Storage space adjustment method, device and storage medium

Abstract

The present invention discloses a storage space adjustment method, device and storage medium, comprising: receiving a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit; determining whether the model meets the threshold requirement based on the performance evaluation result of the model and the model performance threshold set for the model training task; and adjusting the storage space of the base station centralized unit based on whether the model meets the threshold requirement. The base station centralized unit adjusts the storage space of the base station centralized unit according to the returned instruction. The present invention overcomes the problem of low model performance due to storage space, and improves the overall utilization efficiency of storage resources.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular to a storage space adjustment method, device, and storage medium. Background Art

[0002] For network performance analysis, AI (Artificial Intelligence) model training technology is currently used.

[0003] The shortcoming of existing technology is that due to the small storage capacity of network elements, there is little performance data, which is very limited for AI model training and affects the model accuracy. Summary of the Invention

[0004] The present invention provides a storage space adjustment method, device and storage medium to solve the problem that due to the small storage capacity of network elements, performance data is scarce, which is very limited for AI model training and affects model accuracy.

[0005] The present invention provides the following technical solutions:

[0006] A storage space adjustment method, comprising:

[0007] Receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0008] Determine whether the model meets the threshold requirements based on the model performance evaluation results and the model performance threshold set for the model training task;

[0009] The storage space of the base station centralized unit is adjusted according to whether the model meets the threshold requirement.

[0010] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0011] During implementation, the storage space of the base station centralized unit is adjusted based on whether the model meets the threshold requirements, including:

[0012] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0013] In implementation, the storage space is reduced in a step-by-step manner; and / or,

[0014] Increase the storage space in steps.

[0015] In implementation, the base station centralized unit is instructed to adjust the storage space via a Storage Space Change Request message.

[0016] During implementation, for a base station centralized unit that meets the threshold requirement, the method further includes:

[0017] The online reasoning module of the centralized unit of the base station is instructed to adopt the best performing model among the models that meet the threshold requirement.

[0018] In implementation, for a base station centralized unit that does not meet the threshold requirement, one of the following indications or a combination thereof is further included:

[0019] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0020] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0021] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0022] During implementation, it further includes:

[0023] receiving a storage space allocation amount sent by a data storage module of a central unit of a base station;

[0024] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0025] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0026] A storage space adjustment method, comprising:

[0027] The base station centralized unit sends the model and the performance evaluation results corresponding to the model, where the model is the network performance AI training model executed on the centralized unit;

[0028] The base station central unit adjusts the storage space of the base station central unit according to the returned instruction.

[0029] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0030] During implementation, the storage space of the base station centralized unit is adjusted in a step-size manner according to the instruction.

[0031] In implementation, the instruction to adjust the storage space is received via a Storage Space Change Request message.

[0032] During implementation, it further includes:

[0033] The online reasoning module of the base station centralized unit adopts the model with the best performance among the indicated models according to the instruction.

[0034] During implementation, one or a combination of the following instructions is further included:

[0035] The online reasoning module of the base station centralized unit adopts the model with the best performance among the indicated models according to the instruction;

[0036] configuring the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning according to the instruction;

[0037] Migrate the model from the migration source as instructed.

[0038] During implementation, it further includes:

[0039] The data storage module of the central unit of the base station sends the storage space allocation amount.

[0040] A storage space adjustment device, comprising:

[0041] The processor reads the program from the memory and performs the following steps:

[0042] Receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0043] Determine whether the model meets the threshold requirements based on the model performance evaluation results and the model performance threshold set for the model training task;

[0044] Adjusting the storage space of the base station centralized unit based on whether the model meets the threshold requirement;

[0045] A transceiver is used to receive and send data under the control of the processor.

[0046] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0047] During implementation, the storage space of the base station centralized unit is adjusted based on whether the model meets the threshold requirements, including:

[0048] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0049] In implementation, the storage space is reduced in a step-by-step manner; and / or,

[0050] Increase the storage space in steps.

[0051] In implementation, the base station centralized unit is instructed to adjust the storage space via a Storage Space Change Request message.

[0052] During implementation, for a base station centralized unit that meets the threshold requirement, the method further includes:

[0053] The online reasoning module of the centralized unit of the base station is instructed to adopt the best performing model among the models that meet the threshold requirement.

[0054] In implementation, for a base station centralized unit that does not meet the threshold requirement, one of the following indications or a combination thereof is further included:

[0055] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0056] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0057] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0058] During implementation, it further includes:

[0059] receiving a storage space allocation amount sent by a data storage module of a central unit of a base station;

[0060] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0061] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0062] A storage space adjustment device, comprising:

[0063] A receiving unit, configured to receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0064] A determination unit, configured to determine whether the model meets the threshold requirement based on the performance evaluation result of the model and the model performance threshold set for the model training task;

[0065] The indicating unit is used to adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement.

