Prediction model training method, system, device, computer equipment and storage medium

Abstract

The present application relates to a prediction model training method, system, device, computer equipment and storage medium. The method includes: obtaining the prediction model of the target base station group as the target prediction model of each base station; for any base station, determining the first energy-saving control loss value of the target prediction model, and adjusting the target model parameters of the target prediction model based on the first energy-saving control loss value; determining the second energy-saving control loss value of the adjusted prediction model, and adjusting the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, using the adjusted prediction model as the target prediction model corresponding to each base station, and jumping to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until the preset conditions are met; respectively using the current target prediction model of each base station as the trained prediction model of each base station. The use of this method can improve the overall training speed of the prediction model of each base station.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular to a prediction model training method, system, device, computer equipment and storage medium. Background Art

[0002] With the rapid deployment of 5G in recent years, base station electricity costs have accounted for nearly half of operators' costs. With many global operators now aiming for carbon neutrality, implementing energy-saving control at base stations and reducing their power consumption have become increasingly pressing challenges.

[0003] In existing technology, mathematical models are typically used to predict the optimal energy-saving control parameters for each base station. However, given the relatively small number of samples accumulated by a base station, training a model for each base station can result in slow or even non-convergence of the model for some base stations, resulting in a longer total training time for each base station model. Summary of the Invention

[0004] Based on this, it is necessary to provide a prediction model training method, system, device, computer equipment and storage medium to address the above technical problems.

[0005] In a first aspect, the present application provides a prediction model training method. The method comprises:

[0006] Obtaining a prediction model corresponding to the target base station group as a target prediction model corresponding to each of the base stations;

[0007] For any of the base stations in the target base station group, determining a current first energy-saving control loss value of the target prediction model, and adjusting a current target model parameter of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize an energy-saving effect of an energy-saving operation performed by the base station based on the target prediction model;

[0008] Determine the adjusted second energy-saving control loss value of the prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until a preset condition is met;

[0009] The current target prediction model of each base station is respectively used as the trained prediction model of each base station.

[0010] In one embodiment, determining the current first energy-saving control loss value of the target prediction model includes:

[0011] Acquiring current operating parameters of the base station, and predicting energy-saving control parameters of the base station based on the operating parameters and the target prediction model;

[0012] Sending the energy-saving control parameter to the base station, and receiving a feedback parameter returned by the base station;

[0013] An actual reward value is determined according to the reward parameter, and a current first energy-saving control loss value of the target prediction model is determined according to the actual reward value and the target reward value corresponding to the base station.

[0014] In one embodiment, adjusting the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station and using the adjusted prediction model as the target prediction model corresponding to each base station includes:

[0015] determining a total energy-saving control loss value corresponding to the prediction model according to the second energy-saving control loss value of each base station;

[0016] When the total energy-saving control loss value is greater than the loss value threshold, the initial model parameters are adjusted according to the total energy-saving control loss value and the adjustment step, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

[0017] In one embodiment, the method further comprises:

[0018] When the total energy-saving control loss value is less than or equal to the loss value threshold, for any of the base stations, the current target model parameters of the target prediction model corresponding to the base station are adjusted according to the second energy-saving control loss value corresponding to the base station, and the adjusted prediction model is used as the target prediction model corresponding to each of the base stations.

[0019] In one embodiment, before obtaining the prediction model corresponding to the target base station group, the method further includes:

[0020] performing clustering processing on each of the base stations according to the operating parameters of each of the base stations to obtain at least one clustering result;

[0021] For any of the clustering results, when the number of the base stations in the clustering result is greater than a preset number, the base stations in the clustering result are regarded as a base station group.

[0022] In one embodiment, the method further comprises:

[0023] In a case where a base station to be activated is detected, obtaining the operating parameters of the base station to be activated, and determining a first base station group to which the base station to be activated belongs based on the operating parameters of the base station to be activated;

[0024] The base station to be activated is added to the first base station group, and the current prediction model of the first base station group is used as the target prediction model corresponding to the base station to be activated.

[0025] In a second aspect, the present application further provides a prediction model training system, the system comprising a server and at least one base station group, the base station group comprising at least one base station, wherein:

[0026] The server is configured to obtain a prediction model corresponding to a target base station group as the target prediction model corresponding to each base station, obtain current operating parameters of each base station, and predict, for any base station, an energy-saving control parameter of the base station based on the operating parameters and the target prediction model, and send the energy-saving control parameter to the base station;

[0027] The base station is configured to receive the energy-saving control parameter, operate according to the energy-saving control parameter, obtain a report parameter corresponding to the energy-saving control parameter, and send the report parameter to the server;

[0028] The server is further configured to determine, based on each of the feedback parameters, a current first energy-saving control loss value of the target prediction model of each of the base stations, and adjust the current target model parameters of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to represent the energy-saving effect of the energy-saving operation performed by the base station based on the target prediction model;

[0029] The server is also used to determine the second energy-saving control loss value of the adjusted prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of predicting the energy-saving control parameters of the base station based on the operating parameters and the target prediction model, and sending the energy-saving control parameters to the base station until the preset conditions are met, and use the current target prediction model of each base station as the trained prediction model of each base station.

