Method and apparatus for predicting site value attributes based on machine learning

Abstract

This invention discloses a method and apparatus for predicting the value attributes of planned sites based on machine learning. The method includes: collecting user feature data within a target area corresponding to the target location information of the site to be planned, obtaining a user feature data set; reconstructing a trained target prediction model based on the traffic demand characteristics corresponding to each user feature data; and inputting each user feature data from the user feature data set into the reconstructed target prediction model to obtain a prediction result for the site to be planned. The prediction result represents the predicted site value information of the site to be planned, including site traffic information and / or site user capacity information. Therefore, implementing this invention can improve the prediction accuracy of the target prediction model and the prediction flexibility for predicting the value attributes of different sites to be planned, thereby improving the scientific nature of planned base station layout.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and apparatus for predicting the value attributes of planning sites based on machine learning. Background Technology

[0002] Base stations provide technical support for people's information exchange and communication guarantees for economic and social development. Therefore, the scientific planning of base station layout is of great significance to the development of human society.

[0003] Scientific planning of base station layout is partly reflected in the coordination between the investment value and revenue value attributes of planned base stations. This requires assessing the value attributes of the planned base station sites before construction, thus ensuring a scientific and effective site layout. Currently, the main method for assessing the value attributes of planned sites is expert scoring. Experts score the surrounding environment of the planned site based on reference data such as the surrounding population, urbanization level, and high-value commercial targets, as well as their subjective experience, thereby predicting the value attributes of the planned site. Although the results of expert scoring predictions are heavily influenced by personal subjectivity, this method is simple and efficient for scenarios with limited reference data or simple scenarios requiring assessment by local experts. However, as mobile communication networks continue to expand, the environments of planned sites are becoming increasingly complex, and the corresponding reference data volume is also constantly increasing. The low predictive accuracy of the expert scoring method will become more prominent, thus compromising the scientific nature of planned base station layout.

[0004] Therefore, it is particularly important to propose a technical solution to improve the accuracy of predicting the value attributes of planning sites. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and apparatus for predicting the value attributes of planning sites based on machine learning, which can help improve the accuracy of predicting the value attributes of planning sites.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a method for predicting the value attributes of planning sites based on machine learning, the method comprising:

[0007] Based on the target location information of the site to be planned, user feature data within the target area corresponding to the target location information is collected to obtain a user feature data set.

[0008] Based on the traffic demand characteristics corresponding to each user feature data, reconstruct the trained target prediction model;

[0009] Each user feature data in the user feature data set is input into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the predicted site value information of the site to be planned. The site value information includes site traffic information and / or site user capacity information.

[0010] As an optional implementation, in the first aspect of the present invention, the step of collecting user feature data within the target area corresponding to the target location information of the site to be planned, to obtain a user feature data set, includes:

[0011] Based on the target location information of the site to be planned, determine the target area corresponding to the target location information;

[0012] Based on the coverage area of ​​the target region, at least one data collection point is determined;

[0013] For each data acquisition point, the data acquisition range corresponding to the data acquisition point is determined, and the initial feature data within the data acquisition range corresponding to the data acquisition point is collected to obtain the initial feature data set.

[0014] For each initial feature data set, it is determined whether there is abnormal feature data in the initial feature data set. When it is determined that there is abnormal feature data in the initial feature data set, the abnormal feature data is removed to obtain a subset of user feature data of the data collection point corresponding to the initial feature data set.

[0015] Based on the subset of user feature data from all the data collection points, determine the set of user feature data within the target area corresponding to the target location information.

[0016] As an optional implementation, in the first aspect of the present invention, before reconstructing the trained target prediction model based on the traffic demand features corresponding to each of the user feature data, the method further includes:

[0017] Train the original prediction model to obtain the target prediction model;

[0018] The original prediction model includes a first set of original models and a second original model. The first set of original models includes at least one base model, which includes at least one of gradient boosting regression model, extreme tree regression model, random forest regression model, and adaptive boosting regression model. The second original model includes a linear regression model.

[0019] And, the training of the original prediction model to obtain the target prediction model includes:

[0020] The training sample dataset is divided according to the preset cross-validation conditions;

[0021] Each training sample data in the divided training sample dataset is input into the corresponding base model to obtain a first prediction result corresponding to each training sample data and a first target model set corresponding to the first original model set.

[0022] The first prediction result corresponding to all the training sample data is input into the second original model to obtain the second prediction result corresponding to all the training sample data and the second target model corresponding to the second original model. The second prediction result is used to represent the predicted planning site value information corresponding to all the training sample data.

[0023] Based on the second prediction results corresponding to all the training sample data, the first target model set and the second target model are determined as prediction models to be verified.

[0024] All validation sample data in the validation sample dataset are input into the prediction model to be validated to obtain validation results corresponding to all the validation sample data. The validation results are used to represent the planning site value information corresponding to all the predicted validation sample data.

[0025] Determine whether the verification result meets the preset verification completion conditions. If it is determined that the verification result meets the preset verification completion conditions, then determine that the prediction model to be verified is the target prediction model.

[0026] As an optional implementation, in the first aspect of the present invention, when it is determined that the verification result does not meet the preset verification completion condition, the method further includes:

[0027] Based on the first loss function of the training sample dataset and the second loss function of the validation sample dataset, it is determined whether the fit of the prediction model to be validated meets the abnormal fit condition. When it is determined that the fit meets the abnormal fit condition, the model correction strategy of the prediction model to be validated is determined based on the fit.

[0028] According to the model correction strategy, the prediction model to be verified is adjusted, and the operation of inputting all the verification sample data in the verification sample dataset into the prediction model to be verified to obtain the verification results corresponding to all the verification sample data is triggered, as well as the operation of determining whether the verification results meet the preset verification completion conditions is triggered.

[0029] As an optional implementation, in the first aspect of the present invention, determining the first target model set and the second target model as prediction models to be verified based on the second prediction results corresponding to all the training sample data includes:

[0030] The second prediction result corresponding to all the training sample data is compared with the actual planned site value information corresponding to all the training sample data to obtain the comparison result.

[0031] Determine whether the comparison result meets the preset training completion conditions. If it is determined that the comparison result meets the preset training completion conditions, then the first target model set and the second target model are determined as the prediction models to be verified.

[0032] When it is determined that the comparison result does not meet the preset training completion condition, the first target model set and the second target model are adjusted according to the comparison result and the preset training completion condition, and the adjusted first target model set and the second target model are determined as the prediction model to be verified.

[0033] As an optional implementation, in a first aspect of the present invention, adjusting the first target model set and the second target model according to the comparison result and the preset training completion condition includes:

[0034] Based on the data deviation characteristics corresponding to the comparison results, at least one deviation influence factor is determined. The deviation influence factor is used to represent the target training sample data and target feature parameters that cause the comparison results to fail to meet the preset training completion conditions. The training sample dataset includes the target training sample data, and the target feature parameters include feature parameters in the first target model set and / or feature parameters in the second target model. The feature parameters include weights and biases.

