Mobile base station backhaul device and method

Abstract

The present invention relates to the field of mobile communication technology, and in particular to a backhaul device and method for a mobile base station, the method comprising the following steps: deploying a high-gain cluster antenna, a plurality of CPE units and an intelligent control unit on a vehicle where the mobile base station is located; wherein an AI large model is deployed in the intelligent control unit, and the working indicators of the plurality of CPE units are continuously collected in real time by the intelligent control unit, and the working indicators of the plurality of CPE units are comprehensively analyzed and sorted by using the trained AI large model; according to the comprehensive analysis and sorting results, the optimal CPE unit is automatically switched to provide a backhaul channel for the mobile base station, so as to realize intelligent communication management and make the backhaul channel more stable. The present invention is applicable to a base station system, and also to a base station plus a repeater extension coverage system, and is particularly applicable to a distributed wireless system, including a fiber-optic remote distributed system, a cable-remote frequency-shifted repeater system, a frequency-shifting system, a sea area communication, a low-altitude communication, and the like.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technology, and in particular to a backhaul device and method for a mobile base station. Background Art

[0002] At present, most 4G / 5G base stations for mobile communications are deployed in fixed locations and rely on fixed optical fiber and other methods for backhaul. For areas where it is difficult to select a site, or in a shielded vehicle, due to physical limitations, it is impossible to receive signals from fixed base stations outside. In places far away from fixed base stations, such as sea or airspace, ordinary mobile phones have limited transmission power and receiving gain capabilities, resulting in poor reception of base station signals and inability to effectively use 4G / 5G networks. In response to these communication blind spots, there is an urgent need for a technology that can deploy mobile base stations on vehicles (such as ship-borne base stations, ship-borne base stations, vehicle-borne base stations, base stations on aircraft, etc.) and provide them with stable and controllable backhaul, so as to achieve good signal coverage in these areas and meet people's communication needs in special scenarios. Summary of the invention

[0003] The purpose of the present invention is to provide a backhaul device and method for a mobile base station, to provide a stable and reliable backhaul channel for a base station on a mobile vehicle (referred to as a mobile base station), so that the mobile base station can be connected to a core network to perform business.

[0004] To achieve the purpose of the present invention, the following technical solutions are adopted:

[0005] A first aspect of the present invention provides a backhaul method for a mobile base station, which is used to provide a backhaul channel for the mobile base station on a vehicle, and includes the following steps:

[0006] A high-gain cluster antenna, multiple CPE units and an intelligent control unit are deployed on the vehicle where the mobile base station is located; wherein a large AI model is deployed in the intelligent control unit, and multiple CPE units share one high-gain cluster antenna to obtain macro base station signals and communicate with the core network through the macro base station;

[0007] After the initial deployment, the intelligent control unit randomly selects a CPE unit connected to the core network as a backhaul channel for the mobile base station;

[0008] Performing model training on the AI ​​big model in the intelligent control unit;

[0009] The intelligent control unit continuously collects the working indicators of the multiple CPE units in real time and uses the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units;

[0010] Based on the comprehensive analysis and sorting results, the system automatically switches to the best CPE unit to provide a backhaul channel for the mobile base station.

[0011] A further improvement is that the high-gain beam-forming antenna is an omnidirectional antenna having multiple radiation surfaces, and multiple CPE units are connected to different radiation surfaces of the high-gain beam-forming antenna.

[0012] A further improvement is that the core network allocates a fixed IP address to each CPE unit.

[0013] A further improvement is that the AI ​​big model adopts a deep neural network model, and the specific method of training the AI ​​big model in the intelligent control unit includes:

[0014] Data collection: Collect time data, obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract the time features of hours, days, weeks, and months; collect location data, use the GPS module to obtain the longitude and latitude information of the intelligent control unit, and obtain the approximate location information through network positioning technology when there is no GPS signal and it is unavailable; collect CPE working index data, each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit; collect CPE fault information data, the CPE unit itself has a fault detection mechanism, by identifying hardware faults, connection interruptions, signal loss faults, and feeding back the fault type and occurrence time to the intelligent control unit;

[0015] Preprocessing of collected data: removing outliers and erroneous data from the collected data, standardizing numerical data and converting them into data with a mean of 0 and a standard deviation of 1, and performing one-hot encoding or label encoding on discrete features and categorical data;

[0016] Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used for model training, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the final performance of the model.

