Method for adjusting cell load and related device

Abstract

The application provides a cell load adjustment method and a related device applied to the field of artificial intelligence, which can accurately adjust the load degree of a target cell to keep various performance indexes of the target cell within a proper range and provide better network service for users. The method of the application comprises the following steps: obtaining characteristic information of a target cell and a load index of the target cell, wherein the load index is used to indicate the load degree of the target cell; processing the characteristic information through a first target model to obtain model parameters of a second target model; obtaining the second target model based on the model parameters; and processing the load index through the second target model to obtain an operation instruction, wherein the operation instruction is used to adjust the load index.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to a method for adjusting cell load and related equipment. Background Technology

[0002] In wireless cellular networks, uneven user distribution leads to uneven load distribution among cells (i.e., areas covered by the wireless network). High-load cells, with a large number of users and high service demand, are prone to insufficient network resources, making it difficult to guarantee the quality of service for users. Conversely, low-load cells, with fewer users and lower service demand, do not fully utilize their network resources.

[0003] Currently, the main method for adjusting the load level of a cell is manual modification. However, manual modification relies on the experience of experts, and the factors considered are often too simplistic and not comprehensive enough. This makes it impossible to accurately adjust the cell's load level and thus maintain the cell's various performance indicators within an appropriate range to provide users with sufficiently high-quality network services. Summary of the Invention

[0004] This application provides a method for adjusting cell load and related equipment, which can accurately adjust the load level of a target cell to keep the various performance indicators of the target cell within a suitable range, thereby providing users with better network services.

[0005] A first aspect of this application provides a method for adjusting cell load, the method comprising:

[0006] When it is necessary to adjust the load level of a target cell, the characteristic information and load index of the target cell can be collected. The characteristic information of the target cell can be used to indicate the current scenario of the target cell, and the load index of the target cell can be used to indicate the load level of the target cell (also known as the load balancing level).

[0007] After obtaining the feature information and load index of the target cell, a first target model can be obtained, which is a trained neural network. Then, the feature information of the target cell is input into the first target model to perform a series of processing on the feature information of the target cell (e.g., feature extraction, etc.) to obtain the model parameters of the second target model.

[0008] A second target model can be constructed based on the model parameters of the second target model; for example, the second target model can be a multilayer perceptron. Thus, the whole consisting of the first target model and the second target model can be used to indicate the functional relationship between the load index of the target cell and the operational instructions for the target cell. Generally, the operational instructions for the target cell are a monotonically increasing function of the load index of the target cell, and the rate of increase is determined by the characteristic information of the target cell.

[0009] After constructing the second target model, the load index of the target cell can be input into the second target model to process the load index of the target cell and obtain the operation instructions for the target cell. The operation instructions for the target cell can be used to adjust the load index of the target cell.

[0010] As can be seen from the above method, after obtaining the feature information and load indicators of the target cell, the feature information can be processed using the first target model to obtain the model parameters of the second target model. Then, based on the model parameters of the second target model, a second target model is constructed, and the load indicators are processed using the second target model to obtain operation instructions for the target cell. These operation instructions can be used to adjust the load indicators of the target cell. In the aforementioned process, the model parameters of the second target model are obtained based on the feature information of the target cell. Since the feature information of the target cell can be used to characterize various features of the target cell (i.e., the scenario in which the target cell is located), when processing the load indicators of the target cell using the second target model, various factors such as the various features of the target cell are considered. Therefore, the operation instructions for the target cell output by the second target model can be used to accurately adjust the load level of the target cell to keep the various performance indicators of the target cell within a suitable range, thereby providing users with better network services.

[0011] Furthermore, the whole consisting of the first target model and the second target model can be used to indicate the functional relationship between the load index of the target cell and the operation instructions for the target cell. This functional relationship is usually monotonically increasing or monotonically decreasing, which conforms to the constraints of expert experience (equivalent to integrating expert experience into the model). The control strategy learned by the model is more in line with the business logic.

[0012] In one possible implementation, the second target model includes a first sub-model and a second sub-model. Processing the load index using the second target model to obtain the operation instruction includes: processing the load index using the first sub-model to obtain a first operation; if the load index equals a preset index threshold, the first operation is used to not adjust the load index; processing the load index using the second sub-model to obtain a second operation; and performing a weighted sum of the first and second operations to obtain the operation instruction. In the aforementioned implementation, after obtaining the feature information and load index of the target cell, the first target model can be obtained. Then, the feature information of the target cell is input into the first target model to perform a series of processing steps on the feature information of the target cell, obtaining the model parameters of the first sub-model and the model parameters of the second sub-model. Next, the first sub-model can be constructed based on the model parameters of the first sub-model, and the second sub-model can be constructed based on the model parameters of the second sub-model. Subsequently, the load indicators of the target cell can be input into the first sub-model and the second sub-model respectively. The first sub-model processes the load indicators of the target cell to obtain the first operation for the target cell, and the second sub-model processes the load indicators of the target cell to obtain the second operation for the target cell. Finally, the first operation and the second operation for the target cell are weighted and summed to obtain the operation instruction for the target cell. In the above process, the first and second sub-models are constructed through a teacher-student network architecture. The first sub-model adds expert constraints (i.e., determining whether the target cell absorbs or releases users based on preset indicator thresholds) to ensure the security of the control strategy output by the model. The second sub-model relaxes the expert constraints, learns from the data, and outputs the corresponding control strategy. The weighted sum of the outputs of the two models is used as the final strategy, thereby balancing the security and efficiency of the control strategy.

[0013] In one possible implementation, a first target model and a first sub-model are used to indicate a first functional relationship between load indicators and operation instructions. The first target model and the second sub-model are used to indicate a second functional relationship between load indicators and operation instructions. The second functional relationship is obtained by shifting the first functional relationship. Both the first and second functional relationships are monotonically increasing or monotonically decreasing. In the aforementioned implementation, since both the first and second functional relationships are monotonically increasing or monotonically decreasing, the operation instructions for the target cell obtained by weighted summation of the first and second operations also have a monotonically increasing or monotonically decreasing functional relationship with the load indicators of the target cell, which is consistent with expert experience strategies.

[0014] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0015] In one possible implementation, the operation instructions for the target cell are used to modify the target cell's configuration parameters to adjust load metrics. The target cell's characteristic information includes at least one of the following: the target cell's configuration parameters; the target cell's call statistics data; the configuration parameters of the target cell's neighboring cells; and the neighboring cells' call statistics data. In this way, the target cell's characteristic information encompasses multi-dimensional features, comprehensively characterizing the scenario in which the target cell exists.

[0016] In one possible implementation, the configuration parameters of the target cell may include at least one of the following: the antenna transmit power of the target cell, the antenna downtilt angle of the target cell, the antenna horizontal azimuth angle of the target cell, the reference signal receiving power (RSRP) threshold for user initiation of inter-frequency handover measurement in the target cell, the RSRP threshold for user cessation of inter-frequency handover measurement in the target cell, the specific frequency RSRP offset for user triggering inter-frequency handover procedure in the target cell, the specific neighboring cell RSRP offset for user triggering inter-frequency handover procedure in the target cell, the RSRP threshold for user initiation of intra-frequency handover measurement in the target cell, the RSRP threshold for user cessation of intra-frequency handover measurement in the target cell, the specific frequency RSRP offset for user triggering intra-frequency handover procedure in the target cell, and the specific neighboring cell RSRP offset for user triggering intra-frequency handover procedure in the target cell, etc. The configuration parameters of a neighboring cell may include at least one of the following: the antenna transmit power of the neighboring cell, the antenna downtilt angle of the neighboring cell, the antenna horizontal azimuth angle of the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating inter-frequency handover procedures in the neighboring cell, the specific neighboring cell RSRP offset for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, and the specific neighboring cell RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, etc.

[0017] In one possible implementation, the call statistics data of the target cell may include at least one of the following: the average number of users in the target cell per unit time period, the average number of active users in the target cell per unit time period, the uplink traffic in the target cell per unit time period, the proportion of low channel quality indicator (CQI) reports in the target cell per unit time period, the proportion of data packets with a length less than a preset length threshold in the target cell per unit time period (also referred to as the proportion of small packets in the target cell per unit time period), and the average length of data packets in the target cell per unit time period, etc. The call statistics data of neighboring cells may include at least one of the following: the average number of users in the neighboring cells per unit time period, the average number of active users in the neighboring cells per unit time period, the uplink traffic in the neighboring cells per unit time period, the proportion of CQI reports in the neighboring cells per unit time period, the proportion of data packets with a length less than a preset length threshold in the neighboring cells per unit time period (also referred to as the proportion of small packets in the neighboring cells per unit time period), and the average length of data packets in the neighboring cells per unit time period, etc.

[0018] A second aspect of this application provides a model training method, comprising: acquiring feature information and load indicators of a first cell, wherein the load indicators of the first cell are used to indicate the load level of the first cell; processing the feature information using a first model to be trained to obtain model parameters of a second model to be trained; obtaining the second model to be trained based on the model parameters; processing the load indicators of the first cell using the second model to be trained to obtain operation instructions for the first cell, wherein the operation instructions for the first cell are used to adjust the load indicators of the first cell; processing the operation instructions for the first cell and the operation instructions for the second cell using a third target model to obtain a first score and a second score, wherein the first score is used to evaluate the impact of the operation instructions for the first cell on the load indicators of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance indicators of the entire network, and the operation instructions for the second cell are used to adjust the load indicators of the second cell, wherein the first cell and the second cell are different cells; and updating the model parameters of the first model to be trained based on the first score and the second score until the model training conditions are met to obtain a first target model.

[0019] As can be seen from the above method, when using the critic model (i.e., the third objective model) to train the actor model (i.e., the whole composed of the first objective model and the second objective model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the ability to balance local and global performance, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the whole network, and realize the coordination between multiple cells.

[0020] In one possible implementation, the second model to be trained includes a first sub-model and a second sub-model. The second target model processes the load index of the first cell to obtain the operation instruction for the first cell, which includes: processing the load index of the first cell through the first sub-model to obtain a first operation; if the load index of the first cell is equal to a preset index threshold, the first operation is used to not adjust the load index of the first cell; processing the load index of the first cell through the second sub-model to obtain a second operation; and performing a weighted summation of the first operation and the second operation to obtain the operation instruction for the first cell.

[0021] In one possible implementation, a first training model and a first sub-model are used to indicate a first functional relationship between the load index of the first cell and the operation instruction for the first cell. The first training model and the second sub-model are used to indicate a second functional relationship between the load index of the first cell and the operation instruction for the first cell. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0022] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0023] In one possible implementation, the operation instruction for the first cell is used to modify the configuration parameters of the first cell to adjust the load index of the first cell; the characteristic information of the first cell includes at least one of the following: the configuration parameters of the first cell; the call statistics data of the first cell; the configuration parameters of the neighboring cells of the first cell; and the call statistics data of the neighboring cells.

[0024] In one possible implementation, the configuration parameters of the first cell may include at least one of the following: the antenna transmit power of the first cell, the antenna downtilt angle of the first cell, the antenna horizontal azimuth angle of the first cell, the RSRP threshold for users initiating inter-frequency handover measurements in the first cell, the RSRP threshold for users initiating inter-frequency handover measurements in the first cell, the specific frequency RSRP offset for users initiating inter-frequency handover procedures in the first cell, the specific neighboring cell RSRP offset for users initiating inter-frequency handover procedures in the first cell, the RSRP threshold for users initiating intra-frequency handover measurements in the first cell, the RSRP threshold for users initiating intra-frequency handover measurements in the first cell, the specific frequency RSRP offset for users initiating intra-frequency handover procedures in the first cell, and the specific neighboring cell RSRP offset for users initiating intra-frequency handover procedures in the first cell, etc. The configuration parameters of the neighboring cell may include at least one of the following: the antenna transmit power of the neighboring cell, the antenna downtilt angle of the first cell, the antenna horizontal azimuth angle of the first cell, the RSRP threshold for users in the neighboring cell to initiate inter-frequency handover measurement, the RSRP threshold for users in the neighboring cell to stop inter-frequency handover measurement, the specific frequency RSRP offset for users in the neighboring cell to trigger inter-frequency handover procedures, the specific neighboring cell RSRP offset for users in the neighboring cell to trigger inter-frequency handover procedures, the RSRP threshold for users in the neighboring cell to initiate intra-frequency handover measurement, the RSRP threshold for users in the neighboring cell to stop intra-frequency handover measurement, the specific frequency RSRP offset for users in the neighboring cell to trigger intra-frequency handover procedures, and the specific neighboring cell RSRP offset for users in the neighboring cell to trigger intra-frequency handover procedures, etc.

