Model optimization method and apparatus, and storage medium

Abstract

The application provides a model optimization method and device and a storage medium, relates to the technical field of communication, and can enable data of various enterprises to be shared and improve the accuracy of local models of the various enterprises. The method comprises the following steps: determining the sample quantity and the local model parameter of each first industry device in a plurality of first industry devices; the parameter difference degree between any two first industry devices in the plurality of first industry devices is greater than or equal to a preset threshold; determining the parameter of a shared model based on the sample quantity and the local model parameter of each first industry device; and sending the parameter of the shared model to each first industry device, so that each first industry device optimizes the local model based on the parameter of the shared model. The application is used in the model optimization process.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a model optimization method, apparatus and storage medium. Background Technology

[0002] Currently, data operations across various industries remain confined to their own industry-specific networks. Data cannot be shared between companies within the same industry, nor can it be shared with data from telecom operators. This hinders the development of integrated and innovative applications across these private networks. If each company performs independent modeling within its own network, the model can only be optimized based on that company's data, rather than on data from the entire industry.

[0003] Based on the above, how to enable data sharing among various enterprises is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] This application provides a model optimization method, apparatus, and storage medium that enables data sharing among enterprises and improves the accuracy of local models in each enterprise.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] In a first aspect, this application provides a model optimization method, which includes: determining the number of samples and local model parameters for each of a plurality of first industry devices; the parameter difference between any two first industry devices being greater than or equal to a preset threshold; determining the parameters of a shared model based on the number of samples and local model parameters of each first industry device; and sending the parameters of the shared model to each first industry device so that each first industry device optimizes its local model based on the parameters of the shared model.

[0007] In one possible implementation, the method further includes: receiving data summaries from each of a plurality of second-industry devices; the second-industry devices being industry devices among a plurality of first-industry devices that support data sharing; the data summaries including: the number of samples, and information about each sample among the plurality of samples; the sample information including: the mean of the samples, the variance of the samples, the median of the samples, the maximum value of the samples, and the minimum value of the samples; and determining a plurality of first-industry devices from the plurality of second-industry devices based on the data summaries from the plurality of second-industry devices.

[0008] In one possible implementation, determining multiple first-industry devices from multiple second-industry devices based on data summaries of multiple second-industry devices includes: determining multiple pairs of second-industry devices based on the multiple second-industry devices; each pair of second-industry devices includes any two second-industry devices from the multiple second-industry devices; performing the following operations on each pair of second-industry devices to obtain the device difference degree of the multiple pairs of second-industry devices: determining the device difference degree of a target second-industry device pair based on data summaries of two second-industry devices in a target second-industry device pair; the target second-industry device pair is any one of the multiple pairs of second-industry devices; determining multiple preset second-industry device pairs from the multiple pairs of second-industry devices; the preset second-industry device pairs are second-industry device pairs whose device difference degree is greater than or equal to a preset threshold; determining any one second-industry device as a third-industry device from each of the multiple preset second-industry device pairs; and performing deduplication processing on the multiple third-industry devices to obtain multiple first-industry devices.

[0009] In one possible implementation, the parameters of the shared model satisfy the following formula:

[0010]

[0011] Where, F(θ) target For the parameters of the shared model; K is the number of devices in the multiple first-industry systems; n is the sum of the sample sizes of the devices in the multiple first-industry systems; n k F represents the sample size of the k-th industry equipment among multiple first-industry equipment; k (θ) represents the local model parameters of the kth industry device among multiple first-industry devices.

[0012] Secondly, this application provides a model optimization apparatus, which includes: a communication unit and a processing unit; the processing unit is configured to determine the sample size and local model parameters of each of a plurality of first industry devices; the parameter difference between any two of the plurality of first industry devices is greater than or equal to a preset threshold; the processing unit is further configured to determine the parameters of a shared model based on the sample size and local model parameters of each first industry device; the communication unit is configured to send the parameters of the shared model to each first industry device, so that each first industry device optimizes its local model based on the parameters of the shared model.

[0013] In one possible implementation, the communication unit is further configured to receive data briefings from each of a plurality of second industry devices; the second industry devices are industry devices that support data sharing among a plurality of first industry devices; the data briefings include: the number of samples, and information about each sample among the plurality of samples; the sample information includes: the mean of the samples, the variance of the samples, the median of the samples, the maximum value of the samples, and the minimum value of the samples; the processing unit is further configured to determine a plurality of first industry devices from the plurality of second industry devices based on the data briefings of the plurality of second industry devices.

[0014] In one possible implementation, the processing unit is further configured to determine multiple pairs of second-industry equipment based on multiple pairs of second-industry equipment; a pair of second-industry equipment includes any two second-industry equipment from the multiple pairs of second-industry equipment; the processing unit is further configured to perform the following operation on each pair of second-industry equipment to obtain the equipment difference degree of the multiple pairs of second-industry equipment: determine the equipment difference degree of the target second-industry equipment pair based on the data summary of two second-industry equipment in the target second-industry equipment pair; the target second-industry equipment pair is any one of the multiple pairs of second-industry equipment; the processing unit is further configured to determine multiple preset pairs of second-industry equipment from the multiple pairs of second-industry equipment; the preset pairs of second-industry equipment are pairs of second-industry equipment with an equipment difference degree greater than or equal to a preset threshold; the processing unit is further configured to determine any one of the second-industry equipment from each preset pair of preset pairs of second-industry equipment as a third-industry equipment; the processing unit is further configured to perform deduplication processing on the multiple third-industry equipment to obtain multiple first-industry equipment.

