Model training method and device, information determination method and device, equipment, storage medium and computer program product

Abstract

The embodiment of the invention discloses a model training method, which comprises the following steps of: acquiring first requirements of sample equipment in a plurality of first time periods and first attribute information of the sample equipment, determining requirement feature information of the sample equipment based on the plurality of first time periods and the first requirements, and then determining the requirement feature information of the sample equipment based on the first attribute information and the plurality of first requirements. Determining a second demand of each type of sample equipment in each first time period, and training the initial gradient decision model based on the first attribute information of the sample equipment, the plurality of first time periods, the demand feature information and the second demand to obtain a target gradient decision model, the problem that the demand prediction amount is inaccurate when the demand amount of the cloud host is predicted in the prior art is solved. The embodiment of the invention further discloses an information determination method and device, equipment, a storage medium and a computer program product.

Description

Technical Field

[0001] The present application relates to information determination technology in the computer field, and in particular to model training, information determination methods, devices, equipment, storage media and computer program products. Background Art

[0002] With the rapid development of cloud computing, cloud hosts deployed on cloud computing platforms have gradually replaced traditional virtual machines. Therefore, in the face of growing demand, how to accurately predict the demand for cloud hosts in order to reasonably schedule cloud hosts has become extremely important. At present, the related technology uses an autoregressive integrated moving average (ARIMA) model or a long short-term recurrent neural network (LSTM) model to predict the demand for cloud hosts. However, when predicting demand through an ARIMA model or LSTM model, only the time factor is considered, and the ARIMA model or LSTM model can only perform short-term predictions, which leads to the problem that the demand forecast for the determined cloud host is inaccurate. Summary of the invention

[0003] To solve the above technical problems, the embodiments of the present application hope to provide a model training, information determination method, device, equipment, storage medium and computer program product, which solves the problem of inaccurate demand prediction when predicting the demand for cloud hosts in the related art.

[0004] The technical solution of this application is implemented as follows:

[0005] A model training method, the method comprising:

[0006] Acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods;

[0007] Based on the multiple first time periods and the first demand, determine demand characteristic information of the sample device; wherein the demand characteristic information represents how the demand of the sample device changes over time;

[0008] Determine, based on the first attribute information and the plurality of the first requirements, a second requirement for each type of sample device in each first time period;

[0009] The initial gradient decision model is trained based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

[0010] In the above solution, the step of determining the demand characteristic information of the sample device based on the multiple first time periods and the first demand includes:

[0011] The target time feature extraction model is used to process the multiple first time periods and the first demand to obtain first feature information, second feature information and third feature information corresponding to the sample device; wherein the third feature information represents the change of the influence range of the target period in the multiple first time periods on the demand for the sample device over time;

[0012] Demand characteristic information is determined based on the first characteristic information, the second characteristic information, and the third characteristic information.

[0013] In the above solution, the target time feature extraction model is used to process the multiple first time periods and the first requirement to obtain the first feature information, the second feature information and the third feature information corresponding to the sample device, including:

[0014] Using the growth trend sub-model in the target time feature extraction model to process the multiple first time periods and the first demand to obtain the first feature information;

[0015] Processing the plurality of first time periods and the first demand using a seasonal trend sub-model in the target time feature extraction model to obtain the second feature information;

[0016] The plurality of first time periods and the first demand are processed using a logical growth sub-model in the target time feature extraction model to obtain the third feature information.

[0017] In the above solution, determining the second requirement of each type of sample device in each first time period based on the first attribute information and the plurality of first requirements includes:

[0018] Classifying the sample devices based on first configuration information of the sample devices; wherein the first attribute information includes the first configuration information;

[0019] Based on the plurality of the first requirements, a second requirement for each type of sample equipment in each first time period is determined.

[0020] In the above solution, the initial gradient decision model is trained based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain the target gradient decision model, including:

[0021] Acquire first basic information of the sample device; wherein the first basic information is information related to the deployment location of the sample device and the user to which the sample device belongs;

[0022] The initial gradient decision model is used to process the category of the sample device, the first basic information, the multiple first time periods and the demand feature information to obtain demand forecast information for each category of sample device; wherein the category is determined based on the first attribute information;

[0023] Determine a target loss value based on a target loss function, the second demand and the demand forecast information; wherein the target loss function is a root mean square error function;

[0024] Based on the target loss value, a target grid search algorithm is used to update the parameters of the initial gradient decision model until the target gradient decision model is obtained.

[0025] In the above scheme, the method further includes:

[0026] Acquire real-time requirements of the first device in multiple second time periods and second attribute information of the first device;

[0027] The target gradient decision model is updated based on the multiple second time periods, the real-time demand and the second attribute information.

[0028] A method for determining information, the method comprising:

[0029] Obtaining third attribute information of the device to be processed;

[0030] The target time period and the third attribute information are processed using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training an initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand characteristic information and the second demand of each type of sample device; the demand characteristic information is determined based on multiple first time periods and the first demand of the sample devices in the multiple first time periods.

[0031] In the above solution, the target gradient decision model is used to process the target period and the third attribute information to obtain the target demand of the device to be processed in the target period, including:

[0032] Acquire second basic information of the device to be processed; wherein the second basic information is information related to the deployment location of the device to be processed and the user to which the device to be processed belongs;

[0033] Determine the target category of the device to be processed based on the third attribute information;

[0034] The target gradient decision model is used to process the target time period, the target category and the second basic information to obtain the target demand.

[0035] A model training device, comprising:

[0036] A first acquisition unit, configured to acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods;

[0037] A first determining unit, configured to determine demand characteristic information of the sample device based on the multiple first time periods and the first demand; wherein the demand characteristic information represents how the demand of the sample device changes over time;

[0038] A second determining unit, configured to determine a second requirement for each type of sample device in each first time period based on the first attribute information and a plurality of the first requirements;

[0039] A training unit is used to train an initial gradient decision model based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

[0040] An information determination device, the device comprising:

[0041] A second acquisition unit, used to acquire third attribute information of the device to be processed;

[0042] A processing unit is used to process the target time period and the third attribute information using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand feature information and the second demand of each type of sample device; the demand feature information is determined based on multiple first time periods and the first demand of the sample devices in the multiple first time periods.

