better capture long-term dependencies, improving its ability to generate accurate predictions or representations.

[0078] The second BiLSTM layer learns the dependencies of the changed data output from the self-attention layer. The output of each BiLSTM block of the second BiLSTM layer is obtained as the output of the second BiLSTM layer.

[0079] The second DNN makes predictions based on the output of the second BiLSTM layer. The second DNN implements a time-distributed wrapper to extract predictions at the relevant time and the relevant time step. The time-distributed wrapper works as follows:

[0080] The time-distributed dense layer is used as a time-distributed wrapper. The time-distributed dense layer is particularly useful when processing variable-length sequences or when applying a dense layer to each time step of a sequence separately. Mathematically, the time-distributed dense layer can be expressed as follows: Input shape: (batch_size,time_steps,input_dim) Applying a time-distributed dense layer: output=TimeDistributed(Dense(units))(input) Output shape: (batch_size,time_steps,units)

[0081] where batch_size represents the number of sequences in a batch, time_steps is the length of each sequence, input_dim is the dimensionality of the input at each time step, and units is the number of units or neurons in the dense layer.

[0082] The time-variant dense layer reshapes the input tensor to (batch_size*time_steps,input_dim), applies a standard dense layer with unit neurons, and then reshapes the output to (batch_size,time_steps,units).

[0083] The use of time-distributed dense layers allows the network to learn different weights at each time step, capturing temporal patterns and dependencies in the sequence, which is particularly useful when the relationship between the input and output varies at different time steps.

[0084] FIG. 10 illustrates the architecture of model 5 in an implementation example. The architecture includes a first bidirectional long short-term memory (BiLSTM) layer (for receiving data on which the associated predictions are based), at least one DNN, in this case two deep neural networks (DNNs) (DNN1 and DNN2), a self-attention network, a second BiLSTM layer, and a second at least one DNN (for outputting prediction data). The first pair of BiLSTM layers receives input data. Model 5 is configured to receive one input data set, so the architecture of model 5 includes only the first BiLSTM layer. Input "C-1" represents the input data set in FIG. 10. Each stage of the architecture operates in a similar manner to that described above with reference to FIG. 9, so a duplicated description will be omitted. The main difference between the architecture of FIG. 10 and the architecture of FIG. 9 is that FIG. 10 has only one first BiLSTM layer and therefore no concatenated layer.

[0085] FIG. 11 illustrates the architecture of model 2 in an implementation example. The architecture includes a first bidirectional long short-term memory (BiLSTM) layer (for receiving data on which relevant predictions are based), at least one DNN, in this case two deep neural networks (DNNs) (DNN1 and DNN2), a recurrent vector layer, a pair of self-attention networks, a second BiLSTM layer pair, and a second at least one DNN pair (for outputting prediction data). The first BiLSTM layer pair receives input data. Model 2 is configured to receive one input dataset, so the architecture of model 2 includes only the first BiLSTM layer. Input "C-1" represents the input dataset of FIG. 22.

[0086] The recurrent vector layer replicates the output of the first DNN so that two processing branches are executed. Each branch includes a self-attention network / layer, a second BiLSTM layer, and a second at least one DNN. Model 2 is configured to output two prediction data sets, so the architecture includes two processing branches. Of course, the two prediction data sets can be considered as being combined so that the model is finally configured to output one prediction data set (combined network usage).

[0087] Each stage of the architecture operates similarly to that described above with reference to Figure 9, so duplicated descriptions will be omitted. The main differences between the architecture of Figure 11 and the architecture of Figure 9 are that there is only one first BiLSTM layer in Figure 11, and therefore no connection layer, and the architecture of Figure 11 includes two parallel branches: a recurrent vector layer and a subsequent stage.

[0088] It will be appreciated that the above architecture is not required and variations in the architecture may be used.

[0089] It can be said that the first to sixth models are encoder-decoder (network) models, and / or the first to sixth models each include a deep neural network (DNN), and / or the first to sixth models are self-attention based models, and / or the first to sixth models are self-attention based encoder-decoder (network) models, and / or the first to sixth models each include a self-attention network, and / or the first to sixth models each include a BiLSTM layer. It can be said that the first to sixth models each include a first BiLSTM layer, a first at least one deep neural network (DNN), a self-attention network, a second BiLSTM layer, and a second at least one DNN.

[0090] 12-17 are diagrams corresponding to Models 1-6, respectively, showing example implementation architectures used for each model and example data input to and output from the models. The input and output data each include N records and the same time steps, but this is not required. That is, the input and output data may have different numbers of records and / or time steps, which may also vary between models.

[0091] The prediction process can predict any number of variables and does not require predicting all variables in the input data.

[0092] The training and prediction process can in some aspects be considered to be divided into stages 1, 2, and 3.

[0093] Stage 1 includes training models 1 to 6. That is, if "A" represents non-network data, "B" represents non-5G network usage data, and "C" represents 5G usage data, then stage 1 includes the following for at least one reference region: Case 1: Use the correlation between “A” and “B” and their time-spatial relationship up to time “T” to predict “C” from time “T”+1 to time “T”+“K” (“K”>1) (Model 1). Case 2: Use “A” and its temporal-spatial relationship up to time “T” to predict “B” and “C” from time “T”+1 to time “T”+“K” (“K”>1) (Model 2). Case 3: Use the correlation between "A" and "B" and the time-spatial relationship up to time "T" to predict "B" from time "T"+1 to time "T"+"K" ("K">1) (Model 3) Case 4: Use the correlation between “A” and “B” and the time-spatial relationship from time “T”+1 to “T”+“K” to predict “C” from time “T”+1 to time “T”+“K” (“K”>1) (Model 6).

