A method, device, equipment and medium for improving steady-state performance of a working machine

Abstract

This invention relates to the field of machine learning, and proposes a method, apparatus, device, and medium for improving the steady-state performance of a worker machine. The method includes: establishing a training set and kernel function corresponding to the worker machine; calculating the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain a local prediction model of the worker machine; sending the local prediction model to a server; aggregating the local prediction model based on an aggregation algorithm to obtain a global prediction model; and performing a fusion calculation on the worker machine of the local prediction model and the global prediction model to obtain a prediction model that approximates the objective function. The method disclosed in this invention can guarantee the steady-state performance improvement of each worker machine after fusing local and global predictions; from the perspective of steady-state performance improvement, the mean squared error can be directly used for comparison and quantification of performance improvement; and the performance improvement can be demonstrated through experiments.

Description

Technical Field

[0001] This invention relates to the field of machine learning, and in particular to a method, apparatus, device, and medium for improving the steady-state performance of a machine. Background Technology

[0002] The Internet of Things (IoT) generates massive amounts of distributed data. A typical training method involves storing this data on servers and training models through these servers. However, this approach suffers from significant communication and computational inefficiencies. For example, the hundreds of gigabytes of data generated by a car over several hours represent a tremendous burden during transmission and computation. Distributed machine learning effectively addresses the challenges of excessive computation, large amounts of training data, and large model sizes. Current technologies are primarily limited to offline learning, while in practical applications, machines receive data streams. Therefore, online learning is an effective way to solve this problem. Existing methods for improving the learning performance of predictive workers compare the width of the confidence intervals between the global prediction error and the local prediction error of the machine, lacking quantifiable error metrics. Therefore, a method for improving the steady-state performance of the machine that can quantify error is needed. Summary of the Invention

[0003] In view of this, the present invention proposes a method, apparatus, device, and medium for improving the steady-state performance of the working machine. The proposed method first processes streaming data using Gaussian process regression (GPR) to predict the test output. Then, the expectation and variance of the local predictions are sent to a server. The server aggregates the expected and variance data from all the working machines, calculates the expectation and variance of the global prediction, and sends the resulting global prediction model back to each working machine. This proposed method improves the steady-state performance of the working machine, resulting in a prediction model with better predictive performance under steady-state conditions after the fusion of global and local predictions.

[0004] To achieve the above objectives, one aspect of the present invention provides a method for improving the steady-state performance of a working machine. The method includes the following steps: establishing a training set and a kernel function corresponding to the working machine; calculating the Gaussian posterior probability distribution of the working machine on the training set based on the kernel function to obtain a local prediction model of the working machine; sending the local prediction model to a server; aggregating the local prediction model based on an aggregation algorithm to obtain a global prediction model; and performing a fusion calculation on the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function.

[0005] In some embodiments, the method further includes: establishing a validation dataset and distributing the validation dataset equally among worker machines; calculating the mean square error of the prediction model of the approximation objective function for the worker machines using the prediction model of the approximation objective function; comparing the mean square error of the prediction model of the approximation objective function with the mean square error of the local prediction model; and determining that the prediction model of the approximation objective function is valid in response to the fact that the mean square error of the prediction model of the approximation objective function is less than the mean square error of the local prediction model.

[0006] In some embodiments, establishing a training set and kernel function corresponding to the worker machine, and calculating the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain the local prediction model of the worker machine includes: establishing a projection data set corresponding to the training data, and obtaining the neighborhood of the projection data in the projection data set; constructing a training set corresponding to the worker machine based on the neighborhood of the projection data; calculating the Gaussian posterior probability distribution of the worker machine on the training set according to the selected kernel function, and obtaining the expectation and variance of the local prediction of the worker machine.

[0007] In some embodiments, sending the local prediction model to the server and aggregating the local prediction model based on an aggregation algorithm to obtain a global prediction model includes: sending the expectation and variance of the local prediction to the server, and using the server's aggregation algorithm to calculate the expectation and variance of the local prediction as the expectation and variance of the global prediction.

[0008] In some embodiments, the step of fusing the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function includes: sending the expectation and variance of the global prediction to the working machine, and setting the working machine to perform fusing calculations on the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function.

[0009] In some embodiments, sending the expectation and variance of the global prediction to the worker machine, and setting the worker machine to perform a fusion calculation of the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function includes: comparing the variance of the local prediction of the worker machine's test data with the variance of the global prediction, and constructing a set of test data for the worker machine based on the comparison result; in response to the existence of test data in the set of test data for the worker machine where the variance of the local prediction is not greater than the variance of the global prediction, using the global prediction to construct a prediction model that approximates the objective function.