[0030] In a third aspect, the present application also provides a prediction model training device. The device comprises:

[0031] A first acquisition module is used to acquire a prediction model corresponding to a target base station group as a target prediction model corresponding to each base station;

[0032] a first adjustment module, configured to determine, for any base station in the target base station group, a current first energy-saving control loss value of the target prediction model, and adjust a current target model parameter of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize an energy-saving effect of an energy-saving operation performed by the base station based on the target prediction model;

[0033] a second adjustment module, configured to determine a second energy-saving control loss value of the adjusted prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until a preset condition is met;

[0034] The first processing module is used to use the current target prediction model of each base station as the trained prediction model of each base station.

[0035] In one embodiment, the first adjustment module is further configured to:

[0036] Acquiring current operating parameters of the base station, and predicting energy-saving control parameters of the base station based on the operating parameters and the target prediction model;

[0037] Sending the energy-saving control parameter to the base station, and receiving a feedback parameter returned by the base station;

[0038] An actual reward value is determined according to the reward parameter, and a current first energy-saving control loss value of the target prediction model is determined according to the actual reward value and the target reward value corresponding to the base station.

[0039] In one embodiment, the second adjustment module is further configured to:

[0040] determining a total energy-saving control loss value corresponding to the prediction model according to the second energy-saving control loss value of each base station;

[0041] When the total energy-saving control loss value is greater than the loss value threshold, the initial model parameters are adjusted according to the total energy-saving control loss value and the adjustment step, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

[0042] In one embodiment, the apparatus further comprises:

[0043] The third adjustment module is used to adjust the current target model parameters of the target prediction model corresponding to any of the base stations according to the second energy-saving control loss value corresponding to the base station when the total energy-saving control loss value is less than or equal to the loss value threshold, and use the adjusted prediction model as the target prediction model corresponding to each of the base stations.

[0044] In one embodiment, the apparatus further comprises:

[0045] a clustering module, configured to perform clustering processing on each of the base stations according to an operating parameter of each of the base stations to obtain at least one clustering result;

[0046] The second processing module is configured to, for any of the clustering results, treat the base stations in the clustering result as a base station group when the number of the base stations in the clustering result is greater than a preset number.

[0047] In one embodiment, the apparatus further comprises:

[0048] A second acquisition module is configured to, when a base station to be activated is detected, acquire the operating parameters of the base station to be activated, and determine the first base station group to which the base station to be activated belongs based on the operating parameters of the base station to be activated;

[0049] An adding module is used to add the base station to be activated to the first base station group, and use the current prediction model of the first base station group as the target prediction model corresponding to the base station to be activated.

[0050] In a fourth aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements any of the above methods when executing the computer program.

[0051] In a fifth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements any of the above methods when executed by a processor.

[0052] In a sixth aspect, the present application further provides a computer program product, comprising a computer program, which implements any of the above methods when executed by a processor.

[0053] The above-mentioned prediction model training method, system, device, computer equipment and storage medium divide the base station into multiple base station groups, each base station group corresponds to a prediction model, and during the training process, the prediction model is used as the target prediction model of each base station. First, the target model parameters of each base station are adjusted according to the first energy-saving control loss value for each base station, and then the second energy-saving control loss value is obtained according to the adjusted target model parameters, and the initial model parameters of the prediction model are adjusted accordingly, and the above process is repeated until the preset conditions are met. The embodiment of the present application adjusts the initial model parameters according to the effect of adjusting the initial model parameters of the prediction model (that is, the second energy-saving control loss value) to obtain the optimal initial model parameters for each base station in the base station group. When the final model of each base station is obtained by continuing to train according to the initial model parameters, the training time of each base station model can be made relatively average, and the overall training speed of each base station model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of a flow chart of a prediction model training method in one embodiment;

[0055] Figure 2 104 is a flow chart of step 104 in one embodiment;

[0056] Figure 3 106 is a flow chart of step 106 in one embodiment;

[0057] Figure 4 A schematic diagram of a single training target prediction model in one embodiment;

[0058] Figure 5 Schematic diagram of a flow chart of a prediction model training method in one embodiment;

[0059] Figure 6 Schematic diagram of a flow chart of a prediction model training method in one embodiment;

[0060] Figure 7 A schematic diagram of adding a base station to be activated in one embodiment;

[0061] Figure 8 is a structural block diagram of a prediction model training device in one embodiment;

[0062] Figure 9 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0064] In one embodiment, Figure 1 As shown, a prediction model training method is provided. This embodiment is described by applying the method to a server, and includes the following steps:

[0065] Step 102: Obtain the prediction model corresponding to the target base station group as the target prediction model corresponding to each base station.

[0066] In an embodiment of the present application, each base station in a region (such as a province or a city) can be divided into at least one base station group, and a base station group includes at least one base station. The base stations can be grouped according to the location of each base station, the relevant parameters of each base station when it is working, etc. For example, multiple base stations that are close to each other can be divided into a base station group, or multiple base stations with similar locations can be divided into a base station group (for example, base stations located along the railway are a base station group, base stations located in urban areas are a base station group, etc.), or multiple base stations with similar traffic within the base station can be divided into a base station group, etc., and this embodiment of the present application does not specifically limit this.