[0066] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0067] In implementation, the instructing unit is further configured to adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement, including:

[0068] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0069] In implementation, the instructing unit is further configured to instruct to reduce the storage space in a step-by-step manner; and / or to increase the storage space in a step-by-step manner.

[0070] In implementation, the instructing unit is further configured to instruct the base station central unit to adjust the storage space via a Storage Space Change Request message.

[0071] During implementation, the instructing unit is further configured to instruct the online reasoning module of the base station centralized unit that meets the threshold requirement to adopt the model with the best performance among the models that meet the threshold requirement.

[0072] In implementation, the instructing unit is further configured to indicate one or a combination of the following instructions to the base station centralized unit that does not meet the threshold requirement:

[0073] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0074] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0075] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0076] In implementation, the receiving unit is further configured to receive the storage space allocation amount sent by the data storage module of the central unit of the base station;

[0077] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0078] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0079] A base station, comprising:

[0080] The processor reads the program from the memory and performs the following steps:

[0081] Send the model and the performance evaluation results corresponding to the model, where the model is a network performance AI training model executed on the centralized unit;

[0082] adjusting the storage space of the base station centralized unit according to the returned instruction;

[0083] A transceiver is used to receive and send data under the control of the processor.

[0084] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0085] During implementation, the storage space of the base station centralized unit is adjusted in a step-size manner according to the instruction.

[0086] In implementation, the instruction to adjust the storage space is received via a Storage Space Change Request message.

[0087] During implementation, it further includes:

[0088] The online inference module adopts the best performing model among the instructed models according to the instruction.

[0089] During implementation, one or a combination of the following instructions is further included:

[0090] The online reasoning module adopts the model with the best performance among the instructed models according to the instructions;

[0091] configuring the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning according to the instruction;

[0092] Migrate the model from the migration source as instructed.

[0093] During implementation, it further includes:

[0094] The data storage module sends the storage space allocation amount.

[0095] A base station, comprising:

[0096] A sending unit, configured to send the model and a performance evaluation result corresponding to the model, wherein the model is a network performance AI training model executed on the centralized unit;

[0097] The adjusting unit is used to adjust the storage space of the base station central unit according to the returned instruction.

[0098] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0099] In implementation, the adjusting unit is further configured to adjust the storage space of the base station centralized unit in a step-size manner according to the instruction.

[0100] In an implementation, the adjustment unit is further configured to receive an instruction to adjust the storage space via a Storage Space Change Request message.

[0101] During implementation, the adjustment unit is further configured to adopt the model with the best performance among the indicated models according to the indication.

[0102] In implementation, the adjustment unit is further configured to execute one of the following instructions or a combination thereof:

[0103] Adopting the best performing model among the indicated models according to the instructions;

[0104] The learning mode of the online reasoning module is configured as autonomous learning and transfer learning;

[0105] Migrate the model from the migration source as instructed.

[0106] During implementation, the sending unit is further configured to send the storage space allocation amount.

[0107] A computer-readable storage medium stores a computer program for executing the storage space adjustment method.

[0108] The beneficial effects of the present invention are as follows:

[0109] The current technology allocates storage space for different model training tasks in a uniform or fixed manner, resulting in abundant storage space for some tasks, while some data storage has little effect on reminding model performance. At the same time, some tasks have low model performance due to insufficient storage of samples, and the overall storage resource utilization efficiency is low. In the technical solution provided by the implementation of the present invention, since it is possible to determine whether the model meets the threshold requirements, that is, whether the model performance meets the standards, based on the model sent by the base station centralized unit and the performance evaluation results corresponding to the model, the storage space of the base station centralized unit is adjusted according to whether the model meets the threshold requirements, thereby overcoming the problem of low model performance due to storage space and improving the overall utilization efficiency of storage resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0111] Figure 1 Schematic diagram of a flow chart of a method for adjusting storage space on a node side indicating adjustment according to an embodiment of the present invention;

[0112] Figure 2 Schematic diagram of the implementation flow of the storage space adjustment method on the base station side in an embodiment of the present invention;

[0113] Figure 3 This is a schematic diagram of interactive messages between modules in an embodiment of the present invention;

[0114] Figure 4 This is a schematic diagram of the architecture 1 of the process interaction diagram of each module in an embodiment of the present invention;

[0115] Figure 5 This is a schematic diagram of the interaction diagram architecture 2 of the modules in the embodiment of the present invention;

[0116] Figure 6 A schematic diagram of a storage space adjustment implementation process according to an embodiment of the present invention;