[0010] In some embodiments, sending the expectation and variance of the global prediction to the worker machine, and setting the worker machine to perform fusion calculation of the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function, further includes: in response to the absence of test data in the set of test data of the worker machine where the variance of the local prediction is not greater than the variance of the global prediction, using the local prediction to construct a prediction model that approximates the objective function.

[0011] In another aspect, the present invention provides an apparatus for improving the steady-state performance of a working machine. The apparatus includes: a first module configured to establish a training set and a kernel function corresponding to the working machine, and to calculate the Gaussian posterior probability distribution of the working machine on the training set based on the kernel function to obtain a local prediction model of the working machine; a second module configured to send the local prediction model to a server, and to aggregate the local prediction model based on an aggregation algorithm to obtain a global prediction model; and a third module configured to perform a fusion calculation on the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function.

[0012] In another aspect of the present invention, a computer device is also provided, including at least one processor; and a memory storing computer instructions executable on the processor, the instructions, when executed by the processor, implementing the steps of any of the methods described above.

[0013] In another aspect of the present invention, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements any of the method steps described above.

[0014] This invention has at least the following beneficial effects: It proposes a method, apparatus, device, and medium for improving the steady-state performance of a working machine. Specifically, the method for improving the steady-state performance of a working machine is constructed by directly comparing and quantifying the mean square error, thus establishing an online fusion method for improving the steady-state performance of working machines. This ensures the steady-state performance improvement of each working machine after fusion using local and global predictions. From the perspective of steady-state performance improvement, the mean square error can be directly used for comparison and quantification of performance improvement. Furthermore, the method proposed in this invention allows for experimental verification of the performance improvement. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0016] Figure 1 A schematic diagram illustrating an embodiment of a method for improving the steady-state performance of a working machine provided by the present invention;

[0017] Figure 2 A schematic diagram of an embodiment of a device for improving the steady-state performance of a working machine provided by the present invention;

[0018] Figure 3 A schematic diagram illustrating an embodiment of a computer device provided by the present invention;

[0019] Figure 4 This is a schematic diagram of an embodiment of a computer-readable storage medium provided by the present invention. Detailed Implementation

[0020] The following describes embodiments of the present invention. However, it should be understood that the disclosed embodiments are merely examples, and other embodiments may take various alternative forms.

[0021] Furthermore, it should be noted that all uses of the terms "first" and "second" in the embodiments of this invention are for the purpose of distinguishing two entities or parameters with the same name but different names. Therefore, "first" and "second" are merely for convenience of expression and should not be construed as limiting the embodiments of this invention. Subsequent embodiments will not elaborate on this further. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements may include not only those elements but also elements not expressly listed or inherent to those processes, methods, articles, or apparatuses.

[0022] One or more embodiments of this application will now be described with reference to the accompanying drawings.

[0023] Based on the above objectives, the first aspect of the present invention provides an embodiment of a method for improving the steady-state performance of a working machine. Figure 1 The diagram shown is a schematic representation of an embodiment of a method for improving the steady-state performance of a working machine provided by the present invention. Figure 1 As shown, a method for improving the steady-state performance of a working machine according to an embodiment of the present invention includes the following steps:

[0024] S1. Establish the training set and kernel function corresponding to the working machine, and calculate the Gaussian posterior probability distribution of the working machine on the training set based on the kernel function to obtain the local prediction model of the working machine.

[0025] S2. Send the local prediction model to the server, and aggregate the local prediction model based on the aggregation algorithm to obtain the global prediction model;

[0026] S3. The local prediction model and the global prediction model are fused and calculated on the working machine to obtain a prediction model that approximates the objective function.

[0027] Based on the above objectives, in the first aspect of the present invention, another embodiment of a method for improving the steady-state performance of a working machine is also proposed.

[0028] Distributed machine learning addresses situations with excessive computational demands, large amounts of training data, and overly large model sizes. For computationally intensive scenarios, multi-threaded or multi-machine parallel computation based on shared memory (or virtual memory) can be employed. For scenarios with excessive training data, the data needs to be partitioned and distributed across multiple worker nodes for training, ensuring that the local data on each worker node remains within acceptable limits. Each worker node trains a sub-model based on its local data and communicates with other worker nodes according to certain rules (primarily regarding sub-model parameters or parameter updates) to ensure the effective integration of training results from various worker nodes to obtain a global machine learning model. For overly large models, the model needs to be partitioned and distributed across different worker nodes for training. Unlike data parallelism, model parallelism involves strong dependencies between sub-models, as the output of one sub-model may be the input of another. Without communication of intermediate computation results, the entire model training cannot be completed.

