Prediction method, device, computer equipment and storage medium

Abstract

The present disclosure proposes a prediction method, apparatus, computer equipment and storage medium. When the method is executed by a first device, it includes: receiving a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by the second device each time training the prediction model; generating a target model parameter according to a plurality of first model parameters; and sending the target model parameter, wherein the target model parameter is used by the second device to adjust the prediction model obtained by training, and the adjusted prediction model is used for the next training of the prediction model. Thus, the target model parameter can be determined in combination with a plurality of first model parameters during the model iterative training process and provided to the prediction model obtained by adjusting the training of the second device for the next round of model training process, thereby realizing deep mining of the data, and then improving the model learning ability, realizing comprehensive improvement of the model performance and further optimization of the accuracy.

Description

Technical Field

[0001] The present disclosure relates to the field of infrastructure and IT support technology, and in particular to a prediction method, apparatus, computer equipment and storage medium. Background Art

[0002] In today's highly competitive business environment, companies are facing the challenge of enhancing customer service experience, and the key is to be able to accurately predict customer satisfaction with the service. By comprehensively collecting and analyzing multi-dimensional feature data related to services or products, such as network connection quality, call clarity, and video playback smoothness, companies can use machine learning algorithms to gain a deep understanding and predict user satisfaction.

[0003] In the existing technology, the user satisfaction model is mainly constructed by using classification models such as decision trees and logistic in machine learning, as well as more complex deep neural networks such as LSTM and ANN.

[0004] This approach cannot fully capture and utilize deep patterns in the data, which limits the algorithm's ability to learn from a wider range of data and hinders the overall improvement of model performance and further optimization of accuracy. Summary of the invention

[0005] The present disclosure aims to solve one of the technical problems in the related art at least to some extent.

[0006] To this end, the purpose of the present disclosure is to propose a prediction method, device, computer equipment and storage medium, whereby, during the model iterative training process, the target model parameters can be determined by combining multiple first model parameters and provided to the second device to adjust the prediction model obtained by training for the next round of model training process, thereby realizing in-depth mining of the data, thereby improving the model learning ability, and achieving a comprehensive improvement in model performance and further optimization of accuracy.

[0007] To achieve the above-mentioned purpose, the prediction method proposed in the embodiment of the first aspect of the present disclosure is executed by the first device; the method includes:

[0008] Receiving a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by each training of the prediction model by the second device;

[0009] generating target model parameters according to a plurality of said first model parameters; and

[0010] The target model parameters are sent, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

[0011] To achieve the above-mentioned purpose, the prediction method proposed in the second aspect of the present disclosure is executed by the second device; the method includes:

[0012] Acquire a plurality of object data of each object, wherein the plurality of object data includes: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and a quantity of the first object data and a quantity of the second object data satisfy a condition;

[0013] Acquire a plurality of data features corresponding to each of the object data;

[0014] Training the prediction model at least once according to the plurality of data features, wherein, during each training, obtaining and sending a first model parameter of the prediction model obtained through the training;

[0015] receiving a target model parameter, wherein the target model parameter is obtained by adjusting the first model parameter;

[0016] The trained prediction model is adjusted according to the target model parameters, and the adjusted prediction model is used for the next training until the trained prediction model converges.

[0017] To achieve the above-mentioned purpose, the prediction device provided in the third aspect of the present disclosure is executed by a first device; the device includes:

[0018] A receiving module, configured to receive a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by training a prediction model of the second device each time;

[0019] A generating module, configured to generate target model parameters according to a plurality of said first model parameters; and

[0020] A sending module is used to send the target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

[0021] To achieve the above-mentioned purpose, the prediction device provided in the fourth aspect of the present disclosure is executed by a second device; the device includes:

[0022] A first acquisition module is used to acquire a plurality of object data of each object, wherein the plurality of object data includes: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and a quantity of the first object data and a quantity of the second object data satisfy a condition;

[0023] A second acquisition module, used for acquiring a plurality of data features corresponding to each of the object data;

[0024] A model training module, used for training the prediction model at least once according to the plurality of data features, wherein, during each training, a first model parameter of the prediction model obtained through training is obtained and sent;

[0025] A receiving module, configured to receive a target model parameter, wherein the target model parameter is obtained by adjusting the first model parameter;

[0026] The parameter adjustment module is used to adjust the prediction model obtained through training according to the target model parameters, and the adjusted prediction model is used for the next training until the prediction model obtained through training converges.

[0027] The computer device proposed in the fifth embodiment of the present disclosure includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the prediction method proposed in the first embodiment of the present disclosure or the prediction method proposed in the second embodiment is implemented.

[0028] The sixth aspect embodiment of the present disclosure proposes a non-temporary computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the prediction method proposed in the first aspect embodiment of the present disclosure or the prediction method proposed in the second aspect embodiment.

[0029] The seventh aspect embodiment of the present disclosure proposes a computer program product. When the instructions in the computer program product are executed by a processor, the prediction method proposed in the first aspect embodiment of the present disclosure or the prediction method proposed in the second aspect embodiment is executed.

[0030] The prediction method, apparatus, computer equipment and storage medium provided by the present disclosure include, when executed by the first device: receiving the first model parameters corresponding to each second device, wherein the first model parameters are the model parameters obtained by the second device each time the prediction model is trained; generating target model parameters according to multiple first model parameters; and sending the target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained by training, and the adjusted prediction model is used for the next training of the prediction model. Thus, the target model parameters can be determined in combination with multiple first model parameters during the model iterative training process and provided to the prediction model obtained by the second device to adjust the training for the next round of model training process, thereby realizing deep mining of data, and then improving the model learning ability, and realizing comprehensive improvement of model performance and further optimization of accuracy.

