Scheduling method and device

Abstract

The invention discloses a scheduling method and device, and the method comprises the steps: inputting a network state parameter to a first classification engine if the network state parameter of a target network meets a network reset condition, and obtaining a first prediction result; if the first prediction result meets a confidence coefficient condition, determining an execution decision result of a target network scheduling strategy based on the first prediction result; the execution decision result comprises execution of the target network scheduling strategy and non-execution of the target network scheduling strategy; wherein the first prediction result is the success rate of executing the target network scheduling strategy; when the first prediction result meets the confidence coefficient condition, it is represented that the confidence coefficient of the first prediction result output by the first classification engine is higher than the target confidence coefficient; the target network scheduling strategy is used for resetting the target network.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of communication technology, and relate to but are not limited to a scheduling method and device. Background Art

[0002] When data stagnation (Data Stall) occurs in an electronic device, it is determined whether the scheduling strategy can be executed based on the current network status. If it can, a scheduling behavior (action) is initiated. Data Stall refers to when the device encounters data transmission stagnation or delay during the network connection process. Determining whether to execute an action is performed based on a linear classifier (if-else) structure, which has poor fitting ability. The scheduling behavior is inefficient, resulting in low scheduling efficiency, which affects the power consumption of electronic devices on the one hand, and increases the load on the network on the other. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a scheduling method and device.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides a scheduling method, including:

[0006] If the network state parameter of the target network meets the network reset condition, the network state parameter is input into the first classification engine to obtain a first prediction result;

[0007] If the first prediction result satisfies the confidence condition, determining an execution decision result of the target network scheduling strategy based on the first prediction result; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy;

[0008] Among them, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result meets the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

[0009] In a second aspect, an embodiment of the present application provides a scheduling device, including:

[0010] A first classification module, configured to input the network state parameters into a first classification engine to obtain a first prediction result if the network state parameters of the target network meet the network reset condition;

[0011] A first determination module is used to determine an execution decision result of the target network scheduling strategy based on the first prediction result if the first prediction result meets the confidence condition; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy;

[0012] Among them, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result meets the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, the memory storing a computer program that can be run on the processor, and when the processor executes the program, if the network status parameters of the target network meet the network reset condition, the network status parameters are input into a first classification engine to obtain a first prediction result; if the first prediction result meets the confidence condition, based on the first prediction result, an execution decision result of the target network scheduling strategy is determined; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy; wherein the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result meets the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

[0014] In a fourth aspect, an embodiment of the present application provides a storage medium storing executable instructions for executing by a processor to implement, if the network state parameters of the target network meet the network reset condition, input the network state parameters to the first classification engine to obtain a first prediction result; if the first prediction result meets the confidence condition, based on the first prediction result, determine the execution decision result of the target network scheduling strategy; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy; wherein, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result meets the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network. .

[0015] In a fifth aspect, a computer program product includes a computer program or an instruction. When the computer program or the instruction is executed by a processor, if the network status parameters of the target network meet the network reset condition, the network status parameters are input into a first classification engine to obtain a first prediction result; if the first prediction result meets the confidence condition, based on the first prediction result, an execution decision result of the target network scheduling strategy is determined; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy; wherein the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result meets the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A schematic diagram of an implementation flow of a scheduling method provided in an embodiment of the present application;

[0017] Figure 2A A schematic diagram of an implementation flow of a scheduling method provided in an embodiment of the present application;

[0018] Figure 2B A schematic diagram of a training encoder provided in an embodiment of the present application;

[0019] Figure 3A A schematic diagram of an implementation flow of a scheduling method provided in an embodiment of the present application;

[0020] Figure 3B A schematic diagram of an implementation flow of a scheduling method provided in an embodiment of the present application;

[0021] Figure 4 A flowchart for implementing a hybrid scheduling strategy provided in an embodiment of the present application;

[0022] Figure 5 A schematic diagram of the structure of a scheduling device provided in an embodiment of the present application;

[0023] Figure 6 A hardware entity schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the specific technical solution of the embodiments of the present application will be further described in detail below in conjunction with the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application, but are not used to limit the scope of the present application.

[0025] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0026] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0028] When an electronic device uses a network to interact with data, due to the poor network environment affecting communication, you can try to restore normal data communication by resetting the network connection of the electronic device. However, restoring data communication by resetting the network is not suitable for all network communication scenarios, that is, resetting the network cannot solve the problem of blocked network communication. In order to avoid power consumption loss caused by unnecessary network reset, the present application embodiment provides a scheduling method, such as Figure 1 As shown, the method includes:

[0029] Step S110: If the network status parameters of the target network meet the network reset condition, the network status parameters are input into the first classification engine to obtain a first prediction result;

[0030] Wherein, the first prediction result is the success rate of executing the target network scheduling strategy;

[0031] Here, the network status parameter is an indicator that characterizes the network connection status and performance of the electronic device. The network status parameter includes at least one of the following: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Signal to Interference plus Noise Ratio (SINR), frequency information, Dual SIM Dual Standby (DSDS), Dual SIM Data Registration State (Data Reg State), Voice State, Cell Received Signal Strength Indication (RSSI), Intra Inter switching, Backoff in Data registration, Data registration state (data reg state) and cause, where,

[0032] The reference signal received power is an important indicator used in mobile communication networks to measure the power strength of the downlink signal from the base station antenna to the user equipment. It indicates the power level of the reference signal sent by the Long Term Evolution (LTE) or 5G base station. This reference signal is sent periodically by the base station and can be used by mobile devices to evaluate the quality and strength of the network connection. The larger the reference signal received power value, the stronger the received signal and the better the signal quality.

[0033] RSRS is an indicator of the quality of LTE reference signal reception. Different LTE candidate cells can be ranked according to signal quality and used as input for handover and cell reselection decisions.

[0034] Signal to Interference and Noise Ratio refers to the ratio of the strength of the received useful signal to the strength of the received interference signal (noise and interference), that is, the "Signal to Noise Ratio". The signal to noise ratio is a standard to measure the quality of the signal. In wireless communication, the higher the signal to noise ratio, the better the signal quality and the stronger the anti-interference ability.

