Methods, devices, and electronic equipment for predicting the remaining available time of image acquisition devices

Abstract

This application provides a method, apparatus, and electronic device for predicting the remaining available time of an image acquisition device, relating to the field of device detection technology. The method includes: acquiring operational status information of the target image acquisition device in at least one dimension; inputting the operational status information of each dimension into an available time prediction model trained using artificial intelligence technology to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model; and obtaining the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension. This application embodiment can obtain the remaining available time of the image acquisition device with high accuracy, and the efficiency of training the available time prediction model is also higher than related technologies.

Description

Technical Field

[0001] This application relates to the field of equipment testing technology, and more specifically, to a method, apparatus, and electronic device for predicting the remaining usable time of an image acquisition device. Background Technology

[0002] In the field of facial recognition, the image acquisition device is a core component, and its quality greatly affects the facial recognition experience. Measuring the usable time of the image acquisition device is an important aspect of its quality.

[0003] Current methods for predicting the usable time of image acquisition devices can only predict whether the usable time of the image acquisition device will end, that is, whether the image acquisition device can continue to be used. They cannot predict the remaining usable time of the image acquisition device, which makes it impossible for users to prepare for the replacement of the image acquisition device in time, resulting in a poor user experience. Summary of the Invention

[0004] This application provides a method, apparatus, and electronic device for predicting the remaining available time of an image acquisition device that overcomes or at least partially solves the above-mentioned problems.

[0005] In a first aspect, embodiments of this application provide a method for predicting the remaining usable time of an image acquisition device, the method comprising:

[0006] Acquire the operational status information of the target image acquisition device in at least one dimension;

[0007] The operational status information of each dimension is input into a pre-trained available time prediction model for the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence of the operational status information of the corresponding dimension on the remaining available time of the target image acquisition device.

[0008] Based on the influence factors of the target image acquisition device in each dimension, the remaining available time of the target image acquisition device is obtained;

[0009] Each available duration prediction model is trained based on the operational status information and influencing factors of the sample image acquisition device in the corresponding dimension.

[0010] Secondly, embodiments of this application provide a training method for an available duration prediction model, comprising:

[0011] At least one training sample is obtained, and each training sample includes the running status of the sample image acquisition device in the corresponding dimension at each prediction time and the actual remaining available time at the corresponding prediction time.

[0012] The operating status of the sample image acquisition device at each prediction time in the corresponding dimension is input into the policy function of the available duration prediction model to be trained, and the probability distribution of the influence factors of the sample image acquisition device at the corresponding prediction time in the corresponding dimension is obtained from the policy function output.

[0013] Sampling is performed based on the probability distribution to obtain the remaining available time for the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and a historical prediction influence factor sequence is constructed for the corresponding prediction time.

[0014] Input the actual remaining available time at the corresponding prediction time and the historical prediction impact factor sequence into the reward function to obtain the reward value at the corresponding prediction time output by the reward function;

[0015] Based on the reward value and probability distribution of influencing factors at each prediction time, the parameters of the policy function are trained, and the trained policy function is used as the training available duration prediction model.

[0016] Thirdly, embodiments of this application provide a device for predicting the remaining usable time of an image acquisition device, comprising:

[0017] The status information acquisition module is used to acquire the operating status information of the target image acquisition device in at least one dimension;

[0018] The available time prediction module for each dimension is used to input the running status information of each dimension into the pre-trained available time prediction model of the corresponding dimension, and obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence on the remaining available time of the target image acquisition device.

[0019] The available time summary module is used to obtain the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension;

[0020] Each available duration prediction model is trained based on the operational status information and influencing factors of the sample image acquisition device in the corresponding dimension.

[0021] Fourthly, embodiments of this application provide a training apparatus for an available duration prediction model, comprising:

[0022] The training sample acquisition module is used to acquire at least one training sample. Each training sample includes the running status information of the sample image acquisition device in the corresponding dimension at each prediction time and the actual remaining available time at the corresponding prediction time.

[0023] The probability distribution acquisition module is used to input the operating status information of the sample image acquisition device in the corresponding dimension at each prediction time into the policy function of the available duration prediction model to be trained, and obtain the probability distribution of the influence factors of the sample image acquisition device in the corresponding dimension at the corresponding prediction time output by the policy function.

[0024] The impact factor sequence module is used to sample according to the probability distribution, obtain the remaining available time for the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and construct the historical prediction impact factor sequence at the corresponding prediction time.

[0025] The reward value acquisition module is used to input the actual remaining available time and the historical prediction impact factor sequence at the corresponding prediction time into the reward function to obtain the reward value at the corresponding prediction time output by the reward function.

[0026] The training module is used to train the parameters of the policy function based on the reward value and the probability distribution of the influencing factors at each prediction time, and the trained policy function is used as the training available duration prediction model.

[0027] Fifthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method provided in the first or second aspect.

[0028] In a sixth aspect, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first or second aspect.

[0029] In a seventh aspect, embodiments of this application provide a computer program that includes computer instructions stored in a computer-readable storage medium. When a processor of a computer device reads the computer instructions from the computer-readable storage medium, the processor executes the computer instructions, causing the computer device to perform steps that implement the method provided in the first aspect.

[0030] The method, apparatus, electronic device, and storage medium for predicting the remaining available time of an image acquisition device provided in this application embodiment obtain the operating status information of the target image acquisition device in at least one dimension, input the operating status of each dimension into the corresponding available time prediction model, and obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. Since this application embodiment constructs a prediction model for each dimension of operating status information, the prediction result is more accurate. Although the number of prediction models to be trained is large, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster to train multiple prediction models than to train a prediction model with multi-dimensional operating status information. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models at the same time, thereby further improving the training efficiency of all models. This application embodiment improves the accuracy and reliability of remaining available time prediction. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0032] Figure 1 A schematic diagram of the implementation environment for the method of predicting the remaining available time of the image acquisition device provided in the embodiments of this application;

[0033] Figure 2 A schematic diagram illustrating an application scenario provided in an embodiment of this application;

[0034] Figure 3 A flowchart illustrating a method for predicting the remaining available time of an image acquisition device provided in an embodiment of this application;

[0035] Figure 4 An optional flowchart is provided for an embodiment of this application;

[0036] Figure 5 A schematic diagram of a CNN model provided for an embodiment of this application;

[0037] Figure 6 A flowchart illustrating a method for predicting the remaining available time of an image acquisition device provided in another embodiment of this application;

[0038] Figure 7 A schematic diagram illustrating an application scenario provided for another embodiment of this application;

[0039] Figure 8 A flowchart illustrating the training process of a reinforcement learning-based model is provided for embodiments of this application.

[0040] Figure 9 This is a schematic diagram of the structure of an available duration prediction model provided in an embodiment of this application;

[0041] Figure 10 This is a schematic diagram of the training process of the available time prediction model in an embodiment of this application;

[0042] Figure 11 A schematic diagram of the structure of an available duration prediction model to be trained, provided for another embodiment of this application;

[0043] Figure 12 A schematic diagram of the structure of a device for predicting the remaining available time of an image acquisition apparatus provided in an embodiment of this application;

[0044] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0045] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting the invention.

[0046] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0048] This application relates to technologies related to artificial intelligence (AI). AI is the theory, methods, technology, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities.

[0049] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0050] This application primarily concerns machine learning within artificial intelligence. Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0051] Supervised learning is a type of machine learning where a function (model parameters) is learned from a given training dataset. When new data arrives, this function can be used to predict the outcome. The training set for supervised learning requires both input and output, or features and labels. The labels in the training set are manually labeled. Supervised learning is most commonly used for classification problems (distinguishing between classification and clustering). It trains an optimal model (a set of functions, where optimal means best under a certain evaluation criterion) using existing training samples (known data and their corresponding outputs). This model is then used to map all inputs to corresponding outputs, and simple judgments are made on the outputs to achieve classification. This gives it the ability to classify unknown data.

[0052] Reinforcement learning, also known as rewarded learning or evaluative learning, is another type of machine learning. It involves an agent learning through trial and error, receiving rewards that guide its behavior based on interactions with the environment. The goal is to maximize the agent's reward value. Reinforcement learning differs from supervised learning in that it does not require labeling input-output pairs or explicitly correcting suboptimal actions.

[0053] Policy gradient method is a type of method in reinforcement learning. Its characteristic is that it can directly output actions based on the current state without having to evaluate the value or potential benefits of each action.

[0054] Episode: Similar to the concept of a "game" in chess, a game is called an episode; similarly, the entire process from the start of a game to victory / defeat is also called an episode. In the prediction of the remaining usable time of the image acquisition device in this application embodiment, an episode is the entire process from when the image acquisition device is first used after leaving the factory until the remaining usable time reaches 0, i.e., it cannot be used or is no longer recommended for use. The total number of times the operating status information is collected during this process is denoted as T.

