Resource optimization method and system based on cloud mobile phone recovery prediction model

Abstract

The invention relates to the technical field of resource optimization. The invention provides a resource optimization method and system based on a cloud mobile phone recovery prediction model, and the method comprises the steps: determining the return visit probability of a current user when it is determined that the resource demand is sharply increased and a trigger condition of an oversell recovery strategy is triggered, and the oversell recovery strategy comprises the construction of the cloud mobile phone recovery prediction model, predicting the return visit probability of the current user by using the constructed cloud mobile phone recovery prediction model, and determining a customized recovery strategy corresponding to the current user to execute resource recovery processing; repeatedly executing resource recovery processing until the current resource quantity is greater than a safety threshold value; and establishing an evaluation mechanism, and monitoring the implementation effect of the oversell recovery strategy so as to adjust the oversell recovery strategy. Customized recovery can be effectively ensured, and effective utilization of resources is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of resource optimization and provides a resource optimization method and system based on a cloud phone recycling prediction model. Background Art

[0002] As a revolutionary service innovation in the digital age, cloud phones are gradually reshaping the mobile life experience. This service model abandons the constraints of traditional physical devices, allowing users to easily enjoy the full functions of the Android operating system without holding a physical phone. With the powerful support of cloud technology, users only need an Internet connection to access a powerful virtual mobile phone environment anytime and anywhere.

[0003] However, in the face of sudden high demand, the management of cloud phone resources is crucial. The existing technology has the problem of missing sales opportunities due to out-of-stock. When the demand for resources surges, the sales strategy cannot be flexibly adjusted, and it cannot ensure that users can smoothly purchase the required products, which leads to reduced user satisfaction.

[0004] Therefore, it is necessary to provide a resource optimization method based on the cloud phone recycling prediction model to solve the above problems. Summary of the invention

[0005] The present invention provides a resource optimization method and system based on a cloud phone recycling prediction model to solve the problem of missed sales opportunities due to out-of-stock in the prior art. When resource demand surges, the sales strategy cannot be flexibly adjusted, and it cannot be ensured that users can smoothly purchase the required products, which leads to reduced user satisfaction and other technical problems. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0006] The first aspect of the present invention proposes a resource optimization method based on a cloud phone recycling prediction model, the resource optimization method comprising: when it is determined that the resource demand has surged and the triggering condition of the oversold recycling strategy is triggered, determining the return probability of the current user, the oversold recycling strategy comprising constructing a cloud phone recycling prediction model, using the constructed cloud phone recycling prediction model to predict the return probability of the current user, determining a customized recycling strategy corresponding to the current user to perform resource recycling processing; repeatedly performing resource recycling processing until the current resource amount is greater than a safety threshold; establishing an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy.

[0007] The second aspect of the present invention proposes a resource optimization system based on a cloud phone recycling prediction model, which executes the resource optimization method based on the cloud phone recycling prediction model described in the first aspect of the present invention, and the resource optimization system includes: a determination module, when it is determined that the resource demand has surged and the triggering condition of the oversold recycling strategy is triggered, determining the return probability of the current user, the oversold recycling strategy includes constructing a cloud phone recycling prediction model, using the constructed cloud phone recycling prediction model to predict the return probability of the current user, and determining a customized recycling strategy corresponding to the current user to perform resource recycling processing; a repeated execution module, repeatedly executing the resource recycling processing until the current resource amount is greater than the safety threshold; an evaluation and adjustment module, establishing an evaluation mechanism, monitoring the implementation effect of the oversold recycling strategy, so as to adjust the oversold recycling strategy.

[0008] The third aspect of the present invention provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the resource optimization method based on the cloud phone recycling prediction model described in the first aspect of the present invention.

[0009] The fourth aspect of the present invention provides a computer-readable medium having a computer program stored thereon, and when the computer program is executed by a processor, the resource optimization method based on the cloud phone recycling prediction model described in the first aspect of the present invention is implemented.

[0010] The embodiments of the present invention include the following advantages:

[0011] Compared with the prior art, the present invention uses the constructed cloud phone recycling prediction model to predict the return probability of the current user, determines the customized recycling strategy corresponding to the current user, and performs resource recycling processing, which can effectively solve the problem of missing sales opportunities due to out-of-stock, especially when resource demand surges; repeatedly performs resource recycling processing until the current resource amount is greater than the safety threshold, which can effectively ensure sufficient current resource availability; establishes an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy, which can effectively ensure user customized recycling, ensure effective resource utilization, and effectively improve user satisfaction and experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the steps of the resource optimization method based on the cloud phone recycling prediction model of the present invention;