[0035] Frequency can represent the absolute frequency value, which is generally the center frequency of the modulated signal. Frequency is a number given to a fixed frequency, which is used to identify and distinguish different frequencies. In mobile communication networks, frequency information is very important for the communication between base stations and user equipment. By selecting the appropriate frequency, interference between different cells can be avoided and the overall performance of the network can be improved.

[0036] Dual-SIM dual-standby means that a mobile user equipment (UE) can be inserted with two Subscriber Identity Module (SIM) cards, and both cards can be in standby mode at the same time.

[0037] Dual SIM card data registration status refers to the data connection registration status of each of the two SIM cards in a dual SIM card dual standby mobile user device. In this state, the mobile user device can identify and manage the data connections of the two SIM cards, allowing users to switch data connections between the two cards, or use the data services of the two cards at the same time (if the mobile user device and the network support it). In a dual SIM card dual standby mobile user device, users can view the data registration status of each SIM card to ensure that they are all properly connected to the network and can transmit data.

[0038] Voice status refers to the status of a mobile user device when making a voice call. In a dual-SIM dual-standby mobile user device, users can view the voice status of each SIM card to ensure that they can make voice calls normally. Voice status may include different states such as in call, in standby, busy, etc. These states can help users understand the current call status of the mobile user device.

[0039] The cell received signal strength indicator reflects the total transmission power received by a specific user. It indicates the signal strength received by the mobile user device from the base station or cell. The higher the RSSI value, the stronger the signal received by the mobile user device; the lower the RSSI value, the weaker the received signal. It is one of the important indicators to measure the network coverage quality and signal strength. In a dual-SIM dual-standby mobile user device, each SIM card may be connected to a different base station or cell, so each SIM card has its own RSSI value. Users can view the RSSI value of each SIM card to understand their respective network connection quality and signal strength.

[0040] Intra-RAT handover refers to the handover of cells within the same Radio Access Technology (RAT). When a mobile user device moves from one cell to another, and both cells belong to the same RAT, an intra-RAT handover occurs. This handover usually occurs in areas with overlapping coverage in order to maintain the quality of the user's network connection.

[0041] In a mobile communication network, when a mobile user device attempts to register with the network, it may encounter some registration failures. To deal with this situation, the network usually adopts a mechanism called "Backoff".

[0042] Backoff in data registration means that during the data registration process, if the device fails to register for the first time (perhaps due to network congestion, poor signal quality, or other reasons), it can wait for a period of time according to the backoff mechanism and try to register again. If it fails again, the waiting time will be longer, and so on. This mechanism helps balance the network load and registration success rate.

[0043] Data Registration State refers to the state of a mobile user device registering a data connection in the network. This state can indicate whether the device has successfully registered with the network, whether it is attempting to register, or whether registration has failed.

[0044] Cause: During the registration process, if the device encounters any problems or fails, the network can usually return a cause code to indicate the specific reason for the registration failure. This cause code can help users and devices understand the reason for the registration failure and take appropriate measures to solve the problem.

[0045] For example, if Data Reg State shows that registration has failed, and Cause indicates that it is caused by network congestion, the mobile user equipment may try to wait for a period of time before retrying registration, or switch to another network environment for retrying.

[0046] During the implementation process, it can be determined based on the network status parameters that there are some problems with the current configuration or status of the target network, and the network connection can be restored or improved by resetting the network settings. The network status parameters of the target network meet the network reset conditions, which may include but are not limited to the following situations: unstable network connection: frequent disconnection, slow connection speed or inability to connect to the network; network configuration errors: such as IP address conflicts, Domain Name System (DNS) settings errors, etc. Malware attacks: such as viruses, Trojans, etc. that cause network settings to be tampered with; system updates or upgrades: after the system is updated, the network settings may be incompatible or need to be reset.

[0047] The first classification engine can classify the acquired network state parameters through machine learning algorithms or predefined classification systems to obtain classification results. For example, the first classification engine can be a logistic regression (LR) classifier model trained based on a deep neural network (DNN). The classification model can implement binary classification or be extended to implement multi-classification. The classification model constructs a linear model, linearly weights the input features, and then maps the results to the (0, 1) interval through a sigmoid function to obtain a probability prediction value. Based on this probability prediction value, the probability that the sample belongs to a certain category can be determined.

[0048] In some embodiments, the network status parameters are input into the LR classifier model to obtain the success rate of executing the target network scheduling strategy, that is, the first prediction result. For example, the obtained network status parameters can be input into the LR classifier model to obtain a probability prediction value of 0.9 for the reset to take effect, and it can be determined that the problem of unstable network connection can be solved by resetting.

[0049] Step S120: If the first prediction result meets the confidence condition, determine the execution decision result of the target network scheduling strategy based on the first prediction result; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy;

[0050] Among them, the first prediction result satisfies the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; and the target network scheduling strategy is used to reset the target network.

[0051] Here, satisfying the confidence condition indicates that the confidence of the first prediction result is higher than the target confidence. The confidence of the prediction result is a measure of the reliability of the first classification engine for the first prediction result, usually expressed in the form of probability, ranging from 0 to 1. A prediction with a confidence of 1 means that the model is very sure that the prediction result is correct, while a confidence of 0 means that the model does not believe the prediction result at all.

[0052] The target network scheduling policy is used to reset the target network, and can be reset by at least one of the following network scheduling policies: disconnecting or cleaning up all network connections (Cleanup all connections), re-registering (Re-register) and restarting the electronic device (Restarting radio).

[0053] During the implementation process, it is determined that the first prediction result meets the confidence condition, indicating that the first prediction result is credible and correct. Based on the first prediction result, it can be determined whether to execute the target network scheduling strategy or not.

[0054] For example, when the first prediction result is that the probability of executing the target action (Action) to reestablish the network connection is 0.9, it is determined that the probability of successful execution 0.9 meets the confidence condition, wherein the target action can be restarting the electronic device, and the electronic device can be restarted to restore the network connection. Since it has been determined that the confidence of the target action meets the confidence condition, the success rate of executing the target action to restore the network connection can be effectively improved.