[0043] A model training device, comprising: a first processor, a first memory and a first communication bus;

[0044] The first communication bus is used to realize the communication connection between the first processor and the first memory;

[0045] The first processor is used to execute the model training program in the first memory to implement the steps of the above-mentioned model training method.

[0046] An information determination device, the information determination device comprising: a second processor, a second memory, and a second communication bus;

[0047] The second communication bus is used to realize the communication connection between the second processor and the second memory;

[0048] The second processor is used to execute the information determination program in the second memory to implement the steps of the above-mentioned information determination method.

[0049] A computer-readable storage medium storing one or more programs, wherein the one or more programs can be executed by one or more processors to implement the steps of the above-mentioned model training method or information determination method.

[0050] A computer program product, comprising a computer program, which implements the steps of the above-mentioned model training method or information determination method when executed by a processor.

[0051] The model training, information determination method, apparatus, device, storage medium and computer program product provided by the embodiments of the present application can obtain the first demand of the sample device in multiple first time periods and the first attribute information of the sample device, and determine the demand characteristic information characterizing the change of the demand of the sample device over time based on the multiple first time periods and the first demand. Then, based on the first attribute information and the multiple first demands, the second demand of each type of sample device in each first time period is determined, and then the initial gradient decision model is trained based on the first attribute information of the sample device, the multiple first time periods, the demand characteristic information and the second demand to obtain the target gradient decision model In this way, the initial gradient decision model can be trained in combination with the first attribute information of the sample device, multiple first time periods and demand characteristic information to obtain a target gradient decision model for predicting the demand for cloud hosts, that is, when training the target gradient decision model, the time factor (i.e., multiple first time periods), the attribute information of the sample device and the demand characteristic information of the sample device are simultaneously considered, rather than only considering the time factor when predicting demand as in the related art; and long-term demand prediction can be performed through the target gradient decision model, thus solving the problem of inaccurate demand prediction for cloud hosts in the related art when predicting the demand for cloud hosts, thereby realizing flexible scheduling of cloud hosts. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of a model training method provided in an embodiment of the present application;

[0053] Figure 2 A feature importance analysis diagram of a model training method provided in an embodiment of the present application;

[0054] Figure 3A flowchart of another model training method provided in an embodiment of the present application;

[0055] Figure 4 A schematic diagram of a flow chart of an initial gradient decision model in another model training method provided in an embodiment of the present application;

[0056] Figure 5 A schematic diagram of the trend of the target loss function in the initial gradient decision model in another model training method provided in an embodiment of the present application;

[0057] Figure 6 A flowchart of an information determination method provided in an embodiment of the present application;

[0058] Figure 7 A schematic diagram for comparing prediction results in an information determination method provided in an embodiment of the present application;

[0059] Figure 8 A schematic diagram of the structure of a model training device provided in an embodiment of the present application;

[0060] Fig. 9 A schematic diagram of the structure of an information determination device provided in an embodiment of the present application;

[0061] Fig.10 A schematic diagram of the structure of a model training device provided in an embodiment of the present application;

[0062] Fig.11 A schematic diagram of the structure of an information determination device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0064] It should be understood that the "embodiments of the present application" or "the aforementioned embodiments" mentioned throughout the specification mean that specific features, structures or characteristics related to the embodiments are included in at least one embodiment of the present application. Therefore, "in the embodiments of the present application" or "in the aforementioned embodiments" appearing throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the various embodiments of the present application, the size of the sequence numbers of the above-mentioned processes does not mean the order of execution. 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 for description only and do not represent the advantages and disadvantages of the embodiments.

[0065] Unless otherwise specified, the electronic device performs any step in the embodiments of the present application, and the processor of the electronic device may perform the step. It is also worth noting that the embodiments of the present application do not limit the order in which the electronic device performs the following steps. In addition, the methods used to process data in different embodiments may be the same method or different methods. It should also be noted that any step in the embodiments of the present application can be independently executed by the electronic device, that is, when the electronic device executes any step in the following embodiments, it may not rely on the execution of other steps.

[0066] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] The present application embodiment provides a model training method, referring to Figure 1 As shown, the method may include the following steps:

[0068] Step 101: Acquire first requirements of multiple sample devices and first attribute information of the sample devices in a first time period.

[0069] In the embodiment of the present application, the first period may refer to a period of time, specifically, a period of time may be a few hours, a few days, a few weeks, or even a few months or even a few years. For example, the first period of time may refer to a day between December 1, 2024 and December 2, 2024.

[0070] In the embodiment of the present application, the sample device may refer to a sample cloud host, i.e., a cloud host used for model training; the first attribute information may include specification information (i.e., configuration information) and creation time of the sample device; the first demand may refer to the historical demand for the sample cloud host in each first time period. The specification information (i.e., configuration information) may include a central processing unit (CPU), memory capacity, hard disk type, and hard disk capacity, etc.

[0071] Step 102: Determine demand characteristic information of sample devices based on multiple first time periods and first demands.

[0072] The demand characteristic information represents how the demand for sample equipment changes over time.

[0073] In the embodiment of the present application, the demand characteristic information may refer to the changing trend of the demand of the sample cloud host over time. Specifically, multiple first time periods may be encoded respectively, and then the time period labels and first demands obtained after encoding may be processed using a time characteristic extraction model to obtain the demand characteristic information of the sample device.

[0074] Step 103: Determine the second requirement of each type of sample device in each first time period based on the first attribute information and the multiple first requirements.

[0075] In one achievable manner, the sample devices may be classified according to the configuration information in the first attribute information, and the requirements (ie, second requirements) of sample devices of the same specification may be determined according to the first requirements of the sample devices.

[0076] In another achievable manner, the sample devices may be classified according to the configuration information and the deployment locations of the sample devices, and then the second requirements for the sample devices of the same specification and located at the same location may be determined according to the first requirements for the sample devices.