[0067] A base station group corresponds to a prediction model, and the prediction model has initial model parameters. The initial model parameters can be set randomly, or can also be determined by those skilled in the art based on experience, and the present application embodiment does not specifically limit this. The prediction model can be any input for the current operating parameters of the base station (such as current time, current uplink and downlink traffic, current number of users, etc.), and output is a model for controlling the energy-saving control parameters for the base station to perform energy-saving operations (such as the time when the base station performs symbol shutdown, carrier shutdown, channel shutdown and other operations, or it can also be an operating parameter threshold for causing the base station to perform energy-saving operations when the operating parameter is less than the operating parameter threshold), which can be a model for directly outputting energy-saving control parameters, or it can be a model for predicting the reward value of each group of energy-saving control parameters, and outputting the energy-saving control parameter with the highest reward value and its corresponding reward value, and the present application embodiment does not specifically limit this.

[0068] The prediction model corresponding to the base station group can be used as the target prediction model for each base station in the base station group. In other words, the parameters of each target prediction model are initial model parameters. Subsequently, the energy-saving control parameters can be predicted based on each target prediction model to obtain the first energy-saving control loss value corresponding to the target prediction model.

[0069] Step 104: For any base station in the target base station group, determine the current first energy-saving control loss value of the target prediction model, and adjust the current target model parameters of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize the energy-saving effect of the energy-saving operation performed by the base station based on the target prediction model.

[0070] In an embodiment of the present application, for any base station, the operating parameters of the base station can be obtained, and the energy-saving control parameters of the base station can be predicted by the target prediction model corresponding to the base station, and the first energy-saving control loss value can be determined based on the energy-saving control parameters. If the prediction model corresponding to the base station is trained by training samples, the first energy-saving control loss value of the target prediction model can be calculated based on the predicted energy-saving control parameters and the actual energy-saving control parameters in the sample. If no training samples are set, but the first energy-saving control loss value is calculated based on the actual performance of the energy-saving control parameters when applied, the server can send the energy-saving control parameters to the corresponding base station, so that the base station performs energy-saving control according to the energy-saving control parameters, and at the same time collects the return parameters corresponding to the energy-saving control parameters (the return parameters represent the effect of energy-saving control according to the energy-saving control parameters. It can include the operating parameters of the base station after energy-saving control, such as power consumption, network indicators, etc., and can also include indicators such as user satisfaction collected from users), and then calculate the first energy-saving control loss value of the target prediction model based on the return parameters and the target return parameters. The embodiment of the present application does not specifically limit the method for calculating the first energy-saving control loss value. Any method for calculating the loss value based on the loss function is applicable to the embodiment of the present application, for example, calculating the first energy-saving control loss value based on the absolute value loss function, calculating the first energy-saving control loss value based on the mean square error function, etc.

[0071] After obtaining the first energy-saving control loss value corresponding to each base station, the current parameters of the target prediction model (i.e., the target model parameters) can be adjusted according to the first energy-saving control loss value. This application does not specifically limit the method for adjusting the target model parameters. Any method for adjusting the model parameters based on the loss is applicable to the embodiments of this application, such as grid search, gradient descent, etc.

[0072] Step 106, determine the second energy-saving control loss value of the adjusted prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until the preset conditions are met.

[0073] In the embodiment of the present application, after adjusting the prediction model, the new energy-saving control parameters of the base station can be predicted using the adjusted prediction model and the operating parameters of the base station, and a second energy-saving control loss value can be determined based on the new energy-saving control parameters. The method for determining the second energy-saving control loss value can refer to the method for determining the first energy-saving control loss value, and the embodiment of the present application will not be repeated here.

[0074] Since the second energy-saving control loss value is obtained based on the adjusted target prediction model and represents the model performance of the adjusted target prediction model, and the target prediction model is adjusted based on the initial model parameters, the second energy-saving control loss value can represent the effect of adjusting the prediction model based on the initial model parameters, that is, the performance of the initial model parameters. Therefore, the initial model parameters can be adjusted based on the second energy-saving control loss value. The embodiment of the present application does not specifically limit the method for adjusting the initial model parameters based on the second energy-saving control loss value. Any method of adjusting the model parameters based on the loss is applicable to the embodiment of the present application.

[0075] After adjusting the initial model parameters of the prediction model, the process of using the prediction model as the target prediction model for each base station, adjusting the target prediction model to determine the performance of the initial model parameters, and adjusting the initial model parameters can be repeated until the preset conditions are met. The preset conditions can be that the first energy-saving control loss value of each base station is less than a preset threshold, the second energy-saving control loss value of each base station is less than a preset threshold, or the number of times the prediction model has been adjusted has reached a preset number. It can be set by those skilled in the art according to actual needs, and the embodiments of the present application do not specifically limit this.

[0076] Step 108: The current target prediction model of each base station is used as the trained prediction model of each base station.