[0025] In one possible implementation, the voice statistics data of the first cell may include at least one of the following: the average number of users in the first cell per unit time period, the average number of active users in the first cell per unit time period, the uplink traffic in the first cell per unit time period, the proportion of low channel quality indication (CQI) reports in the first cell per unit time period, the proportion of data packets with a length less than a preset length threshold in the first cell per unit time period (also referred to as the proportion of small packets in the first cell per unit time period), and the average length of data packets in the first cell per unit time period, etc. The voice statistics data of the neighboring cells of the first cell may include at least one of the following: the average number of users in the neighboring cells per unit time period, the average number of active users in the neighboring cells per unit time period, the uplink traffic in the neighboring cells per unit time period, the proportion of CQI reports in the neighboring cells per unit time period, the proportion of data packets with a length less than a preset length threshold in the neighboring cells per unit time period (also referred to as the proportion of small packets in the neighboring cells per unit time period), and the average length of data packets in the neighboring cells per unit time period, etc.

[0026] A third aspect of this application provides a cell load adjustment device, comprising: a first acquisition module for acquiring feature information of a target cell and a load index of the target cell, the load index indicating the load level of the target cell; a first processing module for processing the feature information through a first target model to obtain model parameters of a second target model; a second acquisition module for acquiring a second target model based on the model parameters; and a second processing module for processing the load index through the second target model to obtain an operation instruction for adjusting the load index.

[0027] As can be seen from the above apparatus, after acquiring the feature information and load indicators of the target cell, the feature information can be processed using the first target model to obtain the model parameters of the second target model. Then, based on the model parameters of the second target model, a second target model is constructed, and the load indicators are processed using the second target model to obtain operation instructions for the target cell. These operation instructions can be used to adjust the load indicators of the target cell. In the aforementioned process, the model parameters of the second target model are obtained based on the feature information of the target cell. Since the feature information of the target cell can be used to characterize various features of the target cell (i.e., the scenario in which the target cell is located), when processing the load indicators of the target cell using the second target model, various factors such as the various features of the target cell are considered. Therefore, the operation instructions output by the second target model for the target cell can be used to accurately adjust the load level of the target cell, so as to keep the various performance indicators of the target cell within a suitable range and provide users with better network services.

[0028] In one possible implementation, the second target model includes a first sub-model and a second sub-model, and a second processing module, which is used to: process the load index through the first sub-model to obtain a first operation; if the load index is equal to a preset index threshold, the first operation is used to not adjust the load index; process the load index through the second sub-model to obtain a second operation; and perform a weighted summation of the first operation and the second operation to obtain an operation instruction.

[0029] In one possible implementation, the first target model and the first sub-model are used to indicate the first functional relationship between the load index and the first operation, and the first target model and the second sub-model are used to indicate the second functional relationship between the load index and the second operation. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0030] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0031] In one possible implementation, the operation instruction is used to modify the configuration parameters of the target cell to adjust the load index; the characteristic information of the target cell includes at least one of the following: the configuration parameters of the target cell; the call statistics data of the target cell; the configuration parameters of the target cell's neighboring cells; and the call statistics data of the neighboring cells.

[0032] In one possible implementation, the configuration parameters of the target cell may include at least one of the following: the antenna transmit power of the target cell, the antenna downtilt angle of the target cell, the antenna horizontal azimuth angle of the target cell, the RSRP threshold for user initiation of inter-frequency handover measurement in the target cell, the RSRP threshold for user initiation of inter-frequency handover measurement in the target cell, the specific frequency RSRP offset for user triggering inter-frequency handover procedure in the target cell, the specific neighboring cell RSRP offset for user triggering inter-frequency handover procedure in the target cell, the RSRP threshold for user initiation of intra-frequency handover measurement in the target cell, the RSRP threshold for user initiation of intra-frequency handover measurement in the target cell, the specific frequency RSRP offset for user triggering intra-frequency handover procedure in the target cell, and the specific neighboring cell RSRP offset for user triggering intra-frequency handover procedure in the target cell, etc. The configuration parameters of a neighboring cell may include at least one of the following: the antenna transmit power of the neighboring cell, the antenna downtilt angle of the neighboring cell, the antenna horizontal azimuth angle of the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating inter-frequency handover procedures in the neighboring cell, the specific neighboring cell RSRP offset for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, and the specific neighboring cell RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, etc.

[0033] In one possible implementation, the call statistics data of the target cell may include at least one of the following: the average number of users in the target cell per unit time period, the average number of active users in the target cell per unit time period, the uplink traffic in the target cell per unit time period, the proportion of low channel quality indication (CQI) reports in the target cell per unit time period, the proportion of data packets with a length less than a preset length threshold in the target cell per unit time period (also referred to as the proportion of small packets in the target cell per unit time period), and the average length of data packets in the target cell per unit time period, etc. The call statistics data of the target cell's neighboring cells may include at least one of the following: the average number of users in the neighboring cells per unit time period, the average number of active users in the neighboring cells per unit time period, the uplink traffic in the neighboring cells per unit time period, the proportion of CQI reports in the neighboring cells per unit time period, the proportion of data packets with a length less than a preset length threshold in the neighboring cells per unit time period (also referred to as the proportion of small packets in the neighboring cells per unit time period), and the average length of data packets in the neighboring cells per unit time period, etc.

[0034] A fourth aspect of this application provides a model training apparatus, comprising: a first acquisition module for acquiring feature information and load index of a first cell, wherein the load index of the first cell indicates the load level of the first cell; a first processing module for processing the feature information of the first cell using a first model to be trained to obtain model parameters of a second model to be trained; a second acquisition module for acquiring the second model to be trained based on the model parameters; a second processing module for processing the load index of the first cell using the second model to be trained to obtain an operation instruction for the first cell, wherein the operation instruction for the first cell is used to adjust the load index of the first cell; a third processing module for processing the operation instruction for the first cell and the operation instruction for the second cell using a third target model to obtain a first score and a second score, wherein the first score is used to evaluate the impact of the operation instruction for the first cell on the load index of the first cell, the second score is used to evaluate the impact of the operation instruction for the first cell on the performance index of the entire network, and the operation instruction for the second cell is used to adjust the load index of the second cell, wherein the first cell and the second cell are different cells; and an update module for updating the model parameters of the first model to be trained based on the first score and the second score until the model training conditions are met to obtain a first target model.

[0035] As can be seen from the above device, when using the critic model (i.e., the third objective model) to train the actor model (i.e., the whole composed of the first objective model and the second objective model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the performance to balance local and global aspects, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the whole network, and realize the coordination between multiple cells.

[0036] In one possible implementation, the second model to be trained includes a first sub-model and a second sub-model, and a second processing module, used to: process the load index of the first cell through the first sub-model to obtain a first operation; if the load index of the first cell is equal to a preset index threshold, the first operation is used to not adjust the load index of the first cell; process the load index of the first cell through the second sub-model to obtain a second operation; and perform a weighted summation of the first operation and the second operation to obtain an operation instruction for the first cell.

[0037] In one possible implementation, a first training model and a first sub-model are used to indicate a first functional relationship between the load index of the first cell and a first operation for the first cell, and a first training model and a second sub-model are used to indicate a second functional relationship between the load index of the first cell and a second operation for the first cell. The second functional relationship is obtained by shifting the first functional relationship, and both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0038] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0039] In one possible implementation, the operation instruction for the first cell is used to modify the configuration parameters of the first cell to adjust the load index of the first cell; the characteristic information of the first cell includes at least one of the following: the configuration parameters of the first cell; the call statistics data of the first cell; the configuration parameters of the neighboring cells of the first cell; and the call statistics data of the neighboring cells.

[0040] In one possible implementation, the configuration parameters of the first cell may include at least one of the following: the antenna transmit power of the first cell, the antenna downtilt angle of the first cell, the antenna horizontal azimuth angle of the first cell, the RSRP threshold for users initiating inter-frequency handover measurements in the first cell, the RSRP threshold for users initiating inter-frequency handover measurements in the first cell, the specific frequency RSRP offset for users initiating inter-frequency handover procedures in the first cell, the specific neighboring cell RSRP offset for users initiating inter-frequency handover procedures in the first cell, the RSRP threshold for users initiating intra-frequency handover measurements in the first cell, the RSRP threshold for users initiating intra-frequency handover measurements in the first cell, the specific frequency RSRP offset for users initiating intra-frequency handover procedures in the first cell, and the specific neighboring cell RSRP offset for users initiating intra-frequency handover procedures in the first cell, etc. The configuration parameters of a neighboring cell may include at least one of the following: the antenna transmit power of the neighboring cell, the antenna downtilt angle of the neighboring cell, the antenna horizontal azimuth angle of the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating inter-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating inter-frequency handover procedures in the neighboring cell, the specific neighboring cell RSRP offset for users initiating inter-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the RSRP threshold for users initiating same-frequency handover measurements in the neighboring cell, the specific frequency RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, and the specific neighboring cell RSRP offset for users initiating same-frequency handover procedures in the neighboring cell, etc.

[0041] In one possible implementation, the voice statistics data of the first cell may include at least one of the following: the average number of users in the first cell per unit time period, the average number of active users in the first cell per unit time period, the uplink traffic in the first cell per unit time period, the proportion of low channel quality indication (CQI) reports in the first cell per unit time period, the proportion of data packets with a length less than a preset length threshold in the first cell per unit time period (also referred to as the proportion of small packets in the first cell per unit time period), and the average length of data packets in the first cell per unit time period, etc. The voice statistics data of the neighboring cells of the first cell may include at least one of the following: the average number of users in the neighboring cells per unit time period, the average number of active users in the neighboring cells per unit time period, the uplink traffic in the neighboring cells per unit time period, the proportion of CQI reports in the neighboring cells per unit time period, the proportion of data packets with a length less than a preset length threshold in the neighboring cells per unit time period (also referred to as the proportion of small packets in the neighboring cells per unit time period), and the average length of data packets in the neighboring cells per unit time period, etc.

[0042] A fifth aspect of this application provides a cell load adjustment apparatus, the apparatus including a memory and a processor; the memory stores code, and the processor is configured to execute the code, wherein when the code is executed, the cell load adjustment apparatus performs the method described in the first aspect or any possible implementation thereof.

[0043] A sixth aspect of this application provides a model training apparatus, which includes a memory and a processor; the memory stores code, and the processor is configured to execute the code. When the code is executed, the model training apparatus performs the method described in the second aspect or any possible implementation thereof.

[0044] A seventh aspect of this application provides a circuit system including a processing circuit configured to perform the method described in the first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect.

[0045] An eighth aspect of this application provides a chip system including a processor for calling a computer program or computer instructions stored in a memory, such that the processor performs the method as described in the first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect.

[0046] In one possible implementation, the processor is coupled to the memory via an interface.

[0047] In one possible implementation, the chip system also includes a memory that stores computer programs or computer instructions.

[0048] A ninth aspect of this application provides a computer storage medium storing a computer program that, when executed by a computer, causes the computer to perform the method as described in the first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect.