[0017] Model training process: define the mean square error loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score; select the Adam optimizer and use its adaptive learning rate feature to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level.

[0018] A further improvement is that the specific method of continuously collecting the working indicators of the multiple CPE units in real time through the intelligent control unit and using the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units includes:

[0019] The intelligent control unit continuously collects time data, location data, CPE working index data, CPE fault information data in real time, and performs real-time preprocessing;

[0020] The pre-processed time data, location data, CPE work indicator data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit.

[0021] The plurality of CPE units are ranked according to the score or selection probability of each CPE unit.

[0022] A second aspect of the present invention provides a backhaul device for a mobile base station, which is used to be deployed on a vehicle to provide a backhaul channel for the mobile base station. The backhaul device includes:

[0023] High-gain cluster antenna;

[0024] A plurality of CPE units, wherein the plurality of CPE units share one high-gain cluster antenna to obtain macro base station signals and communicate with the core network through the macro base station;

[0025] An intelligent control unit, the intelligent control unit is connected to the plurality of CPE units, an AI large model is deployed in the intelligent control unit, wherein the intelligent control unit comprises:

[0026] An initial backhaul channel establishment module is used to randomly select a CPE unit connected to the core network as a backhaul channel for the mobile base station after the backhaul device is initially deployed;

[0027] A model training module, used to train the AI ​​big model in the intelligent control unit;

[0028] A data collection and analysis module, used for continuously collecting the working indicators of the plurality of CPE units in real time and using the trained AI big model to comprehensively analyze and sort the working indicators of the plurality of CPE units;

[0029] The backhaul channel switching module is used to automatically switch to the best CPE unit to provide a backhaul channel for the mobile base station based on the comprehensive analysis and sorting results.

[0030] A further improvement is that the high-gain beam-forming antenna is an omnidirectional antenna having multiple radiation surfaces, and multiple CPE units are connected to different radiation surfaces of the high-gain beam-forming antenna.

[0031] A further improvement is that the core network allocates a fixed IP address to each CPE unit.

[0032] A further improvement is that the AI ​​big model adopts a deep neural network model, and the model training of the AI ​​big model in the intelligent control unit specifically includes:

[0033] Data collection: Collect time data, obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract the time features of hours, days, weeks, and months; collect location data, use the GPS module to obtain the longitude and latitude information of the intelligent control unit, and obtain the approximate location information through network positioning technology when there is no GPS signal and it is unavailable; collect CPE working index data, each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit; collect CPE fault information data, the CPE unit itself has a fault detection mechanism, by identifying hardware faults, connection interruptions, signal loss faults, and feeding back the fault type and occurrence time to the intelligent control unit;

[0034] Preprocessing of collected data: removing outliers and erroneous data from the collected data, standardizing numerical data and converting them into data with a mean of 0 and a standard deviation of 1, and performing one-hot encoding or label encoding on discrete features and categorical data;

[0035] Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used for model training, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the final performance of the model.

[0036] Model training process: define the mean square error loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score; select the Adam optimizer and use its adaptive learning rate feature to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level.

[0037] A further improvement is that the continuous real-time collection of the working indicators of the multiple CPE units and the use of the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units include:

[0038] The data collection and analysis module continuously collects time data, location data, CPE work index data, CPE fault information data in real time, and performs real-time preprocessing;

[0039] The pre-processed time data, location data, CPE work indicator data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit.

[0040] The plurality of CPE units are ranked according to the score or selection probability of each CPE unit.

[0041] The beneficial effects of the present invention are:

[0042] The present invention provides a return channel for a mobile base station by using multiple CPE units with an omnidirectional high-gain cluster antenna on a vehicle, and combining an intelligent control unit (AIBOX) with an AI large model function, which has significant advantages, and is particularly suitable for vehicles such as ships, aircraft, and vehicles with fixed routes. Specifically, the high-gain cluster antenna is matched with multiple CPE units, which not only provides a higher receiving gain for the CPE unit, but also increases the transmission power, effectively enhancing the signal reception and transmission capabilities; the main and standby use of multiple CPE units can always select the optimal signal of the macro base station to ensure the quality and stability of the communication signal; the intelligent control unit can provide feedback and training based on multi-dimensional information such as time, geographic location, signal quality and rate of CPE by relying on the AI ​​large model. During use, the AI ​​large model can automatically select the best CPE unit as the return channel of the mobile base station, realize intelligent communication management, and make the return channel more stable. In addition, even if the deployment or network optimization plan of the macro base station changes, the intelligent control unit can easily respond with its intelligent learning and adaptive capabilities without human intervention, ensuring the continuous and stable operation of communication.