[0077] In an embodiment of the present application, after the preset conditions are met, the current target prediction model of each base station can be used as the prediction model after training of each base station. Depending on the preset conditions, the effect of the training may also be different. For example, if the preset condition is that the first energy-saving control loss value of each base station is less than the preset threshold, the prediction model after training of each base station will be a different prediction model, which is equivalent to training a different prediction model for each base station. If the preset condition is that the second energy-saving control loss value of each base station is less than the preset threshold, the prediction model after training of each base station will be the same prediction model, which is equivalent to training a prediction model for a base station group that can handle all base stations in the base station group.

[0078] After obtaining the trained prediction model for each base station, you can continue to train each prediction model individually. In this case, there is no need to adjust the initial model parameters. Instead, you can follow the normal model training process to determine the loss of each prediction model and adjust the parameters of each prediction model individually to further improve the accuracy of the obtained prediction model.

[0079] The prediction model training method provided in the embodiment of the present application divides the base station into multiple base station groups, each base station group corresponds to a prediction model, and during the training process, the prediction model is used as the target prediction model of each base station. First, the target model parameters of each base station are adjusted according to the first energy-saving control loss value, and then the second energy-saving control loss value is obtained according to the adjusted target model parameters, and the initial model parameters of the prediction model are adjusted accordingly, and the above process is repeated until the preset conditions are met. The embodiment of the present application adjusts the initial model parameters according to the effect of adjusting the initial model parameters of the prediction model (that is, the second energy-saving control loss value) to obtain the optimal initial model parameters for each base station in the base station group. When the final model of each base station is obtained by continuing to train according to the initial model parameters, the training time of each base station model can be made relatively average, which can improve the overall training speed of each base station model.

[0080] In one embodiment, Figure 2 As shown, in step 104, determining the current first energy-saving control loss value of the target prediction model includes:

[0081] Step 202: Acquire the current operating parameters of the base station, and predict the energy-saving control parameters of the base station based on the operating parameters and the target prediction model.

[0082] Step 204: Send the energy-saving control parameter to the base station, and receive the feedback parameter returned by the base station.

[0083] Step 206 : determining an actual reward value according to the reward parameter, and determining a current first energy-saving control loss value of the target prediction model according to the actual reward value and the target reward value corresponding to the base station.

[0084] In embodiments of the present application, the prediction model can be trained while being applied. The server can obtain the operating parameters returned by each base station to the server by sending an operating parameter acquisition request to each base station. The operating parameters can be parameters obtained in real time by the base station upon receiving the operating parameter acquisition request, or they can be operating parameters pre-stored by the base station for a period of time. The server predicts the energy-saving control parameters corresponding to each base station based on the target prediction model and operating parameters of each base station, and sends the energy-saving control parameters to the corresponding base station.

[0085] The base station performs an energy-saving operation for a preset duration according to the energy-saving control parameters (the preset duration can be several hours, a day, etc.), and at the same time collects the return parameters corresponding to the energy-saving control parameters. The return parameters include parameters that can characterize the effect of the energy-saving operation, such as power consumption, network indicators, etc., and the specific parameter types included can be selected by those skilled in the art according to actual needs. After the preset duration ends, the base station sends the return parameters obtained within the preset duration to the server. Based on the return parameters, the server can calculate the actual return value corresponding to the energy-saving control parameter (for example, the difference between each return parameter and the normal operating parameters of the base station when no energy-saving control is performed can be calculated, and the difference can be summed or weighted to obtain the actual return value). According to the actual return value and the target return value pre-set for each base station, the current first energy-saving control loss value of the target prediction model of the base station can be obtained. The embodiment of the present application does not specifically limit the method of calculating the first energy-saving control loss value based on the actual return value and the target return value. Any loss function that calculates the loss using the actual return value and the target return value is applicable to the embodiment of the present application.

[0086] It should be noted that the second energy-saving control loss value of the target prediction model can also be calculated in the above manner, which will not be described in detail in the embodiment of the present application.

[0087] The prediction model training method provided in the embodiment of the present application predicts energy-saving control parameters based on the current operating parameters of the base station, and calculates the actual return value of the base station based on the reward parameters obtained by the base station executing the energy-saving control parameters, and then obtains the first energy-saving control loss value based on the actual reward value and the target reward value. Therefore, the prediction model can be trained while applying the prediction model, without the need to manually construct a training set, thereby reducing the application requirements of the prediction model.

[0088] In one embodiment, Figure 3 As shown, in step 106, the initial model parameters of the prediction model are adjusted according to the second energy-saving control loss value of each base station, and the adjusted prediction model is used as the target prediction model corresponding to each base station, including:

[0089] Step 302: Determine a total energy-saving control loss value corresponding to the prediction model according to the second energy-saving control loss value of each base station.

[0090] Step 304 , when the total energy-saving control loss value is greater than the loss value threshold, the initial model parameters are adjusted according to the total energy-saving control loss value and the adjustment step size, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

[0091] In an embodiment of the present application, the second energy-saving control loss value of each base station can be combined into a total energy-saving control loss value, and the initial model parameters can be adjusted based on the total energy-saving control loss value. For example, the total energy-saving control loss value can be obtained by summing the second energy-saving loss values; or a corresponding weight can be set for each base station, with a base station that is significantly different from other base stations having a smaller weight, and then a weighted sum of the second energy-saving loss values ​​can be performed to prevent some special base stations in the base station group from affecting the adjustment effect of the initial model parameters. This embodiment of the present application does not specifically limit this.