[0049] A tenth aspect of this application provides a computer program product storing instructions that, when executed by a computer, cause the computer to perform the method as described in the first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect.

[0050] In this embodiment, after obtaining the feature information and load indicators of the target cell, the feature information can be processed using a first target model to obtain the model parameters of a second target model. Then, a second target model is constructed based on the model parameters, and the load indicators are processed using the second target model to obtain an operation instruction for the target cell. This operation instruction can be used to adjust the load indicators of the target cell. In the aforementioned process, the model parameters of the second target model are obtained based on the feature information of the target cell. Since the feature information of the target cell can be used to characterize various features of the target cell (i.e., the scenario in which the target cell is located), when processing the load indicators of the target cell using the second target model, various factors such as the various features of the target cell are considered. Therefore, the operation instruction for the target cell output by the second target model can be used to accurately adjust the load level of the target cell to keep its various performance indicators within a suitable range, providing users with better network services.

[0051] Furthermore, the whole consisting of the first target model and the second target model can be used to indicate the functional relationship between the load index of the target cell and the operation instructions for the target cell. This functional relationship is usually monotonically increasing or monotonically decreasing, which conforms to the constraints of expert experience (equivalent to integrating expert experience into the model). The control strategy learned by the model is more in line with the business logic.

[0052] Furthermore, through the teacher-student network architecture, a first sub-model and a second sub-model were constructed. The first sub-model added expert constraints (i.e., using τ as a threshold to determine whether the target cell should absorb or release users) to ensure the security of the control policy output by the model. The second sub-model relaxed the expert constraints and learned from the data to output the corresponding control policy. The outputs of the two models were weighted to obtain the final policy, thereby balancing the security and efficiency of the control policy.

[0053] Furthermore, when using the critic model (i.e., the third objective model) to train the actor model (i.e., the whole composed of the first objective model and the second objective model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the ability to balance local and global performance, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the entire network, thus realizing the coordination between multiple cells. Attached Figure Description

[0054] Figure 1 A structural diagram illustrating the main framework of artificial intelligence;

[0055] Figure 2aA schematic diagram of a cell load adjustment system provided in an embodiment of this application;

[0056] Figure 2b Another structural schematic diagram of the cell load adjustment system provided in this application embodiment;

[0057] Figure 2c A schematic diagram of the related equipment for adjusting cell load provided in an embodiment of this application;

[0058] Figure 3 A schematic diagram of the system 100 architecture provided in the embodiments of this application;

[0059] Figure 4 A schematic flowchart illustrating a cell load adjustment method provided in an embodiment of this application;

[0060] Figure 5 A schematic diagram of the structure of the actor model provided in the embodiments of this application;

[0061] Figure 6 Another flowchart illustrating the cell load adjustment method provided in this application embodiment;

[0062] Figure 7 This is another structural diagram of the actor model provided in the embodiments of this application;

[0063] Figure 8 A schematic flowchart of the model training method provided in the embodiments of this application;

[0064] Figure 9 Another schematic diagram of the model training method provided in the embodiments of this application;

[0065] Figure 10 A schematic diagram of a cell load adjustment device provided in an embodiment of this application;

[0066] Figure 11 A schematic diagram of the structure of the model training apparatus provided in the embodiments of this application;

[0067] Figure 12 A schematic diagram of the structure of the execution device provided in the embodiments of this application;

[0068] Figure 13 A schematic diagram of the structure of the training device provided in the embodiments of this application;

[0069] Figure 14 This is a schematic diagram of the structure of a chip provided in an embodiment of this application. Detailed Implementation

[0070] This application provides a method for adjusting cell load and related equipment, which can accurately adjust the load level of a target cell to keep the various performance indicators of the target cell within a suitable range, thereby providing users with better network services.

[0071] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0072] In wireless cellular networks, uneven user distribution leads to uneven load distribution among cells (i.e., areas covered by the wireless network). High-load cells, with a large number of users and high service demand, are prone to insufficient network resources, making it difficult to guarantee the quality of service for users. Conversely, low-load cells, with fewer users and lower service demand, do not fully utilize their network resources.

[0073] To adjust the load level of a cell, the cell's configuration parameters can be modified to allow the cell to release or absorb users (equivalent to reducing or increasing the cell's load level), thereby optimizing various performance indicators of the cell and providing users with better network services.

[0074] When modifying the configuration parameters of a cell, the method usually involves manual modification. However, manual modification relies on the experience of experts, and the factors considered are often too simplistic and not comprehensive enough. This makes it impossible to accurately modify the cell's configuration parameters to keep the cell's load level within an appropriate range, which in turn makes it impossible to maintain the cell's various performance indicators within a suitable range and provide users with sufficiently high-quality network services.

[0075] To address the aforementioned problems, this application provides a method for adjusting cell load, which can be implemented using artificial intelligence (AI) technology. AI technology is a discipline that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence. AI technology achieves optimal results by perceiving the environment, acquiring knowledge, and using that knowledge. In other words, artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can react in a way similar to human intelligence. Using artificial intelligence for image processing is a common application of AI.

[0076] First, the overall workflow of the artificial intelligence system is described; please refer to [link / reference]. Figure 1 , Figure 1 This is a structural diagram illustrating the main framework of artificial intelligence. The following explanation of the AI ​​framework is based on two dimensions: the "Intelligent Information Chain" (horizontal axis) and the "IT Value Chain" (vertical axis). The "Intelligent Information Chain" reflects a series of processes from data acquisition to processing. For example, it could be the general process of intelligent information perception, intelligent information representation and formation, intelligent reasoning, intelligent decision-making, and intelligent execution and output. In this process, data undergoes a condensation process of "data—information—knowledge—wisdom." The "IT Value Chain" reflects the value that artificial intelligence brings to the information technology industry, from the underlying infrastructure of human intelligence and information (provided and processed by technology) to the industrial ecosystem of the system.

[0077] (1) Infrastructure

[0078] Infrastructure provides computing power to support artificial intelligence systems, enabling communication with the external world and providing support through a basic platform. This communication occurs through sensors; computing power is provided by intelligent chips (hardware acceleration chips such as CPUs, NPUs, GPUs, ASICs, and FPGAs); and the basic platform includes distributed computing frameworks and related platform guarantees and support, which may include cloud storage and computing, interconnected networks, etc. For example, sensors communicate with the outside world to acquire data, and this data is provided to intelligent chips in the distributed computing system provided by the basic platform for computation.

[0079] (2) Data

[0080] The data at the next layer of infrastructure is used to represent the data sources in the field of artificial intelligence. The data involves graphics, images, voice, text, and IoT data from traditional devices, including business data from existing systems and sensor data such as force, displacement, liquid level, temperature, and humidity.

[0081] (3) Data processing

[0082] Data processing typically includes methods such as data training, machine learning, deep learning, search, reasoning, and decision-making.

[0083] Among them, machine learning and deep learning can perform intelligent information modeling, extraction, preprocessing, and training on data, including symbolization and formalization.

[0084] Reasoning refers to the process in which, in a computer or intelligent system, the machine thinks and solves problems by simulating human intelligent reasoning, based on reasoning control strategies and using formalized information. Typical functions include search and matching.

[0085] Decision-making refers to the process of making decisions based on intelligent information after reasoning, and it typically provides functions such as classification, sorting, and prediction.

[0086] (4) General ability

[0087] After the data processing mentioned above, the results of the data processing can be used to form some general capabilities, such as algorithms or a general system, for example, translation, text analysis, computer vision processing, speech recognition, image recognition, etc.

[0088] (5) Smart Products and Industry Applications

[0089] Intelligent products and industry applications refer to products and applications of artificial intelligence systems in various fields. They are the encapsulation of overall artificial intelligence solutions, productizing intelligent information decision-making and realizing practical applications. Their application areas mainly include: intelligent terminals, intelligent transportation, intelligent healthcare, autonomous driving, smart cities, etc.

[0090] The following sections will introduce several application scenarios for this application.

[0091] Figure 2a This is a schematic diagram of a cell load adjustment system provided in an embodiment of this application. The system includes a network control center (also known as a data processing device), which is connected to multiple wireless base stations. The signal coverage area of ​​each wireless base station can be divided into multiple cells, which constitute a wireless network that can provide communication services to users. The network control center can manage all cells. For example, the network control center can collect data from each cell, modify the configuration parameters of the cells, and monitor the status of each cell in real time.

[0092] The network control center receives cell load adjustment requests triggered by itself through an interactive interface. It then uses storage devices for data storage and processors for data processing to adjust the cell load through machine learning, deep learning, search, inference, and decision-making methods. The storage device in the network control center can be a general term, including local storage and a database storing historical data. This database can reside on the network control center or on other network servers.

[0093] exist Figure 2a In the cell load adjustment system shown, the network control center can obtain a cell load adjustment request and collect relevant information about a cell based on the request. Then, the network control center can process this information to obtain operations for adjusting the cell's load level. For example, the network control center can collect characteristic information and load indicators (indicating the cell's load level) of a cell, and then perform a series of processes on this information to obtain operations for adjusting the cell's load indicators.

[0094] exist Figure 2a In this context, the network control center can execute the cell load adjustment method of this application embodiment.

[0095] Figure 2b This is another schematic diagram of the cell load adjustment system provided in the embodiments of this application. Figure 2b The system comprises user equipment and a network control center. User equipment includes smart terminals such as mobile phones, personal computers, or information processing centers. User equipment is the initiator of cell load adjustment requests; these requests are typically initiated by users (e.g., wireless network administrators) through their user equipment.

[0096] exist Figure 2b In the image processing system shown, the user equipment can receive user commands and initiate a cell load adjustment request to the network control center, enabling the network control center to perform cell load adjustment operations. The process of the network control center implementing cell load adjustment is similar to... Figure 2a Similar to the description above, it will not be repeated here.

[0097] exist Figure 2b In addition, the network control center can also execute the cell load adjustment method of the embodiments of this application.

[0098] Figure 2c This is a schematic diagram of the related equipment for adjusting cell load provided in an embodiment of this application.

[0099] The above Figure 2a and Figure 2b The user equipment in the context can specifically be Figure 2c Local device 301 or local device 302 in the system. Figure 2a The data processing equipment in the middle can specifically be Figure 2c The execution device 210 in the process includes a data storage system 250 that can store the data to be processed by the execution device 210. The data storage system 250 can be integrated into the execution device 210 or set up in the cloud or on other network servers.

[0100] Figure 2a and Figure 2b The processor in the system can perform data training / machine learning / deep learning using neural network models or other models (e.g., support vector machine-based models), and use the data to train or learn the model to process relevant information about the cell, thereby obtaining the corresponding processing results.

[0101] Figure 3 A schematic diagram of the system 100 architecture provided in this application embodiment, in Figure 3 In the process, the execution device 110 is configured with an input / output (I / O) interface 112 for data interaction with external devices. Users can input data to the I / O interface 112 through the client device 140. The input data in this embodiment may include various scheduled tasks, callable resources, and other parameters.

[0102] During the preprocessing of input data by the execution device 110, or during the calculation module 111 of the execution device 110 performing calculations and other related processing (such as implementing the neural network function in this application), the execution device 110 may call data, code, etc. in the data storage system 150 for corresponding processing, or store the data, instructions, etc. obtained from the corresponding processing into the data storage system 150.

[0103] Finally, I / O interface 112 returns the processing result to client device 140, thereby providing it to the user.

[0104] It is worth noting that the training device 120 can generate corresponding target models / rules based on different training data for different objectives or tasks. These target models / rules can then be used to achieve the aforementioned objectives or complete the aforementioned tasks, thereby providing the user with the required results. The training data can be stored in the database 130 and originates from training samples collected by the data acquisition device 160.