[0015] In one possible implementation, the parameters of the shared model satisfy the following formula:

[0016]

[0017] Where, F(θ) target For the parameters of the shared model; K is the number of devices in the multiple first-industry systems; n is the sum of the sample sizes of the devices in the multiple first-industry systems; n k F represents the sample size of the k-th industry equipment among multiple first-industry equipment; k (θ) represents the local model parameters of the kth industry device among multiple first-industry devices.

[0018] Thirdly, this application provides a model optimization apparatus, which includes: a processor and a communication interface; the communication interface and the processor are coupled, and the processor is used to run computer programs or instructions to implement the model optimization method as described in the first aspect and any possible implementation thereof.

[0019] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a terminal, cause the terminal to perform the model optimization method as described in the first aspect and any possible implementation thereof.

[0020] Fifthly, this application provides a computer program product containing instructions that, when run on a model optimization apparatus, cause the model optimization apparatus to perform the model optimization method as described in the first aspect and any possible implementation thereof.

[0021] In a sixth aspect, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run computer programs or instructions to implement the model optimization method as described in the first aspect and any possible implementation thereof.

[0022] Specifically, the chip provided in this application also includes a memory for storing computer programs or instructions.

[0023] The above technical solution brings at least the following beneficial effects: The model optimization method provided in this application allows the computing server to determine the sample size and local model parameters of each of the multiple first-industry devices (i.e., the parameter difference between any two first-industry devices is greater than or equal to a preset threshold). Based on the sample size and local model parameters of each first-industry device, the server determines the parameters of a shared model and sends these parameters to each first-industry device, enabling each device to optimize its local model based on the shared model parameters. As can be seen from the above, the computing server can perform joint modeling based on the data (i.e., sample size and local model parameters) of each of the multiple first-industry devices, and distribute the parameters of the shared model obtained from the joint modeling to each of the first-industry devices. This allows each first-industry device to optimize its own model based on the parameters of the shared model obtained from the joint modeling, and enables each device to optimize its model based on the shared data of multiple first-industry devices, thereby improving the accuracy of the local model of each first-industry device. Attached Figure Description

[0024] Figure 1 A structural diagram of a communication system provided in an embodiment of this application;

[0025] Figure 2 A flowchart of a model optimization method provided in an embodiment of this application;

[0026] Figure 3 A flowchart of another model optimization method provided in the embodiments of this application;

[0027] Figure 4 A flowchart of another model optimization method provided in the embodiments of this application;

[0028] Figure 5 This is a schematic diagram of the structure of a model optimization device provided in an embodiment of this application;

[0029] Figure 6 This is a schematic diagram of another model optimization device provided in an embodiment of this application. Detailed Implementation

[0030] The model optimization method, apparatus, and storage medium provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0031] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0032] The terms "first" and "second," etc., used in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0033] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0034] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0035] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0036] like Figure 1 As shown, Figure 1 A schematic diagram of a communication system provided in an embodiment of this application is shown. The communication system includes: a computing server 101, an operator server 102, and multiple first industry devices 103.

[0037] The computing server 101 is used to determine the number of samples and local model parameters of each of the multiple first industry devices 103, determine the parameters of the shared model based on the number of samples and local model parameters of each first industry device 103, and send the parameters of the shared model to each first industry device 103 so that each first industry device 103 optimizes its local model based on the parameters of the shared model.

[0038] Among them, the parameter difference between any two first industry equipment 103 is greater than or equal to a preset threshold.

[0039] Optionally, the computing server 101 may include: an industry server, a capability modeling server, and a converged application server. The industry server, capability modeling server, and converged application server can all communicate and interact with multiple first-industry devices 103 to achieve data sharing and collaborative modeling. The following describes each of these different servers:

[0040] The industry server can perform fusion analysis and modeling of samples shared by multiple first industry devices 103 based on industry business characteristics (or features). Different first industry devices 103 can obtain the results of optimal data utilization through shared samples. In this way, the first industry device 103 can improve its own insufficient samples and inaccurate models based on the above results.

[0041] For example, data sharing in the vehicle-to-everything (V2X) industry can effectively regulate the layout of urban charging stations, achieve overall coordination of urban charging stations, promote the precise construction and deployment of charging stations by enterprises, and drive the healthy and sustainable development of the new energy vehicle industry.

[0042] Another example is that data sharing in the bike-sharing industry can optimize the placement of bicycles by various companies, which helps improve bicycle utilization.

[0043] Capability modeling servers can share data on business scenarios within the video industry.

[0044] As an example, data sharing in the video industry can involve sharing video surveillance data from industrial parks and video surveillance data from the public safety industry, and then using this shared data to build capability models.