[0092] After obtaining the total energy-saving control loss value, if the total energy-saving control loss value is greater than a preset loss value threshold, it indicates that the initial model parameters have not yet been adjusted to the optimal value. In this case, the initial model parameters can be adjusted according to the preset adjustment step size. The adjustment step sizes corresponding to different initial model parameters can be the same or different, and this is not specifically limited in this embodiment of the application.

[0093] The following illustrates this process using the gradient descent algorithm. When applying the gradient descent algorithm, the specific relationship between the loss function and model parameters must be determined. Therefore, a prediction model can be used that uses operating parameters as input and predicts energy-saving control parameters and their corresponding reward values. The reward value predicted by this prediction model is used as the target reward value, and a loss function is constructed based on the target reward value. This establishes a connection between the model parameters and the loss function.

[0094] The total loss function of the base station group can be obtained by summing the loss functions of the target prediction models of each base station. Based on the gradient descent algorithm, the partial derivative of each initial model parameter with respect to the total loss function is obtained. The initial model parameters are adjusted accordingly based on the adjustment step size and partial derivative (refer to formula (1)):

[0095]

[0096] Among them, w is any initial model parameter, w ′ is the adjusted initial model parameter, β is the adjustment step size, L(w i ′ ) is the loss function of the i-th target prediction model, Refers to finding the partial derivative of the initial model parameter w with respect to the loss function of the i-th target prediction model.

[0097] After adjusting the initial model parameters, the prediction model obtained after adjusting the initial model parameters can be used as the target prediction model corresponding to each base station.

[0098] The prediction model training method provided in the embodiment of the present application calculates a total energy-saving control loss value based on the second energy-saving control loss value, and adjusts the initial model parameters according to the total energy-saving control loss value and the adjustment step size. The embodiment of the present application adjusts the initial model parameters according to the effect of adjusting the initial model parameters of the prediction model (that is, the second energy-saving control loss value) to obtain the optimal initial model parameters for each base station in the base station group. When the final model of each base station is obtained by continuing to train according to the initial model parameters, the training time of each base station model can be made relatively average, which can improve the overall training speed of each base station model.

[0099] In one embodiment, the method further includes:

[0100] When the total energy-saving control loss value is less than or equal to the loss value threshold, for any base station, the current target model parameters of the target prediction model corresponding to the base station are adjusted according to the corresponding second energy-saving control loss value of the base station, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

[0101] In the embodiment of the present application, when the total energy-saving control loss value is less than or equal to the loss value threshold, it means that the current initial model parameters have reached a relatively good level. At this time, the adjustment of the initial model parameters can be stopped, and the target prediction model corresponding to each base station can be trained separately. Figure 4 FIG. 4 shows a specific process of the above method.

[0102] Since the second energy-saving control loss value is the loss value corresponding to the adjusted target prediction model, the target prediction model can be adjusted directly according to the second energy-saving control loss value. The method of adjusting the target prediction model according to the second energy-saving control loss value can refer to the method of adjusting the target prediction model according to the first energy-saving control loss value, and the embodiments of the present application will not be repeated here. After adjusting the target prediction model according to the second energy-saving control loss value, the loss value of the adjusted target prediction model can be further determined, and then the target prediction model can be further adjusted according to the loss value until the loss value of the target prediction model is less than the preset threshold.

[0103] The prediction model training method provided in the embodiment of the present application stops adjusting the initial model parameters when the total energy-saving control loss value is less than or equal to the loss value threshold, and independently adjusts each target prediction model based on the second energy-saving control loss value. When the initial model parameters have reached a relatively optimal level, the embodiment of the present application independently trains each target prediction model to improve the prediction accuracy of each target prediction model.

[0104] In one embodiment, Figure 5 As shown, in step 102, before obtaining the prediction model corresponding to the target base station group, the method further includes:

[0105] Step 502: clustering each base station according to its operating parameters to obtain at least one clustering result.

[0106] Step 504: For any clustering result, if the number of base stations in the clustering result is greater than a preset number, the base stations in the clustering result are grouped as a base station group.

[0107] In the embodiments of the present application, the operating parameters of a base station refer to parameters that characterize the characteristics of users within the base station's coverage area, such as the base station's location, the temporal pattern of traffic flow, the time of peak traffic flow, etc. Because base stations with similar operating parameters also correspond to similar optimal energy-saving control parameters, base stations can be clustered based on their operating parameters to obtain base station groups. This ensures that the initial model parameters of the trained prediction model are close to the optimal model parameters of all base stations, thereby accelerating the training of the prediction model.

[0108] Each base station can be clustered according to all the working parameters of the base station, or one parameter can be selected from each working parameter to cluster each base station. This embodiment of the present application does not specifically limit this. After obtaining each clustering result, the clustering result with a number of base stations greater than a preset number can be selected as a base station group. For the clustering results and discrete points that do not form a base station group, the base stations corresponding to these clustering results and discrete points can be added to the base station group closest to it, or these base stations can be prevented from participating in the overall model training based on the base station group, but the model training is performed separately for these base stations. This embodiment of the present application does not specifically limit this.