[0117] Figure 7 This is an example of an implementation of a built-in training module and a storage module in an embodiment of the present invention;

[0118] Figure 8 This is an example of an implementation of an external training module and a storage module in an embodiment of the present invention;

[0119] Figure 9 This is a schematic structural diagram of a storage space adjustment device according to an embodiment of the present invention;

[0120] Figure 10 Schematic diagram of the base station structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0121] During the course of the invention, the inventors noticed that:

[0122] Current mobile communication networks contain a large number of similar network elements (NEs) that share functionality, load balancing, and strong correlations between network performance metrics. For example, NEs within the same pool in the core network share the same set of performance metrics, with even user numbers and load sharing, resulting in very similar performance measurements in most cases. In wireless networks, there are also clusters of base stations that exhibit similar wireless performance data due to similar deployment scenarios.

[0123] During daily network operations and maintenance, performance data is reported and aggregated to the network management platform's servers or stored locally within base stations. Currently, the server's storage capacity only supports storing up to one month's worth of data per network element. Based on a 15-minute data collection cycle, only 96*30=2880 sample points can be extracted for each network performance indicator at a time. This is very limited for AI model training, impacting model accuracy. For base stations, storage capacity is even smaller. Considering that each network element can simultaneously train multiple model tasks requiring different samples, the storage allocated for each task is even smaller.

[0124] Current technology allocates storage space for different model training tasks in a uniform or fixed manner, resulting in abundant storage space for some tasks, while some data storage has little effect on improving model performance. At the same time, some tasks have insufficient stored samples, resulting in low model performance and low overall storage resource utilization efficiency.

[0125] Current technology allocates storage space for different model training tasks within the same network element in a uniform or fixed manner, resulting in abundant storage space for some tasks, and some data storage has little effect on improving model performance. At the same time, some tasks have insufficient stored samples, resulting in low model performance and low overall storage resource utilization efficiency.

[0126] Based on this, in order to improve the utilization of storage resources while ensuring the performance of multiple models, an embodiment of the present invention will provide a dynamic adjustment scheme for data storage space based on transfer learning, which is used to reduce data storage space requirements while ensuring the model effect.

[0127] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0128] During the description, the implementation of the node indicating adjustment and the base station will be explained separately, and then an example of the coordinated implementation of the two will be given to better understand the implementation of the solution provided in the embodiment of the present invention. This description does not mean that the two must be implemented in coordination or separately. In fact, when the node indicating adjustment and the base station are implemented separately, they each solve their own problems. When the two are used in combination, better technical effects will be achieved.

[0129] Figure 1 A flowchart of a method for adjusting storage space on a node side indicating adjustment is implemented, as shown in the figure, which may include:

[0130] Step 101: Receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0131] Step 102: Determine whether the model meets the threshold requirement based on the performance evaluation result of the model and the model performance threshold set for the model training task;

[0132] Step 103: Adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement.

[0133] During implementation, the node that instructs adjustment may be a centralized node in the wireless network, a newly added functional node, or an existing southbound network management device.

[0134] Figure 2The flowchart of the storage space adjustment method on the base station side is shown in the figure, which may include:

[0135] Step 201: The base station centralized unit sends a model and a performance evaluation result corresponding to the model, wherein the model is a network performance AI training model executed on the centralized unit;

[0136] Step 202: The base station centralized unit adjusts the storage space of the base station centralized unit according to the returned instruction.

[0137] Specifically, the solution can be considered to consist of four parts: data storage, model training, online reasoning, and model evaluation. Data storage, model training, and online reasoning are located in a centralized unit at the base station and can be implemented using existing technologies.

[0138] For the sake of ease of description and understanding, the module that implements indication adjustment on the node where indication adjustment is implemented is called the wireless network model centralized evaluation module. During implementation, this module can be located in a centralized node in the wireless network, can be a newly added functional node, or can be placed in the current southbound network management equipment.

[0139] The data storage module allocates storage resources to store sample data for each model training task of a certain network element according to the preset initial default storage space allocation and the minimum storage space threshold, and the resource amount is not less than the minimum storage space threshold.

[0140] Due to differences in the amount of training data, the performance of models trained on different nodes also varies. For example, models trained on nodes with large amounts of training data can capture the monthly and annual cycle characteristics of indicators, large-scale trends, and the impact of holidays, all over a larger time span. Nodes with smaller amounts of training data tend to have poorer model performance.

[0141] The AI ​​training module, based on instructions from the central node, generates a set of usable models for each model training task at each node. These models can be autonomous learning models only, multiple transfer learning models, or a combination of autonomous and transfer learning models. The models are stored in chronological order. In other words, in practice, the models are autonomous learning models and / or transfer learning models.