[0077] Step 104: training the initial gradient decision model based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

[0078] In an achievable manner, the first attribute information, the time period label of the first time period, and the demand characteristic information can be used as input parameters of the initial gradient decision model, that is, the first attribute information, the time period label of the first time period, and the demand characteristic information can be input into the initial gradient decision model for processing, and the parameters of the initial gradient decision model are updated according to the obtained processing result and the second demand until the target gradient decision model is obtained through training. In an achievable manner, the initial gradient decision model can refer to an Extreme Gradient Boosting (XGBOOST) model.

[0079] In another feasible manner, basic information related to the location of the sample device and the user to which it belongs can be obtained, and the basic information, the first attribute information, the time period label and the demand characteristic information can be input into the XGBOOST model for processing until the target gradient decision model is obtained through training.

[0080] It should be noted that the related art usually uses a time series prediction model to process the time period and the corresponding demand to obtain the predicted demand of the cloud host. That is to say, the related art only considers the time factor when predicting the demand, and ignores the specification information of the cloud host itself. However, Figure 2 As shown, the influence of the specification information of the cloud host on the demand is actually greater than the influence of the time factor on the demand, which leads to inaccurate determined demand. In the embodiment of the present application, the prediction model is trained by combining the attribute information, time period and demand characteristic information of the cloud host, which significantly improves the prediction accuracy of the demand for the cloud host.

[0081] The model training method provided in the embodiments of the present application can combine the first attribute information of the sample device, multiple first time periods and demand characteristic information to jointly train the initial gradient decision model to obtain a target gradient decision model for predicting the demand of the cloud host, that is, when training the target gradient decision model, the time factor (that is, multiple first time periods), the attribute information of the sample device and the demand characteristic information of the sample device are simultaneously considered, rather than only considering the time factor when predicting demand as in the related art; and long-term demand prediction can be performed through the target gradient decision model, thus solving the problem of inaccurate demand prediction of the cloud host when predicting the demand of the cloud host in the related art, thereby realizing flexible scheduling of the cloud host.

[0082] Based on the above embodiments, the embodiments of the present application provide a model training method, which can be applied to a model training device, referring to Figure 3 As shown, the method may include the following steps:

[0083] Step 201: The model training device obtains first requirements and first attribute information of sample devices in multiple first time periods.

[0084] In the embodiment of the present application, the first demand may refer to the historical demand of the sample cloud host. Specifically, the historical demand of the sample cloud hosts in a plurality of first time periods and the first attribute information of each sample cloud host may be obtained from the operator's data center.

[0085] It should be noted that after obtaining the first attribute information, the first attribute information can be verified to determine whether the first attribute information lacks information of necessary fields, and the data lacking necessary field information is removed. After that, all information is sorted in ascending order according to the creation time of the sample cloud host, that is, the information of the sample cloud host with the earliest creation time is ranked first, and the information of the sample cloud host with the latest creation time is ranked last. Among them, the necessary fields include the identification, specification information (i.e., configuration information), creation time, and deletion time of the sample cloud host.

[0086] Step 202: The model training device uses the target time feature extraction model to process multiple first time periods and first requirements to obtain first feature information, second feature information, and third feature information corresponding to the sample device.

[0087] The third characteristic information represents how the impact range of the demand for sample devices during the target period in the multiple first time periods changes over time.

[0088] In an embodiment of the present application, the target time feature extraction model may refer to a Prophet model, and the Prophet model is composed of three core sub-models, namely a trend sub-model, a season sub-model, and a holiday sub-model. Specifically, each first time period may be encoded to obtain a time period label in the format of [yy-MM-dd]. After that, each time period label and the first demand of the corresponding time period are combined into a data set, and then the three sub-models in the Prophet model are used to process multiple time period labels and multiple first demands, respectively, to obtain the first feature information, the second feature information, and the third feature information. It should be noted that the target period may refer to the holiday period in the first time period.

[0089] Exemplarily, the data set obtained by combining the time period label and the first requirement may be shown in the following Table 1:

[0090] Time period label First demand 2021-05-12 160 2021-06-08 200 2023-06-02 178 2023-11-13 192

[0091] Table 1

[0092] In the embodiment of the present application, step 202 can be implemented through steps 202a to 202c.

[0093] Step 202a: The model training device uses the growth trend sub-model in the target time feature extraction model to process multiple first time periods and first demands to obtain first feature information.

[0094] In the embodiment of the present application, the basic composition of the Prophet model can be shown as the following formula (1):

[0095] y(t)=g(t)+s(t)+h(t) Formula (1)

[0096] Among them, g(t) represents the trend sub-model, s(t) represents the season sub-model, and h(t) represents the holiday sub-model.

[0097] In the Prophet model, the trend submodel g(t) is the core component of the Prophet model. The trend submodel prediction of Prophet can be divided into two types: saturated growth submodel and piecewise linear submodel. Specifically, the saturated growth submodel can be expressed by the following formula (2); the piecewise linear submodel can be expressed by the following formula (3):

[0098]

[0099] g line (t) = (r + a(t)δ)t + (d + a(t)T γ ) Formula (3)

[0100] Among them, ν represents the model carrying capacity, r represents the predicted growth rate, δ and γ represent the fitness, a(t) represents the number of times the mutation point changes before time T, and d represents the offset.

[0101] In an embodiment of the present application, a data set consisting of multiple time period labels of the first time period and the first demand can be input into the Prophet model as input parameters. After that, the trend sub-model in the Prophet model will process the input data set to obtain the first characteristic information.

[0102] Step 202b: The model training device uses the seasonal trend sub-model in the target time feature extraction model to process the multiple first time periods and the first demands to obtain second feature information.

[0103] In the Prophet model, the seasonal sub-model s(t) is fitted using the Fourier series, which can be expressed as follows:

[0104]

[0105] Wherein, N represents the number of periods T used in the Prophet model; T represents the expected period length. In one achievable manner, T may be 365.

[0106] In an embodiment of the present application, a data set consisting of multiple time period labels of the first time period and the first demand can be input into the Prophet model as input parameters. After that, the seasonal sub-model in the Prophet model will process the input data set respectively to obtain the second characteristic information.

[0107] Step 202c: The model training device uses the logical growth sub-model in the target time feature extraction model to process the multiple first time periods and the first demands to obtain third feature information.