[0105] exist Figure 3In the scenario shown, the user can manually provide input data, which can be done through the interface provided by I / O interface 112. Alternatively, the client device 140 can automatically send input data to I / O interface 112. If user authorization is required for the client device 140 to automatically send input data, the user can set the corresponding permissions in the client device 140. The user can view the output results of the execution device 110 on the client device 140, which can be presented in various forms such as display, sound, or animation. The client device 140 can also act as a data acquisition terminal, collecting the input data and output results of the input I / O interface 112 as new sample data and storing them in the database 130. Alternatively, data can be collected directly from the I / O interface 112 without going through the client device 140, using the input data and output results of the input I / O interface 112 as new sample data and storing them in the database 130.

[0106] It is worth noting that, Figure 3 This is merely a schematic diagram of a system architecture provided in an embodiment of this application. The positional relationships between the devices, components, modules, etc., shown in the diagram do not constitute any limitation. For example, in Figure 3 In this context, the data storage system 150 is an external memory relative to the execution device 110. However, in other cases, the data storage system 150 can also be placed within the execution device 110. For example... Figure 3 As shown, a neural network can be trained using training device 120.

[0107] This application also provides a chip including a neural network processor (NPU). This chip can be configured as follows: Figure 3 The execution device 110 shown is used to perform the calculations of the calculation module 111. This chip can also be placed in, for example... Figure 3 The training device 120 shown is used to complete the training work of the training device 120 and output the target model / rules.

[0108] The Neural Processing Unit (NPU) is a coprocessor mounted on the main central processing unit (CPU) (host CPU), where tasks are assigned by the CPU. The core of the NPU is the computation circuitry, which is controlled by a controller to retrieve data from memory (weight memory or input memory) and perform calculations.

[0109] In some implementations, the arithmetic circuitry includes multiple process engines (PEs). In some implementations, the arithmetic circuitry is a two-dimensional pulsating array. The arithmetic circuitry can also be a one-dimensional pulsating array or other electronic circuitry capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuitry is a general-purpose matrix processor.

[0110] For example, suppose we have an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit retrieves the corresponding data of matrix B from the weight memory and caches it in each PE (Process Equipment) of the arithmetic circuit. The arithmetic circuit retrieves the data of matrix A from the input memory and performs matrix operations with matrix B. The partial or final result of the obtained matrix is ​​stored in the accumulator.

[0111] Vector computation units can further process the output of computational circuits, such as vector multiplication, vector addition, exponentiation, logarithmic operations, size comparisons, etc. For example, vector computation units can be used for computation in non-convolutional / non-FC layers of neural networks, such as pooling, batch normalization, and local response normalization.

[0112] In some implementations, the vector computation unit can store the processed output vector into a unified buffer. For example, the vector computation unit can apply a nonlinear function to the output of the arithmetic circuit, such as a vector of accumulated values, to generate activation values. In some implementations, the vector computation unit generates normalized values, merged values, or both. In some implementations, the processed output vector can be used as activation input to the arithmetic circuit, for example, for use in subsequent layers of a neural network.

[0113] The unified memory is used to store input data and output data.

[0114] The weight data is directly transferred from the external memory to the input memory and / or unified memory, stored in the weight memory, and stored in the unified memory to the external memory through the direct memory access controller (DMAC).

[0115] The bus interface unit (BIU) is used to enable interaction between the main CPU, DMAC, and instruction fetch memory via a bus.

[0116] The instruction fetch buffer, connected to the controller, is used to store the instructions used by the controller.

[0117] The controller is used to invoke instructions cached in the memory to control the operation of the computing accelerator.

[0118] Generally, the unified memory, input memory, weight memory, and instruction fetch memory are all on-chip memories, while external memory is memory outside the NPU. This external memory can be double data rate synchronous dynamic random access memory (DDRSDRAM), high bandwidth memory (HBM), or other readable and writable memories.

[0119] Since the embodiments of this application involve a large number of neural network applications, for ease of understanding, the relevant terms and concepts such as neural networks involved in the embodiments of this application will be introduced below.

[0120] (1) Neural Network

[0121] A neural network can be composed of neural units, which can be operational units that take xs and an intercept of 1 as inputs, and whose output can be:

[0122]

[0123] Where s = 1, 2, ..., n, where n is a natural number greater than 1, Ws is the weight of xs, and b is the bias of the neural unit. f is the activation function of the neural unit, used to introduce nonlinear characteristics into the neural network to convert the input signal in the neural unit into the output signal. The output signal of this activation function can be used as the input of the next convolutional layer. The activation function can be the sigmoid function. A neural network is a network formed by connecting many of the above-mentioned individual neural units together, that is, the output of one neural unit can be the input of another neural unit. The input of each neural unit can be connected to the local receptive field of the previous layer to extract the features of the local receptive field, which can be a region composed of several neural units.

[0124] The work of each layer in a neural network can be described by the mathematical expression y = a(Wx + b). From a physical perspective, the work of each layer in a neural network can be understood as transforming the input space (the set of input vectors) to the output space (i.e., from the row space to the column space of a matrix) through five operations on the input space. These five operations include: 1. Dimensionality increase / decrease; 2. Magnification / scaling; 3. Rotation; 4. Translation; 5. "Bending". Operations 1, 2, and 3 are performed by Wx, operation 4 by +b, and operation 5 by a(). The term "space" is used here because the objects being classified are not individual things, but a class of things, and space refers to the set of all individuals of this class of things. Here, W is the weight vector, and each value in this vector represents the weight value of a neuron in that layer of the neural network. This vector W determines the spatial transformation from the input space to the output space mentioned above; that is, the weights W of each layer control how the space is transformed. The purpose of training a neural network is to ultimately obtain the weight matrix of all layers of the trained neural network (a weight matrix formed by the vectors W of many layers). Therefore, the training process of a neural network is essentially about learning how to control the transformation space, and more specifically, learning the weight matrix.

[0125] Because we want the output of the neural network to be as close as possible to the actual predicted value, we can compare the current network's prediction with the desired target value, and then update the weight vector of each layer of the neural network based on the difference between the two (of course, there is usually an initialization process before the first update, that is, pre-configuring the parameters of each layer in the neural network). For example, if the network's prediction is too high, the weight vector is adjusted to make it predict lower, and this adjustment is continued until the neural network can predict the actual target value. Therefore, it is necessary to predefine "how to compare the difference between the predicted value and the target value," which is the loss function or objective function. These are important equations used to measure the difference between the predicted value and the target value. Taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, so training the neural network becomes the process of minimizing this loss as much as possible.

[0126] (2) Backpropagation algorithm

[0127] Neural networks can employ backpropagation (BP) to correct the parameters of the initial neural network model during training, thereby reducing the reconstruction error loss. Specifically, forward propagation of the input signal to the output generates error loss; this error loss information is then propagated back to update the parameters of the initial neural network model, leading to convergence of the error loss. The backpropagation algorithm is an error-loss-driven backpropagation process aimed at obtaining the optimal parameters of the neural network model, such as the weight matrix.

[0128] The method provided in this application is described below from the perspectives of neural network training and neural network application.

[0129] The model training method provided in this application involves image processing and can be specifically applied to data processing methods such as data training, machine learning, and deep learning. It performs symbolic and formal intelligent information modeling, extraction, preprocessing, and training on training data (such as the feature information and load indicators of the target cell in the model training method of this application), ultimately obtaining a trained neural network (such as the first target model and the second target model in this application). Furthermore, the cell load adjustment method provided in this application can utilize the trained neural network to input input data (such as the feature information and load indicators of the target cell in the cell load adjustment method of this application) into the trained neural network to obtain output data (such as operation instructions in the cell load adjustment method of this application). It should be noted that the model training method and the cell load adjustment method provided in this application are inventions based on the same concept and can be understood as two parts of a system or two stages of an overall process: such as the model training stage and the model application stage.

[0130] It is worth noting that the embodiments of this application can be implemented based on the actor-critic model architecture in reinforcement learning. This model architecture includes an actor model and a critic model. The actor model can be used to implement the cell load adjustment method (i.e., the model application stage) provided in the embodiments of this application. The actor model contains two neural networks: an action network and a weight network. The critic model can be used to implement the model training method (i.e., the model training stage) provided in the embodiments of this application, with the aim of training the aforementioned actor model. The critic model can also contain two neural networks: a local network and a global network. For ease of explanation, the weight network will be referred to as the first target model, the action network as the second target model, and the entire critic model as the third target model.

[0131] The following section will first introduce the model application stage. Figure 4 This is a flowchart illustrating a cell load adjustment method provided in an embodiment of this application. This method can be implemented using an actor model, and its output is an action. Figure 5 A schematic diagram of the actor model provided in the embodiments of this application is shown below. Figure 5 As shown, the actor model contains a first objective model and a second objective model. The output of the first objective model is applied to the second objective model; in other words, the output of the first objective model is used to construct the second objective model. Figure 4 As shown, the method includes:

[0132] 401. Obtain the characteristic information and load index of the target cell. The load index of the target cell is used to indicate the load level of the target cell.

[0133] In this embodiment, when it is necessary to adjust the load level of the target cell, the characteristic information c of the target cell and the load index x of the target cell can be collected. The characteristic information c of the target cell can be used to indicate the current scenario of the target cell, and the load index x of the target cell can be used to indicate the load level of the target cell (also known as the load balancing level).

[0134] Specifically, the characteristic information c of the target cell may include at least one of the following:

[0135] Configuration parameters of the target cell;

[0136] Call statistics data for the target community;

[0137] Configuration parameters of neighboring communities of the target community;

[0138] Statistical data from neighboring communities, etc.

[0139] Furthermore, the configuration parameters of the target cell may include at least one of the following:

[0140] The antenna transmit power of the target cell;

[0141] Antenna downtilt angle of the target cell;

[0142] The horizontal azimuth angle of the antenna in the target cell;

[0143] The reference signal receiving power (RSRP) threshold for user-initiated inter-frequency handover measurement in the target cell;

[0144] The RSRP threshold for stopping inter-frequency handover measurements in the target cell;

[0145] The specific frequency RSRP offset that triggers the inter-frequency handover process for users in the target cell;

[0146] The RSRP offset of a specific neighboring cell that triggers the inter-frequency handover process for users in the target cell;

[0147] The RSRP threshold for initiating intra-frequency handover measurement in the target cell;

[0148] The RSRP threshold for stopping intra-frequency handover measurements in the target cell;

[0149] The specific frequency RSRP offset that triggers the same-frequency handover process for users in the target cell;

[0150] The RSRP offset of a specific neighboring cell that triggers the same-frequency handover process for users in the target cell, etc.

[0151] Furthermore, the call statistics data of the target cell may include at least one of the following:

[0152] The average number of users in the target community per unit time period;

[0153] The average number of active users in the target community within a given time period;

[0154] Uplink traffic of the target cell within a unit of time period;

[0155] The proportion of low channel quality indicator (CQI) reports for a target cell per unit time period;

[0156] The proportion of data packets in the target cell whose length is less than a preset length threshold within a unit time period (also known as the proportion of small packets in the target cell within a unit time period);

[0157] The average length of data packets per unit time period in the target cell, etc.

[0158] Furthermore, the configuration parameters of neighboring cells may include at least one of the following:

[0159] The antenna transmission power of the neighboring community;

[0160] The antenna downtilt angle in the neighboring community;

[0161] The horizontal azimuth angle of the antenna in the neighboring community;

[0162] RSRP threshold for inter-frequency handover measurement initiated by users in neighboring communities;

[0163] The RSRP threshold for measuring when users in neighboring communities stop inter-frequency handover;

[0164] The specific frequency RSRP bias that triggers the inter-frequency handover process for users in neighboring communities;

[0165] The RSRP bias of a specific neighboring cell that triggers the inter-frequency handover process for users in neighboring cells;

[0166] The RSRP threshold measured when users in neighboring communities initiate same-frequency handover;

[0167] The RSRP threshold measured when users in neighboring communities stop switching on the same frequency.