[0055] In an embodiment of the present application, when the network status parameters of the target network meet the network reset condition, the network status parameters are input into the first classification engine to obtain the first prediction result; when it is determined that the first prediction result meets the confidence condition, the execution decision result of the target network scheduling strategy can be determined based on the first prediction result. In this way, the target network scheduling strategy can be determined under the condition that the first prediction result meets the confidence, which effectively improves the success rate of executing the target network scheduling strategy to restore network connection. The effectiveness of the scheduling strategy is improved, that is, the scheduling efficiency is improved, the power consumption of electronic equipment is reduced, and the load on the network is reduced.

[0056] In some embodiments, if the probability value corresponding to the first prediction result is within a first probability interval, the confidence of the first prediction result output by the first classification engine is higher than the target confidence; if the probability value corresponding to the first prediction result is within a second probability interval, the confidence of the first prediction result output by the first classification engine is not higher than the target confidence;

[0057] The first probability interval includes a first sub-probability interval and a second sub-probability interval, and the second probability interval is located between the first sub-probability interval and the second sub-probability interval.

[0058] Here, the first probability interval can be set according to actual needs. The first probability interval includes a first sub-probability interval and a second sub-probability interval. For example, the first sub-probability interval and the second sub-probability interval can be determined using the following formula (1):

[0059] bias = |2p-1.0| (1);

[0060] Among them, bias is the deviation, and p is the probability value corresponding to the first prediction result.

[0061] In the implementation process, a preset deviation threshold value for comparison with the deviation value can be set to 0.6. In this way, when p is 0 to 0.2, or when p is 0.8 to 1, the deviation value is greater than 0.6, then it can be determined that the confidence of the first prediction result is higher than the target confidence, that is, the first sub-probability interval can be set to 0 to 0.2, and the second sub-probability interval can be set to 0.8 to 1.

[0062] During the implementation process, when it is determined that the probability value corresponding to the first prediction result is within the first probability interval, it can be determined that the confidence of the first prediction result is higher than the target confidence, that is, the confidence condition is met. For example, when the probability value corresponding to the first prediction result is 0.9, it belongs to the second sub-probability interval of 0.8 to 1, then it can be determined that the confidence of the first prediction result is higher than the target confidence; when the probability value corresponding to the first prediction result is 0.1, it belongs to the first sub-probability interval of 0 to 0.2, then it can be determined that the confidence of the first prediction result is higher than the target confidence.

[0063] The second probability interval can be set between the first sub-probability interval and the second sub-probability interval. For example, the second probability interval can be determined using the above formula (1). When p is between 0.2 and 0.8, and the deviation value is less than 0.6, it can be determined that the confidence level of the first prediction result is not higher than the target confidence level, that is, the second probability interval can be set to 0.2 to 0.8.

[0064] During implementation, when it is determined that the probability value corresponding to the first prediction result is within the second probability interval, it can be determined that the confidence of the first prediction result is not higher than the target confidence, that is, the confidence condition is not met. For example, when the probability value corresponding to the first prediction result is 0.6, it belongs to the second probability interval of 0.2 to 0.8, then it can be determined that the confidence of the first prediction result is not higher than the target confidence.

[0065] In the embodiment of the present application, if the probability value corresponding to the first prediction result is within the first probability interval, the confidence of the first prediction result is higher than the target confidence; if the probability value corresponding to the first prediction result is within the second probability interval, the confidence of the first prediction result is not higher than the target confidence. In this way, it is possible to determine whether the confidence of the first prediction result meets the target confidence by presetting the probability interval.

[0066] In some embodiments, the above scheduling method is as follows Figure 2A As shown, this can also be achieved by executing the following steps:

[0067] Step S210: if the first prediction result does not meet the confidence condition, input the network state parameter into a target encoder to obtain a network feature code corresponding to the network state parameter;

[0068] Here, the target encoder is used to re-encode the network state parameters so that the output network feature encoding can meet the data features corresponding to the confidence condition.

[0069] In some embodiments, the target encoder can be obtained using Generative Adversarial Networks (GANs). GANs consist of two main parts: a generator and a discriminator. The goal of the generator (encoder) is to generate data similar to real data, while the goal of the discriminator is to distinguish between generated data and real data. Through continuous training and confrontation, the encoder can continuously improve the quality of generated data, and the discriminator can also continuously improve its ability to distinguish between real data and generated data.

[0070] Adopting the adversarial training method of domain adversarial learning to obtain the target encoder can reduce the problem of data distribution mismatch between different domains. For example, as shown in Table 1 above, there is a difference in data distribution between dataset A and dataset B. The data in dataset B can be input into the target encoder to obtain feature encoding that matches the features of dataset A.

[0071] Step S220: input the network feature code into the first classification engine to obtain a second prediction result;

[0072] During implementation, the network feature code can be input into the first classification engine, and classification prediction can be performed again to obtain a second prediction result.

[0073] Step S230: Based on the second prediction result, determine the execution decision result of the target network scheduling strategy.

[0074] During the implementation process, it can be determined based on the second prediction result whether the corresponding confidence meets the confidence condition. When it is determined that the confidence condition is met, the execution decision result of the target network scheduling strategy is determined based on the second prediction result; the execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy.

[0075] In the embodiment of the present application, if the first prediction result does not meet the confidence condition, the network state parameter is input into the target encoder to obtain the network feature code corresponding to the network state parameter; the network feature code is input into the first classification engine to obtain the second prediction result; based on the second prediction result, the execution decision result of the target network scheduling strategy is determined. In this way, the network state parameter is re-encoded to obtain the network feature code close to the distribution that meets the confidence condition, which effectively improves the confidence corresponding to the second prediction result obtained by classifying the network feature code.

[0076] In some embodiments, the target encoder is used to map second data that conforms to characteristics of a second data set to first data that conforms to characteristics of a first data set; a first confidence level of a prediction result output by the first classification engine for the first data is higher than a second confidence level of a prediction result output by the first classification engine for the second data.

[0077] In some embodiments, a soft margin can be pre-set according to a support vector machine (SVM) soft margin to determine whether the probability value corresponding to the first prediction result meets the confidence condition. The soft margin is used to distinguish the first data set from the second data set.

[0078] Table 1 is a statistical table of classification accuracy (Accuracy) of a classifier provided in an embodiment of the present application.