[0109] The prediction model training method provided in the embodiment of the present application clusters each base station according to the working parameters, and regards the base stations belonging to the same clustering result as a base station group, so that the base stations in the base station group are relatively similar, and therefore the target prediction model of each base station is also likely to be similar, which can improve the speed of further training the target prediction model of each base station after obtaining the trained prediction model.

[0110] In one embodiment, Figure 6 As shown, the above method also includes:

[0111] Step 602: When a base station to be activated is detected, operating parameters of the base station to be activated are acquired, and based on the operating parameters of the base station to be activated, a first base station group to which the base station to be activated belongs is determined.

[0112] Step 604: Add the base station to be activated to the first base station group, and use the current prediction model of the first base station group as the target prediction model corresponding to the base station to be activated.

[0113] In an embodiment of the present application, the server can regularly detect the base stations within the coverage area of ​​the server. When a newly appeared base station (a base station to be activated) is detected, the server needs to determine a target prediction model for the base station to be activated. Since it takes a long time to train the target prediction model of the base station to be activated from scratch, it is possible to determine the base station group (the first base station group) to which the base station to be activated belongs, and then use the current prediction model of the base station group as the target prediction model corresponding to the base station to be activated, and add the base station to be activated to the base station group and train together with the existing base stations in the base station group, thereby reducing the training time of the target prediction model of the base station to be activated.

[0114] Reference Figure 7 As shown, it is a schematic diagram of the process after the base station to be activated is added to the base station group. When the prediction model corresponding to the base station group has not been trained, the base station to be activated will be trained together with the original base stations in the base station group after joining the base station group, that is, repeating the process of determining the first energy-saving control loss value in the aforementioned embodiment, adjusting the target model parameters according to the first energy-saving control loss value, re-determining the second energy-saving control loss value, and then allowing the server to adjust the initial model parameters according to the second control loss value. When the prediction model corresponding to the base station group has been trained, after the base station to be activated joins the base station group, the optimal initial model parameters of the prediction model corresponding to the base station group may change. At this time, the above-mentioned process of adjusting the initial model parameters of the prediction model can be restarted, or the prediction model can be not adjusted, but the prediction model corresponding to the base station group is directly used as the target prediction model corresponding to the base station to be activated, and the target prediction model is trained separately. The embodiments of the present application do not make specific limitations on this.

[0115] The prediction model training method provided in the embodiment of the present application determines the first base station group to which the base station to be activated belongs when a base station to be activated appears, and adds the base station to be activated to the first base station group, so that the base station to be activated can be trained together with the base stations in the first base station group, which can improve the training speed of the target prediction model of the base station to be activated.

[0116] In one embodiment, a prediction model training system is provided, the system comprising a server and at least one base station group, the base station group comprising at least one base station, wherein:

[0117] The server is used to obtain the prediction model corresponding to the target base station group as the target prediction model corresponding to each base station, and obtain the current operating parameters of each base station. For any base station, based on the operating parameters and the target prediction model, the energy-saving control parameters of the base station are predicted and sent to the base station.

[0118] The base station is used to receive the energy-saving control parameters, operate according to the energy-saving control parameters, obtain the report parameters corresponding to the energy-saving control parameters, and send the report parameters to the server.

[0119] The server is also used to determine the current first energy-saving control loss value of the target prediction model of each base station based on each feedback parameter, and adjust the current target model parameters of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize the energy-saving effect of the energy-saving operation performed by the base station based on the target prediction model.

[0120] The server is also used to determine the second energy-saving control loss value of the adjusted prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of predicting the energy-saving control parameters of the base station based on the operating parameters and the target prediction model, and sending the energy-saving control parameters to the base station until the preset conditions are met, and use the current target prediction model of each base station as the trained prediction model of each base station.

[0121] In an embodiment of the present application, the server stores a prediction model corresponding to each base station group and a target prediction model corresponding to each base station. The server can set the prediction model corresponding to the base station group as the target prediction model for each base station in the base station group. When training the target prediction model corresponding to each base station, the server can obtain the operating parameters of each base station, predict the energy-saving control parameters of each base station, and send the energy-saving control parameters to each base station, so that each base station executes the energy-saving control parameters and collects the feedback parameters.

[0122] After collecting the feedback parameters, the base station sends the feedback parameters to the server. The server calculates a first energy-saving control loss value based on the feedback parameters and adjusts the target prediction model corresponding to each base station based on the energy-saving control loss value corresponding to each base station. The server re-predicts new energy-saving control parameters based on the adjusted target prediction model and sends the new energy-saving control parameters to each base station, so that each base station executes the new energy-saving control parameters and collects new feedback parameters, and then calculates a second energy-saving control loss value based on the new feedback parameters. The server adjusts the initial model parameters of the prediction model based on the second energy-saving control loss value, resets the target prediction model corresponding to each base station to the prediction model, and repeats the above process until the preset conditions are met, thereby obtaining the trained prediction model of each base station.

[0123] Among them, the methods of determining energy-saving control parameters, collecting feedback parameters, calculating energy-saving control loss values ​​and adjusting model parameters can refer to the relevant descriptions of the aforementioned embodiments, and the embodiments of this application will not be repeated here.