[0142] The online inference module performs online inference on all selected models given under the current training task based on real-time samples, evaluates the model performance based on the inference results, and uploads the model and evaluation results to the model evaluation module.

[0143] The model evaluation module determines which nodes have qualified models available and which nodes do not, based on the model performance evaluation results and the model performance threshold set for each model training task. Based on the storage space allocation reported by the data storage module, the model evaluation module analyzes the relationship between storage allocation, model similarity, and model performance. It also establishes a relationship model between autonomous learning model performance and storage allocation; and it also establishes a relationship model between migration model performance, model similarity, and storage allocation.

[0144] For nodes with qualified models, the online reasoning module is instructed to use the best-performing qualified model. At the same time, the similarity of the models between nodes is calculated. If there is a similar model to the qualified model (the similarity exceeds the threshold), the migration model is added, deleted, and modified in the learning mode of this node with the goal of minimizing storage space. The minimum storage space required for the new learning model is calculated, and the data storage module is instructed to adjust the storage space.

[0145] For nodes without a model that meets the requirements, the online inference module is instructed to use the model with the best performance. At the same time, the learning mode of this node is configured as autonomous learning + transfer learning, in which the n best models that meet the performance requirements in the same model training task of other nodes are selected as migration sources (n is configurable and is a natural number). At the same time, the data storage module is instructed to increase the storage space according to a certain step size.

[0146] Figure 3 The following diagram shows the interaction between modules: Data storage, AI training, online inference, and model evaluation.

[0147] The model evaluation module receives the Model evaluation results update issued by the online reasoning module and indicates the Online model instruction.

[0148] The model training module receives Learn Model instruction, sends and receives Model update request / response to the online inference module, and interacts with the data storage module via Data request and Data updates.

[0149] The data storage module interacts with the model evaluation module via Storage Space Chang request / response.

[0150] Figure 4 The following is a schematic diagram of the module process interaction architecture 1. As shown in the figure, the model evaluation module is located outside the base station centralized unit and is an independent node. The data storage module, model training module, and online inference module are also located in the base station centralized unit.

[0151] Figure 5 This is a schematic diagram of the module process interaction diagram architecture 2. As shown in the figure, the model evaluation module and online inference module are located outside the base station centralized unit and are independent nodes. The data storage module and model training module are also located in the base station centralized unit.

[0152] The following describes the implementation of the centralized model evaluation module for wireless network models.

[0153] The model evaluation module evaluates all models and corresponding performance evaluation results reported by all base station centralized units, generates learning mode recommendations for the AI ​​training module, and generates storage space adjustment recommendations for the data storage module.

[0154] Based on the model performance evaluation results and the model performance threshold set for each model training task, determine whether the model for each task under each base station centralized unit meets the standards and is available. If the model performance evaluation result is better than the threshold, it is judged to meet the standards, otherwise it is not.

[0155] During implementation, the following may be further included:

[0156] receiving a storage space allocation amount sent by a data storage module of a central unit of a base station;

[0157] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0158] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0159] Correspondingly, on the base station side: the data storage module of the base station centralized unit sends the storage space allocation amount.

[0160] Specifically, based on the storage space allocation reported by the data storage module, the relationship between the storage allocation, the similarity of the migration model, and the model performance is analyzed. A mapping relationship between model performance, storage allocation, migration model similarity, and learning mode is generated as a "storage space evaluation model."

[0161] During implementation, the storage space of the base station centralized unit is adjusted based on whether the model meets the threshold requirements, including:

[0162] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0163] In a specific implementation, for a base station centralized unit that meets the threshold requirement, the following steps are further included:

[0164] The online reasoning module of the centralized unit of the base station is instructed to adopt the best performing model among the models that meet the threshold requirement.

[0165] In a specific implementation, for a base station centralized unit that does not meet the threshold requirement, one of the following indications or a combination thereof is further included:

[0166] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0167] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0168] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0169] For the base station side, it further includes:

[0170] The online reasoning module of the base station centralized unit adopts the model with the best performance among the indicated models according to the instruction.

[0171] In a specific implementation, one or a combination of the following instructions is further included:

[0172] The online reasoning module of the base station centralized unit adopts the model with the best performance among the indicated models according to the instruction;

[0173] configuring the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning according to the instruction;

[0174] Migrate the model from the migration source as instructed.

[0175] The following is an example to illustrate.

[0176] Figure 6 This is a schematic diagram of the storage space adjustment implementation process. In the figure, m is the number of qualified models, and n is the number of migration source models. For each task in each base station centralized unit, the following process can be used:

[0177] Get all the model performances for this task;

[0178] Instructs to use the best performing model;

[0179] Determine whether there is a model that meets the standards. If so, proceed to the first part; if not, proceed to the second part.