[0055] The method, apparatus, electronic device, and computer-readable storage medium for predicting the remaining available time of an image acquisition device provided in this application are intended to solve the above-mentioned technical problems of the prior art.

[0056] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0057] The image acquisition device in this application embodiment can be an industrial color camera, a 3D camera, a smartphone, or other devices with shooting functions.

[0058] Optionally, the application scenarios for the prediction method of the remaining available time of the image acquisition device provided in this application include, but are not limited to, at least one of the following:

[0059] 1. In the facial recognition payment scenario, the payment terminal activates the image acquisition device to collect the device's operating status information. This information is then input into a pre-trained available time prediction model. Based on the influencing factors output by the model, the remaining available time is determined. If the remaining available time is not greater than a preset threshold, an alarm is issued, indicating that facial recognition payment is not permitted. If the remaining available time exceeds the preset threshold, the collected facial image data is sent to the payment server, which then performs the facial recognition payment based on the data.

[0060] 2. In access control scenarios, the access control terminal activates the image acquisition device to collect its operating status information. This information is then input into a pre-trained available time prediction model. Based on the influence factor output by the model, the remaining available time is determined. If the remaining available time is not greater than a preset threshold, an alarm is issued, prompting the visitor to verify their identity using other methods, such as fingerprint or voiceprint recognition. If the remaining available time exceeds the preset threshold, the collected facial image data is sent to the identity verification server, which then verifies the visitor's identity based on the facial image data.

[0061] 3. In the autonomous driving scenario, the vehicle terminal activates the image acquisition device to collect the operating status information of the image acquisition device. The operating status information is input into the pre-trained available time prediction model, and the remaining available time is obtained based on the influence factor output by the available time prediction model. If the remaining available time is not greater than a preset threshold, an alarm message is issued to remind the occupants not to engage in autonomous driving. If the remaining available time is greater than the preset threshold, the collected road image data is sent to the server. The server identifies the road conditions based on the road image data and generates an autonomous driving strategy to be fed back to the vehicle terminal.

[0062] 4. In public safety scenarios, the security camera system in public areas activates its image acquisition device to collect operational status information. This information is then input into a pre-trained available time prediction model. Based on the influencing factors output by the model, the remaining available time is determined. If the remaining available time is not greater than a preset threshold, an alarm is issued, prompting relevant personnel to repair the image acquisition device promptly. If the remaining available time exceeds the preset threshold, the collected facial image data is sent to an audit server. The audit server then audits the facial image data and identifies preset individuals within the crowd.

[0063] Of course, the application scenarios of the method for predicting the remaining available time of the image acquisition device provided in this application are not limited to the above situations. Other application scenarios that require the prediction method for the remaining available time of the image acquisition device are also within the protection scope of this application.

[0064] In this application embodiment, the method for predicting the remaining available time of the image acquisition device involves big data processing technology in the field of artificial intelligence. Big data refers to data sets that cannot be captured, managed, and processed within a certain time frame using conventional software tools. It is a massive, rapidly growing, and diverse information asset that requires new processing models to achieve stronger decision-making, insight discovery, and process optimization capabilities. With the advent of the cloud era, big data has attracted increasing attention. Big data requires special technologies to effectively process large amounts of data within a tolerable elapsed time frame.

[0065] This application embodiment uses big data technology to analyze and process the operating status information of a large number of user devices that initiate prediction requests for remaining available time, thereby predicting the remaining available time in a timely manner, which has high application value in the field of device detection.

[0066] The method for predicting the remaining available time of an image acquisition device provided in this application embodiment may specifically include: acquiring the operating status information of the target image acquisition device in at least one dimension; inputting the operating status information of each dimension into a pre-trained available time prediction model of the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model, wherein the influence factor is used to characterize the degree of influence on the remaining available time of the target image acquisition device; and obtaining the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension; wherein each available time prediction model is trained based on the operating status information and influence factor of the sample image acquisition device in the corresponding dimension.

[0067] It should be understood that the method for predicting the remaining available time of the image acquisition device provided in this application embodiment can be applied to data processing devices, such as terminal devices, servers, etc. Specifically, the terminal device can be a smartphone, computer, personal digital assistant (PDA), tablet computer, vehicle terminal, vehicle-mounted device, wearable device, etc. The server can be an application server or a web server. In actual deployment, the server can be a standalone server, a cluster server, or a cloud server.

[0068] When the method for predicting the remaining available time of the image acquisition device provided in this application embodiment is executed by the terminal device, the terminal device can further display the remaining available time and even issue an alarm based on the operating status information of the target image acquisition device of the target user who initiated the remaining available time prediction request in at least one dimension.

[0069] When the method for predicting the remaining available time of the image acquisition device provided in this application embodiment is executed by the server, the server can first obtain the remaining available time based on the obtained operating status information of the target image acquisition device in at least one dimension, and then send the remaining available time to the terminal device for display.

[0070] Please see Figure 1 The illustration provides an exemplary schematic diagram of the implementation environment for the method for predicting the remaining available time of an image acquisition device provided in this application embodiment. Figure 1 As shown, the implementation environment includes: server 100 and multiple user devices 200.

[0071] User equipment refers to the terminal device used by the user. The terminal device can be various electronic devices with image acquisition capabilities, including but not limited to smartphones, tablets, laptops, wearable devices, in-vehicle systems, and vehicle terminals. The user device has a target application installed on it, which can be an application or an operating system. The target application provides the runtime environment for the facial recognition service. For example, the target application can be a payment application, and the facial recognition service can be implemented based on this application, which provides the runtime environment for the facial recognition service. Alternatively, the target application can be the Android operating system, and the facial recognition service can be implemented based on a quick app of the target application, which provides the runtime environment for the quick app used to provide the facial recognition service. The user can send a remaining available time prediction request to the server 100 via the user device 200. This remaining available time prediction request includes the operating status information of the image acquisition device in at least one dimension.

[0072] When any user triggers a remaining available time prediction request on the user device, the user device sends the remaining available time prediction request to the server 100, so that the server 100 can determine the remaining available time of the image acquisition device in the user device based on the remaining available time prediction request.

[0073] The server in this application embodiment can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0074] The server execution method in this application embodiment can be implemented in the form of cloud computing. Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network providing resources is called the "cloud." From the user's perspective, resources in the "cloud" are infinitely scalable, readily available, on-demand, expandable, and pay-as-you-go.

[0075] As a provider of fundamental cloud computing capabilities, a cloud resource pool (referred to as a cloud platform, generally called an IaaS (Infrastructure as a Service) platform) is established. Various types of virtual resources are deployed in the resource pool for external customers to choose from. The cloud resource pool mainly includes: computing devices (virtualized machines containing operating systems), storage devices, and network devices.

[0076] Based on logical function, a PaaS (Platform as a Service) layer can be deployed on top of the IaaS (Infrastructure as a Service) layer, and a SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. Alternatively, SaaS can be deployed directly on top of IaaS. PaaS is a platform for running software, such as databases and web containers. SaaS refers to various types of business software, such as web portals and bulk SMS senders. Generally speaking, SaaS and PaaS are upper layers compared to IaaS.

[0077] Server 100 is used to execute the prediction method for the remaining available time of the image acquisition device provided in the embodiments of this application. Based on the operating status information of the image acquisition device in at least one dimension sent by the user equipment, the server 100 obtains the current remaining available time of the image acquisition device and sends the current remaining available time to the user equipment. If the current remaining available time is not higher than a preset threshold (e.g., 0), the user equipment can terminate the face recognition process or not accept the face recognition result, thereby protecting data security.

[0078] It should be understood that the application scenarios provided above are merely examples. In practical applications, the method for predicting the remaining available time of the image acquisition device provided in this application embodiment can also be applied to other application scenarios that require prediction of the remaining available time. No limitation is made here on the components of the application scenarios of the method for predicting the remaining available time of the image acquisition device provided in this application embodiment.

[0079] Please see Figure 2The application scenario of this application is the prediction stage of the remaining available time of the 3D camera. The terminal can accurately know the distance of each point in the image from the camera by obtaining the data from the 3D camera. By adding the coordinates of the point in the two-dimensional image, the three-dimensional spatial coordinates of each point in the image can be obtained. The real scene can be restored by the three-dimensional coordinates, and applications such as scene modeling and face recognition can be realized.

[0080] The communication module in the terminal is responsible for collecting the operation data of the 3D camera periodically or irregularly. The operation data includes the operation status information of the 3D camera in at least one dimension. The operation data is sent to the backend through the communication module in the terminal.