[0035] Adjust the first target model set and the second target model according to all the aforementioned bias influencing factors and the preset training completion conditions;

[0036] And, adjusting the first target model set and the second target model according to all the said bias influencing factors and the preset training completion conditions includes:

[0037] Determine whether the target training sample data exceeds the target sample interval. If it is determined that the target training sample data exceeds the target sample interval, then perform a removal process on the target training sample data.

[0038] Based on the data characteristics of the training sample dataset after removing the target training sample data, the target feature parameters are adjusted to adjust the first target model set and the second target model.

[0039] As an optional implementation, in the first aspect of the present invention, after adjusting the first target model set and the second target model according to all the said bias influence factors and the preset training completion conditions, the method further includes:

[0040] Determine whether the actual prediction effect corresponding to the adjusted first target model set and the second target model matches the target prediction effect corresponding to the preset training completion condition. When it is determined that the actual prediction effect matches the target prediction effect, the operation of determining the adjusted first target model set and the second target model as the prediction model to be verified is triggered.

[0041] When it is determined that the actual prediction effect does not match the target prediction effect, the operation of adjusting the target feature parameters based on the data features of the training sample dataset after removing the target training sample data is triggered, so as to adjust the first target model set and the second target model.

[0042] As an optional implementation, in the first aspect of the present invention, the step of reconstructing the trained target prediction model based on the traffic demand features corresponding to each of the user feature data includes:

[0043] For each user feature data, determine the user business information corresponding to that user feature data;

[0044] Based on the user's business information, determine the user's corresponding traffic demand characteristics;

[0045] Based on the traffic demand characteristics of all the users, reconstruct the trained target prediction model;

[0046] Furthermore, after determining the user's traffic demand characteristics based on the user's service information, the method further includes:

[0047] Determine whether the traffic demand characteristics corresponding to the user are abnormal demand characteristics. If it is determined that the traffic demand characteristics corresponding to the user are abnormal demand characteristics, then the user is identified as an abnormal user, and an abnormal user information set of the abnormal user is output. The abnormal user information set includes at least one of the abnormal user's user characteristic data, user business information, and traffic demand characteristics.

[0048] A second aspect of this invention discloses a machine learning-based device for predicting the value attributes of planning sites, the device comprising:

[0049] The data acquisition module is used to collect user feature data within the target area corresponding to the target location information of the site to be planned, and obtain a user feature data set.

[0050] The reconstruction module reconstructs the trained target prediction model based on the traffic demand characteristics corresponding to each user feature data.

[0051] The prediction module inputs each user feature data in the user feature data set into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the predicted site value information of the site to be planned. The site value information includes site traffic information and / or site user capacity information.

[0052] As an optional implementation, in the second aspect of the present invention, the specific method by which the acquisition module acquires user feature data set within the target area corresponding to the target location information of the site to be planned, based on the target location information of the site to be planned, includes:

[0053] Based on the target location information of the site to be planned, determine the target area corresponding to the target location information;

[0054] Based on the coverage area of ​​the target region, at least one data collection point is determined;

[0055] For each data acquisition point, the data acquisition range corresponding to the data acquisition point is determined, and the initial feature data within the data acquisition range corresponding to the data acquisition point is collected to obtain the initial feature data set.

[0056] For each initial feature data set, it is determined whether there is abnormal feature data in the initial feature data set. When it is determined that there is abnormal feature data in the initial feature data set, the abnormal feature data is removed to obtain a subset of user feature data of the data collection point corresponding to the initial feature data set.

[0057] Based on the subset of user feature data from all the data collection points, determine the set of user feature data within the target area corresponding to the target location information.

[0058] As an optional implementation, in a second aspect of the invention, the apparatus further includes:

[0059] The training module is used to train the original prediction model before the reconstruction module reconstructs the trained target prediction model based on the traffic demand features corresponding to each user feature data, so as to obtain the target prediction model.

[0060] The original prediction model includes a first set of original models and a second original model. The first set of original models includes at least one base model, which includes at least one of gradient boosting regression model, extreme tree regression model, random forest regression model, and adaptive boosting regression model. The second original model includes a linear regression model.

[0061] Furthermore, the specific methods by which the training module trains the original prediction model to obtain the target prediction model include:

[0062] The training sample dataset is divided according to the preset cross-validation conditions;

[0063] Each training sample data in the divided training sample dataset is input into the corresponding base model to obtain a first prediction result corresponding to each training sample data and a first target model set corresponding to the first original model set.

[0064] The first prediction result corresponding to all the training sample data is input into the second original model to obtain the second prediction result corresponding to all the training sample data and the second target model corresponding to the second original model. The second prediction result is used to represent the predicted planning site value information corresponding to all the training sample data.

[0065] Based on the second prediction results corresponding to all the training sample data, the first target model set and the second target model are determined as prediction models to be verified.

[0066] All validation sample data in the validation sample dataset are input into the prediction model to be validated to obtain validation results corresponding to all the validation sample data. The validation results are used to represent the planning site value information corresponding to all the predicted validation sample data.

[0067] Determine whether the verification result meets the preset verification completion conditions. If it is determined that the verification result meets the preset verification completion conditions, then determine that the prediction model to be verified is the target prediction model.

[0068] As an optional implementation, in the second aspect of the present invention, the training module is further configured to, when it is determined that the verification result does not meet the preset verification completion condition, determine whether the fitting degree of the prediction model to be verified meets the abnormal fitting condition based on the first loss function of the training sample dataset and the second loss function of the verification sample dataset; and when it is determined that the fitting degree meets the abnormal fitting condition, determine the model correction strategy of the prediction model to be verified based on the fitting degree.

[0069] According to the model correction strategy, the prediction model to be verified is adjusted, and the operation of inputting all the verification sample data in the verification sample dataset into the prediction model to be verified to obtain the verification results corresponding to all the verification sample data is triggered, as well as the operation of determining whether the verification results meet the preset verification completion conditions is triggered.

[0070] As an optional implementation, in a second aspect of the present invention, the specific method by which the training module determines the first target model set and the second target model as the prediction model to be verified based on the second prediction results corresponding to all the training sample data includes:

[0071] The second prediction result corresponding to all the training sample data is compared with the actual planned site value information corresponding to all the training sample data to obtain the comparison result.

[0072] Determine whether the comparison result meets the preset training completion conditions. If it is determined that the comparison result meets the preset training completion conditions, then the first target model set and the second target model are determined as the prediction models to be verified.

[0073] When it is determined that the comparison result does not meet the preset training completion condition, the first target model set and the second target model are adjusted according to the comparison result and the preset training completion condition, and the adjusted first target model set and the second target model are determined as the prediction model to be verified.

[0074] As an optional implementation, in a second aspect of the present invention, the specific method by which the training module adjusts the first target model set and the second target model according to the comparison results and the preset training completion conditions includes:

[0075] Based on the data deviation characteristics corresponding to the comparison results, at least one deviation influence factor is determined. The deviation influence factor is used to represent the target training sample data and target feature parameters that cause the comparison results to fail to meet the preset training completion conditions. The training sample dataset includes the target training sample data, and the target feature parameters include feature parameters in the first target model set and / or feature parameters in the second target model. The feature parameters include weights and biases.