[0094] Stage 1 involves, for at least one target region: Case 5: Use correlation between "A" and "B" and time-spatial relationship up to time "T" to predict "B" from time "T"+1 to time "T"+"K" ("K">1) (Model 4) Case 6: Using time series data 'A' up to time 'T', predict 'A' from time 'T'+1 to time 'T'+'K' ('K'>1) (Model 5).

[0095] Stage 2 involves generating an intermediate prediction, i.e., using the target domain data, includes: Case A: Prediction of “A” from time T’+1 to T’+K (first intermediate prediction) using the model trained in “Case 6”. Case B: Using the model trained in “Case 3”, predict “B” from time T’+1 to T’+K (second intermediate prediction). Case C: Prediction of “B” from time T’+1 to T’+K (third intermediate prediction) using the model trained in “Case 5”. Case D: In this case, the difference between two forecasts (i.e., obtained using the models trained in Cases 1 and 2) is used to obtain a forecast for “B” from time T’+1 to T’+K (the sixth intermediate forecast).

[0096] Stage 3 includes generating first through third 5G usage data forecasts and forecast ranges. Stage 3 includes: Prediction 1: Using the results of “Case A” and “Case B” and the predictive model trained on “Case 4”, predict the usage of “C” from time T’+1 to T’+K. Prediction 2: Using the results of “Case A” and “Case C” and the predictive model trained on “Case 4”, predict the usage of “C” from time T’+1 to T’+K. Prediction 3: Using the results of “Case A” and “Case D” and the predictive model trained on “Case 4”, predict the usage of “C” from time T’+1 to T’+K. Final Prediction: Use the minimum and maximum (or closest to the average) of the predicted usage of "C" from time T'+1 to T'+K as the final cold start prediction.

[0097] Methods according to some aspects of the present disclosure can be described as follows. 1) Type 1 (reference area) input data is collected from cities that have 5G connections and have other non-5G connections (2G, 3G, 4G, LTE, etc.) as well as other geographic information, weather information, bus, train, air transport status, and traffic details, etc. 2) Type 2 (target area) input data is collected from cities that do not have 5G connectivity but have non-5G connectivity (2G, 3G, 4G, etc.) as well as other geographic information, weather information, bus, train, air transport status, and traffic details, etc. 3) Distant Supervision-based Pattern Extraction for Time Series Forecasting. We use six methods to collect different types of time series pattern relationships: a) An encoder-decoder network is used on Type 1 data to learn the relationship between (a) “geographical and other data + non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.)”)” and (b) 5G usage data, and use “a” to predict “b” (5G usage). After training, the trained model is saved as “Model 1”. b) Use an encoder-decoder network on the Type 1 data to learn the relationship between (a) “geographic and other data” and (b) “5G usage data + non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.)”)” and use “a” to predict “b” (5G usage). After training, the trained model is saved as “Model 2”. c) Use an encoder-decoder network on the Type 1 data to learn the relationship between (a) “geographic and other data + non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.)”)” and (b) “next time series of non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and use “a” to predict “b” (non-5G usage). After training, the trained model is saved as “Model 3”. d) Use an encoder-decoder network on the Type 2 data to learn the relationship between (a) “geographic and other data + non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and (b) “next time series of non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and use “a” to predict “b” (non-5G usage). After training, the trained model is saved as “Model 4”. e) Use an encoder-decoder network on the Type 2 data to learn the relationship between (a) the “geographic and other data” and (b) the “next time series of non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and predict “b” using “a”. After training, the trained model is saved as “Model 5”. f) Use an encoder-decoder network on the Type 1 data to learn the relationship between (a) “geographic and other data + non-5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and (b) “the next time series of 5G usage data (i.e., other networks (2G, 3G, 4G, LTE, etc.))” and predict “b” using “a”. After training, the trained model is saved as “Model 6”. 4) In the next stage, the learned time series pattern relationships are used for cold-start prediction, which is 5G data usage prediction for areas of non-5G cities (i.e., cities where 5G networks are not available). 5) Use the historical time series data from the non-5G service area and pass it to a model trained on (a) time series data belonging to the 5G service area and (b) time series data belonging to the non-5G service area. This gives us a time series forecast of the non-5G data according to the trend of cities with 5G service and a time series forecast of the non-5G data according to the trend of cities without 5G service. The forecasted data contains very useful relationships. 6) Use "Model 6" to predict the time series geographic data for the "Non-5G Service Areas". Use the predicted non-5G data usage data (as described above) and concatenate it with the time series geographic data for the "Non-5G Service Areas". For all combinations, use "Model 6" to predict 5G data usage for the non-5G service areas. Finally, the model identifies minimum and maximum 5G data usage as a range for the predicted 5G data usage for the "Non-5G Service Areas".

[0098] Aspects disclosed herein may take into account the use of multivariate time series forecasting. In general, multivariate time series forecasting can be described as follows.

[0099] Assumptions: Complex relationships between multiple time series are assumed.

[0100] To highlight the relationships between multiple time series, the problem of multivariate time series forecasting can be formulated based on a data structure called a multivariate time graph (which can be chosen as a case of non-Euclidean learning), which can be expressed as:

number

number

number

number

number

[0101] The Distant Supervision approach can be understood as follows: Traditionally, most machine learning methods require training data. A common approach to collect training data is for humans to label the data. For example, in the case of marriage relationships (e.g., when a model is trained to extract said relationships), a human annotator can label the pair "David Beckham" and "Victoria Beckham" as positive training examples. This approach is costly and does not generate enough data for some ML algorithms, even when the data corpus is large. Moreover, the resulting training data may be noisy due to human error.

[0102] An alternative approach to generating training data, called "Distant Supervision", uses an existing database to collect examples of the relationships to be extracted, and then uses these examples to automatically generate training data. For example, consider a scenario in which there is a database containing the fact that Barack Obama and Michelle Obama are married, then each pair of "Barack Obama" and "Michelle Obama" that appears in the same sentence in any dataset used to train the ML model would be labeled as a positive example of a married relationship. Thus, a large amount of (possibly noisy) training data can be generated. While applying Distant Supervision to obtain positive training examples is considered relatively easy, generating negative examples is more challenging.