[0180] Part I: Implementation of the achievement model.

[0181] Obtain all models of this task from other nodes and calculate model similarity;

[0182] Obtain similarity thresholds and model performance thresholds to match the migration source model;

[0183] Evaluate the amount of storage space, interactive storage space evaluation model. The array shown in the figure is {model ID, learning mode, model similarity, target performance};

[0184] Determine the learning model with the smallest storage space {model ID, learning model, migration source model} and the minimum storage requirement S min ;

[0185] Determine whether the current learning mode contains the above modes. If so, the AI ​​training module generates a new model set. Otherwise, configure the learning mode to increase the minimum storage space.

[0186] Determine the existing storage space S current Is it greater than S min If so, use the step size S min Reduce storage space allocation for the endpoint, otherwise the AI ​​training module generates a new model set.

[0187] Part II: Implementation of the non-compliant model.

[0188] Rank the model performance of other nodes under this task;

[0189] Determine the n best performing models as migration sources, where the value of n is configurable;

[0190] Configure the learning mode to autonomous learning + transfer learning;

[0191] Estimate the maximum amount of storage space S required max ;

[0192] Determine the existing storage space S current Is it less than S? max If so, use the step size S max Increase the storage space allocated for the endpoint, otherwise the AI ​​training module generates a new model set.

[0193] During implementation, the storage space of the base station centralized unit is adjusted based on whether the model meets the threshold requirements, including:

[0194] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0195] In implementation, for the node instructing adjustment, the base station centralized unit is instructed to adjust the storage space via a Storage Space Change Request message.

[0196] For the base station side, it receives an instruction to adjust the storage space through a Storage Space Change Request message.

[0197] In implementation, for the node indicating adjustment, the storage space is reduced in a step-by-step manner; and / or,

[0198] Increase the storage space in steps.

[0199] For the base station side, the storage space of the base station centralized unit is adjusted in a step-by-step manner according to the instruction.

[0200] Specifically, the base station centralized unit may be instructed to reduce or increase data storage space via a Storage Space Change Request message: the storage space indication may be set as a suggested storage space reduction ratio or as an absolute value after the storage space is reduced;

[0201] Storage Space Change Request: The message can be sent by the model evaluation module to the data storage module. The content includes the base station centralized unit name or ID, model task ID, adjustment direction (including increase or decrease), step size (can be a ratio value or an absolute value), adjustment end value ( Figure 6 Medium S min or S max ), if the adjustment direction is to reduce, the step size should be less than or equal to S concentrated unit rrent-S min If the adjustment direction is increasing, the step size should be less than or equal to S max -S centralized unit current.

[0202] After receiving the message, the data storage module increases or decreases the storage space allocated to the model task by an amount equal to the step size according to the adjustment direction and step size.

[0203] The implementation of the data storage module is described below.

[0204] The data storage module can refer to the implementation of existing data storage modules to store data samples of the nodes.

[0205] The data storage module parses and processes the Storage Space Change Request message and adjusts the storage space of the base station centralized unit according to the instructions.

[0206] After adjusting the storage space, send a Storage Space Change Response message:

[0207] Storage Space Change Response: The message can be sent by the data storage module to the model evaluation module. The content includes the base station centralized unit name or ID, model task ID, the absolute value of the storage space after adjustment, timestamp, etc.

[0208] After receiving the message, the model evaluation module records the absolute value of the storage space and the timestamp under the corresponding base station centralized unit name or ID and model task ID, which is used to analyze the mapping relationship between model performance and storage allocation, migration model similarity, and learning mode.

[0209] The following is an example to illustrate.

[0210] Figure 7 This is an example of the implementation of the built-in training module and storage module. Figure 8 The figure is an example of the implementation of the external training module and storage module. It shows a feasible implementation method in a wireless network. As shown in the figure, it is assumed that there are two base station centralized units A and B. A centralized unit wireless network model centralized evaluation node contains a "model evaluation" entity. A and B both send their own model evaluation results to the model evaluation entity and obtain online model instructions and storage space adjustment instructions. Figure 7 In the ,data storage and AI training modules are distributed in each centralized unit. Figure 8 In the AI ​​​​processor, data storage and AI training modules are in centralized servers.

[0211] The following combination Figure 7 、 Figure 8 A feasible and typical application example is described as follows:

[0212] The data storage module stores data samples of network performance indicators (KPIs) (used to assess handover success rates). The model training task is to generate a prediction model for handover success rates. The AI ​​training module in each centralized unit trains multiple prediction models for handover success rates based on these samples.