[0045] For example, the detection requirements for whether factory workers are making or receiving phone calls in the work area or for unauthorized personnel to enter are identification of specific groups of people. The above-mentioned identification of specific groups of people in public safety is the same artificial intelligence (AI) identification model.

[0046] Another example is in location-based application scenarios, where the location data of the Internet of Vehicles can be shared with the location data of industries such as intelligent transportation and shared bicycles, so that subsequent modeling and applications can be carried out based on the shared location data.

[0047] It is understandable that a capability modeling server can act as a capability center, where data from different enterprises can be shared based on the same or similar application scenarios, thereby generating data for capability modeling and improving the accuracy of the modeling.

[0048] Converged application servers can provide a platform for inter-enterprise cooperation and integrated innovative applications. Under the premise of data security, enterprises can explore and innovate cross-domain integrated applications through cross-domain data cooperation, thereby realizing integrated and innovative applications between industry private networks and empowering industry private networks.

[0049] Carrier server 102 is configured with the network architecture of various enterprise-side 5G industry private networks (e.g., virtual network architecture, hybrid network architecture, standalone network architecture, etc.). When computing server 101 distributes tasks, it queries the above-configured network architecture. If the configured network architecture is a virtual network architecture, computing server 101 will also distribute tasks to carrier server 102.

[0050] Multiple industry-specific devices 103 can include multiple industry-specific devices 103 for each of multiple industries. One industry-specific device 103 can correspond to one enterprise in one industry. For example... Figure 1 As shown, the multiple first-industry devices 103 may include: N first-industry devices 103 of industry A, N first-industry devices 103 of industry B, and N first-industry devices 103 of industry C. For example, industry A may be the vehicle-to-everything (V2X) industry, and the N first-industry devices 103 of industry A may include: Tesla... TM First industry equipment 103, and BYD TM The first industry equipment 103. Industry B mentioned above can be the shared bicycle industry, and N first industry equipment 103s in industry B can include: Meituan Bikes. TM The first industry equipment 103, and Qingju bicycles TM The first industry equipment 103. Industry C can be the smart home industry, and the N first industry equipment 103s in industry C can include: Gree air conditioners. TM First industry equipment 103, and Midea air conditioners TM The first industry equipment 103.

[0051] The first industry device 103 is used to provide shared samples to the computing server 101.

[0052] Optional, Figure 1This is just an example framework diagram. Figure 1 The number of nodes included is unlimited, and except for Figure 1 In addition to the functional nodes shown, other nodes may also be included, such as equipment for the second industry, equipment for the third industry, equipment for the fourth industry, etc. This application does not impose any restrictions on this.

[0053] Furthermore, the communication system described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new communication systems, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0054] With the advent of the era of the Internet of Things (IoT), especially after the widespread adoption of 5G networks, the IoT is playing an increasingly important role in communication networks. The IoT enables interconnection between people, between people and things, and between things themselves, creating a new era of ubiquitous connectivity. In this new era, the massive number of connected terminal devices in the IoT, and the vast amounts of data generated by these devices, will provide a huge market for operators to develop IoT connectivity and data services, thus presenting them with a golden opportunity for growth.

[0055] With the increasing volume of data and the new development trend of digital, networked, and intelligent transformation of the current economy and society, enterprises have new expectations for digital operation development. 5G private network solutions have emerged to meet these needs, and operators are beginning to expand private network applications to various industries such as smart manufacturing, smart education, and smart healthcare. Operators' 5G private network products can be customized according to different application scenarios to achieve differentiated capabilities in key factors such as network latency, security isolation, network reliability, network bandwidth, uplink / downlink ratio, and peak rate, meeting the communication service needs of industry users in production, office, and management applications.

[0056] To meet the data transmission isolation needs of users in different industries using 5G private networks, operators primarily employ three network architectures: Virtual Private Network (VPN), Hybrid Private Network (HPB), and Standalone Private Network (SPB). These three architectures are explained in detail below.

[0057] Virtual Private Network (VPN) mode: When initiating data services and forwarding data, the data flow in VPN mode is as follows: 5G private network terminal equipment — access network equipment — core network equipment (e.g., user plane function (UPF)) — enterprise internal application server.

[0058] Hybrid networking configuration: The data flow in the hybrid networking configuration is as follows: 5G private network terminal equipment — access network equipment (e.g., shared access network equipment) — core network equipment (e.g., UPF, multi-access edge computing (MEC)) — enterprise internal application server.

[0059] Independent private network configuration: The data flow in the hybrid network configuration is 5G private network terminal equipment — access network equipment (e.g., private network access network equipment) — core network equipment (e.g., UPF, MEC) — enterprise internal application server.

[0060] Optionally, the content of network service data that operators can collect varies depending on the network topology. Therefore, operators can develop different data operation strategies for industry users based on the collectable network service data and the network topology of the industry users. For example, in a virtual private network (VPN) configuration, data flows through the operator's public network access and core network equipment, allowing operators to collect the data flowing through these channels. Conversely, in hybrid and standalone private network configurations, the network equipment is exclusively for the private network users, and data flow is completely isolated from the operator's public data. This ensures that data flows entirely within the dedicated private network equipment, making it impossible for operators to collect the aforementioned data flow.