[0168] The specific frequency RSRP bias that triggers the same-frequency handover process for users in neighboring communities;

[0169] Users in neighboring communities trigger the same-frequency handover process by setting specific neighboring cell RSRP bias, etc.

[0170] Furthermore, the call statistics data of neighboring communities may include at least one of the following:

[0171] The average number of users in a neighboring community within a given time period;

[0172] The average number of active users in a neighboring community within a given time period;

[0173] Uplink traffic in a neighboring residential area within a given time period;

[0174] The proportion of CQI reports from neighboring communities within a given time period;

[0175] The proportion of data packets in a neighboring cell whose length is less than a preset length threshold within a unit time period (also known as the proportion of small packets in a neighboring cell within a unit time period);

[0176] The average length of data packets in neighboring communities within a unit of time period, etc.

[0177] 402. The feature information of the target cell is processed by the first target model to obtain the model parameters of the second target model.

[0178] 403. Obtain the second target model based on the model parameters of the second target model.

[0179] After obtaining the feature information c and the load index x of the target cell, a first target model can be obtained, which is a trained neural network. Then, the feature information c of the target cell is input into the first target model to perform a series of processing steps (e.g., feature extraction) on the feature information c, resulting in the model parameters (also called the weights) W(c) and b(c) of the second target model. It should be noted that within the first target model, certain operations (e.g., the softplus activation function) can be used to make W(c) a non-negative vector, i.e., W(c) ≥ 0.

[0180] Therefore, based on the model parameters W(c) and b(c) of the second target model, a second target model can be constructed, which can be a multilayer perceptron (MLP). Thus, the whole consisting of the first and second target models (i.e., the entire actor model) can be used to indicate the functional relationship between the load indicators of the target cell and the operational instructions for the target cell (or, in other words, the whole consisting of the first and second target models can be represented by this functional relationship), which can be expressed as:

[0181] a=σ(f M (x, c)) (2)

[0182] f M (x,c)=MLP(W(c),b(c),x) (3)

[0183] In the above formula, 'a' represents the operational instruction for the target cell; f M (x, c) is a monotonically increasing function of the load index x of the target cell, and the model parameter W(c) is f M The slope of (x, c), and the model parameter b(c) are f M A subset of the remaining parameters of (x, c); σ is a monotonically increasing function (e.g., tanh activation function, etc.) used to constrain the range of values ​​for operation a against the target cell to [a...]. L , a H ]. So, because f M (x, c) and σ are both monotonically increasing functions. The operation instruction a for the target cell is also a monotonically increasing function of the load index x of the target cell, and the rate of increase is determined by the characteristic information c of the target cell.

[0184] Therefore, it can be seen that when the load index x of the target cell is larger (i.e., the load level of the target cell is greater), the operation instruction a for the target cell is larger, indicating that the target cell needs to release more users (or the target cell needs to absorb fewer users). When the load index x of the target cell is smaller (i.e., the load level of the target cell is smaller), the operation instruction a for the target cell is smaller, indicating that the target cell needs to release fewer users (or the target cell needs to absorb more users). It can be seen that the form of formula (2) is consistent with the expert experience strategy.

[0185] It should be understood that in this embodiment, the example of the target cell's operation instruction 'a' being a monotonically increasing function of the target cell's load index 'x' is used for illustrative purposes only. In practical applications, the target cell's operation instruction 'a' can also be a monotonically decreasing function of the target cell's load index 'x', simply by controlling the model parameter W(c) < 0 or replacing σ with a monotonically decreasing function. It is understood that in this case, when the target cell's load index 'x' is larger (i.e., the target cell's load level is higher), the operation instruction 'a' for the target cell is smaller, indicating that the target cell needs to release more users (or absorb fewer users). Conversely, when the target cell's load index 'x' is smaller (i.e., the target cell's load level is lower), the operation instruction 'a' for the target cell is larger, indicating that the target cell needs to release fewer users (or absorb more users).

[0186] It should also be understood that the softplus and tanh activation functions are used only for illustrative purposes in this embodiment and do not constitute a limitation on the types of activation functions used in this application. In practical applications, other activation functions with similar properties may also be used.

[0187] 404. The load index is processed through the second target model to obtain the operation instructions for the target cell. The operation instructions for the target cell are used to adjust the load index of the target cell.

[0188] After constructing the second target model, the load index x of the target cell can be input into the second target model to process the load index x of the target cell and obtain the operation instruction a for the target cell. The operation instruction a for the target cell can be used to adjust the load index x of the target cell.

[0189] Specifically, the load metric x of the target cell is typically determined based on call statistics data from both the target cell and neighboring cells. For example, the load metric x can be the ratio of the average number of users in the target cell per unit time period to the average number of users in neighboring cells per unit time period. Another example is the ratio of the average number of active users in the target cell per unit time period to the average number of active users in neighboring cells per unit time period. Yet another example is the ratio of the uplink traffic in the target cell per unit time period to the uplink traffic in neighboring cells per unit time period, and so on.

[0190] The call statistics data of a target cell used to determine its load metric x are often influenced by one or more configuration parameters of the target cell. Therefore, operation instruction 'a' for the target cell can be used to modify the target cell's configuration parameters to indirectly adjust the target cell's load metric x. For example, when the target cell's load metric x is the ratio of the average number of users in the target cell per unit time period to the average number of users in neighboring cells per unit time period, the average number of users in the target cell per unit time period is affected by the RSRP threshold for inter-frequency handover measurement in the target cell (for example, the larger the RSRP threshold for inter-frequency handover measurement in the target cell is set, the smaller the average number of users in the target cell per unit time period). Therefore, operation instruction 'a' for the target cell can be used to modify the RSRP threshold for inter-frequency handover measurement in the target cell, thereby affecting the value of the average number of users in the target cell per unit time period, and thus adjusting the ratio between the average number of users in the target cell per unit time period and the average number of users in neighboring cells per unit time period.

[0191] Furthermore, the operational instruction 'a' for the target cell often represents the extent of modification to the target cell's configuration parameters, that is, the adjustment range of the target cell's load metric 'x'. Continuing with the example above, a larger operational instruction 'a' for the target cell allows for a larger RSRP threshold for initiating inter-frequency handover measurements in the target cell, resulting in fewer average users per unit time period in the target cell (i.e., releasing a sufficient number of users), and a greater reduction in the target cell's load. Therefore, after such adjustments, the adjusted target cell's load metric 'x' can be kept within a suitable range.

[0192] In summary, by adjusting the load index x of the target cell based on the operation instruction a, the adjusted load index x' of the target cell can be kept within a suitable range. Therefore, various performance indicators of the target cell (such as the average downlink perceived rate of users in the target cell per unit time period, the average uplink perceived rate of users in the target cell per unit time period, the average latency of users sending data packets in the target cell per unit time period, and the proportion of users with a rate lower than 5M in the target cell per unit time period, etc.) can also be optimized accordingly, thereby providing better network services for users in the target cell.

[0193] In this embodiment, after obtaining the feature information and load indicators of the target cell, the feature information can be processed using a first target model to obtain the model parameters of a second target model. Then, a second target model is constructed based on the model parameters, and the load indicators are processed using the second target model to obtain an operation instruction for the target cell. This operation instruction can be used to adjust the load indicators of the target cell. In the aforementioned process, the model parameters of the second target model are obtained based on the feature information of the target cell. Since the feature information of the target cell can be used to characterize various features of the target cell (i.e., the scenario in which the target cell is located), when processing the load indicators of the target cell using the second target model, various factors such as the various features of the target cell are considered. Therefore, the operation instruction for the target cell output by the second target model can be used to accurately adjust the load level of the target cell to keep its various performance indicators within a suitable range, providing users with better network services.

[0194] Furthermore, the whole consisting of the first target model and the second target model can be used to indicate the functional relationship between the load index of the target cell and the operation instructions for the target cell. This functional relationship is usually monotonically increasing or monotonically decreasing, which conforms to the constraints of expert experience (equivalent to integrating expert experience into the model). The control strategy learned by the model is more in line with the business logic.

[0195] Figure 6 This is another flowchart illustrating the cell load adjustment method provided in an embodiment of this application. This method can be implemented using an actor model, where the output of the actor model is an operation. Figure 7 Another structural diagram of the actor model provided in the embodiments of this application is shown below. Figure 7 As shown, the actor model includes a first target model, a first sub-model (also known as the teacher network), and a second sub-model (also known as the student network). The output of the first target model acts on the first and second sub-models. In other words, the output of the first target model is used to construct the first and second sub-models.

[0196] like Figure 6 As shown, the method includes:

[0197] 601. Obtain the characteristic information and load index of the target cell. The load index is used to indicate the load level of the target cell.

[0198] For an explanation of step 601, please refer to... Figure 4 The relevant descriptions of step 401 in the illustrated embodiment will not be repeated here.

[0199] 602. The feature information is processed through the first target model to obtain the model parameters of the first sub-model and the model parameters of the second sub-model.

[0200] 603. Obtain the first sub-model based on the model parameters of the first sub-model, and obtain the second sub-model based on the model parameters of the second sub-model.

[0201] After obtaining the feature information c and the load index x of the target cell, a first target model can be obtained, which is a trained neural network. Then, the feature information c of the target cell is input into the first target model to perform a series of processes on the feature information c (e.g., feature extraction), resulting in the model parameters W(c), b(c), and h of the first sub-model. T (c) and v T (c), and the model parameters W(c), b(c), h of the second sub-model. S (c) and v S (c). It should be noted that within the first target model, certain operations (e.g., the softplus activation function, etc.) can be performed to make W(c) a non-negative vector, i.e., W(c)≥0.

[0202] Then, the model parameters W(c), b(c), and h based on the first sub-model T (c) and v T (c) A first sub-model can be constructed, based on the model parameters W(c), b(c), h of the second sub-model. S (c) and v S (c) A second sub-model can be constructed, and both the first and second sub-models can be MLPs. Thus, the whole consisting of the first target model and the first sub-model can be used to indicate the first functional relationship between the load metric of the target cell and the first operation for the target cell (or, in other words, the whole consisting of the first target model and the first sub-model can be represented by this functional relationship), which can be expressed as:

[0203] a T =σ(f M (xh T(c), c)+v T (c)) (5)

[0204] f M (xh T (c), c)+v T (c) = MLP(W(c), b(c), xh) T (c), v T (c)) (6)

[0205] In the above formula, a T This is the first operation targeting the cell; f M (xh T (c), c)+v T (c) is a monotonically increasing function of the load index x of the target cell, and the model parameter W(c) is f M (xh T (c), c)+v T The slope of (c), and the model parameter b(c) is f M Of the remaining parameters of (x, c), it should be noted that f M (xh T (c), c)+v T (c) can be regarded as the f shown in the aforementioned formula (3) M (x, c) is obtained by vertical and horizontal translation, where the model parameter h T (c) represents the magnitude of the horizontal translation, v T (c) represents the magnitude of the vertical translation; σ is a monotonically increasing function (e.g., the tanh activation function, etc.) used to constrain the first operation a for the target cell. T The range of values ​​for is [a L , a H ]. So, because f M (xh T (c), c)+v T (c) and σ are both monotonically increasing functions, and the first operation a for the target cell... T This is a monotonically increasing function of the load index x of the target cell, and the rate of increase is determined by the characteristic information c of the target cell.

[0206] It is worth noting that the first functional relationship must also satisfy the following constraints:

[0207] a0=σ(f M (τ-h T (c), c)+v T (c)) (7)

[0208] In the above formula, τ is a preset index threshold. If the target cell is such that its load index x = τ, then the first operation a for the target cell... T =a0, where the value of a0 is located in [a L , a H In the context of [], it indicates that the target cell's load metrics do not need to be adjusted, meaning the target cell neither needs to absorb nor release users.