[0081] After receiving the operating status information from the 3D camera, the backend communication module saves the operating data in the database of the data storage server.

[0082] At regular intervals, the available time prediction model in the prediction server retrieves the operating data of each 3D camera from the database. Based on the operating data, it is input into the pre-trained available time prediction model. The available time prediction model will predict the influencing factors based on the operating status information of the corresponding dimension of the 3D camera. The prediction server further obtains the remaining available time of the 3D camera based on each influencing factor and returns it to the data storage server for storage.

[0083] After storing the remaining available time of the 3D camera, the data storage server can return the remaining available time to the terminal in real time through the background. When the terminal learns that the remaining available time is not higher than a preset value (e.g., 0), it will display an alarm message that the 3D camera is unavailable.

[0084] Please see Figure 3 The figure illustrates, by way of example, a flowchart of a method for predicting the remaining available time of an image acquisition device provided in an embodiment of this application, as shown in the figure, including:

[0085] S101. Obtain the operating status information of the target image acquisition device in at least one dimension.

[0086] This application embodiment collects operational status information from various dimensions to more accurately predict the remaining lifespan of the image acquisition device. The operational status information can describe the real-time operational status of the image acquisition device, such as operational status information in dimensions such as image quality, startup speed, streaming speed, and shutdown speed. It can also describe the cumulative or average operational status over a period of time, such as operational status information in dimensions such as activation duration (the time span from device activation to the current prediction time), total operating time, average image quality, average startup speed, average streaming speed, average shutdown speed, and total number of failures. This application embodiment does not impose specific limitations.

[0087] S102. Input the running status information of each dimension into the pre-trained available time prediction model of the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence on the remaining available time of the target image acquisition device.

[0088] When determining whether an image acquisition device is malfunctioning, related technologies obtain multi-dimensional operational status information. However, they often construct a feature vector using all dimensions of operational status information. Each row or column of this feature vector represents one dimension of operational status information. Then, using this feature vector, which contains operational status information from all dimensions, and whether the remaining available time is 0, as labels, a neural network model is trained. After the model is trained, the multi-dimensional operational status information of the image acquisition device to be predicted is obtained, and the corresponding feature vector is further obtained. Finally, the feature vector is input into the trained model, and the prediction result of whether the remaining available time of the image acquisition device is 0 is output.

[0089] Through experimentation, the inventors discovered that the method described above, which constructs a feature vector from all dimensions of operational status information and trains a corresponding neural network, does not achieve higher accuracy and efficiency despite integrating all dimensions of operational status information for training. However, this application's embodiment designs a neural network model for each dimension of operational status information. Each neural network model outputs an influence factor of the image acquisition device in the corresponding dimension. This influence factor characterizes the degree to which the operational status information in the corresponding dimension affects the remaining usable time of the image acquisition device. After verification, the method of training the corresponding neural network for each dimension of the running state information adopted in the embodiments of this application can not only obtain a highly accurate prediction result of the remaining available time, but more importantly, it overcomes the bias in related technologies that it is more efficient to train uniformly after the running state information is fused. Although this application has M (M is a positive integer) dimensions of running state, it needs to train M neural network models, which seems to be slower in terms of time cost than training a model with multiple specific dimensions of features. In fact, since the features only have one dimension of running state, the convergence speed during training is faster. Even if M models are trained, it is still faster than training a model with multiple dimensions of features. If parallel training is adopted, the training speed of the model can be further improved, making the training efficiency much greater than that of related technologies.

[0090] It should be understood that before executing step S102, the available time prediction model for each dimension can be pre-trained. Specifically, the available time prediction model for each dimension can be trained in the following way:

[0091] First, a certain number of sample image acquisition devices are collected, along with their corresponding dimensions, at different prediction times. The influence factors for each dimension at each prediction time are then determined. These influence factors characterize the degree to which the operational status information in each dimension affects the remaining usable time of the target image acquisition device. The influence factors can be labeled by maintenance personnel according to preset evaluation criteria. It should be understood that for the same image acquisition device, the corresponding influence factors often change when the prediction time is different. For example, regarding the influence factor for the startup time dimension, as the sample image acquisition devices age, the startup time often gradually increases, and the degree to which the corresponding influence factor affects the remaining usable time also changes.

[0092] Subsequently, the initial model is trained based on the corresponding dimension of the sample image acquisition device's operational status information and influencing factors. The operational status information of the sample image acquisition device in the corresponding dimension is used as training samples, and the influencing factors of the sample image acquisition device in the corresponding dimension are used as sample labels, thereby obtaining the available duration prediction model for the corresponding dimension. The process of training the initial model is the process of determining the model parameters of the initial model. When determining the model parameters, an initial model parameter can be initialized first, and then continuously optimized during subsequent training. The optimized model parameter is then used as the model parameter of the available duration prediction model. The initial model in this embodiment can be a single neural network model or a combination of multiple neural network models.

[0093] S103. Based on the influence factors of the target image acquisition device in each dimension, obtain the remaining available time of the target image acquisition device.

[0094] The impact factor in this application embodiment can be represented in two ways: positive or negative impact on the remaining usable time. When represented as a positive impact, the impact factor is used to characterize the duration for which the image acquisition device can continue to be used due to the corresponding dimension of the operating status information. For example, if the camera's startup time is 0.3 seconds, when the impact factor represents the degree of positive impact, the impact factor may be 3 months, indicating that when the startup time of the image acquisition device is 0.3 seconds, the startup speed information leads to the image acquisition device being able to continue to be used for 3 months. Or, for example, if the startup time is 0.2 seconds, when the impact factor represents the degree of positive impact, the impact factor may be 4 months, indicating that when the startup time of the image acquisition device is 0.2 seconds, the startup speed information leads to the image acquisition device being able to continue to be used for 4 months. Furthermore, the impact factor of each dimension can be summed, and the summation result can be used as the remaining usable time.

[0095] When the impact factor is expressed as the degree of negative impact, the impact factor is used to indicate the amount of available time that the corresponding dimension of operational status information will reduce the rated available time. For example, if the camera's turn-on speed is 0.3 seconds, when the impact factor is used to express the degree of negative impact, the impact factor may be 2 months, indicating that when the turn-on speed is 0.3 seconds, the information in the dimension of turn-on speed will reduce the rated available time by 2 months. Furthermore, the remaining available time can be obtained by summing the impact factors of each dimension and subtracting the summation result from the working years of the image acquisition device.

[0096] Furthermore, the remaining available time in this application embodiment can be an absolute time or a relative time, such as 0-100. When the remaining available time is 0, it means that it cannot be used anymore. When it is 100, it means that the remaining available time is at least the rated usage time. When it is s (s is any number from 0 to 100), it means the rated usage time × s × 100%.

[0097] Please see Figure 4 The figure illustrates an optional flowchart of an embodiment of this application. As shown, in order to accurately determine the remaining usable time (i.e., remaining lifespan) of the image acquisition device (shown as a camera in the figure), nine dimensions of operational status information are collected, including:

[0098] The historical average color quality score is obtained by averaging the color quality scores at each prediction time in history. The color quality score can be obtained by evaluating the color information of the image according to preset rules. The color information can include color saturation, white balance, color reproduction, etc.

[0099] The historical average depth quality score is obtained by averaging the depth quality scores at each prediction time in history. The depth quality score can be obtained by averaging the depth information of the image according to preset rules. The depth information can include detection distance, accuracy level, field of view, frame rate, etc.

[0100] The historical average camera opening speed is obtained by averaging the camera opening speeds at each prediction time in history.

[0101] The historical average camera streaming speed is obtained by averaging the camera streaming speed at each prediction time in history. The streaming speed is the speed at which video streams are acquired from the camera.

[0102] The historical average camera shutdown speed is obtained by averaging the camera shutdown speeds at each predicted time in history.

[0103] The manufacturing date can reflect the natural aging degree of the camera to some extent.

[0104] Total running time.

[0105] Historical error count, which is the total number of times the camera malfunctioned during operation.

[0106] Historical average infrared image quality score is obtained by averaging the infrared image quality scores at each prediction time in history. Commonly used indicators for evaluating infrared image commands include target confusion and target occlusion.

[0107] The deep learning network integrates available time prediction models for each dimension, and can also directly determine the remaining available time of the image acquisition device based on the influence factors output by each available time prediction model. In this embodiment, the remaining available time is expressed as a percentage.

[0108] The method for predicting the remaining available time of an image acquisition device according to the embodiments of this application obtains the operating status information of the target image acquisition device in at least one dimension, inputs the operating status of each dimension into the available time prediction model of the corresponding dimension, and obtains the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. Since the embodiments of this application construct a prediction model for the operating status information of each dimension, the prediction result is more accurate. Although the number of prediction models that need to be trained is large, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster than training a prediction model with multi-dimensional operating status information, even if multiple prediction models need to be trained. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models at the same time, thereby further improving the training efficiency of all models. The embodiments of this application improve the accuracy and reliability of the prediction of the remaining available time.