[0013] Figure 2 It is a schematic diagram of the framework of the cloud phone recycling prediction model in the resource optimization method based on the cloud phone recycling prediction model of the present invention;

[0014] Figure 3It is a schematic diagram of the framework of the prediction process of the cloud phone recycling prediction model in the resource optimization method based on the cloud phone recycling prediction model of the present invention;

[0015] Figure 4 It is a structural block diagram of a resource optimization system based on a cloud phone recycling prediction model of the present invention;

[0016] Figure 5 is a schematic structural diagram of an electronic device embodiment according to the present invention;

[0017] Figure 6 It is a schematic diagram of the structure of a computer-readable medium embodiment according to the present invention. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] In view of the above problems, the present invention provides a resource optimization method based on a cloud phone recycling prediction model. The method uses the constructed cloud phone recycling prediction model to predict the return probability of the current user, determines a customized recycling strategy corresponding to the current user, and performs resource recycling processing, which can effectively solve the problem of missing sales opportunities due to out-of-stock, especially when resource demand surges; repeatedly performs resource recycling processing until the current resource amount is greater than the safety threshold, which can effectively ensure sufficient current resource availability; establishes an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy, which can ensure that while effectively recycling, the recycling cost is reduced, and can also effectively improve user satisfaction and experience.

[0020] It should be noted that the use of overselling strategy means that when the demand for resources surges, the enterprise can flexibly adjust the sales strategy to ensure that users can smoothly purchase the required products. In order to avoid missing sales opportunities due to out-of-stock, the overselling strategy is allowed (i.e., the system temporarily oversells the cloud phone, that is, the sales exceed the actual inventory). This overselling strategy can effectively meet customer needs and enhance the market competitiveness of the enterprise. This helps to improve user satisfaction while optimizing resource utilization. However, while using the overselling strategy, it is necessary to prevent potential inventory risks. The present invention adopts a strategy of recycling "relatively low-risk" devices. These devices generally refer to cloud phones with low frequency of use and low failure rate. Prioritizing the recycling of these devices can quickly respond to the current high demand and ensure that customer orders are met in a timely manner. After the equipment is recycled, replenishment is carried out as soon as possible to restore normal inventory levels. In addition, the overselling strategy also includes dynamic inventory management. When the available amount of resources is less than the safety threshold (i.e., the inventory is lower than the safety line), the recycling measures are automatically triggered to ensure that the inventory returns to a safe level. This mechanism not only helps to maintain the stability of inventory, but also effectively prevents sales losses caused by insufficient inventory.

[0021] It should be noted that the method of the present invention is widely applicable, and is particularly suitable for various products related to resource usage on cloud phones.

[0022] Example 1

[0023] Refer to the following Figure 1 , Figure 2 , Figure 3 , the contents of the present invention will be described in detail.

[0024] Figure 1 It is a step flow chart of the resource optimization method based on the cloud phone recycling prediction model of the present invention. Figure 2 It is a framework schematic diagram of the cloud phone recycling prediction model in the resource optimization method based on the cloud phone recycling prediction model of the present invention. Figure 3 It is a framework schematic diagram of the prediction process of the cloud phone recycling prediction model in the resource optimization method based on the cloud phone recycling prediction model of the present invention.

[0025] like Figure 1 As shown, in step S101, when it is determined that the demand for resources has surged and the triggering conditions of the oversold recovery strategy are triggered, the return probability of the current user is determined. The oversold recovery strategy includes constructing a cloud phone recovery prediction model, using the constructed cloud phone recovery prediction model to predict the return probability of the current user, and determining a customized recovery strategy corresponding to the current user to perform resource recovery processing.

[0026] When the resource demand at the current time is greater than a specific amount and when the resource availability is less than a safety threshold, it is determined that a triggering condition for an oversold recovery strategy is triggered.

[0027] Specifically, the safety threshold refers to that the ratio of the current available amount to the total amount of resources is less than a specified ratio, and the specified ratio is in the range of 5% to 20%.

[0028] In a specific implementation, the specific number is, for example, a surge in the number of single resource requirements, such as more than 1,000 units, or 1,000 to 10,000 units. The specified ratio is 10%.

[0029] Further, when it is determined that the resource demand surges (specifically, the resource demand at the current time is greater than 1,000 units) and the triggering condition of the oversold recovery strategy is triggered (the ratio of the current available amount to the total resource amount is less than 10%), the return visit probability of the current user is determined.

[0030] Use the oversold recovery strategy to determine the return probability of current users.

[0031] Optionally, the return visit probability of the current user is calculated by directly using the historical control time of the user device within a specified historical time period.