[0031] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description or learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and / or additional aspects and advantages of the present disclosure will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0033] Figure 1 is a flowchart of a prediction method proposed in an embodiment of the present disclosure;

[0034] Figure 2 is a flowchart of a prediction method proposed in another embodiment of the present disclosure;

[0035] Figure 3 is a flowchart of a prediction method proposed in another embodiment of the present disclosure;

[0036] Figure 4 is a flowchart of a prediction method proposed in another embodiment of the present disclosure;

[0037] Figure 5 is a flowchart of a prediction method proposed in another embodiment of the present disclosure;

[0038] Figure 6 is a schematic diagram of a user satisfaction prediction method based on federated balanced enhancement and evaluation learning proposed in the present disclosure;

[0039] Figure 7 is a schematic diagram of the structure of a prediction device proposed in an embodiment of the present disclosure;

[0040] Figure 8 is a schematic diagram of the structure of a prediction device proposed in another embodiment of the present disclosure;

[0041] Fig. 9 A block diagram of an exemplary computer device suitable for implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0042] Embodiments of the present disclosure are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present disclosure, and are not to be construed as limitations of the present disclosure. On the contrary, the embodiments of the present disclosure include all changes, modifications, and equivalents that fall within the spirit and connotation of the appended claims.

[0043] Figure 1 It is a flowchart of a prediction method proposed in an embodiment of the present disclosure.

[0044] It should be noted that the executor of the prediction method of this embodiment is a prediction device, which can be implemented by software and / or hardware. The device can be configured in a computer device, and the computer device can include but is not limited to a terminal, a server, etc. For example, the terminal can be a mobile phone, a handheld computer, etc.

[0045] like Figure 1 As shown, the prediction method is executed by the first device and includes:

[0046] S101: Receive a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by the second device each time training a prediction model.

[0047] The first device may refer to a central server in a federated learning system, and the second device may refer to a participant device in the federated learning system.

[0048] That is, in the embodiment of the present disclosure, the first device as a central server in the federated learning system can receive the first model parameters corresponding to each second device, thereby providing reliable data support for the subsequent generation of target model parameters.

[0049] S102: Generate target model parameters according to a plurality of first model parameters.

[0050] The target model parameters may refer to model parameters obtained by the first device after comprehensive analysis based on multiple first model parameters in the embodiment of the present disclosure.

[0051] In the embodiment of the present disclosure, when generating the target model parameters based on multiple first model parameters, the target model parameters may be generated by taking a weighted average of the multiple first model parameters, or the target model parameters may be generated based on the multiple first model parameters based on a method combining numbers and shapes.

[0052] S103: Sending target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

[0053] That is to say, after the first device in the embodiment of the present disclosure generates the target model parameters based on multiple first model parameters, it can send the target model parameters, wherein the target model parameters are used for the second device to adjust the prediction model obtained by training, and the adjusted prediction model is used for the next training of the prediction model. In this way, the prediction model obtained by the second device adjusting the training can be ensured to be continuously updated during the iterative training process.

[0054] In this embodiment, by receiving the first model parameters corresponding to each second device, wherein the first model parameters are the model parameters obtained by the second device each time training the prediction model; generating target model parameters according to multiple first model parameters; and sending the target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained by training, and the adjusted prediction model is used for the next training of the prediction model. Thus, in the process of iterative model training, the target model parameters can be determined by combining multiple first model parameters and provided to the prediction model obtained by adjusting the training of the second device for the next round of model training, thereby realizing deep mining of data, and then improving the model learning ability, realizing comprehensive improvement of model performance and further optimization of accuracy.

[0055] Figure 2 It is a flowchart of a prediction method proposed in another embodiment of the present disclosure.

[0056] like Figure 2 As shown, the prediction method is executed by the first device and includes:

[0057] S201: Acquire global distribution information of a global prediction model, wherein the global distribution information is used to describe output distribution of the global prediction model, and the global prediction model and the prediction model of each second device are jointly trained.

[0058] The global prediction model may refer to a prediction model configured in the first device.

[0059] S202: Send global distribution information.

[0060] That is, after acquiring the global distribution information of the global prediction model, the first device in the embodiment of the present disclosure can send the global distribution information to the second device, thereby providing reliable data support for the parameter adjustment process on the second device side.

[0061] In this embodiment, by acquiring the global distribution information of the global prediction model, wherein the global distribution information is used to describe the output distribution of the global prediction model, the global prediction model and the prediction model of each second device are jointly trained; and the global distribution information is sent. Thus, the output distribution of the global prediction model on the first device side can be determined and sent, thereby providing reliable reference information for the prediction process of the second device.

[0062] Figure 3 It is a flowchart of a prediction method proposed in another embodiment of the present disclosure.

[0063] like Figure 3 As shown, the prediction method, executed by the second device, includes:

[0064] S301: Acquire multiple object data of each object, wherein the multiple object data include: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the number of the first object data and the number of the second object data satisfy a condition.

[0065] The object may refer to an object for which data collection is required in the embodiments of the present disclosure, such as a region, a town, etc., without limitation. Object data may refer to data related to the object.

[0066] The first evaluation tendency and the second evaluation tendency refer to different tendencies of user evaluations, for example, they may refer to satisfaction and dissatisfaction, respectively.

[0067] The condition being satisfied between the quantity of the first object data and the quantity of the second object data may mean that the quantity of the first object data and the quantity of the second object data are relatively balanced.

[0068] Optionally, in some embodiments, when acquiring multiple object data of each object, at least one first object data and multiple second object data may be acquired; and sampling processing is performed on at least one first object data, so that the number of first object data obtained by the sampling processing meets the condition with the number of second object data. Thus, the number of sample data of the first object data can be quickly increased based on the sampling processing technology, so as to ensure that the number of first object data obtained meets the condition with the number of second object data.

[0069] Optionally, in some embodiments, the object data includes at least one of the following:

[0070] Wireless connection rate;

[0071] Radio Resource Control RRC connection rate;

[0072] Uplink physical resource block PRB utilization;

[0073] Downlink PRB utilization;

[0074] Evolved Radio Access Bearer (E-RAB) establishment success rate.

[0075] In this way, it can be ensured that the object data can meet personalized scene requirements.