[0076] Adjust the first target model set and the second target model according to all the aforementioned bias influencing factors and the preset training completion conditions;

[0077] Furthermore, the specific methods by which the training module adjusts the first target model set and the second target model based on all the aforementioned bias influencing factors and the preset training completion conditions include:

[0078] Determine whether the target training sample data exceeds the target sample interval. If it is determined that the target training sample data exceeds the target sample interval, then perform a removal process on the target training sample data.

[0079] Based on the data characteristics of the training sample dataset after removing the target training sample data, the target feature parameters are adjusted to adjust the first target model set and the second target model.

[0080] As an optional implementation, in a second aspect of the invention, the apparatus further includes:

[0081] The first judgment module is used to determine whether the actual prediction effect of the adjusted first target model set and the second target model matches the target prediction effect corresponding to the preset training completion condition after the training module adjusts the first target model set and the second target model according to all the deviation influencing factors and the preset training completion condition; when the first judgment module determines that the actual prediction effect matches the target prediction effect, it triggers the training module to perform the operation of determining the adjusted first target model set and the second target model as the prediction model to be verified; when the first judgment module determines that the actual prediction effect does not match the target prediction effect, it triggers the training module to perform the operation of adjusting the target feature parameters according to the data features of the training sample dataset after removing the target training sample data, so as to adjust the first target model set and the second target model.

[0082] As an optional implementation, in the second aspect of the present invention, the specific method by which the reconstruction module reconstructs the trained target prediction model based on the traffic demand characteristics corresponding to each user feature data includes:

[0083] For each user feature data, determine the user business information corresponding to that user feature data;

[0084] Based on the user's business information, determine the user's corresponding traffic demand characteristics;

[0085] Based on the traffic demand characteristics of all the users, reconstruct the trained target prediction model;

[0086] The device also includes:

[0087] The second judgment module is also used to determine whether the traffic demand characteristics corresponding to the user are abnormal demand characteristics after the reconstruction module determines the traffic demand characteristics corresponding to the user based on the user's user business information.

[0088] The determination module is used to determine that the user is an abnormal user when the second judgment module determines that the user's traffic demand characteristics are abnormal demand characteristics;

[0089] The output module is used to output the abnormal user information set of the abnormal user, the abnormal user information set including at least one of the abnormal user's user characteristic data, user business information, and traffic demand characteristics.

[0090] A third aspect of this invention discloses another machine learning-based device for predicting the value attributes of planning sites, the device comprising:

[0091] Memory containing executable program code;

[0092] A processor coupled to the memory;

[0093] The processor calls the executable program code stored in the memory to execute the machine learning-based site value attribute prediction method disclosed in the first aspect of the present invention.

[0094] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0095] In this embodiment of the invention, user feature data within the target area corresponding to the target location information of the site to be planned is collected based on the target location information to obtain a user feature data set; the trained target prediction model is reconstructed based on the traffic demand features corresponding to each user feature data; each user feature data in the user feature data set is input into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the predicted site value information of the site to be planned, and the site value information includes the site traffic information and / or the number of users the site can carry. As can be seen, implementing this invention can determine the traffic demand characteristics corresponding to each user characteristic data based on the collected user characteristic data within the target area corresponding to the site to be planned. Based on all traffic demand characteristics, the trained target prediction model is reconstructed, enabling the trained target prediction model to be further optimized and improved according to all traffic demand characteristics in the target area where the site to be planned is located. This improves the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned, thereby improving the scientific nature of the planned base station layout. Furthermore, the reconstructed target prediction model can obtain the planned site value information of the site to be planned, thus achieving the purpose of predicting the value attributes of the site to be planned. At the same time, this solution, relying on the target prediction model, not only improves the accuracy of predicting the value attributes of the planned site, but also improves the prediction efficiency and prediction stability for predicting the value attributes of different sites to be planned. Attached Figure Description

[0096] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0097] Figure 1 This is a flowchart illustrating a machine learning-based method for predicting the value attributes of a planning site, as disclosed in an embodiment of the present invention.

[0098] Figure 2 This is a flowchart illustrating another machine learning-based method for predicting the value attributes of planning sites, as disclosed in an embodiment of the present invention.

[0099] Figure 3 This is a schematic diagram of the structure of a machine learning-based site value attribute prediction device disclosed in an embodiment of the present invention.

[0100] Figure 4 This is a schematic diagram of another machine learning-based site value attribute prediction device disclosed in an embodiment of the present invention.

[0101] Figure 5 This is a schematic diagram of the structure of another machine learning-based site value attribute prediction device disclosed in an embodiment of the present invention. Detailed Implementation

[0102] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0103] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0104] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0105] This invention discloses a machine learning-based method and apparatus for predicting the value attributes of planned sites. It can determine the traffic demand characteristics corresponding to each user characteristic data point within the target area corresponding to the planned site, and reconstruct the trained target prediction model based on all traffic demand characteristics. This allows the trained target prediction model to be further optimized and improved based on all traffic demand characteristics of the target area where the planned site is located, thereby improving the prediction accuracy and the flexibility and adaptability of predicting the value attributes of different planned sites. This, in turn, helps to improve the scientific nature of planned base station layout. Furthermore, it can obtain the planned site value information of the planned site based on the reconstructed target prediction model, thus achieving the purpose of predicting the value attributes of the planned site. Simultaneously, this solution, relying on the target prediction model, not only improves the accuracy of predicting the value attributes of planned sites, but also improves the prediction efficiency and stability for predicting the value attributes of different planned sites.

[0106] The following sections will provide detailed explanations.

[0107] Example 1

[0108] Please see Figure 1 , Figure 1 This is a flowchart illustrating a machine learning-based site value attribute prediction method disclosed in an embodiment of the present invention. Wherein, Figure 1 The machine learning-based site value attribute prediction method described herein can be applied to planning / management systems for mobile communication networks / urban and rural areas, as well as to equipment such as satellites, remote sensing, geological exploration, and environmental monitoring. It can also be applied to intelligent terminal devices with protection / monitoring / lighting functions in infrastructure such as bridges, road networks, and buildings. This invention does not limit the application of this method. Figure 1 As shown, this machine learning-based method for predicting the value attributes of planning sites may include the following operations:

[0109] 101. Based on the target location information of the site to be planned, collect user feature data within the target area corresponding to the target location information to obtain a user feature data set.

[0110] In this embodiment of the invention, the target location information can be latitude and longitude data, or it can be a second location information calculated from the known first location information.

[0111] The aforementioned user characteristic data may include POI (Point of Interest) data, population data, and economic data.

[0112] The aforementioned POI data may include feature data of various types of individuals covered within the target area, including but not limited to schools, shopping malls, hospitals, convenience stores, transportation hubs, office buildings, etc.

[0113] In this embodiment of the invention, as an optional implementation, the above-mentioned process of collecting user feature data within the target area corresponding to the target location information of the site to be planned, and obtaining a user feature data set, includes the following operations:

[0114] Based on the target location information of the site to be planned, determine the target area corresponding to the target location information.

[0115] Based on the coverage area of ​​the target region, at least one data collection point shall be identified.

[0116] For each data collection point, determine the data collection range corresponding to that data collection point, and collect the initial feature data within the corresponding data collection range to obtain the initial feature data set.