[0079] Table 1

[0080] quantity Accuracy Total dataset 475636 66.8％ Dataset A 60790 99.1％ Dataset B 414846 62.1％

[0081] As shown in Table 1 above, the total number of data sets is 475636, and the classification accuracy is 66.8%; the number of data sets A (the first data set) is 60790, and the classification accuracy is 99.1%, which can be marked as high-confidence classification data; the number of data sets B (the second data set) is 414846, and the classification accuracy is 62.1%, which can be marked as low-confidence classification data.

[0082] Since the accuracy of the classifier validation data set obtained through training, that is, the total data set, is 66.8%, the following steps can be performed based on the idea of ​​SVM soft interval to determine whether the obtained result is credible, that is, whether the input data is data set A:

[0083] Step 1: Preset a soft interval to 0.6;

[0084] Here, the soft interval is the preset deviation threshold.

[0085] Step 2: Calculate the current deviation using the following formula (1);

[0086] bias = |2p-1.0| (1);

[0087] Among them, bias is the deviation, and p is the probability value corresponding to the first prediction result.

[0088] Step C, defining the data set A with the deviation greater than the soft interval;

[0089] Here, if the deviation is greater than the soft interval, it can be characterized that the confidence of the first prediction result meets the confidence condition. That is, the input network state parameters correspond to data set A as shown in Table 1 above, and the classification accuracy of data set A verified by the classifier is 99.1%, and the prediction accuracy of data set A is high.

[0090] Step 3: Define the data set whose deviation is smaller than the soft margin as data set B.

[0091] Here, if the deviation is greater than the soft interval, it can be characterized that the confidence of the first prediction result does not meet the confidence condition. That is, the input network state parameters correspond to data set B as shown in Table 1 above, and the classification accuracy of data set B verified by classifier is 62.1%, and the prediction accuracy of data set B is low.

[0092] Here, we can adopt the idea of ​​deep adversarial neural network (DANN) domain adversarial learning to learn an encoder.

[0093] Figure 2B A schematic diagram of a training encoder provided in an embodiment of the present application, such as Figure 2B As shown, the schematic diagram includes: an encoder 21, a label predictor 22 and a discriminator 23. In the implementation process, the encoder can be trained by the following steps:

[0094] Step A, training the discriminator 23;

[0095] Get data x from dataset B n , input data x n After being processed by encoder 21, it is converted into a feature vector x f . It can be expressed as the following formula (2):

[0096] x f =Encoder(x n ) (2);

[0097] The loss function can be expressed as the following formula (3):

[0098] L = Discriminator (x f )-Discriminator(x r ) (3);

[0099] Among them, Discriminator (x f ) is the discriminator for the feature vector x f Output of Discriminator(x r ) is the data x obtained by the discriminator for the data set A r The output of the loss function is used to calculate the discriminator's f The output and data x r In implementation, this difference can be used to evaluate model performance and adjust the parameters of the Discriminator network accordingly.

[0100] Step B, training encoder 21;

[0101] Eigenvector x f After being processed by the decoder network, try to decode it back to the original data x n (or as close an approximation as possible). The decoding process can be represented by the following formula (4):

[0102]

[0103] Among them, the loss function can be expressed as the following formula (5):

[0104]

[0105] Among them, Discriminator (x f ) is the discriminator for the feature vector x f Output; x n Input the label predictor to get y n ;Will Input category predictor 22 to get The binary cross entropy loss BCELoss is used to evaluate the output y of the class predictor n and The difference between the distributions.

[0106] The above loss function is used to evaluate the model performance, and the parameters of the encoder 21 are adjusted accordingly. The final output is the trained target encoder. The obtained target encoder can map the second data that meets the characteristics of the second data set to the first data that meets the characteristics of the first data set, wherein the first confidence of the prediction result output by the first classification engine for the first data is higher than the second confidence of the prediction result output by the first classification engine for the second data.

[0107] In the embodiment of the present application, the target encoder can realize mapping the second data that meets the characteristics of the second data set to the first data that meets the characteristics of the first data set. In this way, the distribution difference between the second data and the first data can be effectively reduced, and after the second data is encoded by the target encoder, the confidence of the prediction result of the encoded data input into the first classification engine can be effectively improved.

[0108] In some embodiments, the above scheduling method is as follows Figure 3A As shown, this can also be achieved by performing the following steps:

[0109] Step S301: If the first prediction result does not meet the confidence condition, input the time parameter corresponding to the network state parameter into the second prediction model to obtain a third prediction result;

[0110] Here, the second prediction model can perform prediction based on the time parameters corresponding to the network parameters to obtain a third prediction result.

[0111] For example, as shown in Table 2 below, the probability of successful execution of corresponding actions at different times is dynamically updated.

[0112] Table 2

[0113] time Probability of success 1 0.9 2 0.8 3 0.1 … …

[0114] When the success probability P is greater than the threshold (0.5 for example), the model output is 1, and when P is less than the threshold, the model output is 0.

[0115] The update rule of P is as follows:

[0116]

[0117] Among them, t represents the time when Data Stall occurs; λ is a pre-set hyperparameter, which can be a decimal between 0 and 1; p tRepresents the probability of successful scheduling at time t.

[0118] Step S302: Based on the third prediction result, determine the execution decision result of the target network scheduling strategy.

[0119] During the implementation process, the execution decision result can be determined based on the third prediction result output by the second prediction model. For example, as shown in Table 2 above, when the time parameter 3 is input into the second prediction model and the success probability is 0.1, it is determined not to execute the target network scheduling strategy. In this scenario, there may be other factors that cause network instability, and the network connection cannot be restored or improved by executing the target network scheduling strategy. For example, when a concert is in progress or on the subway during rush hour, the network instability is caused by an overload of users using the network, and resetting the electronic device will not restore the network connection.

[0120] In the embodiment of the present application, if the first prediction result does not meet the confidence condition, the time parameter corresponding to the network state parameter is input into the second prediction model to obtain a third prediction result; based on the third prediction result, the target network scheduling strategy execution decision result is determined. In this way, when the first prediction result is unreliable, the second prediction model can be used to obtain the execution decision result based on the time parameter. Since the prediction of the time parameter is added, the efficiency of the execution decision result can be effectively improved.

[0121] In some embodiments, the above scheduling method is as follows Figure 3B As shown, this can also be achieved by performing the following steps:

[0122] Step S311: if the first prediction result does not meet the confidence condition, input the time parameter corresponding to the network state parameter into the second prediction model to obtain a third prediction result;

[0123] In the implementation process, the above step S311 is performed in the same manner as step S301, and the second prediction model, ie, the time model, can be used to obtain the third prediction result based on the time parameter.