[0043] The present invention is applicable to base station systems, and also to base station plus repeater extended coverage systems, and is particularly applicable to distributed wireless systems, including optical fiber remote distributed systems, cable remote frequency shift repeater systems, frequency shift systems, sea area communications, low-altitude communications, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of a backhaul method of a mobile base station according to the present invention;

[0045] Figure 2 The present invention is a schematic diagram of the component connection structure of a backhaul device of a mobile base station. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0047] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0048] The following is an explanation of some technical terms involved in the present invention:

[0049] CPE (Customer Premises Equipment): Customer Premises Equipment. CPE is the communication intermediary between mobile base stations and macro base stations, responsible for signal reception, conversion and return.

[0050] SINR (Signal to Interference and Noise Ratio) reflects the quality of the signal in an interference environment.

[0051] RSRP (Reference Signal Received Power) directly reflects the signal strength received by the CPE.

[0052] Please refer to the attached Figure 1 , Attachment Figure 2 In a first aspect of an embodiment of the present invention, a backhaul method of a mobile base station is provided, which is used to provide a backhaul channel for the mobile base station on a vehicle, such as Figure 1 As shown, the method comprises the following steps:

[0053] Step S1: deploy a high-gain cluster antenna, multiple CPE units and an intelligent control unit (AIBOX) on the vehicle where the mobile base station is located; wherein a large AI model is deployed in the intelligent control unit, and multiple CPE units share one high-gain cluster antenna to obtain macro base station signals and connect to the core network through the macro base station.

[0054] It is understandable that the vehicle can be a train, a ship, an aircraft, etc. The high-gain cluster antenna is used to more effectively receive the signal sent by the macro base station and improve the strength and quality of the signal reception. The high-gain cluster antenna and multiple CPE units are installed on the outer top of the vehicle.

[0055] Step S2: After the initial deployment, the intelligent control unit randomly selects a CPE unit connected to the core network as the backhaul channel of the mobile base station.

[0056] Step S3: Perform model training on the AI ​​large model in the intelligent control unit.

[0057] It is understandable that the AI ​​big model deployed in the intelligent control unit is trained using certain data sets and algorithms, allowing the model to learn the working characteristics of the CPE unit in different situations, various factors related to the backhaul of the mobile base station, etc., so that it can more accurately analyze and evaluate the working conditions of the CPE unit.

[0058] Step S4: The intelligent control unit continuously collects the working indicators of the multiple CPE units in real time and uses the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units.

[0059] It is understandable that the comprehensive analysis and ranking of the AI ​​big model can evaluate the performance of each CPE unit more comprehensively and objectively, providing a scientific basis for selecting the best return channel.

[0060] Step S5: Automatically switch to the best CPE unit to provide a backhaul channel for the mobile base station based on the comprehensive analysis and sorting results.

[0061] It is understandable that automatically switching to the best CPE unit can optimize the backhaul channel of the mobile base station in real time, improve the efficiency and stability of data transmission, reduce latency and packet loss rate, and provide users with a better communication experience. At the same time, this automatic switching function can be dynamically adjusted according to actual conditions to adapt to different working environments and network conditions.

[0062] In a preferred solution of this embodiment, the high-gain cluster antenna is an omnidirectional antenna having multiple radiating surfaces. In this embodiment, there are three radiating surfaces, each of which can radiate 120 degrees, thereby achieving a radiation direction of 360 degrees. Multiple CPE units are connected to different radiating surfaces of the high-gain cluster antenna.