[0032] The oversold recycling strategy includes building a cloud phone recycling prediction model.

[0033] Based on the self-attention mechanism, a cloud phone recycling prediction model is constructed. The temporal features and non-temporal features of the current user are input into the cloud phone recycling prediction model, and the probability of return of the user device of the current user within a time period of a specific time calculated backward from the current time is output. The specific time includes one day, two days, seven days, fourteen days, twenty days, and thirty days.

[0034] Specifically, the cloud phone recycling prediction model includes a first part and a second part. The first part is a Transformer module, which is used to process time series features; the second part is a feedforward network module, which is used to process non-time series features.

[0035] The construction of the cloud phone recycling prediction model includes establishing a training data set.

[0036] In a specific implementation, the following time series features are extracted: device dimension, user dimension. The following non-time series features are extracted: user primary key feature and device primary key feature.

[0037] For example, the user_pad_id time-step of redfinger_log_to_rf_user_pad divides a day into 24 hour segments (i.e., action time segments), and uses the following auxiliary table to extract time series features: redfinger_log.redfinger_log_t_hour_wjw.

[0038] Specifically, extract device-dimensional features from the app activity log table, app download log table, app installation log table, app uninstall log table, and device push flow control log table in the device to form device-dimensional features. For details, see Figure 3 For example, redfinger_base.redfinger_dws_pad_app_duration is used to represent the app activity log table in the device. redfinger_log.redfinger_log_app_install_detail_wjw is used to represent the app download log table, app installation log table, and app uninstall log table in the device. redfinger_log.redfinger_log_dwd_toc_master_log_pad_control_inc is used to represent the push flow control log table in the device.

[0039] Furthermore, extract the features of the user dimension, specifically extract the features from the user embedding table and the user order table to form the user dimension features. For details, see Figure 3 For example, redfinger_log.redfinger_log_bc_stat_info is used to represent the user tracking table. Redfinger_log.redfinger_log_to_rf_order1 is used to represent the user order table.

[0040] The extraction of non-time series features specifically involves extracting user primary key features and device primary key features, user primary key features with user_id as the primary key, and device primary key features with pad_id as the primary key.

[0041] Next, the time series features and non-time series features marked with whether the device has been revisited are used to establish a training data set for training the cloud phone recycling prediction model. The classification label is characterized by whether the device has been revisited within a specified action time period, such as 24 hours, 48 ​​hours, seven days, fourteen days, twenty days, thirty days, three months, six months, etc., calculated from the current time to the past.

[0042] The timing features include whether there is streaming within the specified action time period, whether the APP is installed or uninstalled, whether the device screen changes, whether there is a return visit to the device, whether there is an active APP on the cloud phone, etc.

[0043] The non-time-series features include user features and product features. User features include user identification, and product features include device identification, etc.

[0044] The cloud phone recycling prediction model includes an encoder, a decoder, a feedforward network, and a classification part.

[0045] Specifically, the time series features of the user to be processed are input into the encoder of the cloud phone recycling prediction model, and then output from the decoder. After the sequence features output by the decoder are merged with the non-time series features (specifically including the device feature matrix and the user feature matrix), they are input into the feedforward network to calculate the final output of the classification part. Among them, the time series features are input into the encoder for the first time, and the first time step output y1 is output. When input for the second time, the time features and the first time step output y1 are input into the encoder together, that is, the decoder contains an autoregressive inference process.

[0046] When using the cloud phone recycling prediction model and splicing non-time series features (such as user features, product features) to the output vector of the decoder, the forward propagation formula and loss function formula use the following expressions:

[0047] H = Transformer Encoder (X)

[0048] Among them, H represents the encoder of the cloud phone recycling prediction model, that is, the output of the Transformer Encoder; X represents the time series features formed by whether there is streaming, whether the APP is installed or uninstalled, whether the device screen changes, whether there is a return visit to the device, whether there is an active APP on the cloud phone, etc. within the specified action time period, that is, the context representation of the above time series features.

[0049] Z = Transformer Decoder (Y input , H)

[0050] Among them, Z represents the decoder of the cloud phone recycling prediction model, that is, the output of the Transformer Decoder, which is specifically represented by a time series feature sequence; Y input represents the input of the decoder; H represents the output of the encoder;

[0051] Furthermore, the following expression is used to concatenate non-time series features:

[0052] C=[Z;U;F]

[0053] Among them, C represents the concatenated features of the time series feature sequence representation output by the decoder of the cloud phone recycling prediction model and the non-time series features such as user features and product features; Z represents the decoder of the cloud phone recycling prediction model, that is, the output of the Transformer Decoder, which is specifically the time series feature sequence representation; U represents user features; F represents product features.