[0209] Similarly, the whole consisting of the first target model and the second sub-model can be used to indicate the second functional relationship between the load index of the target cell and the second operation for the target cell (it can also be understood that the whole consisting of the first target model and the first sub-model can be represented by this functional relationship), and the second functional relationship can be expressed as:

[0210] a S =σ(f M (xh S (c), c)+v S (c)) (8)

[0211] f M (xh S (c), c)+v S (c) = MLP(W(c), b(c), xh) S (c), v S (c)) (9)

[0212] In the above formula, a S This is the second operation targeting the cell; f M (xh S (c), c)+v S (c) is a monotonically increasing function of the load index x of the target cell, and the model parameter W(c) is f M (xh S (c), c)+v S The slope of (c), and the model parameter b(c) is f M Of the remaining parameters of (x, c), it should be noted that f M (xh S (c), c)+v S (c) can be regarded as the f shown in the aforementioned formula (3) M The model is obtained by vertical and horizontal translation of (x, c), where the model parameter h is... S (c) represents the magnitude of the horizontal translation, v S (c) represents the magnitude of the vertical translation; σ is a monotonically increasing function (e.g., the tanh activation function, etc.) used to constrain the second operation a for the target cell. S The range of values ​​for is [aL , a H ]. So, because f M (xh S (c), c)+v S (c) and σ are both monotonically increasing functions, and the second operation a for the target cell... T This is a monotonically increasing function of the load index x of the target cell, and the rate of increase is determined by the characteristic information c of the target cell.

[0213] It is worth noting that f M (xh T (c), c)+v T (c) and f M (xh S (c), c)+v S (c) are all based on f M The result is obtained by translating (x, c). Then, f M (xh S (c), c)+v S (c) can also be considered as being based on f M (xh T (c), c)+v T (c) The result of translation, and the translation range is determined based on the characteristic information c of the target cell.

[0214] Therefore, the operation instruction 'a' for the target cell can be derived from the first operation 'a' for the target cell. T and the second operation a targeting the cell S Sure:

[0215] a=w1*a T +w2*a S (10)

[0216] In the above formula, w1 and w2 are preset weight values, the size of which can be set according to actual needs, and there is no limitation here.

[0217] Based on formula (10), the operational instruction 'a' for the target cell is also a monotonically increasing function of the target cell's load index x. That is, the larger the target cell's load index x (i.e., the greater the load level of the target cell), the larger the operational instruction 'a' for the target cell, indicating that the target cell needs to release more users (or absorb fewer users). When the target cell's load index x is smaller (i.e., the smaller the load level of the target cell), the smaller the operational instruction 'a' for the target cell, indicating that the target cell needs to release fewer users (or absorb more users). It can be seen that the form of formula (2) is consistent with the expert experience strategy.

[0218] It should be understood that in this embodiment, the first and second functional relationships are only illustrated illustratively. In practical applications, both the first and second functional relationships can also be monotonically decreasing, simply by controlling the model parameter W(c) < 0 or replacing σ with a monotonically decreasing function. It can be understood that in this case, the operation instruction 'a' for the target cell is a monotonically decreasing function of the target cell's load index x. That is, the larger the target cell's load index x (i.e., the greater the load level of the target cell), the smaller the operation instruction 'a' for the target cell, indicating that the target cell needs to release more users (or absorb fewer users). Conversely, the smaller the target cell's load index x (i.e., the smaller the load level of the target cell), the larger the operation instruction 'a' for the target cell, indicating that the target cell needs to release fewer users (or absorb more users).

[0219] It should also be understood that the softplus and tanh activation functions are used only for illustrative purposes in this embodiment and do not constitute a limitation on the types of activation functions used in this application. In practical applications, other activation functions with similar properties may also be used.

[0220] 604. The load index of the target cell is processed through the first sub-model to obtain the first operation.

[0221] 605. The load index of the target cell is processed through the second sub-model to obtain the second operation.

[0222] 606. Perform a weighted sum of the first and second operations to obtain the operation instructions.

[0223] After constructing the first submodule and the second submodel, the load index x of the target cell can be input into the first submodel and the second submodel respectively, so that the load index x of the target cell can be processed by the first submodel to obtain the first operation a for the target cell. T The second sub-model processes the load index x of the target cell to obtain the second operation a for the target cell. S Then, the first operation a targeting the cell will be performed. T and the second operation a targeting the cell S Perform a weighted summation to obtain the operation instruction a for the target cell.

[0224] For instructions on how to adjust the load metric x of the target cell using the operation instruction 'a' for the target cell, please refer to [link / reference]. Figure 4 The relevant descriptions of step 404 in the illustrated embodiment will not be repeated here.

[0225] In this embodiment of the application, a first sub-model and a second sub-model are constructed through a teacher-student network architecture. The first sub-model adds expert constraints (i.e., using τ as a threshold to determine whether the target cell absorbs or releases users) to ensure the security of the control policy output by the model. The second sub-model relaxes the expert constraints, learns from data, and outputs the corresponding control policy. The outputs of the two models are weighted to obtain the final policy, thereby balancing the security and efficiency of the control policy.

[0226] The above is a detailed introduction to the model application stage. The following will introduce the model training stage. Figure 8 A schematic flowchart of the model training method provided in the embodiments of this application is shown below. Figure 8 As shown, the method includes:

[0227] 801. Obtain the characteristic information and load index of the first cell. The load index of the first cell is used to indicate the load level of the first cell.

[0228] When it is necessary to train the first model to be trained (i.e., the neural network to be trained), a batch of training samples can be obtained, namely the feature information of the first cell and the load index of the first cell used for training.

[0229] In one possible implementation, the feature information of the first cell includes at least one of the following: configuration parameters of the first cell; call statistics data of the first cell; configuration parameters of the neighboring cells of the first cell; and call statistics data of the neighboring cells.

[0230] In one possible implementation, the configuration parameters include at least one of the following: antenna transmit power; antenna downtilt angle; antenna horizontal azimuth angle; RSRP threshold for user-initiated inter-frequency handover measurement; RSRP threshold for user-stopped inter-frequency handover measurement; RSRP offset for a specific frequency when user-triggered inter-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered inter-frequency handover procedure; RSRP threshold for user-initiated intra-frequency handover measurement; RSRP threshold for user-stopped intra-frequency handover measurement; RSRP offset for a specific frequency when user-triggered intra-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered intra-frequency handover procedure.

[0231] In one possible implementation, the call statistics data includes at least one of the following: the average number of users per unit time period; the average number of active users per unit time period; the uplink traffic per unit time period; the proportion of CQI reports per unit time period; the proportion of data packets with a length less than a preset length threshold per unit time period; and the average length of data packets per unit time period.

[0232] For details regarding the characteristics and load metrics of the first cell, please refer to [link / reference]. Figure 4The relevant descriptions of step 401 in the illustrated embodiment will not be repeated here.

[0233] 802. The feature information of the first cell is processed by the first model to be trained to obtain the model parameters of the second model to be trained.

[0234] 803. Obtain the second model to be trained based on the model parameters of the second model to be trained.

[0235] 804. The load index of the first cell is processed by the second model to be trained to obtain the operation instructions for the first cell. The operation instructions for the first cell are used to adjust the load index of the first cell.

[0236] For explanations of steps 802 to 804, please refer to... Figure 4 The relevant descriptions of steps 402 to 404 in the illustrated embodiment will not be repeated here.

[0237] 805. The operation instructions for the first cell and the operation instructions for the second cell are processed by the third target model to obtain a first score and a second score. The first score is used to evaluate the impact of the operation instructions for the first cell on the load index of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance index of the entire network. The operation instructions for the second cell are used to adjust the load index of the second cell.

[0238] After obtaining the operational instructions for the first cell, operational instructions for the second cell can also be obtained. It should be noted that the operational instructions for the second cell are used to adjust the load indicators of the second cell. The process of obtaining the operational instructions for the second cell is the same as that for the first cell, and will not be repeated here. It is important to note that the first and second cells are different cells within the entire network, and there can be one or more second cells.

[0239] Then, a third target model (i.e., the aforementioned critic model) can be obtained. The third target model is a trained neural network. This third target model contains two parts: a local network and a global network. The feature information of the first cell, the load index of the first cell, and the operational instructions for the first cell can be input into the local network. This local network processes this information to obtain a first score, which is used to evaluate the impact of the operational instructions for the first cell on its load index. Simultaneously, the feature information of the first cell, the operational instructions for the first cell, the feature information of the second cell, and the operational instructions for the second cell can be input into the global network. This global network processes this information to obtain a second score, which is used to evaluate the impact of the operational instructions for the first cell on the performance indicators of the entire network (e.g., rate, throughput, edge user ratio, etc.).

[0240] 806. Based on the first score and the second score, update the model parameters of the first model to be trained until the model training conditions are met, and obtain the first target model.

[0241] Based on the first and second scores, the model parameters of the first model to be trained are updated, and the updated model is trained using the next batch of training samples (i.e., steps 802 to 806 are re-executed) until the model training conditions are met, and the model is obtained. Figure 4 The first target model in the illustrated embodiment is equivalent to obtaining Figure 4 The second target model in the illustrated embodiment.

[0242] It's worth noting that the model training condition can be: stop training when the weighted sum of the first and second scores decreases to 80% of the peak value (or any other percentage, as specified here). Alternatively, the model training condition can be: stop training when the first score decreases to 80% of the peak value (or any other percentage, as specified here). Of course, other similar conditions are also possible.

[0243] In this embodiment of the application, when using the critic model (i.e., the third target model) to train the actor model (i.e., the whole composed of the first target model and the second target model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the performance of balancing local and global aspects, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the entire network, and realize the coordination between multiple cells.

[0244] Figure 9Another schematic diagram of the model training method provided in the embodiments of this application is shown below. Figure 9 As shown, the method includes:

[0245] 901. Obtain the characteristic information and load index of the first cell. The load index of the first cell is used to indicate the load level of the first cell.

[0246] For an explanation of step 901, please refer to... Figure 8 The relevant descriptions of step 801 in the illustrated embodiment will not be repeated here.

[0247] 902. The feature information of the first cell is processed by the first model to be trained to obtain the model parameters of the first sub-model and the model parameters of the second sub-model.

[0248] 903. Obtain the first sub-model based on the model parameters of the first sub-model, and obtain the second sub-model based on the model parameters of the second sub-model.

[0249] In one possible implementation, a first training model and a first sub-model are used to indicate a first functional relationship between the load index of the first cell and the operation instruction for the first cell. The first training model and the second sub-model are used to indicate a second functional relationship between the load index of the first cell and the operation instruction for the first cell. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0250] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0251] 904. The load index of the first cell is processed through the first sub-model to obtain the first operation.

[0252] 905. The load index of the first cell is processed through the second sub-model to obtain the second operation.

[0253] 906. Perform a weighted summation of the first operation and the second operation to obtain the operation instruction for the first cell. The operation instruction for the first cell is used to adjust the load index of the first cell.

[0254] For explanations of steps 902 to 906, please refer to... Figure 6 The relevant descriptions of steps 602 to 606 in the illustrated embodiment will not be repeated here.

[0255] 907. The operation instructions for the first cell and the operation instructions for the second cell are processed by the third target model to obtain a first score and a second score. The first score is used to evaluate the impact of the operation instructions for the first cell on the load index of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance index of the entire network. The operation instructions for the second cell are used to adjust the load index of the second cell.

[0256] 908. Based on the first score and the second score, update the model parameters of the first model to be trained until the model training conditions are met, and obtain the first target model.

[0257] The first target model trained in this embodiment is Figure 6 Similarly, obtaining the first target model in the illustrated embodiment is equivalent to obtaining... Figure 6 The first sub-model and the second sub-model in the illustrated embodiment.