[0063] It should be understood that since the antenna of the macro base station is basically multi-sector transmitting, multiple CPE units are connected to different radiating surfaces of this high-gain beam antenna (for example, CPE unit A is connected to the radiating unit of radiating surface 1, CPE unit B is connected to the radiating unit of radiating surface 2, and CPE unit C is connected to the radiating unit of radiating surface 3). This ensures that at least one CPE unit can obtain the signal of the macro base station, thereby not losing the macro base station signal in any direction in space.

[0064] In a preferred solution of this embodiment, in order to ensure the stability of the backhaul route, it is necessary to avoid unstable transmission in the backhaul channel due to changes in the IP address. Therefore, in this embodiment, the core network allocates a fixed IP address to each CPE unit.

[0065] In this embodiment, the AI ​​big model adopts a deep neural network model (DNN). DNN has a strong nonlinear fitting ability and can process complex input data and mapping relationships. It can learn the optimal decision-making strategy in a dynamic environment. In step S3, the specific method of training the AI ​​big model in the intelligent control unit includes:

[0066] Step S31: Data collection:

[0067] Collect time data and obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract hour, day, week, and month time features at the same time, because different time dimensions may affect the working status of the CPE unit. For example, there will be obvious differences in network usage during peak and off-peak hours on weekdays.

[0068] Collect location data and use the GPS (Global Positioning System) module to obtain the latitude and longitude information of the intelligent control unit. When there is no GPS signal and it is unavailable, use network positioning technology to obtain approximate location information. Because different locations are different, the surrounding signal interference, base station coverage, etc. will vary.

[0069] Collect CPE working index data. Each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit. These indicators directly reflect the working performance and network quality of the CPE unit.

[0070] Collect CPE fault information data. The CPE unit itself has a fault detection mechanism. By identifying hardware failures, connection interruptions, and signal loss failures, it feeds back the fault type and occurrence time to the intelligent control unit.

[0071] It should be understood that by collecting multiple types of data, the working environment and status of the CPE unit can be described from different perspectives, providing rich information for model training, so that the model can learn more complex patterns and laws. Data collection based on time, location, working indicators and fault information can accurately capture the key factors that affect the performance of the CPE unit, helping the model to make more accurate predictions and decisions.

[0072] Step S32: preprocessing the collected data: remove outliers and erroneous data (such as obviously unreasonable values ​​of RSRP or SINR) from the collected data, standardize numerical data such as RSRP, SINR and rate, and convert them into data with a mean of 0 and a standard deviation of 1, so that the model can treat each feature more fairly to speed up the training of the model, and perform one-hot encoding or label encoding on discrete features such as hours, days, weeks, months in time data and categorical data such as CPE fault types so that the model can process them.

[0073] Step S33: Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used to train the model so that the model can learn the patterns and rules in the data; the validation set is used to adjust the hyperparameters of the model, such as learning rate, number of hidden layer neurons, etc., to find the optimal model configuration; the test set is used to evaluate the final performance of the model and test the generalization ability of the model on unseen data.

[0074] Step S34: Model training process: define the mean square error (MSE) loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score. For the reinforcement learning model, use the cumulative reward as the optimization target; select the Adam (Adaptive Moment Estimation) optimizer, and use its adaptive learning rate characteristics to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, and then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level.

[0075] It should be understood that the definition of the loss function provides a clear goal for model training, allowing the model to optimize in the direction of minimizing error or maximizing reward. The adaptive learning rate feature of the Adam optimizer can speed up the convergence of the model, reduce training time, and improve training efficiency. Through continuous iterative training and parameter updates, the model can gradually learn the complex patterns and laws in the data, thereby improving the model's prediction accuracy and decision-making ability.

[0076] Specifically, in this embodiment, in step S4, the specific method of continuously collecting the working indicators of the multiple CPE units in real time by the intelligent control unit and using the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units includes:

[0077] Step S41: The intelligent control unit continuously collects time data, location data, CPE work index data, and CPE fault information data in real time, and performs real-time preprocessing.

[0078] It is understandable that continuous real-time data collection can reflect the latest status of the CPE unit at every moment, ensuring that subsequent analysis is based on the latest network conditions. For example, when a vehicle moves to an area with weak signals, real-time data collection can reflect this change in a timely manner.

[0079] Step S42: The pre-processed time data, location data, CPE work index data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit.