[0054] Next, we use the classification part, specifically the softmax layer, to calculate the probability distribution using the following expression:

[0055] Prob = softmax(WC+b)

[0056] Among them, Prob represents the output of the cloud phone recycling prediction model, specifically the return visit probability corresponding to the user to be processed (such as the current user); softmax represents the classification function of the classification part of the cloud phone recycling prediction model; W and b represent the weight matrix and bias term of the classification function respectively; C represents the concatenated features after concatenating the temporal feature sequence output by the decoder of the cloud phone recycling prediction model with non-temporal features such as user features and product features.

[0057] Assuming the target sequence is Y and the goal is to minimize the cross entropy loss, it is expressed as follows:

[0058]

[0059] in, represents the loss value of the cloud phone recycling prediction model; N represents the batch size; TY represents the length of the output sequence; |V| represents the number of categories, specifically the predicted categories are "revisit" / "no revisit", V| is 2, Y k,t Prob is the vector representation (e.g. one-hot representation) of the true label value of the k-th category at time step t, where k is a positive integer, specifically 1, 2, ..., |V|; k,t Represents the model predicted probability value of the kth category at time step t; represents the L2 regularization term, and W represents the network parameters of the cloud phone recycling prediction model.

[0060] It should be noted that when using the cross entropy loss function of the pytorch framework, the output of the neural network does not need to be processed by softmax in advance, because the nn.CrossEntropyLoss() function will perform a softmax operation internally. If the output has been processed by softmax in advance, the nn.NLLLoss() (negative log-likelihood loss) function needs to be used to calculate the loss.

[0061] The convergence of the cloud phone recycling prediction model is judged by monitoring the iterative gradient change value, so as to determine the completion of the training process of the cloud phone recycling prediction model and obtain the trained cloud phone recycling prediction model.

[0062] Monitor the gradient value of each iteration round and use the following expression to calculate the gradient change value to determine the completion of the training process of the cloud phone recycling prediction model.

[0063]

[0064] in, It represents the gradient change value between the gradient value of the previous iteration t-1 and the gradient value of the current iteration t, t represents the current iteration round, and t-1 represents the previous iteration round. During the model convergence process of the cloud phone recycling prediction model, the gradient change value, that is, the amplitude, will gradually decrease. When the gradient change value is close to zero and stabilizes and no longer changes, the model is determined to have converged.

[0065] Optionally, when When it is in the range of 0.001 to 0.008, the cloud phone recycling prediction model converges.

[0066] For testing, a test data set is established by marking the timing characteristics and non-timing characteristics of the return visit device.

[0067] Optionally, the ratio of the number of training data sets to the number of test data sets is 8:1 or 9:1.

[0068] The training dataset is used to train the cloud phone recycling prediction model, and the test dataset is used to test the cloud phone recycling prediction model.

[0069] For example, when the model is tested using a test set, when the model accuracy reaches 87% to 95%, the model is determined to have converged, that is, the model training process is completed.

[0070] After the above training process including model optimization, a trained cloud phone recycling prediction model is obtained.

[0071] In a specific implementation, when the triggering condition of the oversold recycling strategy is triggered, the constructed cloud phone recycling prediction model is used to make a prediction to start recycling the device.

[0072] Specifically, before using the constructed cloud phone recycling prediction model, a judgment is first made on whether to recycle the device based on user importance.

[0073] When the user's device is in the whitelist, the device will not be recycled, that is, there is no need to use the constructed cloud phone recycling prediction model to make a prediction to start recycling the device.

[0074] The whitelist is a whitelist established using the data of its own users or other public user data.

[0075] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0076] Next, in step S102, the resource recycling process is repeatedly performed until the current resource amount is greater than the safety threshold.

[0077] When it is determined that the resource demand amount surges, the resource recycling process is repeatedly performed until the current resource amount is greater than the safety threshold.

[0078] In an optional implementation, when it is determined that the resource demand is greater than 1,000 units and the user is not in the whitelist, the resource recycling process is repeatedly performed until the current resource amount is greater than the safety threshold.

[0079] Specifically, the cloud phone recycling prediction model constructed in step S101 is used to predict the return visit probability corresponding to the current user, and the resource recycling process is performed according to the predicted return visit probability. Then determine whether the current resource amount is greater than the safety threshold. If the current resource amount is greater than the safety threshold (for example, the specified ratio is 10%), stop executing the resource recycling process. If the current resource amount is less than or equal to the safety threshold (for example, the specified ratio is 10%), stop executing the resource recycling process.

[0080] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0081] Next, in step S103, an evaluation mechanism is established to monitor the implementation effect of the oversold recovery strategy to adjust the oversold recovery strategy.