[0061] Typically, industry clients store their business data on industry application servers. Due to security and privacy concerns, this data is often self-generated and used independently, creating isolated data silos. As network service providers for industry users, operators can build data-intelligent operation systems for these dedicated network users. This enables data sharing and fusion applications within similar business scenarios across the same industry, significantly improving data utilization. The full utilization of industry data can also powerfully drive technological innovation and economic growth.

[0062] Currently, data operations across various industries remain confined to their own industry-specific networks. Data cannot be shared between companies within the same industry, nor can it be shared with data from telecom operators. This hinders the development of integrated and innovative applications across these private networks. If each company performs independent modeling within its own network, the model can only be optimized based on that company's data, rather than on data from the entire industry.

[0063] Furthermore, the varying sizes of enterprises result in different amounts of data that can be provided. If an enterprise's data is small or incomplete, it will lead to unavailability, incompleteness, and inaccuracy in data operations and applications, which will also hinder model optimization across various industries.

[0064] Based on the above, how to enable data sharing among various enterprises is a problem that urgently needs to be solved by those skilled in the art.

[0065] To address the problems existing in the prior art, this application proposes a model optimization method that enables data sharing among enterprises and improves the optimization level of local model parameters for each enterprise. For example... Figure 2 As shown, the method includes:

[0066] S201. The computing server determines the number of samples and local model parameters for each of the multiple first-industry devices.

[0067] Among them, the parameter difference between any two first industry devices is greater than or equal to a preset threshold.

[0068] In one possible implementation, the multiple samples shared by the aforementioned equipment in the first industry can satisfy the following conditions 1 to 4: Condition 1: Since data skew affects the complexity of model training (or model optimization), the overall balance of positive and negative samples among the multiple samples must be ensured. Condition 2: The parameter difference between the samples shared by each piece of equipment in the first industry needs to be greater than or equal to a preset threshold, so as to ensure that the fusion of data samples provides more information for model optimization. Condition 3: The fields corresponding to the multiple samples shared by the aforementioned equipment in the first industry need to be within a reasonable range and outliers should not appear. Condition 4: The null value rate of the fields corresponding to the multiple samples shared by the aforementioned equipment in the first industry needs to be within a reasonable range (e.g., greater than 30%).

[0069] Optionally, the computing server can set the above-mentioned preset threshold according to the actual situation. For example, the computing server can set the above-mentioned preset threshold to 10. The above is only an exemplary description of the preset threshold, and this application does not impose any limitations on it.

[0070] S202. The computing server determines the parameters of the shared model based on the number of samples for each first-industry device and the local model parameters.

[0071] As an optional implementation, the above-mentioned S202 implementation process can be as follows: the computing server can perform joint modeling based on the sample size of each first industry device and the local model parameters, as well as the following formula 1, to determine the parameters of the shared model.

[0072] In one possible implementation, the parameters of the shared model satisfy the following formula 1:

[0073]

[0074] Where, F(θ) targetThese are the parameters for the shared model. K represents the number of devices in the multiple primary industries. n represents the sum of the sample sizes for the multiple primary industry devices. k Let F be the sample size of the k-th industry equipment among multiple first-industry equipment. k (θ) represents the local model parameters of the kth industry device among multiple first-industry devices.

[0075] S203. The computing server sends the parameters of the shared model to each first-industry device, enabling each first-industry device to optimize its local model based on the parameters of the shared model. Correspondingly, the first-industry device receives the parameters of the shared model from the computing server.

[0076] Understandably, to avoid data leakage during data sharing among multiple industry-specific devices, the shared samples provided by each industry-specific device are stored in its local database and not uploaded to the computing server. The industry-specific devices use a federated learning algorithm to transform the raw data into a data model. Each device then calculates its own model parameters (i.e., local model parameters) using its shared data and updates these parameters to the server. This allows the computing server to aggregate the global model (i.e., joint modeling) based on these local model parameters. Next, the computing server averages the local model parameters of each industry-specific device and distributes the aggregated global model back to all industry-specific devices at the end of each iteration. This federated learning method reduces the risk of data leakage and provides technical assurance for data sharing among industry-specific devices.

[0077] Optionally, after S203 above, each first industry device can learn the model based on the obtained global model and its own data. This data sharing can improve the model accuracy of each first industry device.

[0078] The above technical solution brings at least the following beneficial effects: The model optimization method provided in this application allows the computing server to determine the sample size and local model parameters of each of the multiple first-industry devices (i.e., the parameter difference between any two first-industry devices is greater than or equal to a preset threshold). Based on the sample size and local model parameters of each first-industry device, the server determines the parameters of a shared model and sends these parameters to each first-industry device, enabling each device to optimize its local model based on the shared model parameters. As can be seen from the above, the computing server can perform joint modeling based on the data (i.e., sample size and local model parameters) of each of the multiple first-industry devices, and distribute the parameters of the shared model obtained from the joint modeling to each of the first-industry devices. This allows each first-industry device to optimize its own model based on the parameters of the shared model obtained from the joint modeling, and enables each device to optimize its model based on the shared data of multiple first-industry devices, thereby improving the accuracy of the local model of each first-industry device.