[0103] Aspects disclosed herein can be thought of as applying Distant Supervision techniques in the sense of learning multiple important relationships from diverse source data, which may be useful in forecasting 5G data usage demand in target areas with different distributions (i.e., without 5G infrastructure and without 5G usage history). These learned relationships include various aspects such as, but are not limited to, the following: (a) The correlation between geography, non-5G data usage, and 5G data usage, revealing how these factors affect each other. (b) The impact of geographic data on 5G usage demand, which provides insight into how geographic factors shape demand patterns. (c) The impact of geographic data on non-5G usage demand. This highlights the connection between geographic factors and non-5G data consumption. (d) Impact of non-5G network presence on 5G data usage. This shows how the availability of non-5G infrastructure affects 5G data usage. (e) The relationship between changes in geographic characteristics and overall usage requirements for both 5G and non-5G data, which shows how changes in geographic characteristics will affect demand for data services. (f) Other correlations between geographic data characteristics, non-5G data usage, and 5G data usage, which capture additional interconnection patterns between these variables.

[0104] By integrating these relationship patterns and correlations with changes in geographic infrastructure and non-5G usage, we can gain a clearer understanding of the expected demand for 5G data usage in the near future in target areas where no 5G infrastructure and usage history exists.

[0105] Aspects disclosed herein can also be applied to other domains, such as e-commerce, to forecast sales of new products (e.g., products for which data on similar predecessor products is not available). That is, aspects can use Distant Supervision to learn distant time series patterns from different e-commerce providers for sales of different products (and user bases, etc.), and use the distantly learned time series patterns to forecast time series of new products on different e-commerce platforms.

[0106] Aspects disclosed herein can include Distant Supervision based cold-start, multi-step, multivariate, and spatio-temporal time series prediction. Distant Supervision is used to learn distant patterns of time series prediction from different distributions. The system uses the distantly learned patterns for cold-start prediction of differently distributed time series data.

[0107] Aspects disclosed herein include aspects of cold-start spatiotemporal, multivariate, and multi-step time series forecasting. According to one aspect of the invention, the system gives direct forecast values ​​and ranges of minimum and maximum values ​​as forecasts (wherein for each variable and each time step, the minimum and maximum values ​​are given). The source dataset (i.e., used to train the time series forecasting system) and the target dataset (used for time series forecasting) may belong to different distributions. Due to the different distributions, common / useful patterns cannot be obtained a priori using traditional strategies. The concept of Distant Supervision can be applied to learn useful patterns (useful for time series forecasting) distantly. Thus, the system can be considered to present a true cold-start forecasting scenario.

[0108] In other words, the constraints in the scenario considered here indicate that the training data belongs to multiple different distributions for areas with 5G service, and the test data belongs to a different distribution (without 5G service). A distant supervision based strategy is used to indirectly learn patterns from the different distributions so that they can be used for 5G time series forecasting in the different distributions. The system uses the distant learning patterns to forecast 5G requirements at the cell level (cells are a grid-type structure used by mobile operators to divide the entire geographical area into smaller sub-areas to efficiently manage their networks).

[0109] As mentioned above, conventional methods include the first comparison method and the second comparison method. These methods perform cold start predictions, such as sales of a new product (in a previously defined domain or in any new domain), or weather information for a new area. However, these methods use a common distribution for all predictions. With this configuration, all such sales prediction methods benefit from trends in sales, user behavior, etc. Similarly, for weather prediction, such systems use many common features (i.e., common to both existing and cold start areas), other weather-related ground realities, and information, etc. Such approaches do not work (to achieve cold start predictions) when training and testing time series data using different time series data distributions. In the proposed scenario (5G predictions in areas without 5G usage data (history)), there are no (useful) dependencies between features, common metadata, and common background information.

[0110] Aspects disclosed herein can include using Distant Supervision to identify useful patterns from different data distributions (i.e., source training data sets belonging to different domains with minimal overlap of features and target test data sets belonging to different domains) for cold-start multi-stage, multivariate, and spatio-temporal time series prediction. Some aspects disclosed herein can perform cold-start, multi-stage, multivariate, and spatio-temporal time series prediction for data associated with new distributions. Some aspects disclosed herein can effectively predict cold-start time series data in domains that do not have a history of that event and that belong to a different distribution (different from the distribution used to train the system). Distant Supervision-based approaches can be applied to many different time series applications.

[0111] 18 is a block diagram of an information processing appliance 10 or computing device 10, such as a data storage server, that may be used to embody the present invention, to perform some or all of the operations of the methods embodying the present invention, and to perform some or all of the tasks of the apparatus of the embodiments. The computing device 10 may be used to implement any of the method steps or processes described above.

[0112] Computing device 10 includes a processor 993 and a memory 994. Optionally, the computing device also includes a network interface 997 for communicating with other such computing devices, such as other computing devices of embodiments of the present invention. Optionally, the computing device also includes one or more input mechanisms 996, such as a keyboard and mouse, and a display unit 995, such as one or more monitors. These elements may facilitate user operation. The components may be connected to each other via a bus 992.

[0113] The memory 994 may include a computer-readable medium, which term may refer to a single medium or multiple media (e.g., centralized or distributed databases and / or associated caches and servers) configured to carry computer-executable instructions. Computer-executable instructions may include, for example, instructions and data that are accessible by a computer (e.g., one or more processors) and cause it to perform one or more functions or operations. For example, computer-executable instructions may include instructions for performing any of the method steps or processes described above. Thus, the term "computer-readable storage medium" may include any medium that can store, encode or carry a set of instructions for execution by a machine and cause the machine to perform any of the method steps or processes described above. For example, without limitation, the term "computer-readable storage medium" may include solid-state memory, optical media, and magnetic media. By way of example, and not limitation, such computer-readable media may include non-transitory computer-readable media including random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory devices (e.g., solid-state memory devices).