[0076] Optionally, in some embodiments, when acquiring multiple object data of each object, multiple initial data of each object may be collected; the feature missing rate of each initial data is determined; the feature missing rate of the multiple feature missing rates that is less than the missing rate threshold is determined; and the initial data to which the feature missing rate that is less than the missing rate threshold belongs is used as the object data. In this way, the integrity of the obtained object data can be guaranteed, thereby improving the practicality and reliability of the object data in the prediction process.

[0077] The initial data refers to the unprocessed data after data collection for multiple objects.

[0078] Among them, the feature missing rate can be used to describe the completeness of the description of the corresponding initial data on the features of the relevant objects.

[0079] The missing rate threshold refers to a threshold configured in advance for the feature missing rate. The specific value of the missing rate threshold can be flexibly adjusted according to the application scenario, for example, it can be 30%.

[0080] It is understandable that there may be missing features in the initial data collected. If these data with missing features are directly applied to the subsequent calculation process, it may affect the prediction accuracy. Therefore, in the embodiment of the present disclosure, multiple initial data of each object can be collected; the feature missing rate of each initial data is determined; the feature missing rate of multiple feature missing rates that is less than the missing rate threshold is determined; the initial data to which the feature missing rate is less than the missing rate threshold belongs is used as the object data, thereby achieving data cleaning of the initial data.

[0081] S302: Acquire multiple data features corresponding to each object data.

[0082] Among them, data features can be used to describe relevant features corresponding to object data.

[0083] In the embodiments of the present disclosure, when acquiring multiple data features corresponding to each object data, corresponding feature extraction methods may be flexibly selected according to different types of object data.

[0084] S303: Train the prediction model at least once according to the multiple data features, wherein during each training, obtain and send the first model parameter of the prediction model obtained through training.

[0085] That is, after acquiring multiple data features corresponding to each object data, the second device in the embodiment of the present disclosure can train the prediction model at least once based on the multiple data features, and then acquire the first model parameter of the trained prediction model.

[0086] S304: Receive target model parameters, where the target model parameters are obtained by adjusting the first model parameters.

[0087] That is to say, in the embodiment of the present disclosure, after the second device obtains and sends the first model parameters of the trained prediction model during each training, it can receive the target model parameters generated by the first device side based on the first model parameters, thereby providing data support for subsequent model parameter adjustments.

[0088] S305: adjusting the prediction model obtained through training according to the target model parameters, and using the adjusted prediction model for the next training, until the prediction model obtained through training converges.

[0089] In this embodiment, by obtaining multiple object data of each object, wherein the multiple object data include: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the number of the first object data and the number of the second object data satisfy a condition; obtaining multiple data features corresponding to each object data; training the prediction model at least once according to the multiple data features, wherein, in each training, the first model parameters of the prediction model obtained by training are obtained and sent; receiving the target model parameters, wherein the target model parameters are obtained by adjusting the first model parameters; adjusting the prediction model obtained by training according to the target model parameters, and using the adjusted prediction model for the next training until the trained prediction model converges. Thus, the target model parameters can be determined in combination with the multiple first model parameters during the model iterative training process and provided to the second device to adjust the prediction model obtained by training for the next round of model training process, thereby realizing deep mining of the data, and then improving the model learning ability, realizing comprehensive improvement of the model performance and further optimization of the accuracy.

[0090] Figure 4 It is a flowchart of a prediction method proposed in another embodiment of the present disclosure.

[0091] like Figure 4 As shown, the prediction method, executed by the second device, includes:

[0092] S401: Acquire multiple object data of each object, wherein the multiple object data include: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the quantity of the first object data and the quantity of the second object data satisfy a condition.

[0093] S402: Acquire multiple data features corresponding to each object data.

[0094] S403: Train the prediction model at least once according to the multiple data features, wherein during each training, a first model parameter of the prediction model obtained through training is obtained and sent.

[0095] S404: Receive target model parameters, where the target model parameters are obtained by adjusting the first model parameters.

[0096] The description of S401 - S404 can be specifically referred to the above embodiment, which will not be repeated here.

[0097] S405: Receive global distribution information, where the global distribution information is used to describe the output distribution of the global prediction model, and the global prediction model and the prediction model are jointly trained.

[0098] That is, in the embodiment of the present disclosure, the second device can receive global distribution information, wherein the global distribution information is used to describe the output distribution of the global prediction model, and the global prediction model and the prediction model are jointly trained. Thus, data support for the output distribution of the global prediction model on the first device side can be provided for the subsequent determination of the divergence information.

[0099] S406: Obtain output distribution information of the trained prediction model.

[0100] Among them, the output distribution information can be used to indicate the output distribution of the prediction model arranged on the second device side.

[0101] In the embodiment of the present disclosure, when the output distribution information of the prediction model obtained through training is obtained, data support of the output distribution information of the prediction model obtained through training on the second device side can be provided for the subsequent determination of divergence information.

[0102] S407: Determine divergence information according to the global distribution information and the output distribution information.

[0103] The divergence information can be used to indicate the difference between different output distribution information and the global distribution information.

[0104] In the disclosed embodiment, when the divergence information is determined based on the global distribution information and the output distribution information, a reliable execution basis can be provided for subsequent model parameter adjustment.

[0105] S408: Process the target model parameters according to the divergence information to obtain second model parameters.

[0106] That is, in the embodiment of the present disclosure, after the divergence information is determined based on the global distribution information and the output distribution information, the target model parameters can be processed based on the divergence information to obtain the second model parameters to ensure the practicality of the obtained second model parameters.

[0107] S409: adjusting the prediction model obtained through training according to the second model parameters, and using the adjusted prediction model for next training until the prediction model obtained through training converges.

[0108] That is, in the embodiment of the present disclosure, the output distribution information of the trained prediction model can be obtained; the divergence information can be determined based on the global distribution information and the output distribution information; the target model parameters can be processed based on the divergence information to obtain the second model parameters; and the trained prediction model can be adjusted based on the second model parameters. Thus, the obtained divergence information can accurately indicate the difference between the output distribution of the global prediction model and each prediction model on the second device side, thereby ensuring the reliability of the obtained second model parameters.