[0029] Typical distributed machine learning uses deep neural networks as the machine learning model, mainly applied to pattern classification and pattern recognition. However, it is limited to offline learning. In practical applications, the working machine receives data streams in real-time, making online learning a means to solve this problem. Gaussian process regression is one effective method. The Gaussian process model can be equivalent to existing machine learning models, including Bayesian linear models and multilayer neural networks. According to the central limit theorem, assuming the weights in the neural network follow a Gaussian normal distribution, as the width of the neural network approaches infinity, such a neural network is equivalent to Gaussian process regression. However, Gaussian process regression is a non-hyperparametric statistical probability model, unlike traditional learning models such as linear regression, logistic regression, and neural networks, which require solving an optimization problem to minimize the loss function to obtain the optimal model parameters. Gaussian process regression does not require solving an optimization problem. Given training data and test input, Gaussian process regression prediction consists of two steps: inference and prediction. The inference process assumes that the function to be learned follows a Gaussian process, giving the model's Gaussian prior probability distribution, and then using the observed values ​​and the Bayesian rule to calculate the model's Gaussian posterior probability distribution. After completing the local model predictions, each worker machine sends its local predictions (expectation and variance) to the server, allowing the server to calculate the global model, for example, using an average aggregation algorithm. Finally, the server sends the calculated global model (global expectation and variance) back to each worker machine, enabling the worker machine to fuse the global model with its own trained local models to obtain an updated prediction of the objective function that more closely approximates the true value of the function.

[0030] This invention proposes a method to improve the steady-state performance of the working machine by directly comparing the mean squared error of the global prediction from the server with the mean squared error of the local prediction from the working machine, and quantifying the improvement in prediction performance. When the constructed input set is not empty, the fused prediction is more accurate than the local prediction of the working machine; when the input set is empty, the working machine uses local prediction. This performance improvement can be guaranteed both theoretically and experimentally.

[0031] Define the objective function as in is n x The input space is one-dimensional. Without loss of generality, we assume the output is one-dimensional, i.e., At time t, given The corresponding output is

[0032] y=f(x)+ε (1)

[0033] ε follows a pattern with a mean of 0 and a variance of . Gaussian noise with a Gaussian probability distribution, i.e. Define a training set in the following form in The input data set is y = [y(1), y(2), ..., y(n)]. s )] T It is a column vector that aggregates the outputs. The goal of Gaussian process regression is to utilize the training set... In the test dataset The function f is approximated.

[0034] Define a symmetric positive semi-definite kernel function. That is, ∫k(x,x')f(x)f(x')dν(x)dν(x')≥0, where ν is a measure. Let Return a column vector such that its i-th element equals f(x(i)). Assume the function f is a sample from a Gaussian process prior probability distribution with mean μ and kernel function k. Then the training output and test output... Follows a joint probability distribution

[0035]

[0036] in and Return the values ​​of μ(x(i)) and μ(x). * (i) is a vector. Return a matrix such that the element in the i-th row and j-th column is k(x(i), x). * (j)).

[0037] Gaussian process regression utilizes the properties of Gaussian processes and the training set. Predict test dataset The output. This output It follows a normal distribution, that is... here

[0038]

[0039] In distributed machine learning, consider a network with n worker machines. Define this set as... At every moment t, every working machine Utilizing local training data To predict the function for the test input The output. y i (t)=[y i (1),...,y i The local prediction value for each worker machine during training is (t)

[0040]

[0041] In a federated learning framework, each worker machine will generate its trained local predictions. cov i (f * Send it to the server.

[0042] The following are the specific steps for distributed training and fusion.

[0043] (1) Construct a training subset based on the projection of the training set, and define the distance between two training data points x and x' as d(x,x')=||x-x'||, where x is the distance from the data point x to the set. The distance is Define data point x to set The projection set is

[0044] Consider each working machine and its local training dataset For a test data x * Calculate the test data x * To training set The projection is labeled as:

[0045]

[0046] At each time t, this set of local projections Includes One projection data, namely

[0047] For each working machine and its projection set Take out each projection point and label it as... Here, the subscript j represents the j-th projection point. Then, for each projection point... Find one of its neighborhoods Make And targeting δ>0. It is important to note here that the number of neighbors is adjustable, allowing for a fixed selection.