[0114] The processor 993 is configured to control the computing device and perform processing operations such as executing computer program code stored in the memory 994 to perform any of the method steps or processes described above. The memory 994 stores data that is read and written by the processor 993 and can store the training data and / or test data and / or network weights and / or intermediate predictions and / or predictions and / or values ​​and / or other data and / or programs for performing any of the method steps or processes described above. As referred to herein, a processor may include one or more general-purpose processing devices such as a microprocessor, a central processing unit, etc. A processor may include a processor that implements a Complex Instruction Set Computing (CISC) microprocessor, a Reduced Instruction Set Computing (RISC) microprocessor, a Very Long Instruction Word (VLIW) microprocessor, or other instruction set or processor that implements a combination of instruction sets. A processor may also include one or more special-purpose processing devices such as an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), a network processor, etc. In one or more embodiments, the processor is configured to execute the operations and instructions for performing the operations discussed herein. The processor 993 may be considered to include any of the modules described above. Any operation described as being performed by a module may be implemented as a method by a computer, such as the processor 993.

[0115] The display unit 995 may display representations of data stored by the computing device, such as training data and / or predictions and / or GUI windows and / or interactive representations that allow a user to interact with the device 10, for example by drag-and-drop or selection interactions, and / or any other output described above, and may display cursors and dialog boxes and screens that allow interaction between a user and programs and data stored on the computing device. The input mechanism 996 may enable a user to input data and instructions into the computing device, such as allowing a user to enter any of the user inputs described above.

[0116] The network interface (network I / F) 997 can be connected to a network such as the Internet, and can be connected to other such computing devices via the network. The network I / F 997 can control input and output data to and from other devices via the network.

[0117] Other peripheral devices such as microphones, speakers, printers, power supply units, fans, cases, scanners, tracker balls, etc. may also be included in the computing device.

[0118] The method / process embodying the present invention may be performed on a computing device / apparatus 10 as shown in FIG. 18. Such a computing device need not have all the components shown in FIG. 18, but may be comprised of a subset of these components. For example, the device 10 includes a processor 993 and a memory 994 coupled to the processor 993. Alternatively, the device 10 includes a processor 993, a memory 994 coupled to the processor 993, and a display 995. The method embodying the present invention may be performed by a single computing device in communication with one or more data storage servers over a network. The computing device may be a data storage device that stores at least a portion of the data.

[0119] Methods / processes embodying the invention may be performed by a number of computing devices operating in cooperation with one another, one or more of which may be data storage servers that store at least a portion of the data.

[0120] The invention can be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations of them. The invention can be implemented as a computer program or computer program product, i.e. a computer program embodied in a non-transitory information carrier, for example a machine-readable storage device or a propagated signal, for execution by or controlling the operation of one or more hardware modules.

[0121] A computer program may be in the form of a stand-alone program, computer program portions, or multiple computer programs, may be written in any type of programming language, including compiled or interpreted languages, and may be deployed in any form, such as as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a data processing environment. A computer program may be deployed to be executed on one module or on multiple modules that are located at one site or distributed across multiple sites and interconnected by a communication network.

[0122] The steps of the methods of the present invention may be performed by one or more programmable processors executing computer programs to perform the functions of the present invention by operating on input data and generating output. The apparatus of the present invention may be implemented as programmed hardware or as special purpose logic circuitry including Field Programmable Gate Arrays (FPGAs) or Application-Specific Integrated Circuits (ASICs), etc.

[0123] Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Typically, a processor receives instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a processor for executing instructions coupled to one or more memories that store instructions and data.

[0124] The above-described embodiments of the invention may be advantageously used independently of any other embodiment or in any feasible combination with one or more other embodiments of the embodiments.

[0125] The present disclosure extends to the following statements:

[0126] (S1) 1. A computer-implemented method comprising: executing a prediction process to predict 5G usage data for a target geographic area (having no 5G connection / no historical 5G usage data exists or is available), the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data of the target geographic region for a future time period based on non-network data of the target geographic region for a past time period using a first model trained using data of the target geographic region (to predict non-network data of the second time period based on non-network data of the first time period prior to the second time period); generating a second intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a second model trained using data of at least one reference geographic region (which has or had a 5G connection) (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the first 5G usage data prediction by predicting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the second interim prediction using a third model trained using data of the at least one reference geographic region (to predict 5G usage data for the second time period based on non-network data and non-5G network usage data for the second time period); Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using data for the target geographic region (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating the fourth interim prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data of the at least one reference geographic region (to predict combined network usage data for the second time period based on non-network data for the first time period, the combined network usage data including usage data related to 5G and non-5G networks); generating the fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region (to predict 5G usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The computer-implemented method, wherein the non-network data includes any of location data, (geographic data), demographic data, weather data, infrastructure data, and (vehicle) traffic data.

[0127] (S2) The computer-implemented method of S1, wherein the non-network data includes a population or population density of the region of interest.

[0128] (S3) The computer-implemented method of S1, wherein the non-network data includes a population or population density of the region of interest and location data indicating a location and extent of the region of interest.

[0129] (S4) The computer-implemented method of any of S1 to S3, wherein the non-network data, the 5G usage data, and the non-5G network usage data include time series data.

[0130] (S5) The computer-implemented method according to any one of S1 to S4, wherein the weather data includes at least one of temperature, wind speed / volume, rainfall frequency, and rainfall amount.