[0108] In the Prophet model, the holiday sub-model h(t) represents the impact of non-periodic irregular holidays, which can be expressed as follows:

[0109] h(t)=Z(t)k(t) Formula (5)

[0110] Among them, Z(t) is the indicative function, k(t) represents the influence range function of holidays, and k(t)~N(0,V), that is, the influence range function obeys the normal distribution N(0,V). It should be noted that V is an empirical value that can be set by yourself. The larger the V value, the greater the impact of holidays on the model; the smaller the V value, the smaller the impact of holidays on the model.

[0111] In the embodiment of the present application, a logistic growth sub-model can be used to replace the original holiday sub-model, and the logistic growth sub-model can be represented by a logistic growth function that changes with time characteristics (years), that is, at this time,

[0112] In the embodiment of the present application, compared with the original holiday sub-model, the logistic growth sub-model can better reflect the impact range of holidays on the demand for cloud hosts, that is, the impact range of demand follows the logistic growth curve, which is more in line with the actual laws, rather than directly reflecting the impact range of holidays on demand based on the self-set normal distribution value as before. In other words, the present application improves the original Prophet model, modifies the impact range function of the holiday item from a random variable to a logistic growth function that changes with time characteristics, further extracts time characteristics, and improves the accuracy of the extracted time characteristics.

[0113] In an embodiment of the present application, the input data set composed of the time period label of the first time period and the first demand can be processed by the logical growth sub-model in the Prophet model to obtain third characteristic information representing the changing trend of the impact of holiday information in the first time period on the demand for sample cloud hosts over time.

[0114] Step 203: The model training device determines the demand feature information based on the first feature information, the second feature information and the third feature information.

[0115] In the embodiment of the present application, the first characteristic information, the second characteristic information and the third characteristic information may be aggregated to obtain final demand characteristic information.

[0116] Step 204: The model training device classifies the sample devices based on the first configuration information of the sample devices.

[0117] Among them, the first attribute information includes first configuration information.

[0118] In the embodiment of the present application, the first configuration information is the specification information of the sample device, and the first configuration information may include information such as the CPU main frequency, CPU quantity, memory capacity, hard disk data and hard disk capacity of the sample device; the categories of the sample device may include general type, computing type, memory type and network enhanced type. Specifically, the category to which the sample device belongs can be determined from multiple categories based on the first configuration information of each sample device.

[0119] Step 205: The model training device determines the second requirement for each type of sample device in each first time period based on the multiple first requirements.

[0120] In the embodiment of the present application, the demand for sample devices belonging to the same category in each first time period, ie, the second demand, can be determined based on the attribute information and historical demand of the sample devices (ie, the first demand).

[0121] Step 206: The model training device obtains first basic information of the sample device.

[0122] The first basic information is information related to the deployment location of the sample device and the user to which the sample device belongs.

[0123] In the embodiment of the present application, the deployment location may refer to the geographical location where the sample cloud host is deployed, for example, the deployment location may be Suzhou City, Jiangsu Province, or Shanghai City; the user to whom the sample device belongs may refer to the user who rents / purchases the cloud host. Specifically, the deployment location of each sample device and the information related to the renter / purchaser corresponding to each sample device may be obtained from the operator's data center.

[0124] Step 207: The model training device uses the initial gradient decision model to process the category, first basic information, multiple first time periods and demand characteristic information of the sample equipment to obtain demand forecast information for each category of sample equipment.

[0125] The category is determined based on the first attribute information.

[0126] In an embodiment of the present application, the week number information of each first time period can be determined according to the time period label of the time period, and the holiday information of the week and the month can be determined. Then, the original time feature is generated according to the time period label, the week number information and the holiday information. Furthermore, the category and deployment location of each sample device are converted into a dictionary, and a first data set is constructed according to the converted information, the first time period and the demand characteristic information. Then, the XGBOOST model is used to process the first data set to obtain the demand forecast for each type of sample equipment.

[0127] Specifically, the first data set can be divided into a training data set and a test data set in a ratio of 9:1. After that, xgb.DMatrix() can be used to encapsulate the data in the data set, that is, the data in the training data set and the test data set are loaded as DMatrix objects. Further, the encapsulated training data set is input as an input parameter into the initial gradient decision model for processing, so as to obtain the demand forecast amount for each type of sample equipment, that is, the demand forecast information.

[0128] It should be noted that the XGBOOST model is essentially an integration of the results of many weak learners, where each weak learner is a decision tree, so it is also called an integrated model. In the XGBOOST model, there is a sequence between decision trees, that is, the result of the previous decision tree will affect the generation of the next decision tree. In the algorithm, the degree of this influence can be determined by the parameter learning rate (learning rate or eta). When integrating the results of many weak learners, XGBOOST also supports weighting the results of weak learners according to different weights. The specific process of the XGBOOST model can be as follows: Figure 4 shown.

[0129] In each weak learner of XGBOOST, there is an objective function, which is usually obj(θ) = L(θ) + Ω(θ). L(θ) is the loss function, and Ω(θ) is the regularization function. At present, in related technologies, the mean-square error (MSE) function or AUC (the area under the ROC curve and the coordinate axis) is usually used as the loss function. The formula of the MSE function is shown in the following formula (6):

[0130]

[0131] The regularization function is used to control the complexity of the model. When the model has too many parameters, the regularization function can reduce redundant parameters and prevent the model from overfitting. The regularization function can be shown as the following formula (7):

[0132]

[0133] Among them, j is the number of leaf nodes, and ω is the value of the leaf tree node.

[0134] In the embodiment of the present application, the parameters of the XGBOOST model are pre-set, wherein the highest simulation parameter is 50, the learning rate is 30, the maximum depth of the decision tree is 8, the minimum subtree weight is 0.1, and the subsampling ratio is 0.3.

[0135] Step 208: The model training device determines a target loss value based on the target loss function, the second demand, and the demand forecast information.

[0136] Among them, the target loss function is the root mean square error function.

[0137] In the embodiment of the present application, the loss function in the XGBOOST function can be set to a root mean square error function (RMSE). The RMSE can be shown in the following formula (8):

[0138]

[0139] Where n is the number of samples; y i is the actual value; is the predicted value.