[0109] In this embodiment, by receiving global distribution information, wherein the global distribution information is used to describe the output distribution of the global prediction model, the global prediction model and the prediction model are jointly trained. Thus, data support for the output distribution of the global prediction model on the first device side can be provided for the subsequent determination of divergence information. By obtaining the output distribution information of the trained prediction model; determining the divergence information according to the global distribution information and the output distribution information; processing the target model parameters according to the divergence information to obtain the second model parameters; and adjusting the trained prediction model according to the second model parameters. Thus, the obtained divergence information can accurately indicate the difference in the output distribution between the global prediction model and each prediction model on the second device side, thereby ensuring the reliability of the obtained second model parameters.

[0110] Figure 5 It is a flowchart of a prediction method proposed in another embodiment of the present disclosure.

[0111] like Figure 5 As shown, the prediction method, executed by the second device, includes:

[0112] S501: Acquire multiple object data of each object, wherein the multiple object data include: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the quantity of the first object data and the quantity of the second object data satisfy a condition.

[0113] S502: Acquire multiple data features corresponding to each object data.

[0114] S503: Train the prediction model at least once according to the multiple data features, wherein during each training, obtain and send the first model parameter of the prediction model obtained through training.

[0115] S504: Receive target model parameters, where the target model parameters are obtained by adjusting the first model parameters.

[0116] The description of S501 - S504 can be specifically referred to the above embodiment, which will not be repeated here.

[0117] S505: In the current round of iterative training, a first weight corresponding to the first object data and a second weight corresponding to the second object data are obtained.

[0118] The first weight and the second weight may refer to the initial weights of the first object data and the second object data during the iterative training process of the model.

[0119] S506: Determine a first prediction error corresponding to the first object data according to the initial prediction model, and determine a second prediction error corresponding to the second object data according to the initial prediction model.

[0120] The first prediction error and the second prediction error refer to the errors between the first object data and the second object data during the model training process.

[0121] Optionally, in some embodiments, when determining the first prediction error corresponding to the first object data according to the initial prediction model, and determining the second prediction error corresponding to the second object data according to the initial prediction model, the first prediction tendency may be obtained by predicting the first object data according to the initial prediction model; the second prediction tendency may be obtained by predicting the second object data according to the initial prediction model; the first prediction error may be determined according to the first prediction tendency and the first evaluation tendency, and the second prediction error may be determined according to the second prediction tendency and the second evaluation tendency. Thus, the description accuracy of the obtained first prediction error and the second prediction error may be effectively improved.

[0122] The first prediction tendency and the second prediction tendency may be used to indicate the correlation tendency of the prediction results of the first object data and the second object data.

[0123] S507: The first weight is adjusted according to the first prediction error, and the second weight is adjusted according to the second prediction error. The adjusted first weight and second weight are used for the next round of iterative training until the trained prediction model converges, and the trained prediction model is used as the target prediction model.

[0124] That is to say, in the embodiment of the present disclosure, after receiving the target model parameters, the first weight corresponding to the first object data and the second weight corresponding to the second object data can be obtained in the current round of iterative training; the first prediction error corresponding to the first object data is determined according to the initial prediction model, and the second prediction error corresponding to the second object data is determined according to the initial prediction model; the first weight is adjusted according to the first prediction error, and the second weight is adjusted according to the second prediction error, and the adjusted first weight and second weight are used for the next round of iterative training until the prediction model obtained by training converges, and the prediction model obtained by training is used as the target prediction model. Thus, the weight information and prediction error of the object data can be combined in the model training process to accurately and quickly adjust the parameters to ensure the reliability of the obtained target prediction model.

[0125] In this embodiment, in the current round of iterative training, the first weight corresponding to the first object data and the second weight corresponding to the second object data are obtained; the first prediction error corresponding to the first object data is determined according to the initial prediction model, and the second prediction error corresponding to the second object data is determined according to the initial prediction model; the first weight is adjusted according to the first prediction error, and the second weight is adjusted according to the second prediction error, and the adjusted first weight and second weight are used for the next round of iterative training until the prediction model obtained by training converges, and the prediction model obtained by training is used as the target prediction model. In this way, the weight information and prediction error of the object data can be combined in the model training process to accurately and quickly adjust the parameters to ensure the reliability of the obtained target prediction model.

[0126] In summary of the above embodiments, the present disclosure proposes a federated learning system framework for provincial and municipal satisfaction prediction, called a Federated Enhancement Learning System (FELS). The system aims to achieve higher accuracy user satisfaction prediction in the field of data processing and analysis technology by comprehensively utilizing adaptive resampling (AR) technology, variational mode decomposition (VMD) technology, dynamic weight adjustment (DWA) technology, Jensen-Shannon divergence evaluation (JSD) technology and federated learning (FL) technology.

[0127] The FELS model pre-processes the collected user data through adaptive resampling (AR) technology to balance the sample distribution and retain the representativeness of the data. Then, the variational mode decomposition (VMD) technology is used to further process the AR data, extract key features from the dynamic time series, and generate a comprehensive enhanced data set.

[0128] During the model training process, FELS introduced a new concept named "coordinator" (Agent). FELS uses dynamic weight adjustment (DWA) technology to dynamically adjust weights according to the category of samples to increase the sub-model's attention to minority samples. At the same time, FELS uses Jensen-Shannon divergence evaluation (JSD) as a coordinator to measure the difference between the output distribution of each prefecture-level model and the global model. By calculating the JSD value, the system can ensure that the model training of each node not only conforms to the regional characteristics, but also maintains overall consistency and accuracy. As a coordinator, JSD dynamically adjusts model parameters to reduce the differences between prefecture-level and provincial models, thereby ensuring coordination and consistency in the federated learning process.

[0129] Through federated learning (FL) technology, FELS allows regions or departments to share model parameters rather than original data while protecting data privacy, thereby avoiding the problem of data silos and enhancing the model's adaptability to various market changes.