[0080] It is understandable that the AI ​​big model can comprehensively consider multiple factors such as time, location, working indicators and fault information to conduct a comprehensive and objective evaluation of the CPE unit, avoiding the limitation of judging based on only a single indicator. For example, a CPE unit may have a high current signal strength, but often experience connection interruption failures. A comprehensive evaluation can more accurately reflect its actual availability. The score or selection probability provides a clear quantitative basis for subsequent channel selection, making the decision-making process more scientific and reasonable.

[0081] Step S43: Sort the multiple CPE units according to the score or selection probability of each CPE unit.

[0082] It can be understood that after sorting, it can be intuitively seen which CPE unit has the best performance, which facilitates the intelligent control unit to make a quick decision and automatically switch to the best CPE unit to provide a backhaul channel for the mobile base station.

[0083] In addition, the intelligent control unit will record the actual communication effect after each switching of the CPE unit, such as the actual data transmission rate, packet loss rate, etc., and use it as feedback data. According to the changes in the feedback data, the new feedback data will be added to the training data set regularly to retrain the AI ​​large model to adapt to changes in the network environment and improve the performance of the model. As the model is continuously trained and adjusted based on the feedback data, it can gradually avoid some misjudgments caused by insufficient or inaccurate initial training data.

[0084] The intelligent control unit has exception handling and fault tolerance mechanisms:

[0085] Fault detection: The intelligent control unit monitors the status of the CPE unit in real time. When a CPE unit is detected to be faulty, it is promptly excluded from the optional list and the input data of the model is updated.

[0086] Backup strategy: If all CPE units fail or the model cannot make an effective decision, the intelligent control unit can adopt a backup strategy, such as selecting CPE according to a preset priority or trying to reinitialize the CPE.

[0087] The present invention provides a return channel for a mobile base station by using multiple CPE units with an omnidirectional high-gain cluster antenna on a vehicle, and combining an intelligent control unit (AIBOX) with an AI large model function, which has significant advantages, and is particularly suitable for vehicles such as ships, aircraft, and vehicles with fixed routes. Specifically, the high-gain cluster antenna is matched with multiple CPE units, which not only provides a higher receiving gain for the CPE unit, but also increases the transmission power, effectively enhancing the signal reception and transmission capabilities; the main and standby use of multiple CPE units can always select the optimal signal of the macro base station to ensure the quality and stability of the communication signal; the intelligent control unit can provide feedback and training based on multi-dimensional information such as time, geographic location, signal quality and rate of CPE by relying on the AI ​​large model. During use, the AI ​​large model can automatically select the best CPE unit as the return channel of the mobile base station, realize intelligent communication management, and make the return channel more stable. In addition, even if the deployment or network optimization plan of the macro base station changes, the intelligent control unit can easily respond with its intelligent learning and adaptive capabilities without human intervention, ensuring the continuous and stable operation of communication.

[0088] like Figure 2 As shown, a second aspect of an embodiment of the present invention provides a backhaul device of a mobile base station, which is used to be deployed on a vehicle to provide a backhaul channel for the mobile base station. The backhaul device of the mobile base station is used to perform a backhaul method of the mobile base station. Specifically, the backhaul device includes:

[0089] High gain cluster antenna.

[0090] Multiple CPE units share one high-gain cluster antenna to obtain macro base station signals and communicate with the core network through the macro base station.

[0091] An intelligent control unit, the intelligent control unit is connected to the plurality of CPE units, an AI large model is deployed in the intelligent control unit, wherein the intelligent control unit comprises:

[0092] An initial backhaul channel establishment module is used to randomly select a CPE unit connected to the core network as a backhaul channel for the mobile base station after the backhaul device is initially deployed;

[0093] A model training module, used to train the AI ​​big model in the intelligent control unit;

[0094] A data collection and analysis module, used for continuously collecting the working indicators of the plurality of CPE units in real time and using the trained AI big model to comprehensively analyze and sort the working indicators of the plurality of CPE units;

[0095] The backhaul channel switching module is used to automatically switch to the best CPE unit to provide a backhaul channel for the mobile base station based on the comprehensive analysis and sorting results.

[0096] Specifically, the high-gain beam-forming antenna is an omnidirectional antenna having multiple radiation surfaces, and the multiple CPE units are connected to different radiation surfaces of the high-gain beam-forming antenna.