[0117] For each initial feature data set, determine whether there is abnormal feature data in the initial feature data set. If abnormal feature data is found in the initial feature data set, perform a removal process on the abnormal feature data to obtain a subset of user feature data of the data collection point corresponding to the initial feature data set.

[0118] Based on the subset of user characteristic data from all data collection points, determine the set of user characteristic data within the target area corresponding to the target location information.

[0119] The abnormal features of the aforementioned abnormal feature data may include at least one of the following: data is empty, duplicate, garbled, or exceeds the data range corresponding to the initial feature data.

[0120] As can be seen, implementing this optional embodiment can determine the layout of data collection points and the data collection range of each data collection point based on the coverage of the target area corresponding to the site to be planned. This improves the comprehensiveness of the initial feature data collection, thereby improving the comprehensiveness of the user feature data determination. Furthermore, this is beneficial for improving the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned. Simultaneously, implementing this optional embodiment can further perform abnormal feature data screening and removal operations on each initial feature data set, thereby further improving the data quality of the initial feature data, and further contributing to improving the prediction accuracy of the target prediction model.

[0121] 102. Based on the traffic demand characteristics corresponding to each user's feature data, reconstruct the trained target prediction model.

[0122] In this embodiment of the invention, as another optional implementation, the above-mentioned reconstruction of the trained target prediction model based on the traffic demand features corresponding to each user feature data may include the following operations:

[0123] For each user feature data, determine the user business information corresponding to that user feature data.

[0124] Based on the user's business information, determine the user's corresponding traffic demand characteristics.

[0125] Reconstruct the trained target prediction model based on the traffic demand characteristics of all users.

[0126] It is evident that implementing this optional embodiment can determine the corresponding user service information based on each user's user characteristic data, and determine the traffic demand characteristics of each user based on each user's user service information. Thus, based on the traffic demand characteristics of all users, the trained target prediction model can be reconstructed. This can improve the prediction accuracy, flexibility, and adaptability of the value attribute prediction of the planned sites for different target areas, thereby improving the scientific nature of the planned base station layout.

[0127] Furthermore, after determining the user's traffic demand characteristics based on the user's business information, the method may also include the following operations:

[0128] Determine whether the traffic demand characteristics corresponding to the user are abnormal demand characteristics. If it is determined that the traffic demand characteristics corresponding to the user are abnormal demand characteristics, then the user is identified as an abnormal user, and an abnormal user information set is output. The abnormal user information set includes at least one of the abnormal user's user characteristic data, user business information, and traffic demand characteristics.

[0129] The aforementioned abnormal demand characteristics are used to indicate that the user's actual traffic demand characteristics exceed the available traffic demand characteristics corresponding to the user's user service information.

[0130] As can be seen, implementing this optional embodiment can determine whether the traffic demand characteristics of each user are abnormal demand characteristics based on the user service information of each user, thereby enabling early warning of abnormal traffic risks of planned sites during the site value attribute planning stage, thus ensuring the long-term availability of planned base stations.

[0131] 103. Input each user feature data in the user feature data set into the reconstructed target prediction model to obtain the prediction results of the site to be planned.

[0132] In this embodiment of the invention, the prediction results are used to represent the predicted site value information of the site to be planned. The site value information includes site traffic information and / or site user capacity information.

[0133] As can be seen, implementing this embodiment of the invention can determine the traffic demand characteristics corresponding to each user characteristic data based on the collected user characteristic data within the target area where the site to be planned is located. Based on all traffic demand characteristics, the trained target prediction model is reconstructed, enabling the trained target prediction model to be further optimized and improved according to all traffic demand characteristics within the target area where the site to be planned is located. This improves the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned, thereby improving the scientific nature of the planned base station layout. Furthermore, based on the reconstructed target prediction model, the planned site value information of the site to be planned can be obtained, thus achieving the purpose of predicting the value attributes of the site to be planned. At the same time, this solution, relying on the target prediction model, not only improves the accuracy of predicting the value attributes of the planned site, but also improves the prediction efficiency and stability for predicting the value attributes of different sites to be planned.

[0134] Example 2

[0135] Please see Figure 2 , Figure 2 This is a flowchart illustrating a machine learning-based site value attribute prediction method disclosed in an embodiment of the present invention. Wherein, Figure 2 The machine learning-based site value attribute prediction method described herein can be applied to planning / management systems for mobile communication networks / urban and rural areas, as well as to equipment such as satellites, remote sensing, geological exploration, and environmental monitoring. It can also be applied to intelligent terminal devices with protection / monitoring / lighting functions in infrastructure such as bridges, road networks, and buildings. This invention does not limit the application of this method. Figure 2 As shown, this machine learning-based method for predicting the value attributes of planning sites may include the following operations:

[0136] 201. Based on the target location information of the site to be planned, collect user feature data within the target area corresponding to the target location information to obtain a user feature data set.

[0137] 202. Train the original prediction model to obtain the target prediction model.

[0138] In this optional embodiment, the original prediction model includes a first set of original models and a second original model. The first set of original models includes at least one base model, which includes at least one of gradient boosting regression model, extreme tree regression model, random forest regression model, and adaptive boosting regression model. The second original model includes a linear regression model.

[0139] In this optional embodiment, as an optional implementation, training the original prediction model to obtain the target prediction model may include the following operations:

[0140] The training sample dataset is divided according to the preset cross-validation conditions.

[0141] Each training sample in the divided training sample dataset is input into the corresponding base model to obtain the first prediction result corresponding to each training sample and the first target model set corresponding to the first original model set.

[0142] The first prediction result corresponding to all training sample data is input into the second original model to obtain the second prediction result corresponding to all training sample data and the second target model corresponding to the second original model. The second prediction result is used to represent the planned site value information corresponding to all training sample data.

[0143] Based on the second prediction results corresponding to all training sample data, the first target model set and the second target model are determined as the prediction models to be verified.

[0144] All validation sample data in the validation sample dataset are input into the prediction model to be validated to obtain the validation results corresponding to all validation sample data. The validation results are used to represent the planning site value information corresponding to all the predicted validation sample data.

[0145] Determine whether the verification result meets the preset verification completion conditions. If the verification result meets the preset verification completion conditions, then determine the prediction model to be verified as the target prediction model.

[0146] In this optional embodiment, the number of cross-validations corresponding to the above-mentioned preset cross-validation conditions and the logic for dividing the sample dataset can be selected according to the actual application scenario, and the present invention does not impose specific limitations.

[0147] It is evident that implementing this optional embodiment can provide a model architecture for a target prediction model and a model training method for a target prediction model. In this optional embodiment, by jointly predicting multiple basic models in the first target model set and the second target model, it is beneficial to improve the prediction accuracy of the target prediction model, thereby improving the scientific nature of planning base station layout.

[0148] 203. Based on the traffic demand characteristics corresponding to each user's feature data, reconstruct the trained target prediction model.

[0149] 204. Input each user feature data in the user feature dataset into the reconstructed target prediction model to obtain the prediction results of the site to be planned.

[0150] In this optional embodiment, the prediction results are used to represent the predicted site value information of the site to be planned, which includes site traffic information and / or site user capacity information.