[0130] The FELS system of the present invention improves the accuracy and reliability of user satisfaction prediction throughout the province, helping enterprises to better care for users and thus improve user satisfaction.

[0131] like Figure 6 As shown, Figure 6 It is a schematic diagram of the user satisfaction prediction method based on federated balanced enhancement and evaluation learning proposed in the present disclosure.

[0132] 2. Data Collection

[0133] In this study, nine cities in Province A independently completed the collection and preprocessing of user data. These nine cities include City A, City B, City C, City D, City E, City F, City G, City H, and City I. Each city collects data according to unified standards and processes based on local conditions to ensure the comprehensiveness and consistency of the data, and the data remain independent and confidential between cities.

[0134] The collection of user satisfaction label data is mainly carried out through two channels: first, using artificial intelligence (AI) outbound calling technology, and second, through traditional manual customer service telephone interviews. Through AI outbound calling technology, the system can automatically dial user phones, conduct satisfaction surveys and record user feedback. On the other hand, traditional manual customer service telephones are manually dialed by users to collect user feedback information in detail. These two methods together collect a series of data features that reflect network performance and user satisfaction.

[0135] The collected data features include key indicators such as wireless connection rate, average RRC connection rate, average uplink PRB utilization, average downlink PRB utilization, and E-RAB establishment success rate. These data features fully reflect the user experience and satisfaction in the process of using the network, providing rich information for subsequent analysis and modeling.

[0136] In the data cleaning phase, the missing feature rate of each user's data is calculated to identify and remove user records with more missing data. Specifically, if the missing data rate of a user exceeds 30%, the user's data record is deleted. This method not only ensures that the collected data can accurately reflect the user's real behavior and satisfaction changes, but also ensures the accuracy and reliability of the data through strict data cleaning steps.

[0137] Through the above methods, the nine cities in Province A were able to effectively collect and process user data, laying a solid data foundation for the model training of the Federal Reinforcement Learning System (FELS). This systematic data collection and processing method ensures the efficiency and accuracy of the model in predicting user satisfaction.

[0138] 3. Adaptive Resampling AR

[0139] In user satisfaction prediction, data sets often have the problem of class imbalance, that is, the amount of data in some categories is much larger than that in other categories. In order to improve the accuracy and stability of the prediction model, the adaptive resampling (AR) technology is used to process the data set. The following takes the data of city A as an example to explain in detail the specific implementation process of the adaptive resampling technology and the SMOTE algorithm.

[0140] Assume that the user satisfaction data set of city a contains two categories: satisfied and dissatisfied. The category distribution of the initial data set is as follows: the number of records in the satisfied category is 5.7 million, and the number of records in the dissatisfied category is 300,000. Obviously, the number of samples in the "satisfied" category in the data set is much larger than that in the "unsatisfied" category. This imbalance will cause the model to tend to predict the "satisfied" category during training, affecting the overall prediction performance.

[0141] First, calculate the number of samples and their proportions for each category. For the data set of city a, the category distribution ratio is as follows:

[0142]

[0143] In order to balance the data set, it is necessary to determine a target number of samples so that the number of samples in each category tends to be consistent. Assume that the target number of samples for each category is set to the number of the largest category, that is, 5.7 million records. For categories with a small number of samples (unsatisfactory), the number of samples is increased through oversampling technology. Specific methods include randomly copying existing samples and synthesizing minority class samples (using the SMOTE algorithm) to generate more samples of unsatisfactory categories.

[0144] Adaptive resampling mainly automatically adapts to the distribution of the model, and uses different strategies for data generation in the early, middle, and late stages of SMOTE synthetic data generation:

[0145] ① Use SMOTE more often in the early stage

[0146] ②Medium term combined with random undersampling

[0147] ③After the model effect tends to be stable, gradually reduce the use of SMOTE

[0148] The SMOTE (Synthetic Minority Over-sampling Technique) algorithm is a method for synthesizing minority class samples, which aims to balance the category distribution of a data set by generating new samples. The basic idea of ​​the algorithm is to generate new samples by linearly interpolating minority class samples to increase the number of minority class samples.

[0149] In this way, the number of samples of the unsatisfactory category can be gradually increased until it reaches 5.7 million records respectively. For the category with a large number of samples (satisfactory), the number of samples is reduced by undersampling technology. 5.7 million records are randomly selected from the samples of the satisfactory category to form a new sample set of the satisfactory category.

[0150] After the above steps, the number of samples in each category of the user satisfaction data set of city A is 5.7 million. Similarly, the data adaptive resampling method for the other nine cities in province A is similar to this method.

[0151] Through adaptive resampling technology, the category distribution of the data set is balanced, the representativeness of the data set is enhanced, and it is ensured that the model pays balanced attention to samples of each category during training, thereby improving the overall prediction performance.

[0152] 4. VMD variational mode decomposition and dynamic weight adjustment

[0153] The variational mode decomposition (VMD) technique is an adaptive decomposition method for signal processing. By decomposing the signal into several intrinsic mode functions (IMFs), the key features of time series data can be extracted. The following takes the characteristic data collected in city A as an example to explain the VMD decomposition process in detail.

[0154] The data set of city a includes the following features: wireless connection rate, average RRC connection rate, average uplink PRB utilization, average downlink PRB utilization, and E-RAB establishment success rate. These features reflect the user experience in the process of using the network.

[0155] VMD decomposes the original signal f(t) into K modal functions u k combination.

[0156]

[0157] Each modal function u k The calculation process of (t) can be described by the following optimization problem:

[0158]

[0159] in, represents the time derivative, δ(t) is the unit impulse function, j is the imaginary unit, * represents the convolution operation, ω k is the center frequency of the kth mode function.

[0160] By iteratively optimizing the above formula, each modal function u can be obtained k (t) and its corresponding center frequency ω k These mode functions reflect the different frequency components in the original signal, thereby extracting the key features of the signal.