[0109] Convolutional Neural Networks (CNNs) are feedforward neural networks that incorporate convolutional computations and have a deep structure; they are one of the representative algorithms of deep learning. The following uses a CNN model as an example to illustrate the process of training the initial model in this application's embodiments.

[0110] A CNN model includes convolutional layers, pooling layers, and fully connected layers. This application does not specify the exact number of convolutional layers and fully connected layers in a CNN model. In practice, these can be configured as needed. Please refer to... Figure 5 The example shown is a schematic diagram of a CNN model. Figure 5The CNN model shown includes two convolutional layers, one pooling layer, and one fully connected layer. This application does not specifically limit the number of convolutional kernels and bias matrices in each convolutional layer, or the number of weight matrices and bias vectors in each fully connected layer. Similarly, this application does not limit the dimensions of each convolutional kernel and bias matrix, or the dimensions of each weight matrix and bias vector. In practice, the number and dimensions of the convolutional kernels and bias matrices in each convolutional layer, as well as the number and dimensions of the weight matrices and bias vectors in each fully connected layer, can all be taken from empirical values.

[0111] When the operational status information of one dimension of the sample image acquisition device is input into the first-level convolutional layer, the first-level convolutional layer performs dimensionality reduction and feature extraction on the input operational status information through convolution operations to obtain the first feature. This first feature is then input into the second-level convolutional layer, which performs dimensionality reduction and feature extraction on the input first feature through convolution operations to obtain the second feature. This second feature is then input into the pooling layer to reduce the dimensionality of the features and the parameters of the model. The pooling layer primarily replaces a region of a feature with a single value, such as the maximum or average value. If the maximum value is used, it is called max pooling; if the average value is used, it is called mean pooling. The fully connected layer can use the softmax function to map the output of the pooling layer to the corresponding influence factor of the sample image acquisition device.

[0112] Please see Figure 6 The example illustrates a flowchart of a method for predicting the remaining available time of an image acquisition device according to another embodiment of this application, such as... Figure 6 As shown, the operating status information of the target image acquisition device in multiple dimensions is obtained. Figure 6 The system specifically displays the running status information in three dimensions: total running time, number of failures, and image quality. The running status information of each dimension is input into the available time prediction model of the corresponding dimension. The available time prediction model consists of convolutional layers, pooling layers, and fully connected layers. The system obtains the influence factor of the corresponding dimension output by the available time prediction model. Then, the remaining available time is obtained by summing the influence factors of all dimensions.

[0113] Based on the above embodiments, as an optional embodiment, the remaining available time of the target image acquisition device is obtained according to the influence factor of the target image acquisition device in each dimension, including:

[0114] S201. Obtain the weight corresponding to the running status information of each dimension.

[0115] The embodiments of this application can predetermine the weight of different dimensions of operation status information for the remaining available time, and record the weights of different dimensions of operation status information in a preset weight table, so that the weight corresponding to each dimension of operation status information can be obtained by looking up the table.

[0116] S202. Based on the weights, the influence factors of the target image acquisition device in the corresponding dimension are weighted, and the remaining available time of the target image acquisition device in the corresponding dimension is obtained based on the weighting results.

[0117] Since the impact factor also corresponds to the operating status information, the remaining available time of the target image acquisition device in the corresponding dimension can be obtained by weighted summation of the impact factor and the weight of the corresponding dimension.

[0118] As can be seen from the above embodiments, the influence factor can represent either the degree of positive or negative impact on the remaining available time. Therefore, if the influence factor represents the positive impact on the remaining available time, the weighted result of the influence factor is the corresponding remaining available time. If the influence factor represents the negative impact on the remaining available time, the weighted result of the influence factor is the time lost from the rated available time in the corresponding dimension. By subtracting the lost time from the rated available time in the corresponding dimension, the remaining available time of the target image acquisition device in the corresponding dimension can be obtained. In an optional scheme, the influence factor in this application embodiment is used to characterize the degree of positive impact of the operating status information in the corresponding dimension on the remaining available time of the target image acquisition device. Accordingly, obtaining the remaining available time of the target image acquisition device in the corresponding dimension based on the weighted result includes: using the weighted result as the remaining available time of the target image acquisition device in the corresponding dimension.

[0119] Specifically, if the impact factor represents the degree of positive impact on the remaining available time, then the remaining available time of the target image acquisition device in the corresponding dimension can be expressed as:

[0120] T a =w a *Y a

[0121] Among them, T a w represents the remaining available time of the target image acquisition device in dimension a. a Y represents the weight corresponding to the running state information of dimension a. a This represents the influence factor of the target image acquisition device in dimension a.

[0122] In another alternative approach, the influence factor is used to characterize the negative impact of operational status information in the corresponding dimension on the remaining available time of the target image acquisition device. Accordingly, the remaining available time of the target image acquisition device in the corresponding dimension is obtained based on the weighted result, including:

[0123] Based on the rated available time of the target image acquisition device and the weights corresponding to the operating status information of the corresponding dimension, the rated available time of the target image acquisition device in the corresponding dimension is obtained.

[0124] The remaining available time of the target image acquisition device in the corresponding dimension is determined by the difference between the rated available time of the target image acquisition device in the corresponding dimension and the weighted result.

[0125] If the impact factor identifier has a negative impact on the remaining available time, then the remaining available time of the target image acquisition device in the corresponding dimension can be expressed as:

[0126] T' a =T e *w a

[0127] T a =T' a -w a *Y a

[0128] Among them, T e T' represents the rated available time of the target image acquisition device. a This indicates the rated available time of the target image acquisition device in dimension a.

[0129] S203. Accumulate the remaining available time of the target image acquisition device in all dimensions to obtain the remaining available time of the target image acquisition device.

[0130] This application embodiment obtains the weights corresponding to the running status information of each dimension, uses the weights to weight the influence factors of the corresponding dimensions, and obtains the remaining available time of the corresponding dimension based on the weighting result, laying the foundation for obtaining more accurate remaining available time.

[0131] Based on the above embodiments, as an optional embodiment, after obtaining the remaining available time of the target image acquisition device, the method further includes:

[0132] If it is determined that the remaining available time of the target image acquisition device is not higher than the preset threshold, the operating result of the target image acquisition device will be obtained after the preset time.

[0133] If the target image acquisition device operates normally, the influence factor of the target image acquisition device in each dimension is updated. Combined with the operating status information of the target image acquisition device in the corresponding dimension, the available duration prediction model in the corresponding dimension is trained to obtain the updated available duration prediction model.

[0134] Specifically, in this embodiment of the application, when the remaining available time of the target image acquisition device is not higher than a preset threshold (e.g., 0), the operating results of the target image acquisition device can be checked after a certain period of time. If the operating results of the target image acquisition device are normal, the previously predicted influence factors will be updated. The specific update can be completed by manual annotation by maintenance personnel. Then, the available time prediction model is retrained using the previously acquired operating status information and the updated influence factors to continuously improve the prediction ability of the prediction model.

[0135] Please see Figure 7 The illustration shows a schematic diagram of another application scenario provided by the embodiments of this application, which includes the construction stage of the available time prediction model and the prediction stage of the remaining available time.

[0136] In the construction phase of the available duration prediction model, the operation information of the sample image acquisition device at different prediction times is first obtained. The operation information includes at least one dimension of operation status information. Then, the influence factors of the corresponding dimension are determined by manual annotation. Using machine learning, the operation status information of the sample image acquisition device in the corresponding dimension is used as samples, and the influence factors of the corresponding dimension are used as labels to train the available duration prediction model of the corresponding dimension.

[0137] In the remaining available time prediction stage, the operating status information of the target sample acquisition device in at least one dimension is obtained. The operating status information of each dimension is input into the available time prediction model of the corresponding dimension to predict the influence factor of the corresponding dimension. Based on the influence factor of the target image acquisition device in each dimension, the remaining available time of the target image acquisition device is obtained. If the remaining available time is not higher than a preset threshold, the operating result of the target image acquisition device is obtained after a preset time. If the operating result of the target image acquisition device is normal, the influence factor of the target image acquisition device in each dimension is updated. Combining the operating status information of the target image acquisition device in the corresponding dimension, the available time prediction model of the corresponding dimension is retrained.

[0138] Based on the above embodiments, as an optional embodiment, the remaining available time of the target image acquisition device is obtained according to the influence factor of the target image acquisition device in each dimension, including:

[0139] S301. Update the remaining available time of the target image acquisition device according to the preset duration.