[0151] In this embodiment of the invention, for other descriptions of steps 201, 203, and 204, please refer to the detailed description of steps 101-103 in Embodiment 1. These descriptions will not be repeated in this embodiment of the invention.

[0152] As can be seen, implementing this optional embodiment can determine the traffic demand characteristics corresponding to each user characteristic data based on the collected user characteristic data within the target area where the site to be planned is located. Based on all traffic demand characteristics, the trained target prediction model is reconstructed, enabling the trained target prediction model to be further optimized and improved according to all traffic demand characteristics within the target area where the site to be planned is located. This improves the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned, thereby improving the scientific nature of the planned base station layout. Furthermore, the reconstructed target prediction model can obtain the planned site value information of the site to be planned, thus achieving the purpose of predicting the value attributes of the site to be planned. At the same time, this solution, relying on the target prediction model, not only improves the accuracy of predicting the value attributes of the planned site, but also improves the prediction efficiency and prediction stability for predicting the value attributes of different sites to be planned.

[0153] In this optional embodiment, as another optional implementation, when it is determined that the verification result does not meet the preset verification completion conditions, the method may further include the following operations:

[0154] Based on the first loss function of the training sample dataset and the second loss function of the validation sample dataset, it is determined whether the fit of the prediction model to be validated meets the abnormal fit condition. If it is determined that the fit meets the abnormal fit condition, the model correction strategy of the prediction model to be validated is determined according to the fit.

[0155] According to the model correction strategy, the prediction model to be verified is adjusted, and the operation of inputting all the verification sample data in the verification sample dataset into the prediction model to be verified is triggered to obtain the verification results corresponding to all the verification sample data, and the operation of determining whether the verification results meet the preset verification completion conditions is triggered.

[0156] The above-mentioned abnormal fit condition is used to indicate whether the target prediction model is overfitted or underfitted.

[0157] As can be seen, implementing this optional embodiment can determine whether the fitting degree of the prediction model to be validated meets the abnormal fitting condition based on the first loss function and the second loss function, thereby improving the detection accuracy of abnormal fitting in the prediction model to be validated, and thus improving the detection accuracy of abnormal fitting in the target prediction model. Simultaneously, it can determine the model correction strategy for the prediction model to be validated based on its fitting degree, and adjust the prediction model according to the model correction strategy, thereby improving the prediction accuracy of the prediction model to be validated, and thus improving the prediction accuracy of the target prediction model.

[0158] In this optional embodiment, as yet another optional implementation, the above-mentioned determination of whether the fit of the prediction model to be validated satisfies the abnormal fit condition based on the first loss function of the training sample dataset and the second loss function of the validation sample dataset includes:

[0159] Determine the trend of the first loss function in the training sample dataset and verify whether the trend of the second loss function in the sample dataset is decreasing. If the determination result is yes, then the fit of the prediction model to be verified is determined to meet the abnormal fit condition.

[0160] As can be seen, implementing this optional embodiment can determine whether the fitting degree of the prediction model to be verified meets the abnormal fitting condition based on the changing trends of the first loss function and the second loss function, which can further improve the detection accuracy of abnormal fitting of the prediction model to be verified, and further improve the detection accuracy of abnormal fitting of the target prediction model.

[0161] In this optional embodiment, as an optional implementation, when the abnormal fitting condition is determined to be overfitting, the validation sample dataset is cleaned and the feature dimension of the validation sample data is reduced.

[0162] When the abnormal fitting condition is determined to be underfitting, the feature dimension of the validation sample data is increased, and the number of validations is increased.

[0163] It is evident that implementing this optional embodiment can provide a solution when the prediction model to be verified exhibits an abnormal fit, thereby improving the prediction accuracy of the prediction model to be verified and further improving the prediction accuracy of the target prediction model.

[0164] In this optional embodiment, as yet another optional implementation, determining the first target model set and the second target model as the prediction model to be verified based on the second prediction results corresponding to all training sample data may include the following operations:

[0165] The second prediction results corresponding to all training sample data are compared with the actual planned site value information corresponding to all training sample data to obtain the comparison results.

[0166] Determine whether the comparison results meet the preset training completion conditions. If it is determined that the comparison results meet the preset training completion conditions, then the first target model set and the second target model are determined as the prediction models to be verified.

[0167] If it is determined that the comparison result does not meet the preset training completion conditions, then the first target model set and the second target model are adjusted according to the comparison result and the preset training completion conditions, and the adjusted first target model set and the second target model are determined as the prediction models to be verified.

[0168] As can be seen, implementing this optional embodiment can determine whether the first target model set and the second target model are prediction models to be verified based on the comparison results, which can improve the accuracy of determining the prediction models to be verified, and thus provide a guarantee for the subsequent verification of the prediction models to be verified.

[0169] In this optional embodiment, as another optional implementation, adjusting the first target model set and the second target model based on the comparison results and preset training completion conditions may include the following operations:

[0170] Based on the data deviation characteristics corresponding to the comparison results, at least one deviation influence factor is determined. The deviation influence factor is used to represent the target training sample data and target feature parameters that cause the comparison results to fail to meet the preset training completion conditions. The training sample dataset includes the target training sample data, and the target feature parameters include the feature parameters in the first target model set and / or the feature parameters in the second target model. The feature parameters include weights and biases.

[0171] Based on all bias influencing factors and preset training completion conditions, adjust the first target model set and the second target model.

[0172] As can be seen, implementing this optional embodiment can adjust the first target model set and the second target model from two dimensions: training sample data and model feature parameters, in order to improve the determination accuracy of the prediction model to be verified, thereby providing a guarantee for subsequent verification of the prediction model to be verified.

[0173] Furthermore, the aforementioned adjustment of the first target model set and the second target model based on all bias influencing factors and preset training completion conditions may include the following operations:

[0174] Determine whether the target training sample data exceeds the target sample interval. If it is determined that the target training sample data exceeds the target sample interval, then perform a removal process on the target training sample data.

[0175] Based on the data characteristics of the training sample dataset after removing the target training sample data, the target feature parameters are adjusted to adjust the first target model set and the second target model.

[0176] As can be seen, implementing this optional embodiment can "cleanse" the training sample data to eliminate the possibility of poor model training quality due to the training data, thereby improving the accuracy of adjusting the first target model set and the second target model, thus improving the determination accuracy of the prediction model to be verified, and providing a guarantee for subsequent verification of the prediction model to be verified.

[0177] In an optional embodiment, after adjusting the first target model set and the second target model according to all bias influencing factors and preset training completion conditions, the method may further include the following operations:

[0178] Determine whether the actual prediction effect of the adjusted first target model set and the second target model matches the target prediction effect corresponding to the preset training completion conditions. If it is determined that the actual prediction effect matches the target prediction effect, then the operation of determining the adjusted first target model set and the second target model as the prediction model to be verified is triggered.

[0179] When it is determined that the actual prediction effect does not match the target prediction effect, the operation of adjusting the target feature parameters based on the data characteristics of the training sample dataset after removing the target training sample data is triggered, so as to adjust the first target model set and the second target model.

[0180] As can be seen, implementing this optional embodiment can verify and adjust the trained first target model set and second target model multiple times to ensure training quality, thereby improving the prediction accuracy of the prediction model to be verified, which in turn helps to improve the prediction accuracy of the target prediction model.