[0097] Specifically, the core network allocates a fixed IP address to each CPE unit.

[0098] Specifically, the AI ​​big model adopts a deep neural network model, and the model training of the AI ​​big model in the intelligent control unit specifically includes:

[0099] Data collection: Collect time data, obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract time features of hours, days, weeks, and months; collect location data, use the GPS module to obtain the longitude and latitude information of the intelligent control unit, and obtain approximate location information through network positioning technology when there is no GPS signal and it is unavailable; collect CPE work indicator data, each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit; collect CPE fault information data, the CPE unit itself has a fault detection mechanism, by identifying hardware faults, connection interruptions, signal loss faults, and feeding back the fault type and occurrence time to the intelligent control unit.

[0100] Preprocess the collected data: remove outliers and erroneous data from the collected data, standardize the numerical data and convert it into data with a mean of 0 and a standard deviation of 1, and perform one-hot encoding or label encoding on discrete features and categorical data.

[0101] Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used for model training, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the final performance of the model.

[0102] Model training process: define the mean square error loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score. For the reinforcement learning model, use the cumulative reward as the optimization target; select the Adam optimizer and use its adaptive learning rate feature to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, and then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level.

[0103] Specifically, the continuously real-time collection of the working indicators of the plurality of CPE units and the use of the trained AI big model to comprehensively analyze and sort the working indicators of the plurality of CPE units specifically include:

[0104] The data collection and analysis module continuously collects time data, location data, CPE work index data, CPE fault information data in real time, and performs real-time preprocessing.

[0105] The pre-processed time data, location data, CPE work indicator data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit.

[0106] The plurality of CPE units are ranked according to the score or selection probability of each CPE unit.

[0107] Since a backhaul device of a mobile base station provided in an embodiment of the present invention corresponds to a backhaul method of a mobile base station provided in the above-mentioned embodiment of the present invention, the implementation method of the above-mentioned backhaul method of a mobile base station is also applicable to a backhaul device of a mobile base station provided in this embodiment. Therefore, this embodiment will no longer provide a more detailed description of the backhaul device of a mobile base station, and those skilled in the art may specifically refer to the above-mentioned description of the backhaul method of a mobile base station.

[0108] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A backhaul method for a mobile base station, used to provide a backhaul channel for the mobile base station on a vehicle, characterized in that: The following steps are involved: A high-gain cluster antenna, multiple CPE units and an intelligent control unit are deployed on the vehicle where the mobile base station is located; wherein an AI large model is deployed in the intelligent control unit, and the AI ​​large model adopts a deep neural network model. Multiple CPE units share one high-gain cluster antenna to obtain macro base station signals and communicate with the core network through the macro base station; After the initial deployment, the intelligent control unit randomly selects a CPE unit connected to the core network as a backhaul channel for the mobile base station; Performing model training on the AI ​​big model in the intelligent control unit includes: Data collection: Collect time data, obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract the time features of hours, days, weeks, and months; collect location data, use the GPS module to obtain the longitude and latitude information of the intelligent control unit, and obtain the approximate location information through network positioning technology when there is no GPS signal and it is unavailable; collect CPE working index data, each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit; collect CPE fault information data, the CPE unit itself has a fault detection mechanism, by identifying hardware faults, connection interruptions, signal loss faults, and feeding back the fault type and occurrence time to the intelligent control unit; Preprocessing of collected data: removing outliers and erroneous data from the collected data, standardizing numerical data and converting them into data with a mean of 0 and a standard deviation of 1, and performing one-hot encoding or label encoding on discrete features and categorical data; Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used for model training, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the final performance of the model. Model training process: define the mean square error loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score; select the Adam optimizer and use its adaptive learning rate feature to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level; The intelligent control unit continuously collects the working indicators of multiple CPE units in real time and uses the trained AI big model to comprehensively analyze and sort the working indicators of multiple CPE units; Based on the comprehensive analysis and sorting results, the system automatically switches to the best CPE unit to provide a backhaul channel for the mobile base station.

2. A backhaul method for a mobile base station according to claim 1, characterized in that: The high-gain cluster antenna is an omnidirectional antenna having multiple radiation surfaces, and multiple CPE units are connected to different radiation surfaces of the high-gain cluster antenna.