[0140] In the embodiment of the present application, the target loss function, that is, the loss value of the RMSE function (that is, the target loss value) can be determined directly based on the second demand, the demand forecast amount and the RMSE function.

[0141] In a feasible way, as the number of training times increases, the RMSE functions of the training dataset and the test dataset show a consistent downward trend, which can be specifically described as follows: Figure 5 As shown, in the embodiment of the present application, by adopting the RMSE function as the loss function, the standard deviation of the true error can be reflected to better identify and punish large prediction errors, thereby obtaining an accurate gradient decision model.

[0142] Step 209: The model training device updates the parameters of the initial gradient decision model based on the target loss value using a target grid search algorithm until a target gradient decision model is obtained.

[0143] In an embodiment of the present application, if the target loss value meets the target threshold, it means that the error of the model at this time is within the preset error range, and the initial gradient decision model at this time can be directly determined as the target gradient decision model; if the target loss value does not meet the target threshold, it means that the error of the model has not reached the preset error range. At this time, the grid search algorithm can be used to adjust (i.e., update) the maximum number of simulations, learning rate, maximum depth of the decision tree, minimum subtree weight and subsampling ratio of the initial gradient decision model until the final target gradient decision model is obtained.

[0144] It should be noted that the purpose of adjusting the maximum number of simulations is to prevent the problem of overfitting of the model due to too many simulations; the purpose of adjusting the learning rate is to maintain the degree of mutual influence between weak learners within a reasonable range to prevent unstable fitting results; the purpose of adjusting the maximum depth of the decision tree and the minimum subtree weight is to reduce the complexity of the model to prevent the model from being too complex and overfitting; the purpose of adjusting the subsampling ratio is to obtain the model parameters for the best prediction effect.

[0145] In another feasible manner, if the target loss value does not meet the target threshold, the parameters of the initial gradient decision model may be adjusted according to a 5-fold cross validation algorithm until the target gradient decision model is obtained.

[0146] In the embodiment of the present application, when using the XGBOOST model to predict demand, time characteristics and other information such as the type and deployment location of the cloud host are taken into consideration, thereby ensuring the accuracy of the prediction results and ensuring that the model is scalable and modifiable.

[0147] In other embodiments of the present application, steps 210 to 211 may be further performed after step 209 .

[0148] Step 210: The model training device obtains the real-time demand of the first device and the second attribute information of the first device in multiple second time periods.

[0149] In the embodiment of the present application, the first device may refer to a cloud host that has been rented or purchased. Specifically, the latest demand for cloud hosts in multiple second time periods (i.e., real-time demand) and attribute information of each first device (i.e., second attribute information) may be obtained in real time from an online database.

[0150] Step 211: The model training device updates the target gradient decision model based on multiple second time periods, real-time requirements and second attribute information.

[0151] In an embodiment of the present application, the target gradient decision model can be updated according to multiple second time periods, real-time demand and second attribute information to achieve adaptive iteration of the model, solve the model lag problem existing in the related technology, and thus improve the accuracy of the determined cloud host demand.

[0152] The model training method provided in the embodiments of the present application can combine the first attribute information of the sample device, multiple first time periods and demand characteristic information to jointly train the initial gradient decision model to obtain a target gradient decision model for predicting the demand of the cloud host, that is, when training the target gradient decision model, the time factor (that is, multiple first time periods), the attribute information of the sample device and the demand characteristic information of the sample device are simultaneously considered, rather than only considering the time factor when predicting demand as in the related art; and long-term demand prediction can be performed through the target gradient decision model, thus solving the problem of inaccurate demand prediction of the cloud host when predicting the demand of the cloud host in the related art, thereby realizing flexible scheduling of the cloud host.

[0153] Based on the above embodiments, the embodiments of the present application provide an information determination method, which can be applied to an information determination device, referring to Figure 6 As shown, the method may include the following steps:

[0154] Step 301: The information determination device obtains the third attribute information of the device to be processed.

[0155] In the embodiment of the present application, the device to be processed may refer to a cloud host for which demand needs to be predicted; the third attribute information may include configuration information (i.e., specification information), name, creation time, etc. of the device to be processed. Specifically, the third attribute information of the device to be processed may be obtained from a data center (i.e., a database).

[0156] Step 302: The information determination device uses a target gradient decision model to process the target time period and the third attribute information to obtain the target demand of the device to be processed in the target time period.

[0157] Among them, the target gradient decision model is obtained after training the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand characteristic information and the second demand of each type of sample device; the demand characteristic information is determined based on multiple first time periods and the first demands of the sample devices in multiple first time periods.

[0158] In one feasible method, the category of the equipment to be processed can be determined based on the third attribute information, and the target time period and target category can be input into the target gradient decision model for processing to obtain the predicted demand for the equipment to be processed in the specified time period (ie, the target time period).

[0159] In another achievable manner, information related to the location of the device to be processed and the user to which it belongs may be obtained first, and then the target gradient decision model is used to process the third attribute information, the target time period and the obtained information to obtain the target demand. In this case, step 302 may be implemented through steps 302a to 302c.

[0160] Step 302a: The information determination device obtains second basic information of the device to be processed.

[0161] The second basic information is information related to the deployment location of the device to be processed and the user to which the device to be processed belongs.

[0162] In the embodiment of the present application, the deployment location of the device to be processed may refer to the geographical location where the cloud host to be predicted is deployed; the user to whom the device to be processed belongs may refer to the user who rents / purchases the cloud host. Specifically, the deployment location of each device to be processed and the information related to the renter / purchaser corresponding to each device to be processed can be obtained from the operator's data center.

[0163] Step 302b: The information determination device determines the target category of the device to be processed based on the third attribute information.

[0164] In an embodiment of the present application, the category to which the device to be processed belongs, ie, the target category, can be determined from multiple categories (general type, computing type, memory type, etc.) according to the configuration information of each device to be processed.

[0165] Step 303c: The information determination device uses a target gradient decision model to process the target period, target category and the second basic information to obtain the target demand.

[0166] In the embodiment of the present application, the target period may refer to the period when the cloud host is needed. Specifically, the target period may be encoded to obtain a period label of the target period, and the period label, target category and second basic information are used as input parameters, that is, the period label, target category and second basic information are input into the target gradient decision model, and then the target gradient decision model may process the input parameters to obtain the predicted demand for the target category of the equipment to be processed (i.e., the target demand).