[0131] (S6) The computer-implemented method of any of S1-S5, wherein the traffic data includes at least one of the following: number / density of vehicles, average speed of vehicles, average travel time of vehicles, level of congestion, number of traffic congestions, level of utilization of the road network, and maximum capacity (of the road network).

[0132] (S7) The computer-implemented method of any of S1-S6, wherein the demographic data includes at least one of population / population density and economic background.

[0133] (S8) The computer-implemented method of any of S1-S7, wherein the infrastructure data includes at least one of: a number of rail lines, a number of bus lines, a number of tram lines, a capacity of a nearest airport, a distance to a nearest airport, a frequency of bus service, a frequency of train service, and a frequency of tram service.

[0134] (S9) The computer-implemented method of any of S1 to S8, wherein the position data includes information indicating a position and / or extent of the region of interest.

[0135] (S10) The computer-implemented method of any of S1 to S9, wherein the geographical data includes information indicative of a position and / or extent of the region of interest and / or information indicative of geographical features of the region of interest.

[0136] (S11) The computer-implemented method of any of S1 to S10, wherein the 5G usage data includes values ​​of at least one variable over time, the at least one variable including any of the following: number of active users over time; number and / or length of video streams over time; and / or amount of data / bandwidth used for video streaming over time; number and / or length of calls over time; and / or amount of data / bandwidth used for calls over time; number and / or size of SMS messages over time; and / or amount of data / bandwidth used for SMS messages over time; Internet usage over time; and / or amount of data / bandwidth used for Internet-related processing over time; and / or amount of data / bandwidth exchanged over the Internet over time.

[0137] (S12) The computer-implemented method of any of S1 to S11, wherein the non-5G network usage data includes values ​​of at least one variable over time, the at least one variable including any of the following: number of active users over time; number and / or length of video streams over time; and / or amount of data / bandwidth used for video streaming over time; number and / or length of calls over time; and / or amount of data / bandwidth used for calls over time; number and / or size of SMS messages over time; and / or amount of data / bandwidth used for SMS messages over time; Internet usage over time; and / or amount of data / bandwidth used for Internet-related processing over time; and / or amount of data / bandwidth exchanged over the Internet over time.

[0138] (S13) The computer-implemented method of any of S1-S12, wherein the combined network usage data includes values ​​of at least one variable over time, the at least one variable including any of: number of active users over time; number and / or length of video streams over time; and / or amount of data / bandwidth used for video streaming over time; number and / or length of calls over time; and / or amount of data / bandwidth used for calls over time; number and / or size of SMS messages over time; and / or amount of data / bandwidth used for SMS messages over time; Internet usage over time; and / or amount of data / bandwidth used for Internet related processing over time; and / or amount of data / bandwidth exchanged over the Internet over time.

[0139] (S14) The computer-implemented method of any of S1-S13, wherein the non-5G network usage data includes usage data of at least one non-5G communication network.

[0140] (S15) The computer-implemented method of any of S1-S14, wherein the non-5G network usage data includes usage data of one or more non-5G communication networks present in the region of interest.

[0141] (S16) The computer-implemented method of any of S1-S15, wherein the non-5G network usage data includes usage data of at least one non-5G network among a 2G, 3G, 4G, and LTE network.

[0142] (S17) The computer-implemented method of any of S1-S16, wherein the predictive 5G usage data includes a predicted value of at least one variable over time, the at least one variable including any of: number of active users over time; number and / or length of video streams over time; and / or amount of data / bandwidth used for video streaming over time; number and / or length of calls over time; and / or amount of data / bandwidth used for calls over time; number and / or size of SMS messages over time; and / or amount of data / bandwidth used for SMS messages over time; Internet usage over time; and / or amount of data / bandwidth used for Internet related processing over time; and / or amount of data / bandwidth exchanged over the Internet over time.

[0143] (S18) The computer-implemented method of any of S1-S17, wherein the prediction process includes generating at least two of the first through third 5G usage data predictions and combining the at least two 5G usage data predictions to generate a final 5G prediction.

[0144] (S19) The computer-implemented method of S18, wherein combining the at least two 5G usage data forecasts includes calculating an average 5G usage data forecast (including calculating an average of each variable of the at least two 5G usage data forecasts).

[0145] (S20) The computer-implemented method of any of S1-S19, wherein the prediction process includes generating at least two of the first through third 5G usage data predictions and combining the at least two 5G usage data predictions to generate a prediction range for 5G usage data.

[0146] (S21) The computer-implemented method of any of S1-S20, wherein the prediction process includes generating the first to third 5G usage data predictions and combining the first to third 5G usage data predictions to generate a final 5G prediction.

[0147] (S22) The computer-implemented method of S21, wherein combining the first through third 5G usage data forecasts includes calculating an average 5G usage data forecast (including calculating an average of each variable of the first through third 5G usage data forecasts).

[0148] (S23) A computer-implemented method as described in any of S1 to S22, wherein the prediction process includes steps of generating the first to third 5G usage data predictions and combining the first to third 5G usage data predictions to generate a prediction range of 5G usage data.