[0213] The model-related data stored in the centralized evaluation node of the centralized unit wireless network model can be in the following forms:

[0214]

[0215] If the performance threshold of the model evaluation module in the centralized unit of the wireless network model centralized evaluation node is set to 90%, then the centralized unit_A and the centralized unit_B have qualified models.

[0216] For centralized unit_B, the model similarity between m_B_s_1 and m_A_s_1 is calculated to be 0.95, which is higher than the threshold of 0.8, and m_A_s_1 is also a qualified model. Therefore, m_A_s_1 is matched as the migration source model of m_B_s_1, and the migration model number is m_B_A_s_1.

[0217] The storage space required to evaluate {m_B_s_1, autonomous learning, 90%} is 1.5 GB, and the storage space required for {m_B_A_s_1, transfer learning, 0.95, 90%} is 1 GB. Therefore, the model with the smallest storage space is {m_B_A_s_1, transfer learning}.

[0218] Since the current storage space allocation of centralized unit_B is 2GB > 1GB, the centralized unit model evaluation node sends a storage space adjustment request message to centralized unit_B:

[0219] For example, the message format is: {Node: Centralized Unit_B, Task: Predict Handoff Success Rate, Direction: Decrease, Step Size: 250MB, Target: 1GB}. After receiving this message, Centralized Unit_B adjusts the storage space to 1.75GB and informs the Centralized Unit Model Evaluation Node through a response message. After receiving this message, the Centralized Unit Model Evaluation Node supplements the new model information generated by Centralized Unit_B and selects m_B_A_s_1 as the online inference model (indicated by the bold part in the table):

[0220]

[0221]

[0222] If the model performance of centralized unit_A is 85%, which is not up to standard, as shown in the following table (indicated by bold):

[0223]

[0224] For centralized unit_A, select centralized unit_B's model m_B_s_1 as the migration source, and assign the migration model number m_A_B_s_1. Configure centralized unit_A's learning mode to autonomous + transfer learning through a learning mode indication message.

[0225] The storage space required to evaluate {m_A_s_1, autonomous learning, 90%} is 1.5GB, and the storage space required for {m_A_B_s_1, transfer learning, 0.6, 90%} is 1GB, then S max =1.5GB;

[0226] Since the current storage space allocated to centralized unit_A is 1GB < 1.5GB, the centralized unit model evaluation node sends a storage space adjustment request message to centralized unit_A:

[0227] For example, the message format is: {Node: Centralized Unit_A, Task: Predict Handoff Success Rate, Direction: Increase, Step Size: 250MB, Target: 1.5GB}. After receiving this message, Centralized Unit_A adjusts the storage space to 1.25GB and notifies the Centralized Unit Model Evaluation Node via a response message. After receiving this message, the Centralized Unit Model Evaluation Node supplements the new model information generated by Centralized Unit_A and selects m_A_B_s_1 as the online inference model at this time (indicated by the bold part in the table):

[0228]

[0229] Based on the same inventive concept, an embodiment of the present invention also provides a storage space adjustment device, a base station and a computer-readable storage medium. Since the principles of solving problems by these devices are similar to those of the storage space adjustment method, the implementation of these devices can refer to the implementation of the method, and the repeated parts will not be repeated.

[0230] When implementing the technical solution provided by the embodiment of the present invention, it can be implemented as follows.

[0231] Figure 9 This is a schematic diagram of the storage space adjustment device structure. As shown in the figure, the device includes:

[0232] The processor 900 is configured to read the program in the memory 920 and execute the following process:

[0233] Receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0234] Determine whether the model meets the threshold requirements based on the model performance evaluation results and the model performance threshold set for the model training task;

[0235] Adjusting the storage space of the base station centralized unit based on whether the model meets the threshold requirement;

[0236] The transceiver 910 is configured to receive and send data under the control of the processor 900 .

[0237] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0238] During implementation, the storage space of the base station centralized unit is adjusted based on whether the model meets the threshold requirements, including:

[0239] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0240] In implementation, the storage space is reduced in a step-by-step manner; and / or,

[0241] Increase the storage space in steps.

[0242] In implementation, the base station centralized unit is instructed to adjust the storage space via a Storage Space Change Request message.

[0243] During implementation, for a base station centralized unit that meets the threshold requirement, the method further includes:

[0244] The online reasoning module of the centralized unit of the base station is instructed to adopt the best performing model among the models that meet the threshold requirement.

[0245] In implementation, for a base station centralized unit that does not meet the threshold requirement, one of the following indications or a combination thereof is further included:

[0246] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0247] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0248] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0249] During implementation, it further includes:

[0250] receiving a storage space allocation amount sent by a data storage module of a central unit of a base station;

[0251] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0252] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0253] Among them, Figure 9In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 900 and memory represented by memory 920. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. The transceiver 910 may be a plurality of components, i.e., a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 900 is responsible for managing the bus architecture and general processing, and the memory 920 may store data used by the processor 900 when performing operations.