[0048] For each working machine Construct a new training set

[0049] (2) Selecting a kernel function: In practical applications, a kernel function is generally selected.

[0050]

[0051] (3) For each working machine In the new training set The Gaussian posterior probability distribution is calculated above, i.e.:

[0052]

[0053] In the training subset Local predictions are obtained using formula (7). and cov i (f * Then, this local prediction is sent to the server. The steady-state mean square error of the local prediction on the working machine is...

[0054]

[0055] (4) The server uses the following aggregation algorithm to aggregate the local predictions and provide a global prediction:

[0056]

[0057] The steady-state mean square error of the server-side global prediction is:

[0058]

[0059] (5) Based on the global and local prediction variances, design a fusion algorithm for each worker machine to make the fused prediction expectation closer to the true value of function f. Define variables:

[0060]

[0061] in, Depends on kernel function (5), It is the measurement noise variance of the i-th working machine, which depends on ε.

[0062] Construct a test data set x with small uncertainty. * The set is as follows:

[0063]

[0064] If this set is not empty, the global prediction from the server will be used; if this set is empty, the local prediction from the worker will be used. Typically, a constant is given satisfying Ω > 0, and when the worker satisfies... Theoretically, it can be guaranteed that:

[0065]

[0066] (6) Validate the effectiveness of the prediction model using the constructed dataset. Set the objective function as follows:

[0067] f(x) = 5x 2sin(12x)+(x 3 -0.5)sin(3x-0.5)+4cos(2x)

[0068] and Gaussian noise We generate 6000, 10 in the interval [0, 1]. 4 4×10 4 15×10 4 4×10 5 The training data was used, and 120 test points were randomly selected. Assuming there are 40 worker machines in the network, the training data was divided into 40 equal parts, so that each worker machine has 250 training data points. The corresponding data is listed in Table 1, using the experimental results of worker machine 1 and worker machine 6 as examples.

[0069] <![CDATA[Training size n s ]]> 6000 10000 40000 150000 400000 MSE / i = 1 / local 0.0721 0.0681 0.0585 0.0527 0.0582 MSE / i=1 / fused 0.0096 0.0681 0.0094 0.0537 0.0071 MSE / i=6 / local 0.0727 0.0566 0.0564 0.0555 0.0597 MSE / i=6 / fused 0.0727 0.0086 0.0564 0.0076 0.0071

[0070] Table 1

[0071] As shown in Table 1, after using the fusion algorithm, the mean squared errors of predictions for both machines 1 and 6 are no higher than the mean squared errors of local predictions; moreover, the order of magnitude of the prediction error after fusion is reduced to 0.001. Furthermore, the prediction errors decrease as the dataset size increases.

[0072] The method for improving the steady-state performance of the working machine proposed in this invention can guarantee the steady-state performance improvement of each working machine after fusing local and global predictions. From the perspective of steady-state performance improvement, the mean square error can be directly used for comparison and quantification of performance improvement. At the same time, experiments can be used to demonstrate the performance improvement.

[0073] A second aspect of the present invention provides an apparatus for improving the steady-state performance of a machine. Figure 2 The diagram shown is a schematic representation of an embodiment of a device for improving the steady-state performance of a working machine according to the present invention. Figure 2 As shown, the device for improving the steady-state performance of a working machine provided by the present invention includes: a first module 011, configured to establish a training set and kernel function corresponding to the working machine, and calculate the Gaussian posterior probability distribution of the working machine on the training set based on the kernel function to obtain a local prediction model of the working machine; a second module 012, configured to send the local prediction model to a server, and aggregate the local prediction model based on an aggregation algorithm to obtain a global prediction model; and a third module 013, configured to perform fusion calculation on the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function.

[0074] To achieve the above objectives, a third aspect of the present invention provides a computer device. Figure 3The diagram shown is a schematic representation of an embodiment of a computer device provided by the present invention. Figure 3 As shown, an embodiment of a computer device provided by the present invention includes the following modules: at least one processor 021; and a memory 022, the memory 022 storing computer instructions 023 that can be executed on the processor 021, the computer instructions 023 implementing the steps of the method described above when executed by the processor 021.