[0149] (S24) The step of combining the first to third 5G usage data forecasts to generate a final 5G forecast may include, for at least one variable (of the first to third 5G usage data forecasts): calculating an average of the predicted values ​​of the variables (at each time step) in the first to third 5G usage data forecasts; selecting the two predicted values ​​of said variables (at each time step) that are closest to said (corresponding) mean values ​​as the endpoints of a prediction range for said variables; The computer-implemented method of S21 or S23, comprising:

[0150] (S25) the first model is trained based on non-network data for the target geographic region for a first time period and non-network data for the target geographic region for a second time period prior to the first time period (to predict non-network data for the target geographic region for the second time period based on non-network data for the target geographic region for the first time period); and / or the second model is trained based on non-5G network usage data for the at least one reference geographic region for the second time period and based on non-network data and non-5G network usage data for the at least one reference geographic region for the first time period (to predict non-5G network usage data for the at least one reference geographic region for the second time period based on non-network data and non-5G network usage data for the at least one reference geographic region for the first time period); and / or The third model is trained based on 5G usage data, non-network data, and non-5G network usage data for the at least one reference geographic region for the second time period (to predict 5G usage data for the at least one reference geographic region for the second time period based on the non-network data and non-5G network usage data for the at least one reference geographic region for the second time period); and / or the fourth model is trained based on non-5G network usage data for the target geographic region for the second time period and based on non-network data and non-5G network usage data for the target geographic region for the first time period (to predict non-5G network usage data for the target geographic region for the second time period based on non-network data and non-5G network usage data for the target geographic region for the first time period); and / or the fifth model is trained based on the combined network usage data of the at least one reference geographic region for the second time period and based on non-network data of the at least one reference geographic region for the first time period (to predict the combined network usage data of the at least one reference geographic region for the second time period based on the non-network data of the at least one reference geographic region for the first time period); and / or The computer-implemented method of any of S1 to S24, wherein the sixth model is trained based on 5G usage data of the at least one reference geographic region for the second time period and based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period (to predict the 5G usage data of the at least one reference geographic region for the second time period based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period).

[0151] (S26) The computer-implemented method further comprises performing a training process prior to performing the prediction process; The computer-implemented method of any of S1 to S25, wherein the training process includes training at least one of the first to sixth models.

[0152] (S27) The training process includes: training the first model to predict non-network data of the target geographic region for the second time period based on the non-network data of the target geographic region for the first time period and non-network data of the target geographic region for a second time period prior to the first time period; training the second model to predict non-5G network usage data of the at least one reference geographic region for the second time period based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period, and based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period; training the third model to predict 5G usage data for the at least one reference geographic region for the second time period based on 5G usage data, non-network data, and non-5G network usage data for the at least one reference geographic region for the second time period based on the non-network data and the non-5G network usage data for the at least one reference geographic region for the second time period; training the fourth model based on the non-5G network usage data of the target geographic region for the second time period and based on the non-network data and the non-5G network usage data of the target geographic region for the first time period to predict the non-5G network usage data of the target geographic region for the second time period based on the non-network data and the non-5G network usage data of the target geographic region for the first time period; training the fifth model to predict combined network usage data for the at least one reference geographic region for the second time period based on combined network usage data for the at least one reference geographic region for the second time period and based on non-network data for the at least one reference geographic region for the first time period, wherein the combined network usage data includes usage data related to 5G and non-5G networks; training the sixth model to predict the 5G usage data of the at least one reference geographic region for the second time period based on the 5G usage data of the at least one reference geographic region for the second time period and based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period; The computer-implemented method of S26, comprising:

[0153] (S28) The computer-implemented method of any of S1 to S27, wherein the first to sixth models are / include encoder-decoder (network) models.

[0154] (S29) The computer-implemented method of any of S1 to S28, wherein the first to sixth models each include a deep neural network (DNN).

[0155] (S30) The computer-implemented method of any of S1 to S29, wherein the first to sixth models are / include self-attention-based models.

[0156] (S31) The computer-implemented method of any of S1 to S30, wherein the first to sixth models are / include self-attention-based encoder-decoder (network) models.

[0157] (S32) The computer-implemented method of any one of S1 to S31, wherein the first to sixth models each include a self-attention network.

[0158] (S33) The computer-implemented method of any of S1 to S32, wherein the first to sixth models each include a (bidirectional) long short-term memory (LSTM) layer.

[0159] (S34) The computer-implemented method of any of S1 to S33, wherein the first to sixth models each include a first (bidirectional) long short-term memory (LSTM) layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN.

[0160] (S35) The computer-implemented method of any of S1 to S34, wherein the first model includes a first (bidirectional) long short-term memory (LSTM) layer (for receiving data on which the prediction of interest is based), a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting prediction data).

[0161] (S36) The computer-implemented method of any of S1 to S35, wherein the second model includes a first (bidirectional) pair of long short-term memory (LSTM) layers (for receiving data on which the prediction of interest is based), a concatenation layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting prediction data).

[0162] (S37) The computer-implemented method of any of S1 to S36, wherein the third model includes a first (bidirectional) pair of long short-term memory (LSTM) layers (for receiving data on which the prediction of interest is based), a concatenation layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting prediction data).

[0163] (S38) The computer-implemented method of any of S1 to S37, wherein the fourth model includes a first (bidirectional) pair of long short-term memory (LSTM) layers (for receiving data on which the prediction of interest is based), a concatenation layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting prediction data).

[0164] (S39) The computer-implemented method of any of S1 to S38, wherein the fifth model includes a first (bidirectional) long short-term memory (LSTM) layer (for receiving data on which the prediction of interest is based), a first at least one deep neural network (DNN), a recurrent vector layer, a pair of self-attention networks, a second pair of LSTM layers, and a second at least one pair of DNNs (for outputting prediction data).

[0165] (S40) The computer-implemented method of any of S1 to S39, wherein the sixth model includes a first (bidirectional) pair of long short-term memory (LSTM) layers (for receiving data on which the prediction of interest is based), a concatenation layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting prediction data).

[0166] (S41) The computer-implemented method of any of S1 to S40, wherein each of the second, third, fourth, and sixth models includes a first (bidirectional) pair of long short-term memory (LSTM) layers (for receiving data on which a prediction of interest is based), a concatenation layer, a first at least one deep neural network (DNN), a self-attention network, a second LSTM layer, and a second at least one DNN (for outputting predicted data).