3. The backhaul method of a mobile base station according to claim 1, characterized in that: The core network allocates a fixed IP address to each CPE unit.

4. The backhaul method of a mobile base station according to claim 1, characterized in that: The specific method of continuously collecting the working indicators of multiple CPE units in real time by the intelligent control unit and using the trained AI big model to comprehensively analyze and sort the working indicators of multiple CPE units includes: The intelligent control unit continuously collects time data, location data, CPE working index data, CPE fault information data in real time, and performs real-time preprocessing; The pre-processed time data, location data, CPE work indicator data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit. The plurality of CPE units are ranked according to the score or selection probability of each CPE unit.

5. A backhaul device for a mobile base station, used to be deployed on a vehicle to provide a backhaul channel for the mobile base station, characterized in that: The return device comprises: High-gain cluster antenna; Multiple CPE units share one high-gain cluster antenna to obtain macro base station signals and communicate with the core network through the macro base station; An intelligent control unit, wherein the intelligent control unit is connected to a plurality of CPE units, wherein an AI large model is deployed in the intelligent control unit, and the AI ​​large model adopts a deep neural network model, wherein the intelligent control unit includes: An initial backhaul channel establishment module is used to randomly select a CPE unit connected to the core network as a backhaul channel for the mobile base station after the backhaul device is initially deployed; The model training module is used to train the AI ​​big model in the intelligent control unit, including: Data collection: Collect time data, obtain the current timestamp through the system clock, accurate to seconds or minutes, and extract the time features of hours, days, weeks, and months; collect location data, use the GPS module to obtain the longitude and latitude information of the intelligent control unit, and obtain the approximate location information through network positioning technology when there is no GPS signal and it is unavailable; collect CPE working index data, each CPE unit measures its own RSRP, SINR and data transmission rate in real time and feeds back to the intelligent control unit; collect CPE fault information data, the CPE unit itself has a fault detection mechanism, by identifying hardware faults, connection interruptions, signal loss faults, and feeding back the fault type and occurrence time to the intelligent control unit; Preprocessing of collected data: removing outliers and erroneous data from the collected data, standardizing numerical data and converting them into data with a mean of 0 and a standard deviation of 1, and performing one-hot encoding or label encoding on discrete features and categorical data; Training data preparation: Divide the preprocessed data into training set, validation set and test set according to the proportion. The training set is used for model training, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the final performance of the model. Model training process: define the mean square error loss function to measure the difference between the CPE score predicted by the model and the actual optimal CPE score; select the Adam optimizer and use its adaptive learning rate feature to make the model converge to the optimal solution faster; input the training data into the model, calculate the output result through forward propagation, then calculate the loss value according to the loss function, and then update the model parameters through back propagation, and continuously iterate this process until the performance of the model reaches a satisfactory level; A data collection and analysis module, which is used to continuously collect the working indicators of multiple CPE units in real time and use the trained AI big model to comprehensively analyze and sort the working indicators of multiple CPE units; The backhaul channel switching module is used to automatically switch to the best CPE unit to provide a backhaul channel for the mobile base station based on the comprehensive analysis and sorting results.

6. The backhaul device of a mobile base station according to claim 5, characterized in that: The high-gain cluster antenna is an omnidirectional antenna having multiple radiation surfaces, and multiple CPE units are connected to different radiation surfaces of the high-gain cluster antenna.

7. The backhaul device of a mobile base station according to claim 5, characterized in that: The core network allocates a fixed IP address to each CPE unit.

8. The backhaul device of a mobile base station according to claim 5, characterized in that: The continuously real-time collection of the working indicators of the multiple CPE units and the use of the trained AI big model to comprehensively analyze and sort the working indicators of the multiple CPE units specifically include: The data collection and analysis module continuously collects time data, location data, CPE work index data, CPE fault information data in real time, and performs real-time preprocessing; The pre-processed time data, location data, CPE work indicator data, and CPE fault information data are input into the trained AI big model for comprehensive analysis. The AI ​​big model outputs the score or selection probability of each CPE unit. The plurality of CPE units are ranked according to the score or selection probability of each CPE unit.

Images