[0079] In one possible implementation, the computing server needs to pre-select the aforementioned multiple first-industry devices from multiple second-industry devices, so that subsequent joint modeling can be performed based on the data from the first-industry devices. Combined with... Figure 2 ,like Figure 3 As shown, the process by which the computing server pre-selects the aforementioned multiple first-industry devices from multiple second-industry devices can be determined by the following steps S301 to S302.

[0080] S301, the second-industry equipment sends a data summary to the computing server. Correspondingly, the computing server receives data summaries from each of the multiple second-industry equipment.

[0081] Among them, the second-industry equipment refers to the equipment in the first industry that supports data sharing. The data summary includes: the number of samples, and information about each sample. Sample information includes: the sample mean, sample variance, sample median, sample maximum, and sample minimum.

[0082] In one example, the data briefing above may also include the number of positive and negative samples.

[0083] Optionally, prior to step S301 above, the computing server sends task information to each of the multiple fourth-industry devices, and correspondingly, each fourth-industry device receives the task information from the computing server. This task information may include the task to be completed in the joint data modeling, and the key features and fields of the data to be provided for sample sharing. The multiple fourth-industry devices are all or some of the industry devices corresponding to the task information. Next, based on the task information, each fourth-industry device determines whether it will participate in the task. If a fourth-industry device determines to participate in the task, it can be identified as a second-industry device, and this second-industry device can determine the number of shared samples and local model parameters based on the key features and fields of the data in the task information. If a fourth-industry device determines to participate in the task, the process ends.

[0084] Understandably, fourth-industry devices with smaller data volumes may refuse to participate in the aforementioned tasks because they are unable to complete the specific task modeling. However, for fourth-industry devices aiming to expand their local data volume, in order to improve the accuracy of their local models, all of these fourth-industry devices (i.e., second-industry devices) can participate in the aforementioned tasks for joint modeling.

[0085] For example, as shown in Table 1 below, if Task #1 is a task for identifying abnormal cards in the vehicle network industry, then the key features and fields of the data required by Task #1 can be time, traffic, access to a Uniform Resource Locator (URL), and location.

[0086] Table 1

[0087] / / / /

[0088] In some examples, the above tasks may include at least one of the following: an abnormal card identification model in the industry server, a path planning model in the industry server, a trajectory matching model in the industry server, a video visual person recognition model in the capability modeling server, a location-based precise positioning model in the capability modeling server, etc. The fusion application server in the capability modeling server will implement the data fusion application model for electricity meters, as well as water meters and gas meters in the capability modeling server.

[0089] S302, The computing server determines multiple first-industry devices from multiple second-industry devices based on data briefings from multiple second-industry devices.

[0090] In one possible implementation, the above-mentioned S302 process can be as follows: The computing server can determine the parameter difference degree between any two second-industry devices based on information included in the data summary of any two second-industry devices among the plurality of second-industry devices (e.g., the mean, variance, median, maximum, and minimum values ​​of parameters), and determine the parameter difference degree between every two second-industry devices among the plurality of second-industry devices based on the above method. Next, the computing server can sort every two second-industry devices (denoted as second-industry device pairs) based on the parameter difference degree, and determine the top M second-industry device pairs. In this case, the preset threshold is the parameter difference degree between the Mth second-industry device pair. M is a positive integer. Finally, the computing server can determine any one second-industry device from each of the M second-industry device pairs as a third-industry device, and perform deduplication processing on the multiple third-industry devices to obtain multiple first-industry devices.

[0091] The above technical solution brings at least the following beneficial effects: The model optimization method provided in this application involves a computing server receiving data summaries from each of multiple second-industry devices (i.e., the second-industry devices are industry devices that support data sharing among multiple first-industry devices). The data summaries include the number of samples and information about each sample among the multiple samples; the sample information includes the mean, variance, median, maximum value, and minimum value of the sample. Based on the data summaries of the multiple second-industry devices, the computing server determines multiple first-industry devices from among the multiple second-industry devices. This allows the computing server to determine the data of the first-industry devices (e.g., the number of samples and local model parameters) and perform joint modeling based on the data of the first-industry devices.

[0092] In an optional embodiment, as shown in S302, the computing server determines multiple first-industry devices from the multiple second-industry devices based on data summaries from the multiple second-industry devices. Figure 3 Based on the illustrated method embodiments, this embodiment provides a possible implementation, such as... Figure 4 As shown, Figure 4 One possible implementation of S302 provided in this application is that the process of determining multiple first industry devices from multiple second industry devices based on data briefings of multiple second industry devices may include the following steps S401 to S405.

[0093] S401, The computing server is based on multiple second-industry devices, and determines multiple pairs of second-industry devices.

[0094] Among them, a second industry equipment pair includes any two second industry equipment from a plurality of second industry equipment.

[0095] As an optional implementation, the above-mentioned S401 implementation process can be as follows: the computing server can determine every two second industry devices as a second industry device pair from the above-mentioned multiple second industry devices, and then determine multiple second industry device pairs.

[0096] The computation server performs S402 for each of the multiple second-industry device pairs to obtain the parameter differences among the multiple second-industry device pairs:

[0097] S402. The calculation server determines the parameter difference degree of the target second industry equipment pair based on the data summary of the two second industry equipment in the target second industry equipment pair.