[0140] The impact factor can be used to obtain the remaining available time, and conversely, the remaining available time can also be used to update the impact factor. In this application embodiment, if the operation result of the target image acquisition device is normal after a preset time, it means that the remaining available time has at least underestimated the time of the preset time. Therefore, the remaining available time is updated again, and the updated remaining available time can be a preset multiple of the preset time.

[0141] S302. Obtain the weight corresponding to the running status information of each dimension;

[0142] S303. Based on the weights corresponding to the operating status information of each dimension and the remaining available time of the target image acquisition device after the update, update the influence factor of the target image acquisition device in each dimension.

[0143] Specifically, based on the updated remaining available time and the weights corresponding to the running status information of each dimension, the remaining available time of the corresponding dimension can be obtained. Since the remaining available time of the corresponding dimension is obtained based on the influence factor and weight of the corresponding dimension, the updated influence factor of the corresponding dimension can be further deduced.

[0144] Based on the above embodiments, as an optional embodiment, after obtaining the remaining available time of the target image acquisition device, the method further includes:

[0145] The wear rate of the target image acquisition device is determined based on the remaining available time of the target image acquisition device.

[0146] If the wear rate of the target image acquisition device exceeds a preset rate threshold, an alarm message will be generated.

[0147] After obtaining the remaining available time, this embodiment of the application can further determine the attenuation rate based on the remaining available time. If the attenuation rate of the target image acquisition device is found to be greater than a preset rate threshold, an alarm message can be generated. This alarm message can be fed back to the manufacturer of the image acquisition device, so that the manufacturer can optimize the product quality and improve the user experience by addressing the problem of excessively fast attenuation rate.

[0148] Based on the above embodiments, as an optional embodiment, the training method of the available duration prediction model includes: creating at least one worker thread and training the available duration prediction model of the corresponding dimension through the worker thread;

[0149] The number of worker threads is the same as the number of available duration prediction models to be trained, and at least one worker thread is a parallel thread.

[0150] Since the different available duration prediction models in this application embodiment do not have a necessary order of training, this application embodiment creates multiple parallel worker threads according to the number of available duration prediction models to be trained, which can further improve the training efficiency.

[0151] Based on the above embodiments, in order to further improve the prediction accuracy of remaining available time, this application embodiment considers using the actual remaining available time at each prediction time of the sample image acquisition device and the remaining available time predicted at historical prediction times for reinforcement learning, taking into account the impact of the continuous change of the actual remaining available time on the prediction of remaining available time.

[0152] Specifically, the available duration prediction model includes a reward function and a policy function during training.

[0153] The update time for the remaining available time corresponding to the current prediction time is used to obtain the operating status information and the actual remaining available time at the corresponding prediction time for each sample image acquisition device. The actual remaining available time is the actual time when the sample device was scrapped minus the time when the current prediction time was collected. For example, if the actual scrapping time of the sample device is May 1, 2020, and the time when the operating status information was collected is March 1, 2020, then the actual remaining available time for the operating status information on March 1 is 2 months.

[0154] The available time prediction model in this application embodiment is trained using reinforcement learning. Generally, for the training process of a reinforcement learning model, in order to obtain a better policy, the agent needs to continuously interact with the environment. In this application embodiment, the agent can be understood as the available time prediction model, and the environment is the process of predicting the remaining available time.

[0155] Please see Figure 8 This application provides a flowchart of a reinforcement learning-based model training process. An agent outputs an action (actor) through a policy function and applies it to the environment. Upon receiving the action, the environment's state changes, and a reward value is generated according to a reward function. The environment feeds back its current state and reward value to the agent. The agent then outputs the next action based on the reward value and the environment's current state, aiming to increase the probability of receiving a positive reward. The selected action not only affects the current reward value but also influences the environment's state at the next prediction time and the final reward value, thus achieving a cyclical response process.

[0156] State: Taking Go as an example, the current distribution of black and white stones on the board is the current state. Additionally, the previous moves of each player in a game can also be considered the current state. In the prediction of remaining available time in this embodiment, the state can be the running state information at the current prediction time, denoted as X at prediction time t. t Furthermore, the state in the prediction process of the remaining available time in this application embodiment also includes the sequence of historical prediction influence factors given at each previous prediction time, denoted as Y1, Y2, ... Y. T .

[0157] Action: In chess, each move a player makes (the placement of a piece) is an action. In the remaining available time prediction process of this application embodiment, the action is an influencing factor estimated at the current prediction time.

[0158] Policy function: Usually denoted by π(a|s,θ), where s represents the state, a represents the action, and θ is the parameter to be learned. Its output is the probability of action a. In board games, this function determines the next move in the current board state. In the problem of predicting remaining time, this function is used to determine the influencing factor.

[0159] Reward: Rewards are not always obtained only after an action ends. Typically, an action affects not only the current reward but also subsequent rewards, but the reward is discounted over time. That is, the reward obtained after several steps has a weaker impact than the immediate reward; this is called a "discount." This discount is usually denoted by γ, where the reward after prediction time t is γ. t R t When γ = 1, it indicates that the reward value at the current prediction time is as important as the future reward value, i.e., there is no discount. In the prediction process of the remaining available time in this application embodiment, the reward value is obtained through a reward function. The reward value can be divided into two parts: the first part is the accuracy of the estimated influence factor at each prediction time, and the second part is the degree of smooth change of the sequence of influence factors estimated at the current prediction time and influence factors at all previous prediction times.

[0160] During training, the remaining available time prediction model of this application embodiment uses the actual remaining available time of the sample image acquisition device at each prediction time and the historical prediction influence factor sequence as input to the reward function. The output of the reward function is used to characterize the accuracy evaluation result of the prediction influence factor of the sample image acquisition device in the corresponding dimension at each prediction time.

[0161] The historical prediction impact factor sequence includes the prediction impact factors of the sample image acquisition device in the corresponding dimension at least one prediction time prior to each prediction time.

[0162] The predicted impact factor is obtained by the strategy function based on the operational status information at the corresponding prediction time, which predicts the probability distribution of the impact factor of the sample image acquisition device in the corresponding dimension.

[0163] This application embodiment trains the available time prediction model through reinforcement learning. During the training process, the model parameters are trained by continuously calculating the deviation between the predicted value and the actual value of the remaining available time obtained based on the predicted impact factor. This eliminates the need to determine the remaining available time through manual annotation, and the final remaining available time calculated based on the impact factor is closer to the actual value.

[0164] Please see Figure 9 The image shown is a structural diagram of an available duration prediction model provided in an embodiment of this application. Figure 9 As shown, the available time prediction model includes a reward function and a policy function during training. The corresponding dimension of the running state information at each prediction time of the training sample is the state. If there are T prediction times in the training sample, there are T states. Each state is used as the input to the policy function. Based on the input state, the policy function outputs the action: the probability distribution of the influencing factors at each prediction time. Random sampling is performed on the probability distribution of the influencing factors at each prediction time to obtain the sampling result: the predicted influencing factor at each prediction time. Thus, for each prediction time, the predicted influencing factors of at least one prediction time before that prediction time are combined into a historical predicted influencing factor sequence. The actual remaining available time at each time and the historical predicted influencing factor sequence at each prediction time are used as the input to the reward function. Based on the input, the reward function generates an evaluation result to evaluate the accuracy of the predicted influencing factors at each prediction time. Based on this evaluation result, the policy function in the available time prediction model is adjusted to increase the output probability of predicted influencing factors with good evaluation results and decrease the output probability of predicted influencing factors with poor evaluation results. In this way, the trained policy function will learn the correct prediction behavior of influencing factors.

[0165] It should be noted that the input to the reward function in this embodiment includes the actual remaining available time of the training samples (i.e., the sample image acquisition device) and the historical predicted impact factor sequence for each time step. The actual remaining available time provides a basis for evaluating the accuracy of the predicted impact factors at each time step. Furthermore, by constructing the historical predicted impact factor sequence for each time step, the stable change of the impact factors is considered. In other words, the reward function in this embodiment can evaluate the impact factors from both accuracy and stability perspectives, thereby laying the foundation for estimating impact factors with higher accuracy in practical applications.

[0166] Based on the above embodiments, as an optional embodiment, the operating status information of each dimension is input into a pre-trained available duration prediction model for the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available duration prediction model, including:

[0167] S401. Encode the running status information for each dimension to obtain a feature vector. It can be understood that the feature vector is a representation of the running status.

[0168] S402. Input the feature vector into the policy function to obtain the probability distribution of the influence factors of the sample image acquisition device in the corresponding dimension output by the policy function.

[0169] S403. Sample the probability distribution and obtain the influence factor of the target image acquisition device in the corresponding dimension based on the sampling results.