[0167] (S42) A computer program which, when executed on a computer, causes the computer to carry out a method, the method comprising: executing a prediction process to predict 5G usage data for a target geographic area (having no 5G connection / no historical 5G usage data exists or is available), the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data of the target geographic region for a future time period based on non-network data of the target geographic region for a past time period using a first model trained using data of the target geographic region (to predict non-network data of the second time period based on non-network data of the first time period prior to the second time period); generating a second intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a second model trained using data of at least one reference geographic region (which has or had a 5G connection) (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the first 5G usage data prediction by predicting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the second interim prediction using a third model trained using data of the at least one reference geographic region (to predict 5G usage data for the second time period based on non-network data and non-5G network usage data for the second time period); Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using data for the target geographic region (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating the fourth interim prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data of the at least one reference geographic region (to predict combined network usage data for the second time period based on non-network data for the first time period, the combined network usage data including usage data related to 5G and non-5G networks); generating the fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region (to predict 5G usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, (geographic data), demographic data, weather data, infrastructure data, and (vehicle) traffic data.

[0168] (S43) An information processing device comprising: a memory; and a processor coupled to the memory, the processor comprising: configured to execute a prediction process to predict 5G usage data for a target geographic area (which does not have a 5G connection / where no historical 5G usage data exists or is available), the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data of the target geographic region for a future time period based on non-network data of the target geographic region for a past time period using a first model trained using data of the target geographic region (to predict non-network data of the second time period based on non-network data of the first time period prior to the second time period); generating a second intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a second model trained using data of at least one reference geographic region (which has or had a 5G connection) (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the first 5G usage data prediction by predicting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim prediction and the predicted non-5G network usage data of the second interim prediction using a third model trained using data of the at least one reference geographic region (to predict 5G usage data for the second time period based on non-network data and non-5G network usage data for the second time period); Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using data for the target geographic region (to predict non-5G network usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating the fourth interim prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data of the at least one reference geographic region (to predict combined network usage data for the second time period based on non-network data for the first time period, the combined network usage data including usage data related to 5G and non-5G networks); generating the fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region (to predict 5G usage data for the second time period based on the non-network data and non-5G network usage data for the first time period); subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, (geographic data), demographic data, weather data, infrastructure data, and (vehicle) traffic data.

[0169] In addition to the above-described embodiment, the following supplementary notes are further disclosed. (Appendix 1) 1. A computer-implemented method comprising: performing a prediction process for predicting 5G usage data for a target geographic area, the prediction process including generating at least one of first through third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-networked data of the target geographic region for a future time period based on non-networked data of the target geographic region for a past time period using a first model trained using data of the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim forecast and the predicted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third interim prediction by predicting non-5G network usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using the data for the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth interim prediction by forecasting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth interim forecast by forecasting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The computer-implemented method, wherein the non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data. (Appendix 2) 2. The computer-implemented method of claim 1, wherein the non-5G network usage data includes usage data for at least one non-5G communications network. (Appendix 3) 3. The computer-implemented method of claim 1 or 2, wherein the prediction process includes generating at least two of the first, second, and third 5G usage data forecasts and combining the at least two 5G usage data forecasts to generate a final 5G forecast. (Appendix 4) 3. The computer-implemented method of claim 1 or 2, wherein the forecasting process includes generating the first through third 5G usage data forecasts and combining the first through third 5G usage data forecasts to generate a final 5G forecast. (Appendix 5) The step of combining the first to third 5G usage data forecasts to generate a final 5G forecast includes, for at least one variable: calculating an average of the predicted values ​​of the variables in the first through third 5G usage data forecasts; selecting the two predicted values ​​of said variable that are closest to said mean as the endpoints of a prediction range for said variable; 5. The computer-implemented method of claim 4, comprising: (Appendix 6) and performing a training process prior to performing the prediction process; 3. The computer-implemented method of claim 1 or 2, wherein the training process includes training at least one of the first through sixth models. (Appendix 7) The training process includes: training the first model to predict non-network data of the target geographic region for the second time period based on the non-network data of the target geographic region for the first time period and non-network data of the target geographic region for a second time period prior to the first time period; training the second model to predict non-5G network usage data of the at least one reference geographic region for the second time period based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period, and based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period; training the third model to predict 5G usage data for the at least one reference geographic region for the second time period based on 5G usage data, non-network data, and non-5G network usage data for the at least one reference geographic region for the second time period based on the non-network data and the non-5G network usage data for the at least one reference geographic region for the second time period; training the fourth model based on the non-5G network usage data of the target geographic region for the second time period and based on the non-network data and the non-5G network usage data of the target geographic region for the first time period to predict the non-5G network usage data of the target geographic region for the second time period based on the non-network data and the non-5G network usage data of the target geographic region for the first time period; training the fifth model to predict combined network usage data for the at least one reference geographic region for the second time period based on combined network usage data for the at least one reference geographic region for the second time period and based on non-network data for the at least one reference geographic region for the first time period, the combined network usage data including usage data related to 5G and non-5G networks; training the sixth model to predict the 5G usage data of the at least one reference geographic region for the second time period based on the 5G usage data of the at least one reference geographic region for the second time period and based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period; 7. The computer-implemented method of claim 6, comprising: (Appendix 8) 3. The computer-implemented method of claim 1 or 2, wherein the first to sixth models comprise an encoder-decoder model. (Appendix 9) A computer program which, when executed on a computer, causes the computer to carry out a method, the method comprising: performing a prediction process for predicting 5G usage data for a target geographic area, the prediction process including generating at least one of first through third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-networked data of the target geographic region for a future time period based on non-networked data of the target geographic region for a past time period using a first model trained using data of the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the predicted non-network data of the first interim forecast and the predicted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third interim prediction by predicting non-5G network usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using the data for the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth interim prediction by forecasting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth interim forecast by forecasting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data. (Appendix 10) An information processing device comprising: a memory; and a processor coupled to the memory, the processor comprising: configured to perform a prediction process for predicting 5G usage data for a target geographic area, the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic area; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period using a first model trained using data for the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third interim prediction by predicting non-5G network usage data for the target geographic region for the future time period based on the non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using the data for the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth interim prediction by forecasting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth interim prediction by forecasting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data. [Explanation of symbols]

[0170] 10 Computing Equipment 992 Bus 993 Processor 994 Memory 995 Display Unit 996 Input Mechanism 997 Network Interface

Claims

1. 1. A computer-implemented method comprising: executing a prediction process for predicting 5G usage data for a target geographic region, the prediction process including generating at least one of first through third 5G usage data predictions for the target geographic region; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period using a first model trained using data for the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data of the target geographic region for the future time period based on non-network data and non-5G network usage data of the target geographic region for the past time period using a fourth model trained using the data of the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth intermediate prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The computer-implemented method, wherein the non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data.