[0124] Step S312: determining a target prediction result based on the first prediction result and the third prediction result;

[0125] Here, the first prediction result is obtained based on the classification model, and the third prediction result is obtained based on the time model.

[0126] When determining the target prediction result, combining the first prediction result and the third prediction result is a comprehensive consideration process. Combining the two prediction results to determine the target prediction result can improve the accuracy and reliability of the prediction.

[0127] During the implementation process, the first prediction result and the third prediction result can be compared. Determine the similarity, difference and consistency between the two prediction results. If the prediction results are the same, the target execution decision result can be determined based on the same prediction results; if the prediction results differ greatly, it is necessary to analyze the possible reasons. If a prediction result performs better in a specific aspect (such as accuracy, stability or interpretability), it can be given a higher weight. Determine the target prediction result by comparing the weights of the two prediction results. For example, when the weight of the first prediction result is greater than the weight of the third prediction result, the target prediction result can be determined based on the first prediction result.

[0128] Step S313: Based on the target prediction result, determine the target execution decision result of the target network scheduling strategy.

[0129] In the embodiment of the present application, the target prediction result is determined based on the first prediction result and the third prediction result; and the target execution decision result of the target network scheduling strategy is determined based on the target prediction result. In this way, the accuracy and reliability of the prediction can be improved by combining the two prediction results to determine the target prediction result.

[0130] In some embodiments, the above step S312 "determining a target prediction result based on the first prediction result and the third prediction result" can be implemented by the following steps:

[0131] Step 3121: If the first execution decision result determined based on the first prediction result is the same as the second execution decision result determined based on the third prediction result, use the first prediction result or the third prediction result as the target prediction result;

[0132] During implementation, if the first execution decision result determined by the first prediction result is the same as the second execution decision result determined based on the third prediction result, the decision result can be characterized as accurate and reliable, and the first prediction result or the third prediction result is used as the target prediction result. For example, if both the first prediction result and the third prediction result indicate that resetting the electronic device is effective, the operation of resetting the electronic device can be performed.

[0133] Step 3121: if the first execution decision result determined based on the first prediction result is different from the second execution decision result determined based on the third prediction result, respectively obtain a first accumulated reward value corresponding to the first prediction result and a second accumulated reward value corresponding to the third prediction result;

[0134] Here, the first execution decision result determined by the first prediction result is different from the second execution decision result determined by the third prediction result. For example, the first prediction result determines that resetting the electronic device immediately is effective, and the third prediction result determines that resetting the electronic device in the current time period is not effective.

[0135] A first cumulative reward value corresponding to the first prediction result and a second cumulative reward value corresponding to the third prediction result can be obtained, wherein the first cumulative reward value can be determined based on historical statistics of the execution efficiency of the first execution decision result corresponding to the first result under the same circumstances, and the second cumulative reward value can be determined based on the execution efficiency of the second execution decision result corresponding to the third prediction result previously executed under the same circumstances.

[0136] Step 3122: If the first cumulative reward value is higher than the second cumulative reward value, use the first prediction result as the target prediction result;

[0137] During the implementation process, it is determined that the first cumulative reward value is higher than the second cumulative reward value, that is, it can be determined that under the same circumstances, the first execution decision result corresponding to the first prediction result has a higher efficiency, and the first prediction result can be used as the target prediction result.

[0138] Step 3123: If the second cumulative reward value is higher than the first cumulative reward value, use the third prediction result as the target prediction result.

[0139] During the implementation process, it is determined that the second cumulative reward value is higher than the first cumulative reward value, that is, it can be determined that under the same circumstances, the first execution decision result corresponding to the third prediction result has a higher efficiency, and the third prediction result can be used as the target prediction result.

[0140] In the embodiment of the present application, it is first determined whether the execution decision results corresponding to the first prediction result and the third prediction result are the same; if they are determined to be the same, the first prediction result or the third prediction result is used as the target prediction result; if they are determined to be different, the target prediction result can be determined based on the first cumulative reward value corresponding to the first prediction result and the second cumulative reward value corresponding to the third prediction result. In this way, when it is determined that the execution decision results corresponding to the first prediction result and the third prediction result are different, a target prediction result with high efficiency can be determined based on the cumulative reward value obtained by historical statistics.

[0141] In some embodiments, the step 3121 above of “respectively obtaining the first accumulated reward value corresponding to the first prediction result and the second accumulated reward value corresponding to the third prediction result” can be implemented by the following steps:

[0142] Step A, obtaining the target state corresponding to the first prediction result and the third prediction result;

[0143] Here, the target state can be determined as shown in Table 3 below based on the output result (first prediction result) of the classifier (first classification engine) and the time model result (third prediction result).

[0144] Table 3

[0145] state Classifier Results Time model results State 0 0 0 State 1 0 1 State 2 1 0 State 3 1 1

[0146] For example, as shown in Table 3 above, when it is determined that the first prediction result is 0 and the third prediction result is 1, the corresponding target state is determined to be state 1.

[0147] Step B, based on the target state, obtaining a first cumulative reward value corresponding to the first prediction result and a second cumulative reward value corresponding to the third prediction result;

[0148] Wherein, the target states are different, and the first accumulated reward value corresponding to the first prediction result and / or the second accumulated reward value corresponding to the third prediction result are different.

[0149] Here, the first cumulative reward and the second cumulative reward are obtained based on the target state, which can be obtained based on the following Table 4.

[0150] Table 4

[0151] First prediction result The third prediction result State 0 First cumulative reward Second cumulative reward State 1 First cumulative reward Second cumulative reward State 2 First cumulative reward Second cumulative reward State 3 First cumulative reward Second cumulative reward

[0152] When it is determined to be state 0, the first cumulative reward of the first prediction result and the second cumulative reward of the third prediction result can be obtained correspondingly using Table 4 as above.

[0153] In the embodiment of the present application, the target state corresponding to the first prediction result and the third prediction result can be obtained first; then based on the target state, the first cumulative reward value corresponding to the first prediction result and the second cumulative reward value corresponding to the third prediction result can be obtained. In this way, the first cumulative reward value and the second cumulative reward value can be determined by using the target state.