[0167] In the embodiment of the present application, the comparison result of the demand forecasting result by the target gradient boosting model and the demand forecasting result by the ARIMA model can be as follows: Figure 7 shown.

[0168] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0169] The information determination method provided in the embodiments of the present application can adopt a target gradient decision model trained according to the attribute information of the sample device, multiple first time periods, demand feature information characterizing the change of demand over time, and the historical demand (i.e., the first demand) of each sample device, to analyze the attribute information and the target time period of the device to be processed, and obtain the demand forecast of the device to be processed, which solves the problem of inaccurate demand forecast of the cloud host when predicting the demand of the cloud host in the related technology, thereby realizing flexible scheduling of the cloud host.

[0170] Based on the above embodiments, the present invention provides a model training device, which can be applied to Figure 1 and 3 In the model training method provided in the corresponding embodiment, refer to Figure 8 As shown, the model training device 4 may include: a first acquisition unit 41, a first determination unit 42, a second determination unit 43 and a training unit 44, wherein:

[0171] A first acquisition unit 41 is used to acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods;

[0172] A first determining unit 42 is used to determine demand characteristic information of the sample device based on the multiple first time periods and the first demand; wherein the demand characteristic information represents the change of the demand of the sample device over time;

[0173] A second determining unit 43 is used to determine the second requirement of each type of sample device in each first time period based on the first attribute information and the plurality of first requirements;

[0174] The training unit 44 is used to train the initial gradient decision model based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

[0175] In other embodiments of the present application, the first determining unit 42 is further configured to perform the following steps:

[0176] The target time feature extraction model is used to process the multiple first time periods and the first demands to obtain the first feature information, the second feature information and the third feature information corresponding to the sample device; wherein the third feature information represents the change of the influence range of the target period in the multiple first time periods on the demand for the sample device over time;

[0177] Demand characteristic information is determined based on the first characteristic information, the second characteristic information, and the third characteristic information.

[0178] In other embodiments of the present application, the first determining unit 42 is further configured to perform the following steps:

[0179] The growth trend sub-model in the target time feature extraction model is used to process the plurality of first time periods and the first demands to obtain first feature information;

[0180] The seasonal trend sub-model in the target time feature extraction model is used to process the plurality of first time periods and the first demands to obtain second feature information;

[0181] The plurality of first time periods and first demands are processed using the logical growth sub-model in the target time feature extraction model to obtain third feature information.

[0182] In other embodiments of the present application, the second determining unit 43 is further configured to perform the following steps:

[0183] Classifying the sample devices based on the first configuration information of the sample devices; wherein the first attribute information includes the first configuration information;

[0184] Based on the multiple first requirements, a second requirement for each type of sample equipment in each first time period is determined.

[0185] In other embodiments of the present application, the training unit 44 is further configured to perform the following steps:

[0186] Acquire first basic information of the sample device; wherein the first basic information is information related to the deployment location of the sample device and the user to which the sample device belongs;

[0187] The category, the first basic information, the plurality of first time periods and the demand characteristic information of the sample equipment are processed by using the initial gradient decision model to obtain the demand forecast information of each category of the sample equipment; wherein the category is determined based on the first attribute information;

[0188] Determine a target loss value based on the target loss function, the second demand and the demand forecast information; wherein the target loss function is a root mean square error function;

[0189] Based on the target loss value, the target grid search algorithm is used to update the parameters of the initial gradient decision model until the target gradient decision model is obtained.

[0190] In other embodiments of the present application, the training unit 44 is further configured to perform the following steps:

[0191] Acquire the real-time demand of the first device and the second attribute information of the first device in a plurality of second time periods;

[0192] Based on multiple second time periods, real-time demands and second attribute information, the target gradient decision model is updated.

[0193] It should be noted that the specific description of the steps performed by each unit can be found in Figure 1 and 3 The model training method provided in the corresponding embodiment will not be repeated here.

[0194] The model training device provided in the embodiment of the present application can train the initial gradient decision model in combination with the first attribute information of the sample device, multiple first time periods and demand characteristic information to obtain a target gradient decision model for predicting the demand of the cloud host, that is, when training the target gradient decision model, the time factor (that is, multiple first time periods), the attribute information of the sample device and the demand characteristic information of the sample device are simultaneously considered, rather than only considering the time factor when predicting demand as in the related art; and long-term demand prediction can be performed through the target gradient decision model, thus solving the problem of inaccurate demand prediction of the cloud host when predicting the demand of the cloud host in the related art, thereby realizing flexible scheduling of the cloud host.

[0195] Based on the above embodiments, the present invention provides an information determination device, which can be applied to Figure 6 In the information determination method provided in the corresponding embodiment, refer to Fig. 9 As shown, the information determination device 5 may include: a second acquisition unit 51 and a processing unit 52, wherein:

[0196] A second acquisition unit 51 is used to acquire third attribute information of the device to be processed;

[0197] The processing unit 52 is used to process the target time period and the third attribute information using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand feature information and the second demand of each type of sample device; the demand feature information is determined based on multiple first time periods and the first demands of the sample devices in multiple first time periods.

[0198] In other embodiments of the present application, the processing unit 52 is further configured to perform the following steps:

[0199] Acquire second basic information of the device to be processed; wherein the second basic information is information related to the deployment location of the device to be processed and the user to which the device to be processed belongs;

[0200] Determine a target category of the device to be processed based on the third attribute information;

[0201] The target gradient decision model is used to process the target period, target category and the second basic information to obtain the target demand.

[0202] It should be noted that the specific description of the steps performed by each unit can be found in Figure 6 The information determination method provided in the corresponding embodiment will not be repeated here.

[0203] The information determination device provided in the embodiments of the present application can adopt a target gradient decision model trained according to the attribute information of the sample device, multiple first time periods, demand feature information characterizing the change of demand over time, and the historical demand (i.e., the first demand) of each sample device, to analyze the attribute information and the target time period of the device to be processed, and obtain the demand forecast of the device to be processed, which solves the problem of inaccurate demand forecast of the cloud host when predicting the demand of the cloud host in the related technology, thereby realizing flexible scheduling of the cloud host.