[0075] The present invention also provides a computer-readable storage medium. Figure 4 The diagram shown is a schematic representation of an embodiment of a computer-readable storage medium provided by the present invention. Figure 4 As shown, computer-readable storage medium 031 stores a computer program 032 that, when executed by a processor, performs the methods described above. The methods executed by computer program 032 include: establishing a training set and kernel function corresponding to a worker machine; calculating the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain a local prediction model for the worker machine; sending the local prediction model to a server; aggregating the local prediction model based on an aggregation algorithm to obtain a global prediction model; and performing a fusion calculation on the local prediction model and the global prediction model on the worker machine to obtain a prediction model that approximates the objective function.

[0076] In some embodiments, the method further includes: establishing a validation dataset and distributing the validation dataset equally among worker machines; calculating the prediction mean square error corresponding to the worker machines using the prediction model; comparing the prediction mean square error with the mean square error of the local prediction model; and determining that the prediction model is valid in response to the prediction mean square error being less than the mean square error of the local prediction model.

[0077] In some embodiments, establishing a training set and kernel function corresponding to the worker machine, and calculating the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain the local prediction model of the worker machine includes: establishing a projection data set corresponding to the training data, and obtaining the neighborhood of the projection data in the projection data set; constructing a training set corresponding to the worker machine based on the neighborhood of the projection data; calculating the Gaussian posterior probability distribution of the worker machine on the training set according to the selected kernel function, and obtaining the expectation and variance of the local prediction of the worker machine.

[0078] In some embodiments, sending the local prediction model to the server and aggregating the local prediction model based on an aggregation algorithm to obtain a global prediction model includes: sending the expectation and variance of the local prediction to the server, and using the server's aggregation algorithm to calculate the expectation and variance of the local prediction as the expectation and variance of the global prediction.

[0079] In some embodiments, the step of fusing the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function includes: sending the expectation and variance of the global prediction to the working machine, and setting the working machine to perform fusing calculations on the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function.

[0080] In some embodiments, sending the expectation and variance of the global prediction to the worker machine, and setting the worker machine to perform a fusion calculation of the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function includes: comparing the variance of the local prediction of the worker machine's test data with the variance of the global prediction, and constructing a set of test data for the worker machine based on the comparison result; in response to the existence of test data in the set of test data for the worker machine where the variance of the local prediction is not greater than the variance of the global prediction, using the global prediction to construct a prediction model that approximates the objective function.

[0081] In some embodiments, sending the expectation and variance of the global prediction to the worker machine, and setting the worker machine to perform fusion calculation of the expectation and variance of the global prediction and the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function, further includes: in response to the absence of test data in the set of test data of the worker machine where the variance of the local prediction is not greater than the variance of the global prediction, using the local prediction to construct a prediction model that approximates the objective function.

[0082] Finally, it should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program for setting system parameters can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The storage medium for the program can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. The above computer program embodiments can achieve the same or similar effects as any of the corresponding foregoing method embodiments.

[0083] Furthermore, the method disclosed in the embodiments of the present invention can also be implemented as a computer program executed by a processor, which may be stored in a computer-readable storage medium. When the computer program is executed by the processor, it performs the functions defined in the method disclosed in the embodiments of the present invention.

[0084] Furthermore, the above-described method steps and system units can also be implemented using a controller and a computer-readable storage medium for storing a computer program that enables the controller to perform the functions of the above-described steps or units.

[0085] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the functionality of various illustrative components, blocks, modules, circuits, and steps has been generally described. Whether this functionality is implemented as software or as hardware depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the embodiments disclosed herein.

[0086] In one or more exemplary designs, functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functionality may be stored as one or more instructions or code on or transmitted via a computer-readable medium. Computer-readable media include computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one location to another. Storage media may be any available medium accessible to a general-purpose or special-purpose computer. By way of example, and not limitation, computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, or any other medium that may be used to carry or store the required program code in the form of instructions or data structures and is accessible to a general-purpose or special-purpose computer or a general-purpose or special-purpose processor. Furthermore, any connection may be appropriately referred to as computer-readable media. For example, if software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the aforementioned coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are all included in the definition of media. As used herein, disks and optical discs include compact discs (CDs), laser discs, optical discs, digital multifunction discs (DVDs), floppy disks, and Blu-ray discs, wherein disks typically reproduce data magnetically, while optical discs reproduce data optically using lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0087] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.