[0124] The prediction model training system provided by the embodiment of the present application divides the base station into multiple base station groups, each base station group corresponds to a prediction model, and during the training process, the prediction model is used as the target prediction model of each base station. First, the target model parameters of each base station are adjusted according to the first energy-saving control loss value, and then the second energy-saving control loss value is obtained according to the adjusted target model parameters, and the initial model parameters of the prediction model are adjusted accordingly, and the above process is repeated until the preset conditions are met. The embodiment of the present application adjusts the initial model parameters according to the effect of adjusting the initial model parameters of the prediction model (that is, the second energy-saving control loss value) to obtain the optimal initial model parameters for each base station in the base station group. When the final model of each base station is obtained by continuing to train according to the initial model parameters, the training time of each base station model can be made relatively average, which can improve the overall training speed of each base station model.

[0125] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0126] Based on the same inventive concept, the present application also provides a prediction model training device for implementing the prediction model training method described above. The solution to the problem provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more prediction model training device embodiments provided below can be found in the above-mentioned limitations of the prediction model training method and will not be repeated here.

[0127] In one embodiment, Figure 8 As shown, a prediction model training device is provided, including: a first acquisition module 802, a first adjustment module 804, a second adjustment module 806, and a first processing module 808, wherein:

[0128] A first acquisition module 802 is configured to acquire a prediction model corresponding to a target base station group as a target prediction model corresponding to each base station;

[0129] A first adjustment module 804 is configured to determine, for any base station in the target base station group, a current first energy-saving control loss value of the target prediction model, and adjust a current target model parameter of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to represent an energy-saving effect of an energy-saving operation performed by the base station based on the target prediction model;

[0130] A second adjustment module 806 is configured to determine a second energy-saving control loss value of the adjusted prediction model, adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until a preset condition is met;

[0131] The first processing module 808 is configured to use the current target prediction model of each base station as the trained prediction model of each base station.

[0132] The prediction model training device provided by the embodiment of the present application divides the base station into multiple base station groups, each base station group corresponds to a prediction model, and during the training process, the prediction model is used as the target prediction model of each base station. First, the target model parameters of each base station are adjusted according to the first energy-saving control loss value, and then the second energy-saving control loss value is obtained according to the adjusted target model parameters, and the initial model parameters of the prediction model are adjusted accordingly, and the above process is repeated until the preset conditions are met. The embodiment of the present application adjusts the initial model parameters according to the effect of adjusting the initial model parameters of the prediction model (that is, the second energy-saving control loss value) to obtain the optimal initial model parameters for each base station in the base station group. When the final model of each base station is obtained by continuing to train according to the initial model parameters, the training time of each base station model can be made relatively average, which can improve the overall training speed of each base station model.

[0133] In one embodiment, the first adjustment module 804 is further configured to:

[0134] Acquiring current operating parameters of the base station, and predicting energy-saving control parameters of the base station based on the operating parameters and the target prediction model;

[0135] Sending the energy-saving control parameter to the base station, and receiving a feedback parameter returned by the base station;

[0136] An actual reward value is determined according to the reward parameter, and a current first energy-saving control loss value of the target prediction model is determined according to the actual reward value and the target reward value corresponding to the base station.

[0137] In one embodiment, the second adjustment module 806 is further configured to:

[0138] determining a total energy-saving control loss value corresponding to the prediction model according to the second energy-saving control loss value of each base station;

[0139] When the total energy-saving control loss value is greater than the loss value threshold, the initial model parameters are adjusted according to the total energy-saving control loss value and the adjustment step, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

[0140] In one embodiment, the apparatus further comprises:

[0141] The third adjustment module is used to adjust the current target model parameters of the target prediction model corresponding to any of the base stations according to the second energy-saving control loss value corresponding to the base station when the total energy-saving control loss value is less than or equal to the loss value threshold, and use the adjusted prediction model as the target prediction model corresponding to each of the base stations.

[0142] In one embodiment, the apparatus further comprises:

[0143] a clustering module, configured to perform clustering processing on each of the base stations according to an operating parameter of each of the base stations to obtain at least one clustering result;

[0144] The second processing module is configured to, for any of the clustering results, treat the base stations in the clustering result as a base station group when the number of the base stations in the clustering result is greater than a preset number.

[0145] In one embodiment, the apparatus further comprises:

[0146] A second acquisition module is configured to, when a base station to be activated is detected, acquire the operating parameters of the base station to be activated, and determine the first base station group to which the base station to be activated belongs based on the operating parameters of the base station to be activated;

[0147] An adding module is used to add the base station to be activated to the first base station group, and use the current prediction model of the first base station group as the target prediction model corresponding to the base station to be activated.

[0148] Each module in the above-mentioned prediction model training device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0149] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 9 As shown. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a prediction model training method.

[0150] Those skilled in the art will understand that Figure 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0151] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0152] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0153] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0154] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0155] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided in this application may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but are not limited to these.