[0170] Please see Figure 10 The example illustrates a schematic diagram of the training process of the available duration prediction model according to an embodiment of this application, such as... Figure 10 As shown, the process includes:

[0171] S501. Obtain at least one training sample, and obtain the operating status of the sample image acquisition device in the corresponding dimension at each prediction time and the actual remaining available time of the sample image acquisition device at the corresponding prediction time.

[0172] Reinforcement learning training is conducted in "episodes." In this embodiment, each episode represents the complete usage process of the sample image acquisition device from activation to deactivation, i.e., a training sample. A training sample includes the corresponding dimension's operational status information and the actual remaining available time at each prediction moment during the usage process. As can be seen from the above embodiments, the remaining available time for each dimension can be obtained from the influence factor of that dimension. Similarly, the actual remaining available time can also be deduced from the weight of the corresponding dimension.

[0173] S502. Input the operating status of the sample image acquisition device at each prediction time in the corresponding dimension into the policy function of the available duration prediction model to be trained, and obtain the probability distribution of the influence factors of the sample image acquisition device at the corresponding prediction time in the corresponding dimension output by the policy function.

[0174] Input the running state information X at the nth prediction time into the policy function n The policy function outputs the influence factor Y at the nth prediction time. n1 The probability of Y n2 The probability of Y nm The probability of Y, where Y nm Y representsn The m-th predicted value. In this case, the output layer of the neural network functions similarly to softmax regression in a multi-class classification problem, outputting a probability distribution, except that this probability distribution is not used for classification.

[0175] S503. Sampling is performed according to the probability distribution to obtain the remaining available time for the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and a historical prediction influence factor sequence for the corresponding prediction time is constructed.

[0176] By randomly sampling according to the probability distribution of the influencing factors, and using the sampled values ​​as the predicted influencing factors for each prediction time step, the purpose of training in this application embodiment is to adjust the parameters of the policy function so that the sampling results of the influencing factors can be as close as possible to or equal to the influencing factors that optimize the reward value for each step.

[0177] S504. Input the actual remaining available time at the corresponding prediction time and the historical prediction impact factor sequence into the reward function to obtain the reward value at the corresponding prediction time output by the reward function.

[0178] Specifically, step S504 can be represented by the following formula:

[0179] R t =g(Y1,…,Y) t ,T t )

[0180] Among them, R t Y represents the reward value at prediction time t, (.) represents the reward function, and Y1 represents the prediction influence factor at prediction time 1. t Y1 to Y2 represent the prediction influence factor at prediction time t. t This constitutes the historical prediction influence factor sequence at prediction time t, T t This represents the actual remaining available time at prediction time t.

[0181] S505. Based on the reward value and probability distribution of influencing factors at each prediction time, train the parameters of the policy function, and use the trained policy function as the trained available duration prediction model.

[0182] Specifically, step S505 further includes:

[0183] The cumulative reward value for each prediction time is obtained by summing the reward values ​​of all prediction times after each prediction time, and can be expressed by the following formula:

[0184]

[0185] Where G represents the accumulated reward value at prediction time t, k represents the time step from prediction time t+1 to time step T, where time step T refers to the last prediction time of the training sample, at which it is determined that the sample image acquisition device is no longer available, and γ represents the discount.

[0186] Based on the accumulated reward value and the probability distribution of influencing factors at each prediction time, the parameters of the policy function are trained using gradient descent, and can be expressed by the following formula:

[0187]

[0188] Where θ represents the parameters in the policy function, and α represents the learning rate. Represents the gradient, π(Y) t |X t (θ) represents the state information X at prediction time t based on the parameter θ in the policy function. t The probability distribution of the influence factors at prediction time t is obtained.

[0189] Based on the above embodiments, as an optional embodiment, the actual remaining available time at the corresponding prediction time and the historical prediction influence factor sequence are input into the reward function to obtain the reward value at the corresponding prediction time output by the reward function, including:

[0190] S601. Input the prediction impact factor of the sample image acquisition device in the corresponding dimension and the actual remaining available time of the corresponding prediction time into the reward function to obtain the accuracy reward value of the corresponding prediction time output by the reward function; the accuracy reward value is used to characterize the accuracy of the prediction remaining impact factor of the corresponding prediction time.

[0191] In the calculation of the reward value in this application embodiment, the reward value is divided into an accuracy reward value, which is used to characterize the accuracy of the predicted remaining influence factor at the corresponding prediction time, and a stability reward value, which is used to characterize the degree of stable change of the predicted influence factor at the corresponding prediction time relative to the historical predicted influence factor sequence.

[0192] For the accuracy bonus value, it is evaluated based on the prediction impact factor of the sample image acquisition device in the corresponding dimension and the actual remaining available time at the corresponding prediction time for each prediction moment. Specifically, it includes:

[0193] Based on the prediction influence factors and preset weights of the sample image acquisition device in the corresponding dimension at each prediction time, the predicted value of the remaining available time at that prediction time is obtained. The difference between the predicted value of the remaining available time and the actual remaining available time at that prediction time is determined, and the accuracy reward value for each prediction time is obtained based on the difference; or

[0194] Based on the prediction influence factor and preset weight of the sample image acquisition device at each prediction time, the predicted value of the remaining available time at the corresponding prediction time and in the corresponding dimension is obtained. Based on the actual remaining available time at the corresponding prediction time and preset weight, the actual remaining available time at the corresponding prediction time and in the corresponding dimension is obtained. The difference between the actual remaining available time at the corresponding prediction time and the predicted value of the remaining available time is determined. Based on the difference, the accuracy reward value for each prediction time is obtained.

[0195] The smaller the difference, the higher the accuracy of the influencing factor. Therefore, different intervals can be set in this application to determine the accuracy reward value of different differences. For example, the reward value is 1 for a difference of 0-1 days and 0 for a difference greater than 5 days. It should be noted that this application does not make specific limitations on the number of intervals or the specific value of the reward value corresponding to different intervals.

[0196] S602. Input the predicted influence factor of the sample image acquisition device at each prediction time in the corresponding dimension and the historical predicted influence factor sequence at the corresponding prediction time into the reward function to obtain the stationarity reward value of the corresponding prediction time output by the reward function; the stationarity reward value is used to characterize the degree of stationar change of the predicted influence factor at the corresponding prediction time relative to the historical predicted influence factor sequence.

[0197] Ideally, the predicted impact factor should show a decrease in remaining available time of one second for every second of actual time elapsed. However, if the remaining available time increases or remains constant over a period of time as the image acquisition device is used, it provides a poor user experience. Therefore, this application, while evaluating the accuracy of the impact factor prediction, also needs to assess the degree of stability in the historical predicted impact factor sequence. Specifically:

[0198] S6021. For any one of the at least one prediction time before each prediction time, determine the relationship between the prediction influence factor of any one prediction time and the prediction influence factor of the previous prediction time adjacent to any one prediction time.

[0199] S6022. Based on the positive or negative impact of the operational status information of the corresponding dimension on the remaining available time of the target image acquisition device, and combined with the magnitude relationship, determine the stationarity reward sub-value for the prediction time.

[0200] Specifically, if the influence factor represents the positive impact of the corresponding dimension's operational status information on the remaining available time of the target image acquisition device, then the influence factor should theoretically gradually decrease. If it is determined that the predicted influence factor at any prediction time is greater than the predicted influence factor of the previous prediction time adjacent to any prediction time, then the stationarity reward sub-value at any prediction time is the first preset value; if it is determined that the predicted influence factor at any prediction time is not greater than the predicted influence factor of the previous prediction time adjacent to any prediction time, then the stationarity reward sub-value at each prediction time is the second preset value, and the first preset value is less than the second preset value.

[0201] If the influence factor represents the positive impact of the corresponding dimension's operational status information on the remaining available time of the target image acquisition device, then the influence factor should theoretically gradually increase. If it is determined that the predicted influence factor at any prediction time is greater than the predicted influence factor of the previous prediction time adjacent to any prediction time, then the stationarity reward sub-value at any prediction time is the second preset value; if it is determined that the predicted influence factor at any prediction time is not greater than the predicted influence factor of the previous prediction time adjacent to any prediction time, then the stationarity reward sub-value at each prediction time is the first preset value.

[0202] S6023. Obtain the stationarity reward value for each prediction time based on the stationarity reward sub-values ​​of all prediction times in at least one prediction time prior to each prediction time.

[0203] Based on the above embodiments, the negative impact of the corresponding dimension of the operating status information on the remaining available time of the target image acquisition device is characterized by an impact factor. The formula for calculating the reward value at time step t can be:

[0204]

[0205] The function I(x) is an indicator function. When x is true, I(x) = 1, and when x is false, I(x) = 0.