[0082] The establishment of an evaluation mechanism specifically includes calculating the actual number of recycling equipment, recycling accuracy, recycling recall rate, drawing a PR curve, and determining the optimal balance point to optimize the oversold recycling strategy.

[0083] The recovery accuracy is calculated using the following expression:

[0084] Recovery accuracy rate = number of successfully recovered devices / total number of recovered devices;

[0085] The recovery failure rate is calculated using the following expression:

[0086] Recycling error rate = number of incorrectly recycled devices / total number of recycled devices.

[0087] Specifically, the classification threshold of whether to recycle is used as the horizontal coordinate, and the evaluation index F1 of the cloud phone recycling prediction model is used as the vertical coordinate to draw the PR curve, and the highest point in the PR curve is taken as the optimal balance point.

[0088] Furthermore, the classification threshold corresponding to the optimal balance point is in the range of 0.01 to 0.08.

[0089] Specifically, the classification threshold is optimized and adjusted according to the determined optimal balance point. For example, the classification threshold is adjusted to 0.5-0.6. When the classification threshold is 0.5-0.6, the accuracy of the cloud phone recycling prediction model is improved by 5%-15%.

[0090] The establishment of an evaluation mechanism also includes the establishment of a cost calculation model to convert the recycling quantity and recycling accuracy into cost indicators to evaluate the cost of recycling operations. For example, the following expression is used to calculate the recycling cost:

[0091] Recycling cost = (number of recycling equipment × unit recycling cost) + (number of incorrect recycling equipment × error cost)

[0092] Among them, the recycling cost specifically refers to the number of recycled devices in the historical time period of one month from the current time to the past; the unit recycling cost is calculated, for example, based on the cost of the average usage of available devices per unit time; the number of erroneous recycling devices refers to the number of erroneous recycling devices in the historical time period of one month from the current time to the past; the error cost is determined, for example, based on the revenue value of a single available device per unit time used at the current time.

[0093] In another optional implementation, the resources are first recycled according to user importance, ie, user priority, and then according to the probability of future return visits of the user's device, until the current available resources (eg, available devices) are greater than a safety threshold (10%).

[0094] Specifically, the user's priority level includes five priorities: first level, second level, third level, fourth level and fifth level.

[0095] Users with the first level of priority belong to the whitelist. The devices corresponding to the users in the whitelist (i.e. free devices) will not be processed for resource recycling. Specifically, the devices will be recycled after the users in the whitelist have finished using their resources.

[0096] For users with first-level priority, the corresponding customized recycling strategy is to recycle the resources that have ended their use. For example, cloud machine resources that have not been used (not outflowed) for more than x minutes (configurable) are recycled, i.e. resource recycling processing. For example, until the ratio of available devices to the total number of devices is greater than a specified ratio (e.g. 10%).

[0097] Then, the data is backed up within the validity period (12 hours, 24 hours, etc.), and the data is restored when the user returns. After the validity period, the user data will be deleted.

[0098] In another optional implementation, the current inventory is 1 million devices. When the number of unused devices is less than 10% and recycling all 1 million devices cannot meet the resource demand, it is determined to recycle users with the second level priority, i.e., resource recycling processing.

[0099] The second-level priority users are users who pay for using resources. The user devices (e.g., bound devices) whose resources have expired or expired are recycled. The corresponding customized recycling strategy is direct recycling, i.e., resource recycling processing. Until the ratio of available devices to the total number of devices is greater than a specified ratio (e.g., 10%).

[0100] Then, the data is backed up within the validity period (12 hours, 24 hours, etc.), and the data is restored when the user returns. After the validity period, the user data will be deleted.

[0101] In another optional implementation, when the resource demand at the current time is greater than 1,000 devices and the current available resources are less than a safety threshold (i.e., the ratio of available devices to the total number of devices, for example, 10%), it is determined to recycle users with a third level priority, i.e., resource recycling processing, until the ratio of available devices to the total number of devices is greater than a specified ratio (e.g., 10%).

[0102] The user with the third priority refers to a user who has paid for the use of resources but has not used the resources for a long time. The device corresponding to the user or the device bound to the user (for example, a bound device) will be recycled. The corresponding customized recycling strategy is to use the cloud phone recycling prediction model to determine the probability of a revisit for the user, that is, to use the cloud phone recycling prediction model to output the probability of a revisit for the user to control the cloud phone in the future (for example, within the next week or the next half month), and to sort the users from low to high according to the output revisit probability, giving priority to recycling users with low revisit probabilities until the ratio of available devices to the total number of devices is greater than the specified ratio (for example, 10%).