[0088] It should be understood that, as used herein, the singular form “a” is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, “and / or” refers to any and all possible combinations of one or more of the associated listed items.

[0089] The embodiment numbers disclosed in the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0090] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0091] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A method for improving the steady-state performance of a working machine, characterized in that, include: Establish a training set and kernel function corresponding to the working machine, and calculate the Gaussian posterior probability distribution of the working machine on the training set based on the kernel function to obtain the local prediction model of the working machine; The local prediction model is sent to the server, and the local prediction model is aggregated based on the aggregation algorithm to obtain the global prediction model; The process of fusing the local prediction model and the global prediction model on a working machine to obtain a prediction model that approximates the objective function includes: sending the expectation and variance of the global prediction to the working machine; setting the working machine to fuse the expectation and variance of the global prediction with the expectation and variance of the local prediction to obtain a prediction model that approximates the objective function; further including: comparing the variance of the local prediction and the variance of the global prediction of the test data on the working machine, and constructing a set of test data for the working machine based on the comparison result; responding to the existence of test data in the set of test data on the working machine where the variance of the local prediction is not greater than the variance of the global prediction, using the global prediction to construct a prediction model that approximates the objective function; responding to the absence of test data in the set of test data on the working machine where the variance of the local prediction is not greater than the variance of the global prediction, using the local prediction to construct a prediction model that approximates the objective function. Establish a validation dataset and distribute the validation dataset equally among the working machines. Calculate the mean square error of the prediction model of the approximate objective function for the working machines using the prediction model of the approximate objective function. The mean square error of the prediction model that approximates the objective function is compared with the mean square error of the local prediction model. If the mean square error of the prediction model that approximates the objective function is less than the mean square error of the local prediction model, the prediction model that approximates the objective function is valid.

2. The method according to claim 1, characterized in that, The process of establishing a training set and kernel function corresponding to the worker machine, and calculating the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain the local prediction model of the worker machine includes: Establish a set of projected data corresponding to the training data, and obtain the neighborhood of the projected data in the set of projected data. Construct a training set corresponding to the worker based on the neighborhood of the projected data. The Gaussian posterior probability distribution of the worker machine is calculated on the training set based on the selected kernel function, and the expectation and variance of the local predictions of the worker machine are obtained.

3. The method according to claim 2, characterized in that, The step of sending the local prediction model to the server and aggregating the local prediction model based on an aggregation algorithm to obtain the global prediction model includes: The expected value and variance of the local prediction are sent to the server, and the expected value and variance of the local prediction are calculated into the expected value and variance of the global prediction through the aggregation algorithm of the server.

4. A device for improving the steady-state performance of a machine, characterized in that, include: The first module configures the training set and kernel function corresponding to the worker machine, and calculates the Gaussian posterior probability distribution of the worker machine on the training set based on the kernel function to obtain the local prediction model of the worker machine. The second module is configured to send the local prediction model to the server, and aggregate the local prediction model based on the aggregation algorithm to obtain the global prediction model; The third module is configured to perform fusion calculations on the local prediction model and the global prediction model on the working machine to obtain a prediction model that approximates the objective function. And the modules configured for the following functions: Establish a validation dataset and distribute the validation dataset equally among the working machines. Calculate the mean square error of the prediction model of the approximate objective function for the working machines using the prediction model of the approximate objective function. The mean square error of the prediction model that approximates the objective function is compared with the mean square error of the local prediction model. If the mean square error of the prediction model that approximates the objective function is less than the mean square error of the local prediction model, the prediction model that approximates the objective function is valid. The third module is further configured to: send the expected value and variance of the global prediction to the working machine; set the working machine to perform a fusion calculation of the expected value and variance of the global prediction and the expected value and variance of the local prediction to obtain a prediction model that approximates the objective function; and further configured to: compare the variance of the local prediction and the variance of the global prediction of the working machine's test data, and construct a set of test data for the working machine based on the comparison results; in response to the existence of test data in the set of test data of the working machine where the variance of the local prediction is not greater than the variance of the global prediction, construct a prediction model that approximates the objective function using the global prediction; and in response to the absence of test data in the set of test data of the working machine where the variance of the local prediction is not greater than the variance of the global prediction, construct a prediction model that approximates the objective function using the local prediction.

5. A computer device, characterized in that, include: At least one processor; as well as A memory storing computer instructions executable on the processor, which, when executed by the processor, implement the steps of the method according to any one of claims 1-3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-3.

Images