[0098] Among them, the target second industry equipment pair is any one of multiple second industry equipment pairs.

[0099] In one possible implementation, the above-mentioned S402 implementation process can be as follows: the calculation server calculates the parameter difference degree between any two samples in the target second industry equipment pair based on the Mahalanobis distance method, the data summary of any sample in one second industry equipment in the target second industry equipment pair, and the data summary of any sample in the other second industry equipment in the target second industry equipment pair, and determines the sum of the parameter difference degrees between each pair of samples as the parameter difference degree of the target second industry equipment pair.

[0100] The calculation server, based on the Mahalanobis distance method, determines the parameter difference between any two samples in the target second-industry equipment pair using data summaries of any sample from one second-industry equipment and data summaries of any sample from another second-industry equipment in the target second-industry equipment pair. The process can be as follows: First, the server calculates the covariance between the two samples based on the data summaries of the i-th sample from one second-industry equipment (i.e., the mean, variance, median, maximum, and minimum values ​​of the parameters) and the data summaries of the j-th sample from the other second-industry equipment (i.e., the mean, variance, median, maximum, and minimum values ​​of the parameters).

[0101] Optionally, the covariance satisfies the following formula 2:

[0102] P = ∑ ij Cov(u i v j )=E[(u i -μ a (v) j-μ b )] Formula 2

[0103] Where P is the covariance. i For the data summary of the i-th sample of one of the two second-industry equipment mentioned above, this includes information such as the mean, variance, median, maximum, and minimum values ​​of the parameters.

[0104] u i =(u imean ,u ivariance ,u imedian ,u imax ,u imin )). v j For the data summary of the j-th sample of the other equipment in the second industry mentioned above, the information includes the mean, variance, median, maximum, and minimum values ​​of the parameters (i.e., v). j =(v jmean ,v jvariance ,v jmedian ,v jmax ,v jmin )). μ a For u i The result obtained by processing (i.e., μ) a =E(u) i )). μ b For v j The result obtained by processing (i.e., μ) b =E(v) j i and j are positive integers.

[0105] Secondly, the computing device can determine the Mahalanobis distance between the two samples based on the data summary of the i-th sample of a second industry device (i.e., the mean, variance, median, maximum, and minimum values ​​of the parameters), the data summary of the j-th sample of another second industry device (i.e., the mean, variance, median, maximum, and minimum values ​​of the parameters), the covariance, and preset parameters.

[0106] Optionally, the Mahalanobis distance satisfies the following formula 3:

[0107]

[0108] Among them, D ij (u,v) is the Mahalanobis distance between the two samples mentioned above.

[0109] Ultimately, the computing server determines the sum of the parameter differences between each pair of samples as the parameter difference of the target second industry equipment pair.

[0110] Optionally, the parameter difference between the target second industry equipment pairs satisfies the following formula 4:

[0111]

[0112] Among them, TD ab The parameter difference between the second industry equipment pair is the target.

[0113] S403. The computing server determines multiple preset second industry device pairs from multiple second industry device pairs.

[0114] Among them, the preset second industry equipment pair is the second industry equipment pair whose parameter difference is greater than or equal to the preset threshold.

[0115] Understandably, the greater the parameter difference, the smaller the similarity of shared samples between the pre-defined second industry equipment pairs. Conversely, the smaller the parameter difference, the greater the similarity of shared samples between the pre-defined second industry equipment pairs.

[0116] S404. The computing server determines any one of the second industry devices as the third industry device from each of the multiple preset second industry device pairs.

[0117] Optionally, the descriptions of the preset third-industry equipment can be found in the corresponding locations above, and will not be repeated here.

[0118] S405: The computing server performs deduplication on multiple third-industry devices to obtain multiple first-industry devices.

[0119] Understandably, since the computing server selects any one of the multiple preset second industry devices during the process of determining the third industry device, it is easy for duplicate industry devices to appear among the multiple third industry devices selected by the computing server. Therefore, the computing server needs to perform deduplication on the multiple third industry devices to obtain multiple first industry devices.

[0120] The above technical solution brings at least the following beneficial effects: The model optimization method provided in this application calculates multiple pairs of second-industry equipment based on multiple second-industry equipment (i.e., a pair of second-industry equipment includes any two second-industry equipment from the multiple second-industry equipment), and performs the following operation on each pair of second-industry equipment to obtain the equipment difference degree of the multiple second-industry equipment pairs: Based on the data summary of two second-industry equipment in the target second-industry equipment pair (i.e., the target second-industry equipment pair is any one of the multiple second-industry equipment pairs), the equipment difference degree of the target second-industry equipment pair is determined. Next, the calculation server determines multiple preset second-industry equipment pairs from the multiple second-industry equipment pairs (i.e., preset second-industry equipment pairs are second-industry equipment pairs with equipment difference degree greater than or equal to a preset threshold), and determines any one second-industry equipment as a third-industry equipment from each of the multiple preset second-industry equipment pairs, and performs deduplication processing on the multiple third-industry equipment to obtain multiple first-industry equipment. This allows the calculation server to determine the data of the first-industry equipment (e.g., sample size and local model parameters) and perform joint modeling based on the data of the first-industry equipment.