[0206] S603. Combine the accuracy reward value and stability reward value at each prediction time to obtain the reward value for the corresponding prediction time.

[0207] Please see Figure 11 The example illustrates a schematic diagram of the structure of an available duration prediction model to be trained, provided in another embodiment of this application, such as... Figure 11 As shown:

[0208] The input layer 110 of the available time prediction model is used to receive the running status information and the actual remaining available time of each prediction time of the input training sample;

[0209] The available duration prediction model can use the impact factor prediction layer 120 as input to the running status information at each prediction time and output the predicted impact factor at each prediction time. Specifically, the feature extraction layer 1201 extracts the feature vector of the running status information, and then the probability distribution acquisition layer 1202 processes the feature vector of the running status information to obtain the probability distribution of the impact factor at each prediction time. Then, the sampling layer 1203 randomly selects the impact factor at each prediction time according to the probability distribution of the impact factor at each prediction time to obtain the predicted impact factor at each prediction time.

[0210] The reward acquisition layer 130 takes the actual remaining available time and the predicted impact factor at each prediction time as input to obtain the reward value for each prediction time. Specifically: the accuracy reward calculation layer 1301 takes the feature vector of the predicted impact factor and the actual remaining available time at each prediction time as input to output the accuracy reward value of the predicted impact factor at the corresponding prediction time; and the information fusion layer 1302 obtains the historical predicted impact factor sequence for each prediction time based on the predicted impact factor at each prediction time. The historical predicted impact factor sequence for each prediction time includes the predicted impact factors of at least one prediction time before the current prediction time, arranged in the order of prediction time. Further, the stability reward calculation layer 1303 takes the predicted impact factor and the historical predicted impact factor sequence at each prediction time as input to output the stability reward value for each prediction time. Finally, the reward fusion layer 1304 obtains the reward value for each prediction time by combining the accuracy reward value and the stability reward value for each prediction time.

[0211] The parameters in the impact factor prediction layer 120 are adjusted by the hidden layer 140 based on the reward value at each prediction time and the probability distribution of the predicted impact factor. Specifically, the cumulative reward layer 1401 takes the reward values ​​of all prediction times after each prediction time as input and obtains the cumulative reward value at each prediction time by summing them. Then, the gradient training layer 1402 trains the parameters in the impact factor prediction layer using gradient descent based on the cumulative reward value at each prediction time and the probability distribution of the impact factor. After training, the trained usable duration prediction model is obtained by retaining the input layer and the impact factor prediction layer.

[0212] This application provides a device for predicting the remaining available time of an image acquisition device, such as... Figure 12 As shown, the device may include: a status information acquisition module 101, a dimension available time prediction module 102, and an available time aggregation module 103, specifically:

[0213] The status information acquisition module 101 is used to acquire the operating status information of the target image acquisition device in at least one dimension;

[0214] The available time prediction module 102 is used to input the running status information of each dimension into the pre-trained available time prediction model of the corresponding dimension, and obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence on the remaining available time of the target image acquisition device.

[0215] The available time summary module 103 is used to obtain the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension;

[0216] Each available duration prediction model is trained based on the operational status information and influencing factors of the sample image acquisition device in the corresponding dimension.

[0217] The image acquisition device remaining available time prediction device provided in this application embodiment specifically executes the process described in the above method embodiment. For details, please refer to the content of the above image acquisition device remaining available time prediction method embodiment, which will not be repeated here. The image acquisition device remaining available time prediction device provided in this application embodiment obtains the operating status information of the target image acquisition device in at least one dimension, inputs the operating status of each dimension into the available time prediction model of the corresponding dimension, and obtains the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. Since this application embodiment constructs a prediction model for the operating status information of each dimension, the prediction result is more accurate. Although the number of prediction models to be trained is large, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster to train multiple prediction models than to train a prediction model with multi-dimensional operating status information. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models simultaneously, thereby further improving the training efficiency of all models. This application embodiment improves the accuracy and reliability of remaining available time prediction.

[0218] Based on the above embodiments, as an optional embodiment, the available time aggregation module includes:

[0219] The weight acquisition submodule is used to obtain the weights corresponding to the running status information of each dimension;

[0220] The dimension duration acquisition submodule is used to weight the influence factors of the target image acquisition device in the corresponding dimension according to the weight, and obtain the remaining available duration of the target image acquisition device in the corresponding dimension based on the weighting result;

[0221] The duration accumulation unit submodule is used to accumulate the remaining available duration of the target image acquisition device in all dimensions to obtain the remaining available duration of the target image acquisition device.

[0222] Based on the above embodiments, as an optional embodiment, the influence factor is used to characterize the degree of positive influence of the corresponding dimension's operating status information on the remaining available time of the target image acquisition device; the dimension duration acquisition submodule is specifically used to: take the weighted result as the remaining available time of the target image acquisition device in the corresponding dimension.

[0223] Based on the above embodiments, as an optional embodiment, the influence factor is used to characterize the degree of negative impact of the corresponding dimension of the operating status information on the remaining available time of the target image acquisition device;

[0224] The dimension duration acquisition submodule includes:

[0225] The dimension rated duration acquisition unit is used to obtain the rated available duration of the target image acquisition device in the corresponding dimension based on the rated available duration of the target image acquisition device and the weight corresponding to the operating status information of the corresponding dimension.

[0226] The difference calculation unit is used to take the difference between the rated available time of the target image acquisition device in the corresponding dimension and the weighted result as the remaining available time of the target image acquisition device in the corresponding dimension.

[0227] Based on the above embodiments, as an optional embodiment, the prediction device further includes:

[0228] The sampling module is used to obtain the operating results of the target image acquisition device after a preset time if it is determined that the remaining available time of the target image acquisition device is not higher than a preset threshold.

[0229] The model update module is used to update the influence factors of the target image acquisition device in each dimension if the operation result of the target image acquisition device is normal. It combines the operation status information of the target image acquisition device in the corresponding dimension to train the available duration prediction model in the corresponding dimension and obtain the updated available duration prediction model.

[0230] Based on the above embodiments, as an optional embodiment, the model update module includes:

[0231] The remaining duration update submodule is used to update the remaining available duration of the target image acquisition device according to the preset duration.

[0232] The weight acquisition submodule is used to obtain the weights corresponding to the running status information of each dimension;

[0233] The factor update submodule is used to update the influence factor of the target image acquisition device in each dimension based on the weight corresponding to the running status information of each dimension and the remaining available time of the target image acquisition device after the update.

[0234] Based on the above embodiments, as an optional embodiment, the prediction device further includes:

[0235] The attrition rate calculation module is used to determine the attrition rate of the target image acquisition device based on the remaining available time of the target image acquisition device.

[0236] The alarm module is used to generate alarm information if the wear rate of the target image acquisition device exceeds a preset rate threshold.

[0237] Based on the above embodiments, as an optional embodiment, the prediction device further includes:

[0238] The thread creation module is used to create at least one worker thread and train the available duration prediction model for the corresponding dimension through the worker thread;

[0239] The number of worker threads is the same as the number of available duration prediction models to be trained, and at least one worker thread is a parallel thread.

[0240] Based on the above embodiments, as an optional embodiment, the available duration prediction model includes a reward function and a policy function during training;

[0241] The input to the reward function includes the actual remaining available time of the sample image acquisition device at each prediction time and the historical prediction impact factor sequence. The output of the reward function is used to characterize the accuracy evaluation result of the prediction impact factor of the sample image acquisition device in the corresponding dimension at each prediction time.

[0242] The historical prediction impact factor sequence includes the prediction impact factors of the sample image acquisition device in the corresponding dimension at least one prediction time prior to each prediction time.

[0243] The predicted impact factor is obtained by the strategy function based on the operational status information at the corresponding prediction time, which predicts the probability distribution of the impact factor of the sample image acquisition device in the corresponding dimension.

[0244] Based on the above embodiments, as an optional embodiment, the dimension availability duration prediction module includes:

[0245] The encoding submodule is used to encode the runtime status information of each dimension to obtain feature vectors;

[0246] The probability distribution acquisition submodule is used to input the feature vector into the policy function and obtain the probability distribution of the influence factors of the sample image acquisition device in the corresponding dimension of the policy function output;

[0247] The sampling submodule is used to sample the probability distribution and obtain the influence factor of the target image acquisition device in the corresponding dimension based on the sampling results.

[0248] Based on the above embodiments, as an optional embodiment, the prediction device further includes a model training module for training an available duration prediction model, the model training module including:

[0249] The sample acquisition submodule is used to acquire the operating status of the sample image acquisition device in the corresponding dimension and the actual remaining available time at the corresponding prediction time at each prediction time.