[0181] Example 3

[0182] Please see Figure 3 , Figure 3 This is a schematic diagram of a machine learning-based site value attribute prediction device disclosed in an embodiment of the present invention. Figure 3The machine learning-based site value attribute prediction device described herein can be applied to mobile communication networks / urban and rural planning / management systems, as well as to satellite, remote sensing, geological exploration, environmental monitoring, and other equipment. It can also be applied to intelligent terminal devices with protection / monitoring / lighting functions in infrastructure such as bridges, road networks, and buildings. This invention does not limit the application of this method. Figure 3 As shown, the machine learning-based site value attribute prediction device may include:

[0183] The data acquisition module 301 is used to collect user feature data within the target area corresponding to the target location information of the site to be planned, and obtain a set of user feature data.

[0184] In this embodiment of the invention, the target location information can be latitude and longitude data, or it can be a second location information calculated from the known first location information.

[0185] The aforementioned user characteristic data may include POI (Point of Interest) data, population data, and economic data.

[0186] The aforementioned POI data may include feature data of various types of individuals covered within the target area, including but not limited to schools, shopping malls, hospitals, convenience stores, transportation hubs, office buildings, etc.

[0187] Reconstruction module 302 reconstructs the trained target prediction model based on the traffic demand characteristics corresponding to each user's feature data.

[0188] The prediction module 303 inputs each user feature data in the user feature data set into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the planned site value information of the predicted site to be planned. The planned site value information includes planned site traffic information and / or planned site user capacity information.

[0189] As can be seen, implementing this embodiment of the invention can determine the traffic demand characteristics corresponding to each user characteristic data based on the collected user characteristic data within the target area where the site to be planned is located. Based on all traffic demand characteristics, the trained target prediction model is reconstructed, enabling the trained target prediction model to be further optimized and improved according to all traffic demand characteristics within the target area where the site to be planned is located. This improves the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned, thereby improving the scientific nature of the planned base station layout. Furthermore, based on the reconstructed target prediction model, the planned site value information of the site to be planned can be obtained, thus achieving the purpose of predicting the value attributes of the site to be planned. At the same time, this solution, relying on the target prediction model, not only improves the accuracy of predicting the value attributes of the planned site, but also improves the prediction efficiency and stability for predicting the value attributes of different sites to be planned.

[0190] In this embodiment of the invention, as an optional implementation, the above-mentioned acquisition module 301 acquires user feature data within the target area corresponding to the target location information based on the target location information of the site to be planned, and obtains the user feature data set in the following specific ways:

[0191] Based on the target location information of the site to be planned, determine the target area corresponding to the target location information.

[0192] Based on the coverage area of ​​the target region, at least one data collection point shall be identified.

[0193] For each data collection point, determine the data collection range corresponding to that data collection point, and collect the initial feature data within the corresponding data collection range to obtain the initial feature data set.

[0194] For each initial feature data set, determine whether there is abnormal feature data in the initial feature data set. If abnormal feature data is found in the initial feature data set, perform a removal process on the abnormal feature data to obtain a subset of user feature data of the data collection point corresponding to the initial feature data set.

[0195] Based on the subset of user characteristic data from all data collection points, determine the set of user characteristic data within the target area corresponding to the target location information.

[0196] The abnormal features of the aforementioned abnormal feature data may include at least one of the following: data is empty, duplicate, garbled, or exceeds the data range corresponding to the initial feature data.

[0197] As can be seen, implementing this optional embodiment can determine the layout of data collection points and the data collection range of each data collection point based on the coverage of the target area corresponding to the site to be planned. This improves the comprehensiveness of the initial feature data collection, thereby improving the comprehensiveness of the user feature data determination. Furthermore, this is beneficial for improving the prediction accuracy of the target prediction model and the prediction flexibility and adaptability for predicting the value attributes of different sites to be planned. Simultaneously, implementing this optional embodiment can further perform abnormal feature data screening and removal operations on each initial feature data set, thereby further improving the data quality of the initial feature data, and further contributing to improving the prediction accuracy of the target prediction model.

[0198] In an optional embodiment, such as Figure 4 As shown, the device also includes:

[0199] The training module 304 is used to train the original prediction model before the reconstruction module 302 reconstructs the trained target prediction model based on the traffic demand features corresponding to each user feature data, so as to obtain the target prediction model.

[0200] The original prediction model includes a first set of original models and a second set of original models. The first set of original models includes at least one base model, which includes at least one of gradient boosting regression model, extreme tree regression model, random forest regression model, and adaptive boosting regression model. The second set of original models includes a linear regression model.

[0201] Furthermore, the specific methods by which training module 304 trains the original prediction model to obtain the target prediction model include:

[0202] The training sample dataset is divided according to the preset cross-validation conditions.

[0203] Each training sample in the divided training sample dataset is input into the corresponding base model to obtain the first prediction result corresponding to each training sample and the first target model set corresponding to the first original model set.

[0204] The first prediction result corresponding to all training sample data is input into the second original model to obtain the second prediction result corresponding to all training sample data and the second target model corresponding to the second original model. The second prediction result is used to represent the planned site value information corresponding to all training sample data.

[0205] Based on the second prediction results corresponding to all training sample data, the first target model set and the second target model are determined as the prediction models to be verified.

[0206] All validation sample data in the validation sample dataset are input into the prediction model to be validated to obtain the validation results corresponding to all validation sample data. The validation results are used to represent the planning site value information corresponding to all the predicted validation sample data.

[0207] Determine whether the verification result meets the preset verification completion conditions. If the verification result meets the preset verification completion conditions, then determine the prediction model to be verified as the target prediction model.

[0208] In this optional embodiment, the number of cross-validations corresponding to the above-mentioned preset cross-validation conditions and the logic for dividing the sample dataset can be selected according to the actual application scenario, and the present invention does not impose specific limitations.

[0209] It is evident that implementing this optional embodiment can provide a model architecture for a target prediction model and a model training method for a target prediction model. In this optional embodiment, by jointly predicting multiple basic models in the first target model set and the second target model, it is beneficial to improve the prediction accuracy of the target prediction model, thereby improving the scientific nature of planning base station layout.

[0210] In this optional embodiment, as an optional implementation method, the training module 304 is further configured to, when it is determined that the verification result does not meet the preset verification completion conditions, determine whether the fitting degree of the prediction model to be verified meets the abnormal fitting conditions based on the first loss function of the training sample dataset and the second loss function of the verification sample dataset; and when it is determined that the fitting degree meets the abnormal fitting conditions, determine the model correction strategy of the prediction model to be verified based on the fitting degree.

[0211] According to the model correction strategy, the prediction model to be verified is adjusted, and the operation of inputting all the verification sample data in the verification sample dataset into the prediction model to be verified is triggered to obtain the verification results corresponding to all the verification sample data, and the operation of determining whether the verification results meet the preset verification completion conditions is triggered.

[0212] The above-mentioned abnormal fit condition is used to indicate whether the target prediction model is overfitted or underfitted.