[0103] Then, the data is backed up within the validity period (12 hours, 24 hours, etc.), and the data is restored when the user returns. After the validity period, the user data will be deleted.

[0104] In another optional implementation, when the resource demand at the current time is greater than 1,000 devices and the current available resources are less than a safety threshold (i.e., the ratio of available devices to the total number of devices, for example, 10%), it is determined to recycle users with a fourth level priority, i.e., resource recycling processing, until the ratio of available devices to the total number of devices is greater than a specified ratio (e.g., 10%).

[0105] The user with the fourth priority level refers to a user whose user resources have been paid for but have not been used within a specified time. The device corresponding to the user or the device bound to the user (such as a bound device) will be recycled. For example, if the bound device has not been used for X consecutive hours, the bound device will be recycled. The corresponding customized recycling strategy is to use the cloud phone recycling prediction model to determine the probability of a revisit for the user, that is, to use the cloud phone recycling prediction model to output the probability of the user controlling the cloud phone in the future (for example, in the next few hours) to revisit, and to sort the users from low to high according to the output revisit probability, giving priority to recycling users with a low revisit probability. For example, recycle the devices of the top ten users.

[0106] Then, the data is backed up within the validity period (12 hours, 24 hours, etc.), and the data is restored when the user returns. After the validity period, the user data will be deleted.

[0107] In another optional implementation, when the resource demand at the current time is greater than 1,000 devices and the current available resources are less than a safety threshold (i.e., the ratio of available devices to the total number of devices, for example, 10%), it is determined to recycle the user with the first priority level, i.e., resource recycling processing, until the ratio of available devices to the total number of devices is greater than a specified ratio (e.g., 10%).

[0108] The user with the fifth priority level refers to a user who has paid for user resources and has ended using user resources. The device corresponding to the user or the device bound to the user (such as a bound device) is recycled. For example, if the bound device has not been used for X consecutive minutes, the bound device is recycled. The corresponding customized recycling strategy is to use the cloud phone recycling prediction model to determine the probability of a revisit for the user, that is, to use the cloud phone recycling prediction model to output the probability of the user controlling the cloud phone in a certain period of time in the future (for example, within a certain number of minutes in the future), and to sort the users from low to high according to the outputted revisit probability, giving priority to recycling users with a low revisit probability. For example, recycle the devices of the top ten users.

[0109] Then, the data is backed up within the validity period (12 hours, 24 hours, etc.), and the data is restored when the user returns. After the validity period, the user data will be deleted.

[0110] It should be noted that the above is only described as an optional example and should not be understood as a limitation to the present invention.

[0111] Compared with the prior art, the present invention uses the constructed cloud phone recycling prediction model to predict the return probability of the current user, determines the customized recycling strategy corresponding to the current user, and performs resource recycling processing, which can effectively solve the problem of missing sales opportunities due to out-of-stock, especially when resource demand surges; repeatedly performs resource recycling processing until the current resource amount is greater than the safety threshold, which can effectively ensure sufficient current resource availability; establishes an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy, which can effectively ensure user customized recycling, ensure effective resource utilization, and effectively improve user satisfaction and experience.

[0112] Example 2

[0113] The following is a system embodiment of the present invention, which can be used to implement the method embodiment of the present invention. For details not disclosed in the system embodiment of the present invention, please refer to the method embodiment of the present invention.

[0114] Figure 4 This is a schematic diagram of an example of a resource optimization system based on a cloud phone recycling prediction model according to the present invention. Figure 4 , a resource optimization system based on a cloud phone recycling prediction model is described. The resource optimization system is used to execute the resource optimization method based on a cloud phone recycling prediction model described in the first aspect of the present invention.

[0115] like Figure 4 As shown, the resource optimization system 400 based on the cloud phone recycling prediction model includes a determination module 410, a repeated execution module 420 and an evaluation and adjustment module 430.

[0116] In a specific embodiment, the determination module 410 is used to determine the return probability of the current user when it is determined that the resource demand has surged and the triggering condition of the oversold recovery strategy has been triggered. The oversold recovery strategy includes constructing a cloud phone recovery prediction model, using the constructed cloud phone recovery prediction model to predict the return probability of the current user, and determining a customized recovery strategy corresponding to the current user to perform resource recovery processing. The repeated execution module 420 is used to repeatedly execute the resource recovery process until the current resource amount is greater than the safety threshold. The evaluation and adjustment module 430 is used to establish an evaluation mechanism to monitor the implementation effect of the oversold recovery strategy to adjust the oversold recovery strategy.