[0258] For explanations of steps 907 to 908, please refer to... Figure 8 The relevant descriptions of steps 805 to 806 in the illustrated embodiment will not be repeated here.

[0259] In this embodiment of the application, when using the critic model (i.e., the third target model) to train the actor model (i.e., the whole composed of the first target model and the second target model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the performance of balancing local and global aspects, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the entire network, and realize the coordination between multiple cells.

[0260] The above is a detailed description of the cell load adjustment method and model training method provided in the embodiments of this application. The following will introduce the cell load adjustment device and model training device provided in the embodiments of this application. Figure 10 A schematic diagram of the structure of the cell load adjustment device provided in the embodiments of this application is shown below. Figure 10 As shown, the device includes:

[0261] The first acquisition module 1001 is used to acquire the characteristic information of the target cell and the load index of the target cell. The load index is used to indicate the load level of the target cell.

[0262] The first processing module 1002 is used to process the feature information through the first target model to obtain the model parameters of the second target model;

[0263] The second acquisition module 1003 is used to acquire the second target model based on the model parameters;

[0264] The second processing module 1004 is used to process the load index through the second target model to obtain operation instructions, which are used to adjust the load index.

[0265] In this embodiment, after obtaining the feature information and load indicators of the target cell, the feature information can be processed using a first target model to obtain the model parameters of a second target model. Then, a second target model is constructed based on the model parameters, and the load indicators are processed using the second target model to obtain an operation instruction for the target cell. This operation instruction can be used to adjust the load indicators of the target cell. In the aforementioned process, the model parameters of the second target model are obtained based on the feature information of the target cell. Since the feature information of the target cell can be used to characterize various features of the target cell (i.e., the scenario in which the target cell is located), when processing the load indicators of the target cell using the second target model, various factors such as the various features of the target cell are considered. Therefore, the operation instruction for the target cell output by the second target model can be used to accurately adjust the load level of the target cell to keep its various performance indicators within a suitable range, providing users with better network services.

[0266] In one possible implementation, the second target model includes a first sub-model and a second sub-model. The second processing module 1004 is used to: process the load index through the first sub-model to obtain a first operation; if the load index is equal to a preset index threshold, the first operation is used to not adjust the load index; process the load index through the second sub-model to obtain a second operation; and perform a weighted summation of the first operation and the second operation to obtain an operation instruction.

[0267] In one possible implementation, the first target model and the first sub-model are used to indicate the first functional relationship between the load index and the first operation, and the first target model and the second sub-model are used to indicate the second functional relationship between the load index and the second operation. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0268] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0269] In one possible implementation, the operation instruction is used to modify the configuration parameters of the target cell to adjust the load index; the characteristic information of the target cell includes at least one of the following: the configuration parameters of the target cell; the call statistics data of the target cell; the configuration parameters of the target cell's neighboring cells; and the call statistics data of the neighboring cells.

[0270] In one possible implementation, the configuration parameters include at least one of the following: antenna transmit power; antenna downtilt angle; antenna horizontal azimuth angle; RSRP threshold for user-initiated inter-frequency handover measurement; RSRP threshold for user-stopped inter-frequency handover measurement; RSRP offset for a specific frequency when user-triggered inter-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered inter-frequency handover procedure; RSRP threshold for user-initiated intra-frequency handover measurement; RSRP threshold for user-stopped intra-frequency handover measurement; RSRP offset for a specific frequency when user-triggered intra-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered intra-frequency handover procedure.

[0271] In one possible implementation, the call statistics data includes at least one of the following: the average number of users per unit time period; the average number of active users per unit time period; the uplink traffic per unit time period; the proportion of CQI reports per unit time period; the proportion of data packets with a length less than a preset length threshold per unit time period; and the average length of data packets per unit time period.

[0272] Figure 11 A schematic diagram of the model training apparatus provided in the embodiments of this application is shown below. Figure 11 As shown, the device includes:

[0273] The first acquisition module 1101 is used to acquire the feature information of the first cell and the load index of the first cell. The load index of the first cell is used to indicate the load level of the first cell.

[0274] The first processing module 1102 is used to process the feature information through the first model to be trained to obtain the model parameters of the second model to be trained.

[0275] The second acquisition module 1103 is used to acquire the second model to be trained based on the model parameters;

[0276] The second processing module 1104 is used to process the load index of the first cell through the second model to be trained to obtain the operation instructions for the first cell, and the operation instructions for the first cell are used to adjust the load index of the first cell.

[0277] The third processing module 1105 is used to process the operation instructions for the first cell and the operation instructions for the second cell through the third target model to obtain a first score and a second score. The first score is used to evaluate the impact of the operation instructions for the first cell on the load index of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance index of the entire network. The operation instructions for the second cell are used to adjust the load index of the second cell. The first cell and the second cell are different cells.

[0278] The update module 1106 is used to update the model parameters of the first model to be trained based on the first score and the second score until the model training conditions are met, and the first target model is obtained.

[0279] In this embodiment of the application, when using the critic model (i.e., the third target model) to train the actor model (i.e., the whole composed of the first target model and the second target model), the training objective of the actor model is to simultaneously optimize the output values ​​of the local network and the global network. The actor model trained in this way has the performance of balancing local and global aspects, that is, it can balance the impact of the load adjustment of a certain cell on that cell and the impact on the entire network, and realize the coordination between multiple cells.

[0280] In one possible implementation, the second model to be trained includes a first sub-model and a second sub-model. The second processing module 1104 is used to: process the load index of the first cell through the first sub-model to obtain a first operation; if the load index of the first cell is equal to a preset index threshold, the first operation is used to not adjust the load index of the first cell; process the load index of the first cell through the second sub-model to obtain a second operation; and perform a weighted summation of the first operation and the second operation to obtain an operation instruction for the first cell.

[0281] In one possible implementation, a first training model and a first sub-model are used to indicate a first functional relationship between the load index of the first cell and a first operation for the first cell, and a first training model and a second sub-model are used to indicate a second functional relationship between the load index of the first cell and a second operation for the first cell. The second functional relationship is obtained by shifting the first functional relationship, and both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

[0282] In one possible implementation, the magnitude of the translation is determined based on feature information.

[0283] In one possible implementation, the operation instruction for the first cell is used to modify the configuration parameters of the first cell to adjust the load index of the first cell; the characteristic information of the first cell includes at least one of the following: the configuration parameters of the first cell; the call statistics data of the first cell; the configuration parameters of the neighboring cells of the first cell; and the call statistics data of the neighboring cells.

[0284] In one possible implementation, the configuration parameters include at least one of the following: antenna transmit power; antenna downtilt angle; antenna horizontal azimuth angle; RSRP threshold for user-initiated inter-frequency handover measurement; RSRP threshold for user-stopped inter-frequency handover measurement; RSRP offset for a specific frequency when user-triggered inter-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered inter-frequency handover procedure; RSRP threshold for user-initiated intra-frequency handover measurement; RSRP threshold for user-stopped intra-frequency handover measurement; RSRP offset for a specific frequency when user-triggered intra-frequency handover procedure; RSRP offset for a specific neighboring cell when user-triggered intra-frequency handover procedure.

[0285] In one possible implementation, the call statistics data includes at least one of the following: the average number of users per unit time period; the average number of active users per unit time period; the uplink traffic per unit time period; the proportion of CQI reports per unit time period; the proportion of data packets with a length less than a preset length threshold per unit time period; and the average length of data packets per unit time period.

[0286] It should be noted that the information interaction and execution process between the modules / units of the above-mentioned device are based on the same concept as the method embodiment of this application, and the resulting technical effects are the same as those of the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in the embodiment of this application, and it will not be repeated here.

[0287] This application also relates to an execution device. Figure 12 This is a schematic diagram of the execution device provided in an embodiment of this application. Figure 12 As shown, the execution device 1200 can specifically be a mobile phone, tablet, laptop, smart wearable device, server, etc., and is not limited here. Among them, the execution device 1200 can be deployed with... Figure 10 The cell load adjustment device described in the corresponding embodiment is used to implement Figure 4 or Figure 6 This corresponds to the cell load adjustment function in the embodiment. Specifically, the execution device 1200 includes: a receiver 1201, a transmitter 1202, a processor 1203, and a memory 1204 (wherein the execution device 1200 may have one or more processors 1203). Figure 12 (Taking a processor as an example), the processor 1203 may include an application processor 12031 and a communication processor 12032. In some embodiments of this application, the receiver 1201, transmitter 1202, processor 1203, and memory 1204 may be connected via a bus or other means.

[0288] Memory 1204 may include read-only memory and random access memory, and provides instructions and data to processor 1203. A portion of memory 1204 may also include non-volatile random access memory (NVRAM). Memory 1204 stores processor and operation instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof, wherein the operation instructions may include various operation instructions for implementing various operations.

[0289] Processor 1203 controls the operation of the execution device. In specific applications, the various components of the execution device are coupled together through a bus system, which may include not only the data bus, but also power buses, control buses, and status signal buses. However, for clarity, all buses in the diagram are referred to as the bus system.

[0290] The methods disclosed in the embodiments of this application can be applied to or implemented by the processor 1203. The processor 1203 can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 1203 or by instructions in software form. The processor 1203 can be a general-purpose processor, a digital signal processor (DSP), a microprocessor, or a microcontroller, and may further include an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The processor 1203 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 1204. Processor 1203 reads the information in memory 1204 and, in conjunction with its hardware, completes the steps of the above method.

[0291] Receiver 1201 can be used to receive input digital or character information, and to generate signal inputs related to the settings and function control of the execution device. Transmitter 1202 can be used to output digital or character information through the first interface; transmitter 1202 can also be used to send instructions to the disk group through the first interface to modify the data in the disk group; transmitter 1202 may also include a display device such as a display screen.

[0292] In one embodiment of this application, the processor 1203 is used to... Figure 4 or Figure 8 In the first target model of the corresponding embodiment, the information of the target cell is processed to adjust the load level of the target cell.

[0293] This application also relates to a training device. Figure 13 This is a schematic diagram of the structure of a training device provided in an embodiment of this application. Figure 13 As shown, the training device 1300 is implemented by one or more servers. The training device 1300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 1314 (e.g., one or more processors) and memory 1332, and one or more storage media 1330 (e.g., one or more mass storage devices) for storing application programs 1342 or data 1344. The memory 1332 and storage media 1330 can be temporary or persistent storage. The program stored in the storage media 1330 may include one or more modules (not shown in the figure), each module may include a series of instruction operations on the training device. Furthermore, the CPU 1314 may be configured to communicate with the storage media 1330 and execute the series of instruction operations in the storage media 1330 on the training device 1300.

[0294] The training device 1300 may also include one or more power supplies 1326, one or more wired or wireless network interfaces 1350, one or more input / output interfaces 1358; or, one or more operating systems 1341, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0295] Specifically, the training equipment can perform Figure 8 or Figure 9 The model training method in the corresponding embodiment.

[0296] This application also relates to a computer storage medium storing a program for signal processing, which, when run on a computer, causes the computer to perform steps as performed by the aforementioned execution device, or causes the computer to perform steps as performed by the aforementioned training device.

[0297] This application also relates to a computer program product that stores instructions that, when executed by a computer, cause the computer to perform steps as performed by the aforementioned execution device, or to perform steps as performed by the aforementioned training device.

[0298] The execution device, training device, or terminal device provided in this application embodiment can specifically be a chip. The chip includes a processing unit and a communication unit. The processing unit can be, for example, a processor, and the communication unit can be, for example, an input / output interface, pins, or circuits. The processing unit can execute computer execution instructions stored in the storage unit to cause the chip within the execution device to execute the data processing method described in the above embodiments, or to cause the chip within the training device to execute the data processing method described in the above embodiments. Optionally, the storage unit can be a storage unit within the chip, such as a register or cache. Alternatively, the storage unit can be a storage unit located outside the chip within the wireless access device, such as a read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, such as random access memory (RAM).