[0204] Based on the above embodiments, the embodiments of the present application provide a model training device, which can be applied to Figure 1 and 3 In the model training method provided in the corresponding embodiment, refer to Fig.10 As shown, the model training device 6 may include: a first processor 61, a first memory 62 and a first communication bus 63, wherein:

[0205] The first communication bus 63 is used to realize the communication connection between the first processor 61 and the first memory 62;

[0206] The first processor 61 is used to execute the model training program in the first memory 62 to implement the following steps:

[0207] Acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods;

[0208] Determine demand characteristic information of the sample device based on the multiple first time periods and the first demand; wherein the demand characteristic information represents the change of the demand of the sample device over time;

[0209] Determine, based on the first attribute information and the plurality of first requirements, a second requirement for each type of sample device in each first time period;

[0210] The initial gradient decision model is trained based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

[0211] In other embodiments of the present application, the first processor 61 is used to execute the model training program in the first memory 62 to determine the demand feature information of the sample device based on the multiple first time periods and the first demand, so as to implement the following steps:

[0212] The target time feature extraction model is used to process the multiple first time periods and the first demands to obtain the first feature information, the second feature information and the third feature information corresponding to the sample device; wherein the third feature information represents the change of the influence range of the target period in the multiple first time periods on the demand for the sample device over time;

[0213] Demand characteristic information is determined based on the first characteristic information, the second characteristic information, and the third characteristic information.

[0214] In other embodiments of the present application, the first processor 61 is used to execute the model training program in the first memory 62 to process multiple first time periods and first requirements using the target time feature extraction model to obtain the first feature information, the second feature information and the third feature information corresponding to the sample device, so as to implement the following steps:

[0215] The growth trend sub-model in the target time feature extraction model is used to process the plurality of first time periods and the first demands to obtain first feature information;

[0216] The seasonal trend sub-model in the target time feature extraction model is used to process the plurality of first time periods and the first demands to obtain second feature information;

[0217] The plurality of first time periods and first demands are processed using the logical growth sub-model in the target time feature extraction model to obtain third feature information.

[0218] In other embodiments of the present application, the first processor 61 is used to execute the model training program in the first memory 62 to determine the second requirement of each type of sample device in each first time period based on the first attribute information and the multiple first requirements, so as to implement the following steps:

[0219] Classifying the sample devices based on the first configuration information of the sample devices; wherein the first attribute information includes the first configuration information;

[0220] Based on the multiple first requirements, a second requirement for each type of sample equipment in each first time period is determined.

[0221] In other embodiments of the present application, the first processor 61 is used to execute the model training program in the first memory 62 to train the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand feature information and second demand to obtain the target gradient decision model, so as to implement the following steps:

[0222] Acquire first basic information of the sample device; wherein the first basic information is information related to the deployment location of the sample device and the user to which the sample device belongs;

[0223] The category, the first basic information, the plurality of first time periods and the demand characteristic information of the sample equipment are processed by using the initial gradient decision model to obtain the demand forecast information of each category of the sample equipment; wherein the category is determined based on the first attribute information;

[0224] Determine a target loss value based on the target loss function, the second demand and the demand forecast information; wherein the target loss function is a root mean square error function;

[0225] Based on the target loss value, the target grid search algorithm is used to update the parameters of the initial gradient decision model until the target gradient decision model is obtained.

[0226] In other embodiments of the present application, the first processor 61 is used to execute the model training program in the first memory 62 to implement the following steps:

[0227] Acquire the real-time demand of the first device and the second attribute information of the first device in a plurality of second time periods;

[0228] Based on multiple second time periods, real-time demands and second attribute information, the target gradient decision model is updated.

[0229] It should be noted that the specific description of the steps performed by the first processor can be referred to Figure 1 and 3 The model training method provided in the corresponding embodiment will not be repeated here.

[0230] The model training device provided in the embodiment of the present application can train the initial gradient decision model in combination with the first attribute information of the sample device, multiple first time periods and demand characteristic information to obtain a target gradient decision model for predicting the demand of the cloud host, that is, when training the target gradient decision model, the time factor (that is, multiple first time periods), the attribute information of the sample device and the demand characteristic information of the sample device are simultaneously considered, rather than only considering the time factor when predicting demand as in the related art; and long-term demand prediction can be performed through the target gradient decision model, thus solving the problem of inaccurate demand prediction of the cloud host when predicting the demand of the cloud host in the related art, thereby realizing flexible scheduling of the cloud host.

[0231] Based on the above embodiments, the embodiments of the present application provide an information determination device, which can be applied to Figure 6 In the information determination method provided in the corresponding embodiment, refer to Fig.11 As shown, the information determination device 7 may include: a second processor 71, a second memory 72, and a second communication bus 73, wherein:

[0232] The second communication bus 73 is used to realize the communication connection between the second processor 71 and the second memory 72;

[0233] The second processor 71 is used to execute the information determination program in the second memory 72 to implement the following steps:

[0234] Obtaining third attribute information of the device to be processed;

[0235] The target time period and the third attribute information are processed using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand characteristic information and the second demand of each type of sample device; the demand characteristic information is determined based on multiple first time periods and the first demand of the sample devices in multiple first time periods.

[0236] In other embodiments of the present application, the second processor 71 is used to execute the information determination program in the second memory 72 to process the target time period and the third attribute information using the target gradient decision model to obtain the target demand of the device to be processed in the target time period, so as to implement the following steps:

[0237] Acquire second basic information of the device to be processed; wherein the second basic information is information related to the deployment location of the device to be processed and the user to which the device to be processed belongs;

[0238] Determine a target category of the device to be processed based on the third attribute information;

[0239] The target gradient decision model is used to process the target period, target category and the second basic information to obtain the target demand.

[0240] It should be noted that the specific description of the steps performed by the second processor can be referred to Figure 6 The information determination method provided in the corresponding embodiment will not be repeated here.