[0117] According to an optional implementation mode, a cloud phone recycling prediction model is constructed, specifically including:

[0118] Based on the self-attention mechanism, a cloud phone recycling prediction model is constructed. The temporal features and non-temporal features of the current user are input into the model, and the probability of the current user's user device returning within a specific time period calculated backward from the current time is output.

[0119] According to an optional implementation method, the following time series features are extracted: device dimension and user dimension; the following non-time series features are extracted: user primary key features and device primary key features; and a training data set is established using time series features and non-time series features marked with whether the device has been revisited, so as to be used for training the cloud phone recycling prediction model.

[0120] According to an optional implementation manner, the establishment of an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy includes: calculating the actual number of recycling equipment, recycling accuracy, recycling recall rate, drawing a PR curve, and determining the best balance point to optimize the oversold recycling strategy; using the following expression to calculate the recycling accuracy:

[0121] Recovery accuracy rate = number of successfully recovered devices / total number of recovered devices;

[0122] The recovery failure rate is calculated using the following expression:

[0123] Recycling error rate = number of incorrectly recycled devices / total number of recycled devices.

[0124] According to an optional implementation, the classification threshold of whether to recycle is used as the horizontal coordinate, and the evaluation index F1 of the cloud phone recycling prediction model is used as the vertical coordinate to draw a PR curve, and the highest point in the PR curve is used as the optimal balance point; the classification threshold corresponding to the optimal balance point is in the range of 0.01 to 0.08.

[0125] According to an optional implementation manner, when it is determined that a surge in resource demand has occurred and a triggering condition for an oversold recovery strategy has been triggered, determining the return probability of the current user includes: when the resource demand at the current time is greater than a specific amount and when the available resource amount is less than a safety threshold, determining that the triggering condition for the oversold recovery strategy has been triggered, wherein the safety threshold refers to a ratio of the current available amount to the total amount of resources being less than a specified ratio, and the specified ratio being in the range of 5% to 20%.

[0126] It should be noted that due to Figure 4 Resource optimization method based on cloud phone recycling prediction model implemented by resource optimization system based on cloud phone recycling prediction model and Figure 1 The resource optimization method based on the cloud phone recycling prediction model in the example is roughly the same, so the description of the same parts is omitted.

[0127] Compared with the prior art, the present invention uses the constructed cloud phone recycling prediction model to predict the return probability of the current user, determines the customized recycling strategy corresponding to the current user, and performs resource recycling processing, which can effectively solve the problem of missing sales opportunities due to out-of-stock, especially when resource demand surges; repeatedly performs resource recycling processing until the current resource amount is greater than the safety threshold, which can effectively ensure sufficient current resource availability; establishes an evaluation mechanism to monitor the implementation effect of the oversold recycling strategy to adjust the oversold recycling strategy, which can effectively ensure user customized recycling, ensure effective resource utilization, and effectively improve user satisfaction and experience.

[0128] Figure 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

[0129] like Figure 5 As shown, the electronic device is presented in the form of a general computing device. The processor may be one or more and work in coordination. The present invention does not exclude distributed processing, that is, the processor may be dispersed in different physical devices. The electronic device of the present invention is not limited to a single entity, but may also be the sum of multiple physical devices.

[0130] The memory stores a computer executable program, which is usually a machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least part of the steps in the method.

[0131] The memory includes a volatile memory, such as a random access memory unit (RAM) and / or a cache memory unit, and may also be a non-volatile memory, such as a read-only memory unit (ROM).

[0132] Optionally, in this embodiment, the electronic device further includes an I / O interface, which is used for the electronic device to exchange data with an external device. The I / O interface can represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of a variety of bus structures.

[0133] It should be understood that Figure 5 The electronic device shown is only an example of the present invention, and the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as display screens, and some electronic devices also include human-computer interaction elements such as buttons, keyboards, etc. As long as the electronic device can execute the computer-readable program in the memory to implement the method of the present invention or at least part of the steps of the method, it can be considered as an electronic device covered by the present invention.

[0134] Through the above description of the implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by combining software with necessary hardware. Figure 6 As shown, the technical solution according to the embodiment of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of commands to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the above method according to the embodiment of the present invention.

[0135] The software product may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0136] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, device, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0137] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0138] The computer-readable medium carries one or more programs. When the one or more programs are executed by a device, the computer-readable medium implements the data interaction method of the present disclosure.

[0139] Those skilled in the art will appreciate that the above modules can be distributed in the device according to the description of the embodiment, or can be changed accordingly and only used in one or more devices different from the embodiment. The modules of the above embodiments can be combined into one module, or further divided into multiple sub-modules.

[0140] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several commands to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiment of the present invention.