[0154] In some embodiments, the present application also provides a method for updating the accumulated reward value, which can be implemented by the following steps:

[0155] Step C: if the execution based on the target execution decision result is effective, the first cumulative reward value or the second cumulative reward value under the target state is positively updated; the cumulative reward value after the positive update is higher than the cumulative reward value before the positive update;

[0156] During the implementation process, when the target execution decision result is executed and becomes effective, if it is based on the first prediction result, the first cumulative reward value is positively updated; if it is based on the third prediction result, the second cumulative reward value is positively updated. For example, the reward value that takes effect once can be set to 2, and based on the first prediction result, the first cumulative reward value can be increased by 2.

[0157] Step D: If the execution based on the target execution decision result is not effective, the first cumulative reward value or the second cumulative reward value under the target state is reversely updated; the cumulative reward value after the reverse update is lower than the cumulative reward value before the reverse update.

[0158] During the implementation process, if the target execution decision result is not effective, if it is based on the first prediction result, the first cumulative reward value is updated in reverse; if it is based on the third prediction result, the second cumulative reward value is updated in reverse. For example, the reward value for a failure to take effect can be set to -2, and then based on the first prediction result, the first cumulative reward value can be reduced by 2.

[0159] In the embodiment of the present application, if the execution based on the target execution decision result is effective, the first cumulative reward value or the second cumulative reward value under the target state is positively updated; if the execution based on the target execution decision result is not effective, the first cumulative reward value or the second cumulative reward value under the target state is reversely updated. In this way, when a positive reward value is obtained after performing an action, it can be determined that the action is correct, and it is more inclined to choose this action in similar situations, thereby forming good behavioral habits. On the contrary, if a negative reward value is obtained, the execution strategy can be adjusted to avoid performing the action again.

[0160] Figure 4 A flow chart for implementing a hybrid scheduling strategy provided in an embodiment of the present application is as follows: Figure 4 As shown, this can be achieved by following the steps below:

[0161] Step S411, obtaining input data;

[0162] Here, the input data can be at least one of the following: reference signal received power, reference signal received quality, signal to interference plus noise ratio, frequency information, dual SIM dual standby, dual SIM data registration status, voice status, cell received signal strength indication, Inter switching, Backoff in Data registration, data registration status and reason.

[0163] Step S412, input the input data into the classifier to obtain the probability of a positive sample;

[0164] During the implementation process, an LR classifier model can be obtained based on deep neural network training, and the probability of positive samples, that is, the probability of executing scheduling, can be obtained.

[0165] Here, the output of the LR classifier is a probability value between 0 and 1, which is used to indicate the possibility that the sample belongs to a certain category, and a binary classification decision is made based on this probability value.

[0166] For example, the LR classifier can output a probability of whether scheduling is executed. When the probability value is 0.9, it indicates that scheduling is executed; when the probability value is 0.2, it indicates that scheduling is not executed.

[0167] Step S413: determining whether to perform scheduling based on the positive sample probability and the soft interval;

[0168] Here, soft margin is a method introduced by support vector machine (SVM) when dealing with linear inseparable problems or noisy data.

[0169] Table 1 above is a statistical table of the classification accuracy of a classifier provided in an embodiment of the present application. As shown in Table 1 above, the total number of data sets is 475636, and the classification accuracy is 66.8%; the number of data sets A is 60790, and the classification accuracy is 99.1%, which can be marked as high-confidence classification data; the number of data sets B is 414846, and the classification accuracy is 62.1%, which can be marked as low-confidence classification data.

[0170] Since the accuracy of the classifier validation data set obtained through training, that is, the total data set, is 66.8%, the following steps can be performed based on the idea of ​​SVM soft interval to determine whether the obtained result is credible, that is, whether the input data is data set A:

[0171] Step 1: Preset a soft interval interval to 0.6;

[0172] Step 2: Calculate the current deviation using the following formula:

[0173] bias = |2p-1.0| (1);

[0174] Step C, defining the data set A as the data set whose deviation is greater than the soft interval interval;

[0175] Step 3: Define the data set whose deviation is less than the soft interval interval as data set B.

[0176] As shown in Table 1 above, the classification accuracy of dataset A verified by classifier is 99.1%, and the classification accuracy of dataset B verified by classifier is 62.1%.

[0177] If it is determined to be data set A, step S414 is executed; if it is determined to be data set B, step S421 is executed.

[0178] Step S414, executing scheduling;

[0179] Step S421, inputting the input data into the encoder for re-encoding to obtain encoded data;

[0180] Here, we can use the idea of ​​deep adversarial network domain adversarial learning to learn an encoder.

[0181] During the implementation process, for low-confidence data with small soft margins, the data is re-encoded by Encoder and then re-entered into the original classifier for classification, which can effectively improve the robustness of the classifier.

[0182] Figure 2B A schematic diagram of a training encoder provided in an embodiment of the present application, such as Figure 2B As shown, the schematic diagram includes: an encoder 21, a label predictor 22 and a discriminator 23. In the implementation process, the encoder can be trained by the following steps:

[0183] Step A, training the discriminator 23;

[0184] Get data x from dataset B n , input data x n After being processed by encoder 21, it is converted into a feature vector x f . It can be expressed as the following formula (2):

[0185] x f =Encoder(x n ) (2);

[0186] The loss function can be expressed as the following formula (3):

[0187] L = Discriminator (x f )-Discriminator(x r ) (3);

[0188] Among them, Discriminator (x f ) is the discriminator for the feature vector x fOutput of Discriminator(x r ) is the data x obtained by the discriminator for the data set A r The output of the loss function is used to calculate the discriminator's f The output and data x r In implementation, this difference can be used to evaluate model performance and adjust the parameters of the Discriminator network accordingly.

[0189] Step B, training encoder 21;

[0190] Eigenvector x f After being processed by the decoder network, try to decode it back to the original data x n (or as close an approximation as possible). The decoding process can be represented by the following formula (4):

[0191]

[0192] Among them, the loss function can be expressed as the following formula (5):

[0193]

[0194] Among them, Discriminator (x f ) is the discriminator for the feature vector x f Output; x n Input the label predictor to get y n ;Will Input category predictor 22 to get The binary cross entropy loss BCELoss is used to evaluate the output y of the class predictor n and The difference between the distributions.