[0254] An embodiment of the present invention further provides a storage space adjustment device, comprising:

[0255] A receiving unit, configured to receive a model and a performance evaluation result corresponding to the model sent by a base station centralized unit, wherein the model is a network performance AI training model executed on the centralized unit;

[0256] A determination unit, configured to determine whether the model meets the threshold requirement based on the performance evaluation result of the model and the model performance threshold set for the model training task;

[0257] The indicating unit is used to adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement.

[0258] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0259] In implementation, the instructing unit is further configured to adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement, including:

[0260] The base station centralized units that have reached the threshold requirement are instructed to reduce storage space, and the base station centralized units that have not reached the threshold requirement are instructed to increase storage space.

[0261] In implementation, the instructing unit is further configured to instruct to reduce the storage space in a step-by-step manner; and / or to increase the storage space in a step-by-step manner.

[0262] In implementation, the instructing unit is further configured to instruct the base station central unit to adjust the storage space via a Storage Space Change Request message.

[0263] During implementation, the instructing unit is further configured to instruct the online reasoning module of the base station centralized unit that meets the threshold requirement to adopt the model with the best performance among the models that meet the threshold requirement.

[0264] In implementation, the instructing unit is further configured to indicate one or a combination of the following instructions to the base station centralized unit that does not meet the threshold requirement:

[0265] Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance;

[0266] Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning;

[0267] The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station that does not meet the threshold requirement, where n is a natural number.

[0268] In implementation, the receiving unit is further configured to receive the storage space allocation amount sent by the data storage module of the central unit of the base station;

[0269] Analyze the relationship between storage space allocation, model similarity, and model performance;

[0270] Establish a relationship model between the performance of the autonomous learning model used to establish the storage space evaluation model and the storage space allocation amount, and / or establish a relationship model between the performance of the migration model used to establish the storage space evaluation model and the model similarity and storage space allocation amount.

[0271] For the convenience of description, the various parts of the above-mentioned device are divided into various modules or units according to their functions and described separately. Of course, when implementing the present invention, the functions of each module or unit can be realized in the same or multiple software or hardware.

[0272] Figure 10 The following is a schematic diagram of the base station structure. As shown in the figure, the base station includes:

[0273] The processor 1000 is configured to read the program in the memory 1020 and execute the following process:

[0274] Send the model and the performance evaluation results corresponding to the model, where the model is a network performance AI training model executed on the centralized unit;

[0275] adjusting the storage space of the base station centralized unit according to the returned instruction;

[0276] The transceiver 1010 is configured to receive and send data under the control of the processor 1000 .

[0277] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0278] During implementation, the storage space of the base station centralized unit is adjusted in a step-size manner according to the instruction.

[0279] In implementation, the instruction to adjust the storage space is received via a Storage Space Change Request message.

[0280] During implementation, it further includes:

[0281] The online inference module adopts the best performing model among the instructed models according to the instruction.

[0282] During implementation, one or a combination of the following instructions is further included:

[0283] The online reasoning module adopts the model with the best performance among the instructed models according to the instructions;

[0284] configuring the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning according to the instruction;

[0285] Migrate the model from the migration source as instructed.

[0286] During implementation, it further includes:

[0287] The data storage module sends the storage space allocation amount.

[0288] Among them, Figure 10 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 1000 and memory represented by memory 1020. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. The transceiver 1010 may be a plurality of components, i.e., including a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 1000 is responsible for managing the bus architecture and general processing, and the memory 1020 may store data used by the processor 1000 when performing operations.

[0289] An embodiment of the present invention further provides a base station, including:

[0290] A sending unit, configured to send the model and a performance evaluation result corresponding to the model, wherein the model is a network performance AI training model executed on the centralized unit;

[0291] The adjusting unit is used to adjust the storage space of the base station central unit according to the returned instruction.

[0292] In implementation, the model is an autonomous learning model and / or a transfer learning model.

[0293] In implementation, the adjusting unit is further configured to adjust the storage space of the base station centralized unit in a step-size manner according to the instruction.

[0294] In an implementation, the adjustment unit is further configured to receive an instruction to adjust the storage space via a Storage Space Change Request message.

[0295] During implementation, the adjustment unit is further configured to adopt the model with the best performance among the indicated models according to the indication.

[0296] In implementation, the adjustment unit is further configured to execute one of the following instructions or a combination thereof:

[0297] Adopting the best performing model among the indicated models according to the instructions;

[0298] The learning mode of the online reasoning module is configured as autonomous learning and transfer learning;

[0299] Migrate the model from the migration source as instructed.