2. 2. The computer-implemented method of claim 1, wherein the non-5G network usage data includes usage data for at least one non-5G communication network.

3. 3. The computer-implemented method of claim 1, wherein the forecasting process includes generating at least two of the first through third 5G usage data forecasts and combining the at least two 5G usage data forecasts to generate a final 5G forecast.

4. 3. The computer-implemented method of claim 1 or 2, wherein the forecasting process includes generating the first through third 5G usage data forecasts and combining the first through third 5G usage data forecasts to generate a final 5G forecast.

5. The step of combining the first to third 5G usage data forecasts to generate a final 5G forecast includes, for at least one variable: Calculating an average of the predicted values ​​of the variables in the first to third 5G usage data forecasts; selecting the two predicted values ​​of said variable that are closest to said mean as the endpoints of a prediction range for said variable; The computer-implemented method of claim 4 , comprising:

6. and performing a training process prior to performing the prediction process; The computer-implemented method of claim 1 or 2, wherein the training process includes training at least one of the first to sixth models.

7. The training process includes: training the first model to predict non-network data of the target geographic region for the second time period based on the non-network data of the target geographic region for the first time period and non-network data of the target geographic region for a second time period prior to the first time period; training the second model to predict non-5G network usage data of the at least one reference geographic region for the second time period based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period, and based on the non-network data and non-5G network usage data of the at least one reference geographic region for the first time period; training the third model to predict 5G usage data for the at least one reference geographic region for the second time period based on 5G usage data, non-network data, and non-5G network usage data for the at least one reference geographic region for the second time period based on the non-network data and the non-5G network usage data for the at least one reference geographic region for the second time period; training the fourth model based on the non-5G network usage data of the target geographic region for the second time period and based on the non-network data and the non-5G network usage data of the target geographic region for the first time period to predict the non-5G network usage data of the target geographic region for the second time period based on the non-network data and the non-5G network usage data of the target geographic region for the first time period; training the fifth model to predict combined network usage data for the at least one reference geographic region for the second time period based on combined network usage data for the at least one reference geographic region for the second time period and based on non-network data for the at least one reference geographic region for the first time period, the combined network usage data including usage data related to 5G and non-5G networks; training the sixth model to predict the 5G usage data of the at least one reference geographic region for the second time period based on the 5G usage data of the at least one reference geographic region for the second time period and based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period based on the non-network data and the non-5G network usage data of the at least one reference geographic region for the first time period; The computer-implemented method of claim 6 , comprising:

8. The computer-implemented method of claim 1 or 2, wherein the first to sixth models comprise an encoder-decoder model.

9. A computer program which, when executed on a computer, causes the computer to carry out a method, the method comprising: executing a prediction process for predicting 5G usage data for a target geographic region, the prediction process including generating at least one of first through third 5G usage data predictions for the target geographic region; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period using a first model trained using data for the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data of the target geographic region for the future time period based on non-network data and non-5G network usage data of the target geographic region for the past time period using a fourth model trained using the data of the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth intermediate prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth interim prediction by predicting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data.

10. An information processing device comprising: a memory; and a processor coupled to the memory, the processor comprising: configured to perform a prediction process for predicting 5G usage data for a target geographic region, the prediction process including generating at least one of first to third 5G usage data predictions for the target geographic region; Generating the first 5G usage data forecast includes: generating a first intermediate prediction by predicting non-network data for the target geographic region for a future time period based on non-network data for the target geographic region for a past time period using a first model trained using data for the target geographic region; generating a second interim prediction by predicting non-5G network usage data of the target geographic region for the future time period based on the non-network data and non-5G network usage data of the target geographic region for the past time period using a second model trained using data of at least one reference geographic region; generating the first 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the second interim forecast using a third model trained using data of the at least one reference geographic region; Including, Generating the second 5G usage data forecast includes: generating a third intermediate prediction by predicting non-5G network usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a fourth model trained using the data for the target geographic region; generating the second 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the third interim forecast using the third model; Including, Generating the third 5G usage data forecast includes: generating a fourth intermediate prediction by predicting combined network usage data for the target geographic region for the future time period based on non-network data for the target geographic region for the past time period using a fifth model trained with data for the at least one reference geographic region, the combined network usage data including usage data related to 5G and non-5G networks; generating a fifth intermediate prediction by predicting 5G usage data for the target geographic region for the future time period based on non-network data and non-5G network usage data for the target geographic region for the past time period using a sixth model trained with data for the at least one reference geographic region; subtracting the projected 5G usage data of the fifth interim forecast from the combined network usage data of the fourth interim forecast to generate a sixth interim forecast that includes projected non-5G network usage data for the target geographic region for the future time period; generating the third 5G usage data forecast by forecasting 5G usage data for the target geographic region for the future time period based on the forecasted non-network data of the first interim forecast and the forecasted non-5G network usage data of the sixth interim forecast using the third model; Including, The non-network data includes any of location data, demographic data, weather data, infrastructure data, and traffic data.

Images