[0195] The above loss function is used to evaluate the model performance and adjust the encoder parameters accordingly. The final output is the trained encoder.

[0196] Step S422, input the encoded data into the classifier to obtain the probability of a positive sample;

[0197] During the implementation process, the obtained encoded data is input into the classifier again to obtain the probability of positive samples.

[0198] Step S423: determining whether to perform scheduling based on the positive sample probability and the soft interval;

[0199] During implementation, the execution process is as described in step S413 above to determine whether to execute scheduling.

[0200] If it is determined that the scheduling is to be performed, step S414 is executed; if it is determined that the scheduling is not to be performed, step S431 is executed.

[0201] Step S431, obtaining the classification result of the classifier;

[0202] Step S432, obtaining the output result of the time model;

[0203] Here, the above Table 2 can be maintained to dynamically update the probability of successful execution of the corresponding action at different times. When the success probability P is greater than the threshold (for example, 0.5), the model output is 1, and when P is less than the threshold, the model output is 0.

[0204] The update rule of P is as follows:

[0205]

[0206] Among them, t represents the time when Data Stall occurs; λ is a pre-set hyperparameter, which can be a decimal between 0 and 1; p t Represents the probability of successful scheduling at time t.

[0207] Step S433: input the classification result and the input result into the reinforcement learning model for reinforcement learning, so as to determine whether to execute scheduling based on the reinforcement learning result.

[0208] Here, the state table (Q-table) of reinforcement learning is shown in Table 3 above. A state can be determined based on the classifier results and time model results in the state table, and then based on the cumulative reward corresponding to each result, an action with a higher cumulative reward is executed. After execution, the reward value is updated in the positive or negative direction according to whether the action is effective, so as to continuously maintain the state table, so that the actions executed based on the state table are more and more efficient.

[0209] During the implementation process, Q-tabel can be learned based on the actual scheduling results of the current Data Stall. After learning Q-tabel, a scheduling strategy is dynamically selected based on the actual scenario, that is, whether to execute or not execute the network scheduling strategy.

[0210] In the embodiment of the present application, the idea of ​​soft interval is adopted to determine whether the output result of the classifier meets the confidence condition; when it is determined that the confidence condition is not met, the idea of ​​DANN domain adversarial learning is adopted to learn an encoder, and for low-accuracy data with small soft intervals, the encoder is re-encoded and re-input into the original classifier, which improves the robustness of the model and ensures that the model can still output accurate and reliable results under various uncertainties and challenges. Whether to execute the target network scheduling strategy is determined by the time model, which solves the problem of large differences in time distribution in individual scenarios. Further decision-making is achieved through reinforcement learning, which realizes automatic adjustment and optimization strategies and solves complex decision-making problems.

[0211] Based on the foregoing embodiments, an embodiment of the present application provides a scheduling device, which includes the modules included, each module includes each sub-module, each sub-module includes a unit, which can be implemented by a processor in an electronic device; of course, it can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0212] Figure 5 A schematic diagram of the structure of the scheduling device provided in the embodiment of the present application is shown in FIG. Figure 5 As shown, the device 500 includes:

[0213] A first classification module 510, configured to input the network status parameter of the target network into a first classification engine to obtain a first prediction result if the network status parameter of the target network meets the network reset condition;

[0214] A first determination module 520 is configured to determine, if the first prediction result satisfies a confidence condition, an execution decision result of a target network scheduling strategy based on the first prediction result; the execution decision result includes executing the target network scheduling strategy or not executing the target network scheduling strategy;

[0215] Among them, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result satisfies the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

[0216] In some embodiments, if the probability value corresponding to the first prediction result is within a first probability interval, the confidence of the first prediction result output by the first classification engine is higher than the target confidence; if the probability value corresponding to the first prediction result is within a second probability interval, the confidence of the first prediction result output by the first classification engine is not higher than the target confidence; wherein, the first probability interval includes a first sub-probability interval and a second sub-probability interval, and the second probability interval is between the first sub-probability interval and the second sub-probability interval.

[0217] In some embodiments, the device also includes an encoding module, a second classification module and a second determination module, wherein the encoding module is used to input the network state parameters into the target encoder to obtain the network feature encoding corresponding to the network state parameters if the first prediction result does not meet the confidence condition; the second classification module is used to input the network feature encoding into the first classification engine to obtain a second prediction result; the second determination module is used to determine the execution decision result of the target network scheduling strategy based on the second prediction result.

[0218] In some embodiments, the target encoder is used to map second data that conforms to characteristics of a second data set to first data that conforms to characteristics of a first data set; a first confidence level of a prediction result output by the first classification engine for the first data is higher than a second confidence level of a prediction result output by the first classification engine for the second data.

[0219] In some embodiments, the device also includes a prediction module and a third determination module, wherein the prediction module is used to input the time parameter corresponding to the network state parameter into a second prediction model to obtain a third prediction result if the first prediction result does not meet the confidence condition; and the third determination module is used to determine the execution decision result of the target network scheduling strategy based on the third prediction result.

[0220] In some embodiments, the device also includes a prediction module, a fourth determination module and a fifth determination module, wherein the prediction module is used to input the time parameter corresponding to the network state parameter into the second prediction model to obtain a third prediction result if the first prediction result does not meet the confidence condition; the fourth determination module is used to determine the target prediction result based on the first prediction result and the third prediction result; the fifth determination module is used to determine the target execution decision result of the target network scheduling strategy based on the target prediction result.

[0221] In some embodiments, the fifth determination module includes a first submodule, a determination submodule, a second submodule and a third submodule, wherein the first submodule is used to use the first prediction result or the third prediction result as the target prediction result if the first execution decision result determined based on the first prediction result is the same as the second execution decision result determined based on the third prediction result; the determination submodule is used to obtain the first cumulative reward value corresponding to the first prediction result and the second cumulative reward value corresponding to the third prediction result if the first execution decision result determined based on the first prediction result is different from the second execution decision result determined based on the third prediction result; the second submodule is used to use the first prediction result as the target prediction result if the first cumulative reward value is higher than the second cumulative reward value; the third submodule is used to use the third prediction result as the target prediction result if the second cumulative reward value is higher than the first cumulative reward value.