[0300] During implementation, the sending unit is further configured to send the storage space allocation amount.

[0301] For the convenience of description, the various parts of the above-mentioned device are divided into various modules or units according to their functions and described separately. Of course, when implementing the present invention, the functions of each module or unit can be realized in the same or multiple software or hardware.

[0302] An embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program for executing the above storage space adjustment method.

[0303] For specific implementation, please refer to the implementation of the above storage space adjustment method.

[0304] In summary, this solution can reduce data storage space requirements while ensuring the model effect.

[0305] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0306] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0307] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0308] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0309] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A storage space adjustment method, characterized in that: include: Receive a model and a performance evaluation result corresponding to the model sent by the base station centralized unit, wherein the model is a network performance AI training model executed on the base station centralized unit; Determine whether the model meets the threshold requirements based on the model performance evaluation results and the model performance threshold set for the model training task; Adjusting the storage space of the base station centralized unit based on whether the model meets the threshold requirement; The adjusting the storage space of the base station centralized unit according to whether the model reaches the threshold requirement includes: Instructing the base station centralized unit whose model meets the threshold requirement to reduce the storage space, and the base station centralized unit whose model does not meet the threshold requirement to increase the storage space; For a base station centralized unit whose model does not meet the threshold requirement, one or a combination of the following indications is further included: Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance; Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning; The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station whose model does not meet the threshold requirement, where n is a natural number.

2. The method according to claim 1, wherein The model is an autonomous learning model and / or a transfer learning model.

3. The method according to claim 1, wherein Reduce the storage space in steps; or, Increase the storage space in steps.

4. The method according to claim 1, wherein The base station central unit is instructed to adjust the storage space through the storage space adjustment request Storage SpaceChange Request message.

5. The method according to claim 1, wherein For a base station centralized unit whose model meets the threshold requirement, the method further includes: The online reasoning module of the centralized unit of the base station is instructed to adopt the best performing model among the models that meet the threshold requirement.

6. The method according to claim 2, wherein Further including: receiving a storage space allocation amount sent by a data storage module of a central unit of a base station; Analyze the relationship between storage space allocation, model similarity, and model performance; Establish a relationship model between the performance of the autonomous learning model and the amount of storage space allocated, and / or establish a relationship model between the performance of the migration model and the model similarity and the amount of storage space allocated.

7. A storage space adjustment device, characterized in that: include: The processor reads the program from the memory and performs the following steps: Receive a model and a performance evaluation result corresponding to the model sent by the base station centralized unit, wherein the model is a network performance AI training model executed on the base station centralized unit; Determine whether the model meets the threshold requirements based on the model performance evaluation results and the model performance threshold set for the model training task; Adjusting the storage space of the base station centralized unit based on whether the model meets the threshold requirement; a transceiver for receiving and sending data under the control of the processor; The adjusting the storage space of the base station centralized unit according to whether the model reaches the threshold requirement includes: Instructing the base station centralized unit whose model meets the threshold requirement to reduce the storage space, and the base station centralized unit whose model does not meet the threshold requirement to increase the storage space; For a base station centralized unit whose model does not meet the threshold requirement, one or a combination of the following indications is further included: Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance; Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning; The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station whose model does not meet the threshold requirement, where n is a natural number.

8. A storage space adjustment device, characterized in that: include: A receiving unit, configured to receive a model and a performance evaluation result corresponding to the model sent by the base station centralized unit, wherein the model is a network performance AI training model executed on the base station centralized unit; A determination unit, configured to determine whether the model meets the threshold requirement based on the performance evaluation result of the model and the model performance threshold set for the model training task; an indicating unit, configured to adjust the storage space of the base station centralized unit according to whether the model meets the threshold requirement; The indicating unit is further configured to adjust the storage space of the base station centralized unit according to whether the model reaches a threshold, including: Instructing the base station centralized unit whose model meets the threshold requirement to reduce the storage space, and the base station centralized unit whose model does not meet the threshold requirement to increase the storage space; The indicating unit is further configured to indicate one or a combination of the following indications to the base station centralized unit whose model does not meet the threshold requirement: Instructing the online reasoning module of the centralized unit of the base station to adopt the model with the best performance; Configure the learning mode of the online reasoning module of the base station centralized unit to autonomous learning and transfer learning; The best n models whose performance meets the threshold requirement in the same model training task of other base stations are selected as the migration source of the online inference module of the centralized unit of the base station whose model does not meet the threshold requirement, where n is a natural number.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program for executing the method according to any one of claims 1 to 6.

Images