[0213] As can be seen, implementing this optional embodiment can determine whether the fitting degree of the prediction model to be validated meets the abnormal fitting condition based on the first loss function and the second loss function, thereby improving the detection accuracy of abnormal fitting in the prediction model to be validated, and thus improving the detection accuracy of abnormal fitting in the target prediction model. Simultaneously, it can determine the model correction strategy for the prediction model to be validated based on its fitting degree, and adjust the prediction model according to the model correction strategy, thereby improving the prediction accuracy of the prediction model to be validated, and thus improving the prediction accuracy of the target prediction model.

[0214] In this optional embodiment, as another optional implementation, the specific method by which the training module 304 determines the first target model set and the second target model as the prediction model to be verified based on the second prediction results corresponding to all training sample data includes:

[0215] The second prediction results corresponding to all training sample data are compared with the actual planned site value information corresponding to all training sample data to obtain the comparison results.

[0216] Determine whether the comparison results meet the preset training completion conditions. If it is determined that the comparison results meet the preset training completion conditions, then the first target model set and the second target model are determined as the prediction models to be verified.

[0217] If it is determined that the comparison result does not meet the preset training completion conditions, then the first target model set and the second target model are adjusted according to the comparison result and the preset training completion conditions, and the adjusted first target model set and the second target model are determined as the prediction models to be verified.

[0218] As can be seen, implementing this optional embodiment can determine whether the first target model set and the second target model are prediction models to be verified based on the comparison results, which can improve the accuracy of determining the prediction models to be verified, and thus provide a guarantee for the subsequent verification of the prediction models to be verified.

[0219] In this optional embodiment, as another optional implementation, the training module 304 adjusts the first target model set and the second target model according to the comparison results and preset training completion conditions in the following specific ways:

[0220] Based on the data deviation characteristics corresponding to the comparison results, at least one deviation influence factor is determined. The deviation influence factor is used to represent the target training sample data and target feature parameters that cause the comparison results to fail to meet the preset training completion conditions. The training sample dataset includes the target training sample data, and the target feature parameters include the feature parameters in the first target model set and / or the feature parameters in the second target model. The feature parameters include weights and biases.

[0221] Based on all bias influencing factors and preset training completion conditions, adjust the first target model set and the second target model.

[0222] As can be seen, implementing this optional embodiment can adjust the first target model set and the second target model from two dimensions: training sample data and model feature parameters, in order to improve the determination accuracy of the prediction model to be verified, thereby providing a guarantee for subsequent verification of the prediction model to be verified.

[0223] Furthermore, the specific methods by which the aforementioned training module 304 adjusts the first target model set and the second target model based on all bias influencing factors and preset training completion conditions include:

[0224] Determine whether the target training sample data exceeds the target sample range. If it is determined that the target training sample data exceeds the target sample range, then perform a removal process on the target training sample data.

[0225] Based on the data characteristics of the training sample dataset after removing the target training sample data, the target feature parameters are adjusted to adjust the first target model set and the second target model.

[0226] As can be seen, implementing this optional embodiment can "cleanse" the training sample data to eliminate the possibility of poor model training quality due to the training data, thereby improving the accuracy of adjusting the first target model set and the second target model, thus improving the determination accuracy of the prediction model to be verified, and providing a guarantee for subsequent verification of the prediction model to be verified.

[0227] In another alternative embodiment, such as Figure 4 As shown, the device also includes:

[0228] The first judgment module 305 is used to determine whether the actual prediction effect of the adjusted first target model set and second target model matches the target prediction effect corresponding to the preset training completion conditions after the training module 304 adjusts the first target model set and second target model according to all bias influencing factors and preset training completion conditions. When the first judgment module 305 determines that the actual prediction effect matches the target prediction effect, it triggers the training module 304 to determine the adjusted first target model set and second target model as the prediction model to be verified. When the first judgment module 305 determines that the actual prediction effect does not match the target prediction effect, it triggers the training module 304 to adjust the target feature parameters according to the data characteristics of the training sample dataset after removing the target training sample data, so as to adjust the first target model set and second target model.

[0229] As can be seen, implementing this optional embodiment can verify and adjust the trained first target model set and second target model multiple times to ensure training quality, thereby improving the prediction accuracy of the prediction model to be verified, which in turn helps to improve the prediction accuracy of the target prediction model.

[0230] In another optional embodiment, the specific method by which the reconstruction module 302 reconstructs the trained target prediction model based on the traffic demand characteristics corresponding to each user feature data includes:

[0231] For each user feature data, determine the user business information corresponding to that user feature data;

[0232] Based on the user's business information, determine the user's corresponding traffic demand characteristics.

[0233] Reconstruct the trained target prediction model based on the traffic demand characteristics of all users.

[0234] It is evident that implementing this optional embodiment can determine the corresponding user service information based on each user's user characteristic data, and determine the traffic demand characteristics of each user based on each user's user service information. Thus, based on the traffic demand characteristics of all users, the trained target prediction model can be reconstructed. This can improve the prediction accuracy, flexibility, and adaptability of the value attribute prediction of the planned sites for different target areas, thereby improving the scientific nature of the planned base station layout.

[0235] And, such as Figure 4 As shown, the device also includes:

[0236] The second judgment module 306 is also used to determine whether the traffic demand characteristics corresponding to the user are abnormal demand characteristics after the reconstruction module 302 determines the traffic demand characteristics corresponding to the user based on the user's user business information.

[0237] The determination module 307 is used to determine that the user is an abnormal user when the second judgment module 306 determines that the user's traffic demand characteristics are abnormal demand characteristics.

[0238] The output module 308 is used to output a set of abnormal user information for abnormal users. The set of abnormal user information includes at least one of the abnormal user's user characteristic data, user business information, and traffic demand characteristics.

[0239] The aforementioned abnormal demand characteristics are used to indicate that the user's actual traffic demand characteristics exceed the available traffic demand characteristics corresponding to the user's user service information.

[0240] As can be seen, implementing this optional embodiment can determine whether the traffic demand characteristics of each user are abnormal demand characteristics based on the user service information of each user, thereby enabling early warning of abnormal traffic risks of planned sites during the site value attribute planning stage, thus ensuring the long-term availability of planned base stations.

[0241] Example 4

[0242] Please see Figure 5 , Figure 5 This is a schematic diagram of another machine learning-based site value attribute prediction device disclosed in an embodiment of the present invention. Figure 5 As shown, the machine learning-based site value attribute prediction device may include:

[0243] Memory 401 that stores executable program code.

[0244] Processor 402 coupled to memory 401.

[0245] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the machine learning-based site value attribute prediction method described in Embodiment 1 or Embodiment 2 of the present invention.

[0246] Example 5

[0247] This invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute steps in the machine learning-based site value attribute prediction method described in Embodiment 1 or Embodiment 2 of this invention.

[0248] Example 6

[0249] This invention discloses a computer program product, which includes a non-transitory computer read storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the machine learning-based site value attribute prediction method described in Embodiment 1 or Embodiment 2.