[0222] In some embodiments, the determination submodule includes an acquisition unit and a obtaining unit, wherein the acquisition unit is used to obtain the target state corresponding to the first prediction result and the third prediction result; the obtaining unit is used to obtain a first cumulative reward value corresponding to the first prediction result and a second cumulative reward value corresponding to the third prediction result based on the target state; wherein, if the target state is different, the first cumulative reward value corresponding to the first prediction result and / or the second cumulative reward value corresponding to the third prediction result are different.

[0223] In some embodiments, the determination submodule also includes a forward update unit and a reverse update unit, wherein the forward update unit is used to positively update the first cumulative reward value or the second cumulative reward value in the target state if the execution based on the target execution decision result is effective; the cumulative reward value after the forward update is higher than the cumulative reward value before the forward update; the reverse update unit is used to reversely update the first cumulative reward value or the second cumulative reward value in the target state if the execution based on the target execution decision result is not effective; the cumulative reward value after the reverse update is lower than the cumulative reward value before the reverse update.

[0224] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0225] It should be noted that in the embodiment of the present application, if the above method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable an electronic device (which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific combination of hardware and software.

[0226] Correspondingly, an embodiment of the present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the scheduling method provided in the above embodiment are implemented.

[0227] Correspondingly, an embodiment of the present application provides an electronic device, Figure 6 A hardware entity diagram of an electronic device provided in an embodiment of the present application, such as Figure 6 As shown, the hardware entity of the device 600 includes: a memory 601 and a processor 602, wherein the memory 601 stores a computer program that can be run on the processor 602, and the processor 602 implements the steps in the scheduling method provided in the above embodiment when executing the program.

[0228] The memory 601 is configured to store instructions and applications executable by the processor 602, and can also cache data to be processed or processed by the processor 602 and various modules in the electronic device 600 (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0229] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0230] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0231] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0232] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0233] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0234] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0235] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0236] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or partly reflected in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0237] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0238] The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0239] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0240] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A scheduling method, comprising: If the network state parameters of the target network meet the network reset condition, input the network state parameters into the first classification engine to obtain a first prediction result; If the first prediction result satisfies the confidence condition, determining an execution decision result of the target network scheduling strategy based on the first prediction result; The execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy; Among them, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result satisfies the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

2. The method of claim 1, wherein if the probability value corresponding to the first prediction result is within a first probability interval, the confidence level of the first prediction result output by the first classification engine is higher than the target confidence level; If the probability value corresponding to the first prediction result is within the second probability interval, the confidence level of the first prediction result output by the first classification engine is not higher than the target confidence level; in, The first probability interval includes a first sub-probability interval and a second sub-probability interval, and the second probability interval is located between the first sub-probability interval and the second sub-probability interval.

3. The method of claim 1, further comprising: If the first prediction result does not meet the confidence condition, inputting the network state parameter into a target encoder to obtain a network feature code corresponding to the network state parameter; Inputting the network feature code into the first classification engine to obtain a second prediction result; Based on the second prediction result, the execution decision result of the target network scheduling strategy is determined.

4. The method of claim 3, wherein the target encoder is used to map second data that conforms to the characteristics of the second data set to first data that conforms to the characteristics of the first data set; A first confidence level of a prediction result output by the first classification engine for the first data is higher than a second confidence level of a prediction result output by the first classification engine for the second data.

5. The method of claim 1, further comprising: If the first prediction result does not meet the confidence condition, input the time parameter corresponding to the network state parameter into the second prediction model to obtain a third prediction result; Based on the third prediction result, the execution decision result of the target network scheduling strategy is determined.

6. The method of claim 1, further comprising: If the first prediction result does not meet the confidence condition, input the time parameter corresponding to the network state parameter into the second prediction model to obtain a third prediction result; Determining a target prediction result based on the first prediction result and the third prediction result; Based on the target prediction result, a target execution decision result of the target network scheduling strategy is determined.

7. The method according to claim 6, wherein determining a target prediction result based on the first prediction result and the third prediction result comprises: If a first execution decision result determined based on the first prediction result is the same as a second execution decision result determined based on the third prediction result, taking the first prediction result or the third prediction result as the target prediction result; If the first execution decision result determined based on the first prediction result is different from the second execution decision result determined based on the third prediction result, respectively obtaining a first accumulated reward value corresponding to the first prediction result and a second accumulated reward value corresponding to the third prediction result; If the first cumulative reward value is higher than the second cumulative reward value, taking the first prediction result as the target prediction result; If the second cumulative reward value is higher than the first cumulative reward value, the third prediction result is used as the target prediction result.

8. The method according to claim 7, wherein the step of respectively obtaining the first accumulated reward value corresponding to the first prediction result and the second accumulated reward value corresponding to the third prediction result comprises: Obtaining a target state corresponding to the first prediction result and the third prediction result; Based on the target state, obtaining a first cumulative reward value corresponding to the first prediction result and a second cumulative reward value corresponding to the third prediction result; Wherein, the target states are different, and the first accumulated reward value corresponding to the first prediction result and / or the second accumulated reward value corresponding to the third prediction result are different.

9. The method of claim 8, further comprising: If the execution is effective based on the target execution decision result, the first cumulative reward value or the second cumulative reward value under the target state is positively updated; The cumulative reward value after the positive update is higher than the cumulative reward value before the positive update; If the execution based on the target execution decision result is not effective, the first cumulative reward value or the second cumulative reward value under the target state is reversely updated; The accumulated reward value after the reverse update is lower than the accumulated reward value before the reverse update.

10. A scheduling device, comprising: A first classification module, configured to input the network status parameter of the target network into a first classification engine to obtain a first prediction result if the network status parameter of the target network meets the network reset condition; A first determination module, configured to determine an execution decision result of a target network scheduling strategy based on the first prediction result if the first prediction result satisfies a confidence condition; The execution decision result includes executing the target network scheduling strategy and not executing the target network scheduling strategy; Among them, the first prediction result is the success rate of executing the target network scheduling strategy; the first prediction result satisfies the confidence condition, indicating that the confidence of the first prediction result output by the first classification engine is higher than the target confidence; the target network scheduling strategy is used to reset the target network.

Images