[0241] The information determination device provided in the embodiments of the present application can adopt a target gradient decision model trained according to the attribute information of the sample device, multiple first time periods, demand feature information characterizing the change of demand over time, and the historical demand (i.e., the first demand) of each sample device, to analyze the attribute information and the target time period of the device to be processed, and obtain the demand forecast of the device to be processed, which solves the problem of inaccurate demand forecast of the cloud host when predicting the demand of the cloud host in the related technology, thereby realizing flexible scheduling of the cloud host.

[0242] Based on the foregoing embodiments, the embodiments of the present application provide a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement Figure 1 and 3 The model training method provided by the corresponding embodiment, and Figure 6 The corresponding embodiments provide steps of the information determination method.

[0243] Based on the above embodiments, the embodiments of the present application provide a computer program product, the computer program product includes a computer program, the computer program is executed by the first processor 61 and the second processor 71 to implement Figure 1 and 3 The model training method provided by the corresponding embodiment, and Figure 6 The corresponding embodiments provide steps of the information determination method.

[0244] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0245] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0246] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0247] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0248] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the 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 model training method, characterized in that: The method comprises: Acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods; Based on the multiple first time periods and the first demand, determine demand characteristic information of the sample device; wherein the demand characteristic information represents how the demand of the sample device changes over time; Determine, based on the first attribute information and the plurality of the first requirements, a second requirement for each type of sample device in each first time period; The initial gradient decision model is trained based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

2. The method according to claim 1, characterized in that: The determining, based on the plurality of first time periods and the first requirement, requirement characteristic information of the sample device includes: The target time feature extraction model is used to process the multiple first time periods and the first demand to obtain first feature information, second feature information and third feature information corresponding to the sample device; wherein the third feature information represents the change of the influence range of the target period in the multiple first time periods on the demand for the sample device over time; Demand characteristic information is determined based on the first characteristic information, the second characteristic information, and the third characteristic information.

3. The method according to claim 2, characterized in that The adopting the target time feature extraction model to process the multiple first time periods and the first requirement to obtain the first feature information, the second feature information and the third feature information corresponding to the sample device includes: Using the growth trend sub-model in the target time feature extraction model to process the multiple first time periods and the first demand to obtain the first feature information; Processing the plurality of first time periods and the first demand using a seasonal trend sub-model in the target time feature extraction model to obtain the second feature information; The plurality of first time periods and the first demand are processed using a logical growth sub-model in the target time feature extraction model to obtain the third feature information.

4. The method according to claim 1, characterized in that The determining, based on the first attribute information and the plurality of the first requirements, the second requirement of each type of sample device in each first time period comprises: Classifying the sample devices based on first configuration information of the sample devices; wherein the first attribute information includes the first configuration information; Based on the plurality of the first requirements, a second requirement for each type of sample equipment in each first time period is determined.

5. The method according to claim 1, characterized in that: The training of the initial gradient decision model based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain the target gradient decision model includes: Acquire first basic information of the sample device; wherein the first basic information is information related to the deployment location of the sample device and the user to which the sample device belongs; The initial gradient decision model is used to process the category of the sample device, the first basic information, the multiple first time periods and the demand feature information to obtain demand forecast information for each category of sample device; wherein the category is determined based on the first attribute information; Determine a target loss value based on a target loss function, the second demand and the demand forecast information; wherein the target loss function is a root mean square error function; Based on the target loss value, a target grid search algorithm is used to update the parameters of the initial gradient decision model until the target gradient decision model is obtained.

6. The method according to claim 1, characterized in that The method further comprises: Acquire real-time requirements of the first device in multiple second time periods and second attribute information of the first device; The target gradient decision model is updated based on the multiple second time periods, the real-time demand and the second attribute information.

7. A method for determining information, characterized in that: The method comprises: Obtaining third attribute information of the device to be processed; The target time period and the third attribute information are processed using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training an initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand characteristic information and the second demand of each type of sample device; the demand characteristic information is determined based on multiple first time periods and the first demand of the sample devices in the multiple first time periods.

8. The method according to claim 7, characterized in that The adopting the target gradient decision model to process the target period and the third attribute information to obtain the target demand of the device to be processed in the target period includes: Acquire second basic information of the device to be processed; wherein the second basic information is information related to the deployment location of the device to be processed and the user to which the device to be processed belongs; Determine the target category of the device to be processed based on the third attribute information; The target gradient decision model is used to process the target time period, the target category and the second basic information to obtain the target demand.

9. A model training device, characterized in that: The device comprises: A first acquisition unit, configured to acquire first requirements of sample devices and first attribute information of the sample devices in a plurality of first time periods; A first determining unit, configured to determine demand characteristic information of the sample device based on the multiple first time periods and the first demand; wherein the demand characteristic information represents how the demand of the sample device changes over time; A second determining unit, configured to determine a second requirement for each type of sample device in each first time period based on the first attribute information and a plurality of the first requirements; A training unit is used to train the initial gradient decision model based on the first attribute information of the sample device, the multiple first time periods, the demand feature information and the second demand to obtain a target gradient decision model.

10. An information determination device, characterized in that: The device comprises: A second acquisition unit, used to acquire third attribute information of the device to be processed; A processing unit is used to process the target time period and the third attribute information using a target gradient decision model to obtain the target demand of the device to be processed in the target time period; wherein the target gradient decision model is obtained after training the initial gradient decision model based on the first attribute information of the sample device, multiple first time periods, demand feature information and the second demand of each type of sample device; the demand feature information is determined based on multiple first time periods and the first demand of the sample devices in the multiple first time periods.

11. A model training device, characterized in that: The model training device comprises: a first processor, a first memory and a first communication bus; The first communication bus is used to realize the communication connection between the first processor and the first memory; The first processor is used to execute the model training program in the first memory to implement the steps of the model training method as described in any one of claims 1 to 6.

12. An information determination device, characterized in that: The information determination device includes: a second processor, a second memory, and a second communication bus; The second communication bus is used to realize the communication connection between the second processor and the second memory; The second processor is used to execute the information determination program in the second memory to implement the steps of the information determination method according to any one of claims 7 to 8.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the model training method described in any one of claims 1 to 6, or the information determination method described in any one of claims 7 to 8.

14. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6 or 7 to 8.

Images