[0141] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0142] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A resource optimization method based on a cloud phone recycling prediction model, characterized in that: The resource optimization method comprises: When it is determined that the resource demand surges and the triggering condition of the oversold recycling strategy is triggered, the return probability of the current user is determined, and the oversold recycling strategy includes constructing a cloud phone recycling prediction model, using the constructed cloud phone recycling prediction model to predict the return probability of the current user, and determining a customized recycling strategy corresponding to the current user to perform resource recycling processing; Repeat the resource recovery process until the current resource amount is greater than the safety threshold; An evaluation mechanism is established to monitor the implementation effect of the oversold recovery strategy so as to adjust the oversold recovery strategy.

2. The resource optimization method according to claim 1, characterized in that: The cloud phone recycling prediction model is constructed, including: Based on the self-attention mechanism, a cloud phone recycling prediction model is constructed. The temporal features and non-temporal features of the current user are input into the model, and the probability of the current user's user device returning within a specific time period calculated backward from the current time is output.

3. The resource optimization method according to claim 2, characterized in that: Further including: Extract the following time series features: device dimension and user dimension; Extract the following non-time series features: user primary key features and device primary key features; A training data set is established by using time series features and non-time series features marked with whether the device has been revisited, so as to be used for training the cloud phone recycling prediction model.

4. The resource optimization method according to claim 1, characterized in that: The establishment of an evaluation mechanism to monitor the implementation effect of the oversold recovery strategy to adjust the oversold recovery strategy includes: Calculate the actual number of recycling equipment, recycling accuracy, and recycling recall rate, draw a PR curve, and determine the best balance point to optimize the oversold recycling strategy; The recovery accuracy was calculated using the following expression: Recovery accuracy rate = number of successfully recovered devices / total number of recovered devices; The recovery failure rate is calculated using the following expression: Recycling error rate = number of incorrectly recycled devices / total number of recycled devices.

5. The resource optimization method according to claim 4, characterized in that: Further including: Use the classification threshold of whether to recycle as the horizontal coordinate, and use the evaluation index F1 of the cloud phone recycling prediction model as the vertical coordinate to draw the PR curve, and use the highest point in the PR curve as the optimal balance point; The classification threshold corresponding to the optimal balance point is in the range of 0.01 to 0.

08.

6. The resource optimization method according to claim 1, characterized in that: When it is determined that the resource demand surges and the triggering condition of the oversold recovery strategy is triggered, determining the return visit probability of the current user includes: When the resource demand at the current time is greater than a specific amount, and when the resource availability is less than a safety threshold, the triggering condition of the oversold recovery strategy is determined to be triggered. The safety threshold refers to the ratio of the current availability to the total resource amount being less than a specified ratio, and the specified ratio is in the range of 5% to 20%.

7. A resource optimization system based on a cloud phone recycling prediction model, characterized in that: It implements the resource optimization method based on the cloud phone recycling prediction model described in claims 1 to 6, and the resource optimization system includes: A determination module, when it is determined that the resource demand surges and the triggering condition of the oversold recycling strategy is triggered, determines the return probability of the current user, wherein the oversold recycling strategy includes constructing a cloud phone recycling prediction model, using the constructed cloud phone recycling prediction model to predict the return probability of the current user, and determining a customized recycling strategy corresponding to the current user to perform resource recycling processing; Repeat the execution module and repeatedly execute the resource recovery process until the current resource amount is greater than the safety threshold; An evaluation and adjustment module is used to establish an evaluation mechanism to monitor the implementation effect of the oversold recovery strategy so as to adjust the oversold recovery strategy.

8. The resource optimization system according to claim 7, characterized in that: include: Based on the self-attention mechanism, a cloud phone recycling prediction model is constructed, the temporal features and non-temporal features of the current user are input into the cloud phone recycling prediction model, and the probability of the current user's user device returning within a time period calculated from the current time to a specific time is output.

9. The resource optimization system according to claim 7, characterized in that: Further including: Extract the following time series features: device dimension and user dimension; Extract the following non-time series features: user primary key features and device primary key features; A training data set is established by using time series features and non-time series features marked with whether the device has been revisited, so as to be used for training the cloud phone recycling prediction model.

10. The resource optimization system according to claim 7, characterized in that: Further including: Calculate the actual number of recycling equipment, recycling accuracy, and recycling recall rate, draw a PR curve, and determine the best balance point to optimize the oversold recycling strategy; The recovery accuracy was calculated using the following expression: Recovery accuracy rate = number of successfully recovered devices / total number of recovered devices; The recovery failure rate is calculated using the following expression: Recycling error rate = number of incorrectly recycled devices / total number of recycled devices.

Images