[0161] Taking the wireless connection rate as an example, assuming that the original signal is x(t), it can be decomposed into three modal functions u1(t), u2(t) and u3(t), and the corresponding center frequencies are ω1, ω2 and ω3 respectively. Through VMD decomposition, the original signal x(t) can be expressed as:

[0162] x(t)=u1(t)+u2(t)+u3(t)

[0163] This method can improve the model's sensitivity to data and capture the characteristics of data in the frequency domain, thereby improving the prediction accuracy of the sub-model.

[0164] In the model training process of the FELS system, the Dynamic Weight Adjustment (DWA) technology is introduced to improve the model's attention to minority samples. The core idea of ​​the DWA technology is to dynamically adjust the weight according to the category of the sample, so that the model pays more attention to minority samples during the training process, thereby improving the overall prediction performance.

[0165] Suppose there is a data set containing two categories of user satisfaction: satisfied and dissatisfied. Take the sample data of city A as an example, where the number of samples in the satisfied category is 5.7 million and the number of samples in the dissatisfied category is 300,000. The data is seriously unbalanced, so during the training process, the model may be biased towards predicting the satisfied category. To solve this problem, DWA technology can be introduced.

[0166] In each training iteration, the prediction accuracy of the current model for each category is calculated, and the sample weights of each category are dynamically adjusted based on these accuracies. The specific steps are as follows:

[0167] ⑵ Initialize weights: Assign initial weights w to each category 满意 =w 不满意 =1

[0168] ⑵ Calculate prediction error: In each round of training, calculate the prediction error e of the model for each category of samples 满意 and e 不满意 .

[0169] ⑶ Dynamically adjust weights: adjust weights according to prediction errors, increase the weights of categories with larger errors, and reduce the weights of categories with smaller errors. The adjustment formula is as follows:

[0170] w 满意 =w 满意 ×(1-α×e 满意 )

[0171] w 不满意 =w 不满意 ×(1-α×e 不满意 )

[0172] Among them, α is the learning rate, which controls the speed of weight adjustment.

[0173] Through the above steps, the weights of samples in each category are dynamically adjusted during the training process, so that the model pays more attention to samples in the unsatisfactory category, thereby improving the prediction accuracy of minority class samples.

[0174] 5. JSD divergence evaluation and federated learning

[0175] In the FELS system, Jensen-Shannon divergence (JSD) is used as a "coordinator" to measure the difference between the output distribution of each city model and the global model (provincial headquarters model). JSD is used to evaluate the training effect of each city model to ensure that they are consistent with the global model while retaining regional characteristics.

[0176] Assume that there are models of cities a and b and the output distribution P of the provincial headquarters model a , P b and P A It is necessary to calculate the JSD divergence between the models of region a and region b and the provincial headquarters model, and dynamically adjust the parameters of the city models based on the JSD value. The specific steps are as follows:

[0177] ⑴ Calculate the output distribution of the models in each city: Calculate the output distribution P of models a and b a , P b and P A .

[0178] ⑵ Calculate JSD divergence: Calculate the JSD divergence between the Fuzhou and Xiamen models and the provincial headquarters model. The formula for JSD is as follows:

[0179]

[0180] in, is the Kullback-Leibler divergence.

[0181] For city A:

[0182]

[0183] in,

[0184] For city B:

[0185]

[0186] in,

[0187] ⑶ Dynamically adjust model parameters: Dynamically adjust the parameters of models a and b according to the JSD value to ensure the consistency between the models of various cities and the provincial headquarters model.

[0188] If JSD(P a ||P 省总部 ) is large, indicating that there is a large difference between the city a model and the provincial headquarters model. The system will increase the training weight of the city a model to make it pay more attention to the characteristics of the a model.

[0189] If JSD(P b ||P 省总部 ) is smaller, indicating that the model of city b is more consistent with the provincial headquarters model. The system will maintain the existing training weights of the model of city b.

[0190] (4) Return to the training sub-model: a and b continue local training according to the adjusted parameters, and calculate the JSD divergence again in the next round of training, repeating the above steps to gradually narrow the difference with the provincial headquarters model.

[0191] The specific return training process is as follows:

[0192] ① Each city (a, b, etc.) trains the model locally and calculates the local model parameters θ a and θ b .

[0193] ② Shared model parameters: Each city will share its local model parameters θ a and θ b Upload to a central server instead of raw data.

[0194] Since what is uploaded is hyperparameters instead of original data, the data island problem can be effectively solved.

[0195] ③ Aggregate model parameters: The central server aggregates the model parameters of each city to generate the global model parameters θ 省总部 The aggregation method uses strategies such as weighted average, and the formula is as follows:

[0196]

[0197] Among them, N is the number of cities participating in the training.

[0198] ④ Update local model: Each city downloads the updated global model parameters θ 省总部 , and use it to continue local training.

[0199] Through the above steps, the federated learning system with the "coordinator" JSD divergence realizes collaborative training among cities, protects data privacy, solves the data island problem, and improves the overall prediction performance and generalization ability of the model. In the actual application model, Province A uses nine cities to perform the same transformation and calculation.

[0200] That is to say, based on the above embodiments, the present disclosure can propose a user satisfaction prediction method based on federated balanced enhancement and evaluation learning, firstly, user satisfaction data of various cities are collected; then, the collected user data is preprocessed by adaptive resampling (AR) technology, and the variational mode decomposition (VMD) technology is used to further process the data after AR, extract key features from the dynamic time series, and generate a comprehensive enhanced data set; then, dynamic weight adjustment (DWA) is used to enhance the attention to minority class samples in the city sub-model; finally, based on the Jensen-Shannon divergence evaluation (JSD), the difference between the output distribution of the sub-model (the city model) and the global model (the provincial headquarters model) is measured and the training weight of the sub-model with larger difference is increased, and the collaborative training between the cities is realized through the federated learning (FL) technology, which protects data privacy, solves the data island problem, and improves the overall prediction performance and generalization ability of the model.