[0156] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0157] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A prediction model training method, characterized in that: The method comprises: Obtaining a prediction model corresponding to the target base station group as a target prediction model corresponding to each of the base stations; For any of the base stations in the target base station group, determining a current first energy-saving control loss value of the target prediction model, and adjusting a current target model parameter of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize an energy-saving effect of an energy-saving operation performed by the base station based on the target prediction model; For any of the base stations in the target base station group, determine the adjusted second energy-saving control loss value of the prediction model, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any of the base stations in the target base station group until a preset condition is met; wherein the preset condition includes any one of the following: the first energy-saving control loss value of each base station is less than a preset threshold, the second energy-saving control loss value of each base station is less than the preset threshold, and the number of times the prediction model is adjusted reaches a preset number; The current target prediction model of each base station is respectively used as the trained prediction model of each base station.

2. The method according to claim 1, characterized in that Determining the current first energy-saving control loss value of the target prediction model includes: Acquiring current operating parameters of the base station, and predicting energy-saving control parameters of the base station based on the operating parameters and the target prediction model; Sending the energy-saving control parameter to the base station, and receiving a feedback parameter returned by the base station; An actual reward value is determined according to the reward parameter, and a current first energy-saving control loss value of the target prediction model is determined according to the actual reward value and the target reward value corresponding to the base station.

3. The method according to claim 1, characterized in that The adjusting the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, and using the adjusted prediction model as the target prediction model corresponding to each base station, includes: determining a total energy-saving control loss value corresponding to the prediction model according to the second energy-saving control loss value of each base station; When the total energy-saving control loss value is greater than the loss value threshold, the initial model parameters are adjusted according to the total energy-saving control loss value and the adjustment step, and the adjusted prediction model is used as the target prediction model corresponding to each base station.

4. The method according to claim 3, characterized in that The method further comprises: When the total energy-saving control loss value is less than or equal to the loss value threshold, for any of the base stations, the current target model parameters of the target prediction model corresponding to the base station are adjusted according to the second energy-saving control loss value corresponding to the base station, and the adjusted prediction model is used as the target prediction model corresponding to each of the base stations.

5. The method according to claim 1, wherein Before obtaining the prediction model corresponding to the target base station group, the method further includes: performing clustering processing on each of the base stations according to the operating parameters of each of the base stations to obtain at least one clustering result; For any of the clustering results, when the number of the base stations in the clustering result is greater than a preset number, the base stations in the clustering result are regarded as a base station group.

6. The method according to claim 5, characterized in that The method further comprises: In a case where a base station to be activated is detected, obtaining the operating parameters of the base station to be activated, and determining a first base station group to which the base station to be activated belongs based on the operating parameters of the base station to be activated; The base station to be activated is added to the first base station group, and the current prediction model of the first base station group is used as the target prediction model corresponding to the base station to be activated.

7. A prediction model training system, characterized in that: The system includes a server and at least one base station group, wherein the base station group includes at least one base station, wherein: The server is configured to obtain a prediction model corresponding to a target base station group as a target prediction model corresponding to each base station, obtain current operating parameters of each base station, and predict, for any base station, an energy-saving control parameter of the base station based on the operating parameters and the target prediction model, and send the energy-saving control parameter to the base station; The base station is configured to receive the energy-saving control parameter, operate according to the energy-saving control parameter, obtain a report parameter corresponding to the energy-saving control parameter, and send the report parameter to the server; The server is further configured to determine, based on each of the feedback parameters, a current first energy-saving control loss value of the target prediction model of each of the base stations, and adjust the current target model parameters of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to represent the energy-saving effect of the energy-saving operation performed by the base station based on the target prediction model; The server is also used to determine the second energy-saving control loss value of the adjusted prediction model for any base station in the target base station group, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of predicting the energy-saving control parameters of the base station based on the operating parameters and the target prediction model, and sending the energy-saving control parameters to the base station until the preset conditions are met, and use the current target prediction model of each base station as the trained prediction model of each base station; wherein the preset conditions include any one of the following: the first energy-saving control loss value of each base station is less than a preset threshold, the second energy-saving control loss value of each base station is less than the preset threshold, and the number of times the prediction model is adjusted reaches a preset number.

8. A prediction model training device, characterized in that: The device comprises: A first acquisition module is used to acquire a prediction model corresponding to a target base station group as a target prediction model corresponding to each base station; a first adjustment module, configured to determine, for any base station in the target base station group, a current first energy-saving control loss value of the target prediction model, and adjust a current target model parameter of the target prediction model based on the first energy-saving control loss value; the energy-saving control loss value is used to characterize an energy-saving effect of an energy-saving operation performed by the base station based on the target prediction model; A second adjustment module is used to determine the second energy-saving control loss value of the prediction model after adjustment for any base station in the target base station group, and adjust the initial model parameters of the prediction model according to the second energy-saving control loss value of each base station, use the adjusted prediction model as the target prediction model corresponding to each base station, and jump to the step of determining the current first energy-saving control loss value of the prediction model for any base station in the target base station group until a preset condition is met; wherein the preset condition includes any one of the following: the first energy-saving control loss value of each base station is less than a preset threshold, the second energy-saving control loss value of each base station is less than the preset threshold, and the number of times the prediction model is adjusted reaches a preset number; The first processing module is used to use the current target prediction model of each base station as the trained prediction model of each base station.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

11. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images