[0299] For details, please refer to Figure 14 , Figure 14 This is a schematic diagram of the chip provided in an embodiment of this application. The chip can be represented as a neural network processor (NPU) 1400. The NPU 1400 is mounted as a coprocessor on the host CPU, and tasks are assigned by the host CPU. The core part of the NPU is the arithmetic circuit 1403, which is controlled by the controller 1404 to extract matrix data from the memory and perform multiplication operations.

[0300] In some implementations, the arithmetic circuit 1403 internally includes multiple processing engines (PEs). In some implementations, the arithmetic circuit 1403 is a two-dimensional pulsating array. The arithmetic circuit 1403 can also be a one-dimensional pulsating array or other electronic circuitry capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuit 1403 is a general-purpose matrix processor.

[0301] For example, suppose we have an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit retrieves the corresponding data of matrix B from the weight memory 1402 and caches it in each PE of the arithmetic circuit. The arithmetic circuit retrieves the data of matrix A from the input memory 1401 and performs matrix operations with matrix B. The partial result or the final result of the obtained matrix is ​​stored in the accumulator 1408.

[0302] Unified memory 1406 is used to store input and output data. Weight data is directly transferred to weight memory 1402 via Direct Memory Access Controller (DMAC) 1405. Input data is also transferred to unified memory 1406 via DMAC.

[0303] BIU stands for Bus Interface Unit, which is used for interaction between the AXI bus and the DMAC and the Instruction Fetch Buffer (IFB) 1409.

[0304] The Bus Interface Unit (BIU) 1413 is used by the instruction fetch memory 1409 to fetch instructions from external memory, and also by the memory access controller 1405 to fetch the original data of the input matrix A or the weight matrix B from external memory.

[0305] The DMAC is mainly used to move input data from external memory DDR to unified memory 1406, or to weight data to weight memory 1402, or to input data to input memory 1401.

[0306] The vector computation unit 1407 includes multiple processing units that further process the output of the computation circuit 1403 when necessary, such as vector multiplication, vector addition, exponential operations, logarithmic operations, size comparisons, etc. It is mainly used for computation in non-convolutional / fully connected layers of neural networks, such as Batch Normalization, pixel-level summation, and upsampling of the predicted label plane.

[0307] In some implementations, the vector computation unit 1407 can store the processed output vector in the unified memory 1406. For example, the vector computation unit 1407 can apply a linear function, or a nonlinear function, to the output of the computation circuit 1403, such as linearly interpolating the predicted label plane extracted from the convolutional layer, or, for example, accumulating a vector of values ​​to generate activation values. In some implementations, the vector computation unit 1407 generates normalized values, pixel-level summed values, or both. In some implementations, the processed output vector can be used as activation input to the computation circuit 1403, for example, for use in subsequent layers of the neural network.

[0308] The instruction fetch buffer 1409 connected to the controller 1404 is used to store the instructions used by the controller 1404;

[0309] Unified memory 1406, input memory 1401, weighted memory 1402, and instruction fetch memory 1409 are all on-chip memories. External memory is proprietary to this NPU hardware architecture.

[0310] The processor mentioned above can be a general-purpose central processing unit, a microprocessor, an ASIC, or one or more integrated circuits used to control the execution of the above program.

[0311] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0312] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0313] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0314] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for adjusting the load of a residential cell, characterized in that, The method includes: Obtain the characteristic information of the target cell and the load index of the target cell, wherein the load index is used to indicate the load level of the target cell; The feature information is processed by the first target model to obtain the model parameters of the second target model; Obtain the second target model based on the model parameters; The load index is processed by the second target model to obtain an operation instruction, which is used to adjust the load index. The second target model includes a first sub-model and a second sub-model. The step of processing the load index using the second target model to obtain operation instructions includes: The load index is processed by the first sub-model to obtain the first operation. If the load index is equal to a preset index threshold, the first operation is used to not adjust the load index. The load index is processed by the second sub-model to obtain the second operation; The first operation and the second operation are weighted and summed to obtain an operation instruction. The first target model and the first sub-model are used to indicate the first functional relationship between the load index and the first operation. The first target model and the second sub-model are used to indicate the second functional relationship between the load index and the second operation. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

2. The method according to claim 1, characterized in that, The magnitude of the translation is determined based on the feature information.

3. The method according to claim 1 or 2, characterized in that, The operation instructions are used to modify the configuration parameters of the target cell to adjust the load index; the characteristic information of the target cell includes at least one of the following: The configuration parameters of the target cell; The call statistics data of the target cell; Configuration parameters of neighboring cells of the target cell; The call statistics data of the neighboring community.

4. The method according to claim 3, characterized in that, The configuration parameters include at least one of the following: Antenna transmit power; Antenna downtilt angle; Antenna horizontal azimuth angle; The reference signal received power (RSRP) threshold for user-initiated inter-frequency handover measurement; The RSRP threshold for user-stopped inter-frequency switching measurements; The specific frequency RSRP bias that the user triggers the frequency switching process; Specific neighbor cell RSRP bias for user-triggered inter-frequency handover process; The RSRP threshold for user-initiated same-frequency switching measurement; The RSRP threshold for user-stopped same-frequency switching measurements; The specific frequency RSRP bias that the user triggers the same-frequency switching process; The specific neighbor cell RSRP bias that the user triggers the same frequency handover process.

5. The method according to claim 3, characterized in that, The call statistics data includes at least one of the following: Average number of users per unit time period; Average number of active users per unit time period; Uplink traffic per unit time period; The proportion of low channel quality indication (CQI) reports per unit time period; The proportion of data packets whose length is less than a preset length threshold within a unit time period; The average length of data packets per unit time period.

6. A model training method, characterized in that, The method includes: Obtain the characteristic information and load index of the first cell. The load index of the first cell is used to indicate the load level of the first cell. The feature information is processed by the first model to be trained to obtain the model parameters of the second model to be trained. A second model to be trained is obtained based on the model parameters; The load index of the first cell is processed by the second model to be trained to obtain an operation instruction for the first cell, which is used to adjust the load index of the first cell. The operation instructions for the first cell and the operation instructions for the second cell are processed by the third target model to obtain a first score and a second score. The first score is used to evaluate the impact of the operation instructions for the first cell on the load index of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance index of the entire network. The operation instructions for the second cell are used to adjust the load index of the second cell. The first cell and the second cell are different cells. Based on the first score and the second score, the model parameters of the first model to be trained are updated until the model training conditions are met, and the first target model is obtained. The second model to be trained includes a first sub-model and a second sub-model. The step of processing the load indicators of the first cell using the second model to obtain operational instructions for the first cell includes: The load index of the first cell is processed by the first sub-model to obtain the first operation. If the load index of the first cell is equal to the preset index threshold, the first operation is used to not adjust the load index of the first cell. The load index of the first cell is processed by the second sub-model to obtain the second operation; The first operation and the second operation are weighted and summed to obtain the operation instruction for the first cell. The first model to be trained and the first sub-model are used to indicate the first functional relationship between the load index of the first cell and the first operation for the first cell. The first model to be trained and the second sub-model are used to indicate the second functional relationship between the load index of the first cell and the second operation for the first cell. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

7. The method according to claim 6, characterized in that, The magnitude of the translation is determined based on the feature information.

8. The method according to claim 6 or 7, characterized in that, The operation instruction for the first cell is used to modify the configuration parameters of the first cell to adjust the load index of the first cell; the characteristic information of the first cell includes at least one of the following: Configuration parameters of the first cell; Call statistics data for the first cell; Configuration parameters of the neighboring cells of the first cell; The call statistics data of the neighboring community.

9. The method according to claim 8, characterized in that, The configuration parameters include at least one of the following: Antenna transmit power; Antenna downtilt angle; Antenna horizontal azimuth angle; The RSRP threshold for user-initiated inter-frequency handover measurement; The RSRP threshold for user-stopped inter-frequency switching measurements; The specific frequency RSRP bias that the user triggers the frequency switching process; Specific neighbor cell RSRP bias for user-triggered inter-frequency handover process; The RSRP threshold for user-initiated same-frequency switching measurement; The RSRP threshold for user-stopped same-frequency switching measurements; The specific frequency RSRP bias that the user triggers the same-frequency switching process; The specific neighbor cell RSRP bias that the user triggers the same frequency handover process.

10. The method according to claim 8, characterized in that, The call statistics data includes at least one of the following: Average number of users per unit time period; Average number of active users per unit time period; Uplink traffic per unit time period; The proportion of CQI reports per unit time period; The proportion of data packets whose length is less than a preset length threshold within a unit time period; The average length of data packets per unit time period.

11. A device for adjusting the load of a residential area, characterized in that, The device includes: The first acquisition module is used to acquire the feature information of the target cell and the load index of the target cell, wherein the load index is used to indicate the load level of the target cell; The first processing module is used to process the feature information through the first target model to obtain the model parameters of the second target model; The second acquisition module is used to acquire a second target model based on the model parameters; The second processing module is used to process the load index through the second target model to obtain an operation instruction, which is used to adjust the load index. The second target model includes a first sub-model and a second sub-model. The second processing module is used for: The load index is processed by the first sub-model to obtain the first operation. If the load index is equal to a preset index threshold, the first operation is used to not adjust the load index. The load index is processed by the second sub-model to obtain the second operation; The first operation and the second operation are weighted and summed to obtain an operation instruction. The first target model and the first sub-model are used to indicate the first functional relationship between the load index and the first operation. The first target model and the second sub-model are used to indicate the second functional relationship between the load index and the second operation. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

12. A model training device, characterized in that, The device includes: The first acquisition module is used to acquire the feature information of the first cell and the load index of the first cell, wherein the load index of the first cell is used to indicate the load level of the first cell. The first processing module is used to process the feature information through the first model to be trained to obtain the model parameters of the second model to be trained. The second acquisition module is used to acquire a second model to be trained based on the model parameters; The second processing module is used to process the load index of the first cell through the second model to be trained to obtain an operation instruction for the first cell, wherein the operation instruction for the first cell is used to adjust the load index of the first cell. The third processing module is used to process the operation instructions for the first cell and the operation instructions for the second cell through the third target model to obtain a first score and a second score. The first score is used to evaluate the impact of the operation instructions for the first cell on the load index of the first cell, and the second score is used to evaluate the impact of the operation instructions for the first cell on the performance index of the entire network. The operation instructions for the second cell are used to adjust the load index of the second cell. The first cell and the second cell are different cells. The update module is used to update the model parameters of the first model to be trained based on the first score and the second score until the model training conditions are met, and the first target model is obtained. The second model to be trained includes a first sub-model and a second sub-model. The second processing module is used for: The load index of the first cell is processed by the first sub-model to obtain the first operation. If the load index of the first cell is equal to the preset index threshold, the first operation is used to not adjust the load index of the first cell. The load index of the first cell is processed by the second sub-model to obtain the second operation; The first operation and the second operation are weighted and summed to obtain the operation instruction for the first cell. The first model to be trained and the first sub-model are used to indicate the first functional relationship between the load index of the first cell and the first operation for the first cell. The first model to be trained and the second sub-model are used to indicate the second functional relationship between the load index of the first cell and the second operation for the first cell. The second functional relationship is obtained by shifting the first functional relationship. Both the first functional relationship and the second functional relationship are monotonically increasing or monotonically decreasing.

13. A device for adjusting the load of a residential area, characterized in that, The device includes a memory and a processor; the memory stores code, and the processor is configured to execute the code, wherein when the code is executed, the cell load adjustment device performs the method as described in any one of claims 1 to 10.

14. A computer storage medium, characterized in that, The computer storage medium stores one or more instructions that, when executed by one or more computers, cause the one or more computers to perform the method of any one of claims 1 to 10.

15. A computer program product, characterized in that, The computer program product stores instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1 to 10.

Images