[0201] This disclosure also proposes a federated learning method with a "coordinator" JSD divergence:

[0202] Calculate the output distribution of each sub-model; calculate the JSD divergence between each sub-model and the global model; dynamically adjust the parameters of each sub-model according to the JSD value, increase the parameters of the sub-model with a larger JSD value, and ensure the consistency of each sub-model with the global model; each sub-model continues to train locally according to the adjusted parameters, and calculates the JSD divergence again in the next round of training; repeat the above steps until the difference between each sub-model and the global model meets the preset conditions. The federated learning system with the "coordinator" JSD divergence realizes collaborative training between cities, protects data privacy, solves the problem of data islands, and improves the overall prediction performance and generalization ability of the model.

[0203] Compared with the prior art, the federated reinforcement learning system (FELS) proposed in this application has significant technical advantages. The prior art mainly uses classification models such as decision trees and logistic regression, as well as deep neural networks such as LSTM and ANN to build user satisfaction models, which can learn from huge historical data and improve prediction accuracy. However, these methods have significant limitations in dealing with data imbalance and cross-regional data collaborative training, and there are data islands for multiple departments or cities with different data. This solution FELS solves the problem of overfitting or low model accuracy caused by data imbalance in conventional models by combining the variational mode decomposition (VMD) method with the adaptive resampling technology (SMOTE). The "coordinator" mechanism proposed by FELS integrates the federated learning system, and evaluates the difference between the prefecture-level sub-model and the provincial headquarters model in the federated learning by JSD divergence, avoiding the problem that the provincial headquarters model has good results while the prefecture-level model training results are uneven. In addition, FELS integrates dynamic weight adjustment (DWA), adaptive resampling (AR) and VMD decomposition, so that the prefecture-level sub-model can efficiently train minority category samples under the premise of solving the problem of data imbalance. FELS breaks the data barriers of city-level subsidiaries, directly uploads model parameters to the provincial headquarters company model for cyclic optimization, effectively solves the data island problem, and avoids the overfitting problem that may exist in the training of local city-level companies. In addition, FELS is also used in various scenarios with data imbalance problems, such as credit user prediction, weather forecast and other systems. It has strong versatility, and can dynamically adjust the parameters in FELS according to different scenarios, and the model is highly adjustable.

[0204] Figure 7 It is a schematic diagram of the structure of a prediction device proposed in one embodiment of the present disclosure.

[0205] like Figure 7 As shown, the prediction device 70, executed by the first device, includes:

[0206] The receiving module 701 is used to receive a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by the second device each time training the prediction model;

[0207] A generating module 702, configured to generate target model parameters according to a plurality of first model parameters; and

[0208] The sending module 703 is used to send target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

[0209] It should be noted that the above explanation of the prediction method is also applicable to the prediction device of this embodiment and will not be repeated here.

[0210] In this embodiment, by receiving the first model parameters corresponding to each second device, wherein the first model parameters are the model parameters obtained by the second device each time training the prediction model; generating target model parameters according to multiple first model parameters; and sending the target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained by training, and the adjusted prediction model is used for the next training of the prediction model. Thus, in the process of iterative model training, the target model parameters can be determined by combining multiple first model parameters and provided to the prediction model obtained by adjusting the training of the second device for the next round of model training, thereby realizing deep mining of data, and then improving the model learning ability, realizing comprehensive improvement of model performance and further optimization of accuracy.

[0211] Figure 8 It is a schematic diagram of the structure of a prediction device proposed in another embodiment of the present disclosure.

[0212] like Figure 8 As shown, the prediction device 80, executed by the second device, includes:

[0213] A first acquisition module 801 is used to acquire a plurality of object data of each object, wherein the plurality of object data includes: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the quantity of the first object data and the quantity of the second object data satisfy a condition;

[0214] The second acquisition module 802 is used to acquire multiple data features corresponding to each object data;

[0215] A model training module 803 is used to train the prediction model at least once according to the multiple data features, wherein, during each training, a first model parameter of the prediction model obtained through training is obtained and sent;

[0216] A receiving module 804 is used to receive a target model parameter, wherein the target model parameter is obtained by adjusting the first model parameter;

[0217] The parameter adjustment module 805 is used to adjust the prediction model obtained through training according to the target model parameters, and the adjusted prediction model is used for the next training until the prediction model obtained through training converges.

[0218] It should be noted that the above explanation of the prediction method is also applicable to the prediction device of this embodiment and will not be repeated here.

[0219] In this embodiment, by obtaining multiple object data of each object, wherein the multiple object data include: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and the number of the first object data and the number of the second object data satisfy a condition; obtaining multiple data features corresponding to each object data; training the prediction model at least once according to the multiple data features, wherein, in each training, the first model parameters of the prediction model obtained by training are obtained and sent; receiving the target model parameters, wherein the target model parameters are obtained by adjusting the first model parameters; adjusting the prediction model obtained by training according to the target model parameters, and using the adjusted prediction model for the next training until the trained prediction model converges. Thus, the target model parameters can be determined in combination with the multiple first model parameters during the model iterative training process and provided to the second device to adjust the prediction model obtained by training for the next round of model training process, thereby realizing deep mining of the data, and then improving the model learning ability, realizing comprehensive improvement of the model performance and further optimization of the accuracy.

[0220] Fig. 9 A block diagram of an exemplary computer device suitable for implementing embodiments of the present disclosure is shown. Fig. 9 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0221] like Fig. 9 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0222] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnection (PCI) bus.

[0223] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0224] The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Fig. 9 Not shown, often called a "hard drive").

[0225] although Fig. 9 Not shown, a disk drive for reading and writing a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing a removable non-volatile optical disk (e.g., a compact disc read only memory (Compact Disc Read Only Memory; hereinafter referred to as: CD-ROM), a digital versatile disc read only memory (Digital Video Disc Read Only Memory; hereinafter referred to as: DVD-ROM) or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present disclosure.

[0226] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described in the present disclosure.

[0227] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), one or more devices that enable a human body to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed via an input / output (I / O) interface 22. In addition, the computer device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the computer device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0228] The processing unit 16 executes the programs stored in the system memory 28 to perform various functional applications and data processing, such as implementing the prediction method mentioned in the above embodiments.