[0250] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0251] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0252] Finally, it should be noted that the method and apparatus for predicting the value attributes of planning sites based on machine learning disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the value attributes of planning sites based on machine learning, characterized in that, The method includes: Based on the target location information of the site to be planned, user feature data within the target area corresponding to the target location information is collected to obtain a user feature data set. Based on the traffic demand characteristics corresponding to each user feature data, reconstruct the trained target prediction model; Each user feature data in the user feature data set is input into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the predicted site value information of the site to be planned. The site value information includes site traffic information and / or site user capacity information. Furthermore, the target prediction model is trained based on the original prediction model, which includes a first set of original models and a second set of original models. The first set of original models includes at least one base model. Before reconstructing the trained target prediction model based on the traffic demand features corresponding to each user feature data, the method further includes: The training sample dataset is divided according to the preset cross-validation conditions; Each training sample data in the divided training sample dataset is input into the corresponding base model to obtain a first prediction result corresponding to each training sample data and a first target model set corresponding to the first original model set. The first prediction result corresponding to all the training sample data is input into the second original model to obtain the second prediction result corresponding to all the training sample data and the second target model corresponding to the second original model. The second prediction result is used to represent the predicted planning site value information corresponding to all the training sample data. Based on the second prediction results corresponding to all the training sample data, the first target model set and the second target model are determined as prediction models to be verified. All validation sample data in the validation sample dataset are input into the prediction model to be validated to obtain validation results corresponding to all the validation sample data. The validation results are used to represent the planning site value information corresponding to all the predicted validation sample data. Determine whether the verification result meets the preset verification completion condition. If it is determined that the verification result meets the preset verification completion condition, then determine that the prediction model to be verified is the target prediction model. And, determining the first target model set and the second target model as prediction models to be verified based on the second prediction results corresponding to all the training sample data includes: The second prediction result corresponding to all the training sample data is compared with the actual planned site value information corresponding to all the training sample data to obtain the comparison result. Determine whether the comparison result meets the preset training completion conditions. If it is determined that the comparison result meets the preset training completion conditions, then the first target model set and the second target model are determined as the prediction models to be verified. When it is determined that the comparison result does not meet the preset training completion condition, at least one deviation influence factor is determined based on the data deviation characteristics corresponding to the comparison result. The deviation influence factor is used to represent the target training sample data and target feature parameters that cause the comparison result to fail to meet the preset training completion condition. The training sample dataset includes the target training sample data, and the target feature parameters include feature parameters in the first target model set and / or feature parameters in the second target model. The feature parameters include weights and biases. Based on all the aforementioned bias influencing factors and the preset training completion conditions, the first target model set and the second target model are adjusted, and the adjusted first target model set and the second target model are determined as the prediction models to be verified.

2. The method for predicting the value attributes of planning sites based on machine learning according to claim 1, characterized in that, The step involves collecting user feature data within the target area corresponding to the target location information of the site to be planned, thereby obtaining a user feature data set, including: Based on the target location information of the site to be planned, determine the target area corresponding to the target location information; Based on the coverage area of ​​the target region, at least one data collection point is determined; For each data acquisition point, the data acquisition range corresponding to the data acquisition point is determined, and the initial feature data within the data acquisition range corresponding to the data acquisition point is collected to obtain the initial feature data set. For each initial feature data set, it is determined whether there is abnormal feature data in the initial feature data set. When it is determined that there is abnormal feature data in the initial feature data set, the abnormal feature data is removed to obtain a subset of user feature data of the data collection point corresponding to the initial feature data set. Based on the subset of user feature data from all the data collection points, determine the set of user feature data within the target area corresponding to the target location information.

3. The method for predicting the value attributes of planning sites based on machine learning according to claim 1 or 2, characterized in that, The base model includes at least one of the following: gradient boosting regression model, extreme tree regression model, random forest regression model, and adaptive boosting regression model; the second original model includes a linear regression model.

4. The method for predicting the value attributes of planning sites based on machine learning according to claim 3, characterized in that, When it is determined that the verification result does not meet the preset verification completion condition, the method further includes: Based on the first loss function of the training sample dataset and the second loss function of the validation sample dataset, it is determined whether the fit of the prediction model to be validated meets the abnormal fit condition. When it is determined that the fit meets the abnormal fit condition, the model correction strategy of the prediction model to be validated is determined based on the fit. According to the model correction strategy, the prediction model to be verified is adjusted, and the operation of inputting all the verification sample data in the verification sample dataset into the prediction model to be verified to obtain the verification results corresponding to all the verification sample data is triggered, as well as the operation of determining whether the verification results meet the preset verification completion conditions is triggered.

5. The method for predicting the value attributes of planning sites based on machine learning according to claim 1, characterized in that, The step of adjusting the first target model set and the second target model based on all the aforementioned bias influencing factors and the preset training completion conditions includes: Determine whether the target training sample data exceeds the target sample interval. If it is determined that the target training sample data exceeds the target sample interval, then perform a removal process on the target training sample data. Based on the data characteristics of the training sample dataset after removing the target training sample data, the target feature parameters are adjusted to adjust the first target model set and the second target model.

6. The method for predicting the value attributes of planning sites based on machine learning according to claim 5, characterized in that, After adjusting the first target model set and the second target model according to all the aforementioned bias influencing factors and the preset training completion conditions, the method further includes: Determine whether the actual prediction effect corresponding to the adjusted first target model set and the second target model matches the target prediction effect corresponding to the preset training completion condition. When it is determined that the actual prediction effect matches the target prediction effect, the operation of determining the adjusted first target model set and the second target model as the prediction model to be verified is triggered. When it is determined that the actual prediction effect does not match the target prediction effect, the operation of adjusting the target feature parameters based on the data features of the training sample dataset after removing the target training sample data is triggered, so as to adjust the first target model set and the second target model.

7. The method for predicting the value attributes of planning sites based on machine learning according to any one of claims 4-6, characterized in that, The step of reconstructing the trained target prediction model based on the traffic demand features corresponding to each user feature data includes: For each user feature data, determine the user business information corresponding to that user feature data; Based on the user's business information, determine the user's corresponding traffic demand characteristics; Based on the traffic demand characteristics of all the users, reconstruct the trained target prediction model; Furthermore, after determining the user's traffic demand characteristics based on the user's service information, the method further includes: Determine whether the traffic demand characteristics corresponding to the user are abnormal demand characteristics. If it is determined that the traffic demand characteristics corresponding to the user are abnormal demand characteristics, then the user is identified as an abnormal user, and an abnormal user information set of the abnormal user is output. The abnormal user information set includes at least one of the abnormal user's user characteristic data, user business information, and traffic demand characteristics.

8. A machine learning-based site value attribute prediction device, characterized in that, The apparatus is used to perform the machine learning-based site value attribute prediction method as described in any one of claims 1-7, and the apparatus comprises: The data acquisition module is used to collect user feature data within the target area corresponding to the target location information of the site to be planned, and obtain a user feature data set. The reconstruction module reconstructs the trained target prediction model based on the traffic demand characteristics corresponding to each user feature data. The prediction module inputs each user feature data in the user feature data set into the reconstructed target prediction model to obtain the prediction result of the site to be planned. The prediction result is used to represent the predicted site value information of the site to be planned. The site value information includes site traffic information and / or site user capacity information.

9. A machine learning-based site value attribute prediction device, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the machine learning-based site value attribute prediction method as described in any one of claims 1-7.

Images