[0121] It is understood that the above-described model optimization method can be implemented by a model optimization device. To achieve the above functions, the model optimization device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, the embodiments disclosed in this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments disclosed in this application.

[0122] The embodiments disclosed in this application can divide functional modules according to the model optimization device generated by the above method examples. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in the embodiments disclosed in this application is illustrative and is only a logical functional division. In actual implementation, there may be other division methods.

[0123] Figure 5 This is a schematic diagram of a model optimization device provided in an embodiment of the present invention. Figure 5 As shown, the model optimization device 50 can be used to perform... Figures 2-4The model optimization method shown is illustrated. The model optimization device 50 includes a communication unit 501 and a processing unit 502.

[0124] Processing unit 502 is used to determine the number of samples and local model parameters of each of the multiple first industry devices; the parameter difference between any two first industry devices is greater than or equal to a preset threshold; processing unit 502 is also used to determine the parameters of the shared model based on the number of samples and local model parameters of each first industry device; communication unit 501 is used to send the parameters of the shared model to each first industry device so that each first industry device optimizes its local model based on the parameters of the shared model.

[0125] In one possible implementation, the communication unit 501 is further configured to receive data briefings from each of a plurality of second industry devices; the second industry devices are industry devices that support data sharing among a plurality of first industry devices; the data briefings include: the number of samples, and information about each sample among the plurality of samples; the sample information includes: the mean of the samples, the variance of the samples, the median of the samples, the maximum value of the samples, and the minimum value of the samples; the processing unit 502 is further configured to determine a plurality of first industry devices from the plurality of second industry devices based on the data briefings of the plurality of second industry devices.

[0126] In one possible implementation, the processing unit 502 is further configured to determine multiple pairs of second-industry equipment based on multiple pairs of second-industry equipment; a pair of second-industry equipment includes any two second-industry equipment from the multiple pairs of second-industry equipment; the processing unit 502 is further configured to perform the following operation on each pair of second-industry equipment to obtain the equipment difference degree of the multiple pairs of second-industry equipment: determine the equipment difference degree of the target second-industry equipment pair based on the data summary of two second-industry equipment in the target second-industry equipment pair; the target second-industry equipment pair is any one of the multiple pairs of second-industry equipment; the processing unit 502 is further configured to determine multiple preset pairs of second-industry equipment from the multiple pairs of second-industry equipment; the preset pairs of second-industry equipment are pairs of second-industry equipment with an equipment difference degree greater than or equal to a preset threshold; the processing unit 502 is further configured to determine any one of the second-industry equipment from each preset pair of preset pairs of second-industry equipment as a third-industry equipment; the processing unit 502 is further configured to perform deduplication processing on the multiple third-industry equipment to obtain multiple first-industry equipment.

[0127] In one possible implementation, the parameters of the shared model satisfy the following formula:

[0128]

[0129] Where, F(θ) targetFor the parameters of the shared model; K is the number of devices in the multiple first-industry systems; n is the sum of the sample sizes of the devices in the multiple first-industry systems; n k F represents the sample size of the k-th industry equipment among multiple first-industry equipment; k (θ) represents the local model parameters of the kth industry device among multiple first-industry devices.

[0130] In the case of implementing the functions of the integrated modules described above in hardware, this embodiment of the invention provides a possible structural schematic diagram of the model optimization device involved in the above embodiments. For example... Figure 6 As shown, a model optimization device 60, for example, is used to perform... Figures 2-4 The model optimization method is shown. The model optimization device 60 includes a processor 601, a memory 602, a bus 603, and a communication interface 604. The processor 601 and the memory 602 can be connected via the bus 603.

[0131] Processor 601 is the control center of the user equipment and can be a single processor or a collective term for multiple processing elements. For example, processor 601 can be a general-purpose central processing unit (CPU) 602, or other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0132] As one embodiment, processor 601 may include one or more CPUs, for example Figure 6 CPU 0 and CPU 1 are shown in the diagram.

[0133] The memory 602 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0134] In one possible implementation, the memory 602 can exist independently of the processor 601. The memory 602 can be connected to the processor 601 via a bus 603 and is used to store instructions or program code. When the processor 601 calls and executes the instructions or program code stored in the memory 602, it can implement the map drawing method provided in this embodiment of the invention.

[0135] In another possible implementation, the memory 602 can also be integrated with the processor 601.

[0136] Bus 603 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0137] Communication interface 604 is used to connect with other devices via a communication network. This communication network can be Ethernet, a wireless access network, a wireless local area network (WLAN), etc. Communication interface 604 may include a communication unit 501 for receiving data, and may also include an acquisition unit and a receiving unit.

[0138] In one design, the communication interface in the model optimization device 60 provided in this embodiment of the invention can also be integrated into the processor.

[0139] It should be pointed out that, Figure 6 The structure shown does not constitute a limitation on the model optimization device 60. Except... Figure 6 In addition to the components shown, the model optimization device 60 may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0140] As an example, combined Figure 5 The processing unit 502 in the model optimization device performs the same functions as... Figure 6 The processor 601 in it has the same function.