[0229] In order to implement the above embodiments, the present disclosure further proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the prediction method proposed in the above embodiments of the present disclosure is implemented.

[0230] In order to implement the above embodiments, the present disclosure further proposes a computer program product. When an instruction processor in the computer program product executes, the prediction method proposed in the above embodiments of the present disclosure is executed.

[0231] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0232] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

[0233] It should be noted that, in the description of the present disclosure, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present disclosure, unless otherwise specified, the meaning of "plurality" is two or more.

[0234] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present disclosure includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present disclosure belong.

[0235] It should be understood that the various parts of the present disclosure can be implemented in hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0236] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0237] In addition, each functional unit in each embodiment of the present disclosure may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0238] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0239] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0240] Although the embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present disclosure. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present disclosure.

Claims

1. A prediction method, characterized in that: The method is executed by a first device; the method comprising: Receiving a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by each training of the prediction model by the second device; generating target model parameters according to a plurality of said first model parameters; and The target model parameters are sent, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

2. The method according to claim 1, characterized in that The method further comprises: Acquire global distribution information of a global prediction model, wherein the global distribution information is used to describe the output distribution of the global prediction model, and the global prediction model and the prediction model of each second device are jointly trained; The global distribution information is sent.

3. A prediction method, characterized in that: The method is executed by a second device; the method comprising: Acquire a plurality of object data of each object, wherein the plurality of object data includes: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and a quantity of the first object data and a quantity of the second object data satisfy a condition; Acquire a plurality of data features corresponding to each of the object data; Training the prediction model at least once according to the plurality of data features, wherein, during each training, obtaining and sending a first model parameter of the prediction model obtained through the training; receiving a target model parameter, wherein the target model parameter is obtained by adjusting the first model parameter; The trained prediction model is adjusted according to the target model parameters, and the adjusted prediction model is used for the next training until the trained prediction model converges.

4. The method according to claim 3, characterized in that The method further comprises: Global distribution information is received, wherein the global distribution information is used to describe the output distribution of a global prediction model, and the global prediction model and the prediction model are jointly trained.

5. The method according to claim 4, characterized in that The prediction model obtained by training is adjusted according to the target model parameters, including: Obtaining output distribution information of the prediction model obtained through the training; Determining divergence information according to the global distribution information and the output distribution information; Processing the target model parameters according to the divergence information to obtain second model parameters; The trained prediction model is adjusted according to the second model parameters.

6. The method according to claim 3, characterized in that The obtaining of the plurality of object data of each object comprises: Acquire at least one of the first object data and a plurality of the second object data; Sampling processing is performed on at least one of the first object data so that a quantity of the first object data obtained by the sampling processing satisfies a condition with a quantity of the second object data.

7. The method according to claim 3, characterized in that The object data includes at least one of the following: Wireless connection rate; Radio Resource Control RRC connection rate; Uplink physical resource block PRB utilization; Downlink PRB utilization; Evolved Radio Access Bearer (E-RAB) establishment success rate.

8. The method according to claim 3, characterized in that The obtaining of the plurality of object data of each object comprises: Collect multiple initial data for each subject; Determine the feature missing rate of each of the initial data; Determine a feature missing rate that is less than a missing rate threshold among the plurality of feature missing rates; The initial data to which the feature missing rate less than the missing rate threshold belongs is taken as the object data.

9. The method according to claim 3, characterized in that: The step of training an initial prediction model according to the plurality of object data and a plurality of data features corresponding to each of the object data to obtain a target prediction model comprises: In a current round of iterative training, obtaining a first weight corresponding to the first object data and a second weight corresponding to the second object data; Determining a first prediction error corresponding to the first object data according to the initial prediction model, and determining a second prediction error corresponding to the second object data according to the initial prediction model; The first weight is adjusted according to the first prediction error, and the second weight is adjusted according to the second prediction error. The adjusted first weight and second weight are used for the next round of iterative training until the trained prediction model converges, and the trained prediction model is used as the target prediction model.

10. The method according to claim 9, characterized in that The determining a first prediction error corresponding to the first object data according to the initial prediction model, and determining a second prediction error corresponding to the second object data according to the initial prediction model, comprises: Perform prediction based on the first object data according to the initial prediction model to obtain a first prediction tendency; Perform prediction based on the second object data according to the initial prediction model to obtain a second prediction tendency; The first prediction error is determined based on the first prediction tendency and the first evaluation tendency, and the second prediction error is determined based on the second prediction tendency and the second evaluation tendency.

11. A prediction device, characterized in that: The method is executed by a first device; the apparatus comprises: A receiving module, configured to receive a first model parameter corresponding to each second device, wherein the first model parameter is a model parameter obtained by training a prediction model of the second device each time; A generating module, configured to generate target model parameters according to a plurality of said first model parameters; and A sending module is used to send the target model parameters, wherein the target model parameters are used by the second device to adjust the prediction model obtained through training, and the adjusted prediction model is used for the next training of the prediction model.

12. A prediction device, characterized in that: The method is executed by a second device; the apparatus comprises: A first acquisition module is used to acquire a plurality of object data of each object, wherein the plurality of object data includes: first object data of a first evaluation tendency and second object data of a second evaluation tendency, and a quantity of the first object data and a quantity of the second object data satisfy a condition; A second acquisition module, used for acquiring a plurality of data features corresponding to each of the object data; A model training module, used for training the prediction model at least once according to the plurality of data features, wherein, during each training, a first model parameter of the prediction model obtained through training is obtained and sent; A receiving module, configured to receive a target model parameter, wherein the target model parameter is obtained by adjusting the first model parameter; The parameter adjustment module is used to adjust the prediction model obtained through training according to the target model parameters, and the adjusted prediction model is used for the next training until the prediction model obtained through training converges.

13. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 10.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 10 when executed by a processor.

15. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 10 when being executed by a processor.

Images