[0141] Through the above description of the embodiments, those skilled in the art will clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0142] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires; portable computer disks; hard disks; random access memory (RAM); read-only memory (ROM); erasable programmable read-only memory (EPROM); registers; hard disks; optical fibers; portable compact disc read-only memory (CD-ROM); optical storage devices; magnetic storage devices; or any suitable combination thereof; or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may reside in an application-specific integrated circuit (ASIC). In the embodiments of this application, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0143] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model optimization method, characterized in that, include: Determine the sample size and local model parameters for each of the multiple first-industry equipment; The parameter difference between any two of the plurality of first industry equipment is greater than or equal to a preset threshold. Based on the sample size and local model parameters of each first industry device, the parameters of the shared model are determined; The parameters of the shared model are sent to each of the first industry devices so that each of the first industry devices optimizes the local model based on the parameters of the shared model; Receive data briefings from each of the multiple secondary industry devices; The second industry equipment is the industry equipment that supports data sharing among the plurality of first industry equipment; The data summary includes: the number of samples, and information about each sample in the sample set; the sample information includes: the mean of the sample, the variance of the sample, the median of the sample, the maximum value of the sample, and the minimum value of the sample. Based on the plurality of second-industry equipment, a plurality of second-industry equipment pairs are determined; a second-industry equipment pair includes any two of the plurality of second-industry equipment. For each of the plurality of second-industry equipment pairs, the following operation is performed to obtain the equipment difference degree of the plurality of second-industry equipment pairs: Based on the data summary of two second-industry equipment in the target second-industry equipment pair, the equipment difference degree of the target second-industry equipment pair is determined; the target second-industry equipment pair is any one of the plurality of second-industry equipment pairs; From the plurality of second industry equipment pairs, a plurality of preset second industry equipment pairs are determined; the preset second industry equipment pairs are second industry equipment pairs whose equipment difference is greater than or equal to a preset threshold. From each of the plurality of preset second industry equipment pairs, determine any one of the second industry equipment as the third industry equipment; The plurality of third-industry equipment is deduplicated to obtain the plurality of first-industry equipment.

2. The method according to claim 1, characterized in that, The parameters of the shared model satisfy the following formula: Where, F(θ) target Here, K represents the parameters of the shared model; K represents the number of the plurality of first industry devices; n represents the sum of the sample sizes of the plurality of first industry devices; n k F represents the sample size of the k-th industry equipment among the plurality of first industry equipment; k (θ) represents the local model parameters of the kth industry device among the plurality of first industry devices.

3. A model optimization device, characterized in that, include: Communication unit and processing unit; The processing unit is used to determine the number of samples and local model parameters for each of the multiple first industry devices. The parameter difference between any two of the plurality of first industry equipment is greater than or equal to a preset threshold. The processing unit is also used to determine the parameters of the shared model based on the number of samples and local model parameters of each first industry device; The communication unit is used to send the parameters of the shared model to each of the first industry devices, so that each of the first industry devices optimizes the local model based on the parameters of the shared model; The communication unit is also configured to receive data briefings from each of the plurality of second industry devices; The second industry equipment is the industry equipment that supports data sharing among the plurality of first industry equipment; The data summary includes: the number of samples, and information about each sample in the sample set; the sample information includes: the mean of the sample, the variance of the sample, the median of the sample, the maximum value of the sample, and the minimum value of the sample. The processing unit is further configured to determine multiple pairs of second industry equipment based on the multiple second industry equipment; a pair of second industry equipment includes any two of the multiple second industry equipment. The processing unit is further configured to perform the following operation on each of the plurality of second industry equipment pairs to obtain the equipment difference degree of the plurality of second industry equipment pairs: determine the equipment difference degree of the target second industry equipment pair based on the data summary of two second industry equipment in the target second industry equipment pair; the target second industry equipment pair is any one of the plurality of second industry equipment pairs; The processing unit is further configured to determine a plurality of preset second industry equipment pairs from the plurality of second industry equipment pairs; the preset second industry equipment pairs are second industry equipment pairs whose equipment difference is greater than or equal to a preset threshold. The processing unit is further configured to determine any one of the second industry devices as a third industry device from each of the plurality of preset second industry device pairs; The processing unit is also used to perform deduplication processing on the plurality of third-industry equipment to obtain the plurality of first-industry equipment.

4. The apparatus according to claim 3, characterized in that, The parameters of the shared model satisfy the following formula: Where, F(θ) target Here, K represents the parameters of the shared model; K represents the number of the plurality of first industry devices; n represents the sum of the sample sizes of the plurality of first industry devices; n K F represents the sample size of the Kth industry equipment among the plurality of first industry equipment; k (θ) represents the local model parameters of the Kth industry device among the plurality of first industry devices.

5. A model optimization device, characterized in that, include: A processor and a communication interface; the communication interface is coupled to the processor, the processor being used to run computer programs or instructions to implement the model optimization method as described in any one of claims 1-2.

6. A computer-readable storage medium storing instructions, characterized in that, When the computer executes the instruction, the computer performs the model optimization method as described in any one of claims 1-2.

Images