[0250] The strategy function submodule is used to input the operating status of the sample image acquisition device at each prediction time in the corresponding dimension into the strategy function of the available duration prediction model to be trained, and to obtain the probability distribution of the influence factors of the sample image acquisition device at the corresponding prediction time in the corresponding dimension output by the strategy function.

[0251] The sequence construction submodule is used to sample according to the probability distribution, obtain the remaining available time of the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and construct the historical prediction influence factor sequence at the corresponding prediction time.

[0252] The reward function submodule is used to input the actual remaining available time and the historical prediction impact factor sequence at the corresponding prediction time into the reward function to obtain the reward value at the corresponding prediction time output by the reward function.

[0253] The feedback submodule is used to train the parameters of the policy function based on the reward value and the probability distribution of the influencing factors at each prediction time, and to use the trained policy function as the trained available duration prediction model.

[0254] Based on the above embodiments, as an optional embodiment, the reward function submodule includes:

[0255] The accurate reward value calculation unit is used to input the prediction influence factor of the sample image acquisition device in the corresponding dimension and the actual remaining available time of the corresponding prediction time into the reward function to obtain the accuracy reward value of the corresponding prediction time output by the reward function; the accuracy reward value is used to characterize the accuracy of the prediction remaining influence factor at the corresponding prediction time.

[0256] The stationarity reward value calculation unit is used to input the predicted influence factor of the sample image acquisition device in the corresponding dimension and the historical predicted influence factor sequence of the corresponding prediction time into the reward function to obtain the stationarity reward value of the corresponding prediction time output by the reward function; the stationarity reward value is used to characterize the degree of stationarity of the predicted influence factor at the corresponding prediction time relative to the historical predicted influence factor sequence.

[0257] The reward statistics unit is used to combine the accuracy reward value and the stability reward value at each prediction time to obtain the reward value for the corresponding prediction time.

[0258] This application provides a training apparatus for an available duration prediction model, which may include:

[0259] The training sample acquisition module is used to acquire at least one training sample. Each training sample includes the running status information of the sample image acquisition device in the corresponding dimension at each prediction time and the actual remaining available time at the corresponding prediction time.

[0260] The probability distribution acquisition module is used to input the operating status information of the sample image acquisition device in the corresponding dimension at each prediction time into the strategy function of the available duration prediction model to be trained, and obtain the probability distribution of the influence factors of the sample image acquisition device in the corresponding dimension at the corresponding prediction time output by the strategy function.

[0261] The impact factor sequence module is used to sample according to the probability distribution, obtain the remaining available time of the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and construct the historical prediction impact factor sequence at the corresponding prediction time.

[0262] The reward value acquisition module is used to input the actual remaining available time and the historical prediction influence factor sequence at the corresponding prediction time into the reward function to obtain the reward value at the corresponding prediction time output by the reward function.

[0263] The training module is used to train the parameters of the policy function based on the reward value at each prediction time and the probability distribution of the influencing factors, and to use the trained policy function as the trained available duration prediction model.

[0264] This application provides an electronic device comprising: a memory and a processor; at least one program stored in the memory, which, when executed by the processor, can achieve the following compared to the prior art: by acquiring the operating status information of the target image acquisition device in at least one dimension, inputting the operating status of each dimension into the available time prediction model of the corresponding dimension, and obtaining the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. Since this application constructs a prediction model for the operating status information of each dimension, the prediction result is more accurate. Although the number of prediction models to be trained is large, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster than training a prediction model with multi-dimensional operating status information, even if multiple prediction models need to be trained. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models simultaneously, thereby further improving the training efficiency of all models. This application improves the prediction accuracy and reliability of the remaining available time.

[0265] In one alternative embodiment, an electronic device is provided, such as Figure 13 As shown, Figure 13 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may also include a transceiver 4004. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of this electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0266] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0267] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 13 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.

[0268] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic 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 not limited thereto.

[0269] The memory 4003 stores application code that executes the scheme of this application, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0270] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments. Compared with the prior art, by acquiring the operating status information of the target image acquisition device in at least one dimension, and inputting the operating status of each dimension into the corresponding available time prediction model, the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model is obtained. Since this application constructs a prediction model for the operating status information of each dimension, the prediction result is more accurate. Although a large number of prediction models need to be trained, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster to train multiple prediction models than to train a prediction model with multi-dimensional operating status information. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models simultaneously, thereby further improving the training efficiency of all models. This application improves the prediction accuracy and reliability of the remaining available time.

[0271] This application provides a computer program including computer instructions stored in a computer-readable storage medium. When a processor of a computer device reads the computer instructions from the computer-readable storage medium, the processor executes the computer instructions, causing the computer device to perform the content shown in the foregoing method embodiments. Compared with the prior art, by acquiring the operating status information of the target image acquisition device in at least one dimension, and inputting the operating status of each dimension into the available time prediction model of the corresponding dimension, the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model is obtained. Since this application embodiment constructs a prediction model for the operating status information of each dimension, the prediction result is more accurate. Although a large number of prediction models need to be trained, since the training samples only involve the operating status information of one dimension, the convergence speed of a single prediction model during training is faster. This makes it faster than training a prediction model with multi-dimensional operating status information, even if multiple prediction models need to be trained. In particular, since there is no correlation between the training of prediction models of different dimensions, it is more suitable to train multiple prediction models simultaneously, thereby further improving the training efficiency of all models. This application embodiment improves the prediction accuracy and reliability of remaining available time.

[0272] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0273] The above are only some embodiments of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining usable time of an image acquisition device, characterized in that, include: Acquire the operational status information of the target image acquisition device in at least one dimension; The operational status information of each dimension is input into a pre-trained available time prediction model of the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence of the operational status information of the corresponding dimension on the remaining available time of the target image acquisition device. Based on the influence factors of the target image acquisition device in each dimension, the remaining available time of the target image acquisition device is obtained; Each available duration prediction model is trained based on the operational status information and influencing factors of the sample image acquisition device in the corresponding dimension.

2. The method for predicting the remaining available time of the image acquisition device according to claim 1, characterized in that, The step of obtaining the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension includes: Obtain the weights corresponding to the running status information of each dimension; Based on the weights, the influence factors of the target image acquisition device in the corresponding dimension are weighted, and the remaining available time of the target image acquisition device in the corresponding dimension is obtained based on the weighting result. The remaining available time of the target image acquisition device is obtained by summing up the remaining available time of the target image acquisition device in all dimensions.

3. The method for predicting the remaining available time of the image acquisition device according to claim 2, characterized in that, The influence factor is used to characterize the degree of positive impact of the corresponding dimension of the operating status information on the remaining available time of the target image acquisition device; The step of obtaining the remaining available time of the target image acquisition device in the corresponding dimension based on the weighted result includes: The weighted result is taken as the remaining available time of the target image acquisition device in the corresponding dimension.

4. The method for predicting the remaining available time of the image acquisition device according to claim 2, characterized in that, The influence factor is used to characterize the degree of negative impact of the corresponding dimension of the operating status information on the remaining available time of the target image acquisition device; The step of obtaining the remaining available time of the target image acquisition device in the corresponding dimension based on the weighted result includes: Based on the rated available time of the target image acquisition device and the weights corresponding to the operating status information of the corresponding dimension, the rated available time of the target image acquisition device in the corresponding dimension is obtained. The difference between the rated available time of the target image acquisition device in the corresponding dimension and the weighted result is taken as the remaining available time of the target image acquisition device in the corresponding dimension.

5. The method for predicting the remaining available time of the image acquisition device according to any one of claims 1-4, characterized in that, After obtaining the remaining available time of the target image acquisition device, the process further includes: If it is determined that the remaining available time of the target image acquisition device is not higher than a preset threshold, then the running result of the target image acquisition device is obtained after the preset time. If the target image acquisition device operates normally, the influence factor of the target image acquisition device in each dimension is updated. Combined with the operating status information of the target image acquisition device in the corresponding dimension, the available duration prediction model in the corresponding dimension is trained to obtain the updated available duration prediction model.

6. The method for predicting the remaining available time of the image acquisition device according to claim 5, characterized in that, The updating of the influence factors of the target image acquisition device in each dimension includes: The remaining available time of the target image acquisition device is updated according to the preset duration. Obtain the weight corresponding to the running status information of each dimension; Based on the weights corresponding to the operational status information of each dimension and the remaining available time of the target image acquisition device after the update, the influence factor of the target image acquisition device in each dimension is updated.

7. The method for predicting the remaining available time of the image acquisition device according to claim 1, characterized in that, After obtaining the remaining available time of the target image acquisition device, the process further includes: The wear rate of the target image acquisition device is determined based on the remaining available time of the target image acquisition device; If the loss rate of the target image acquisition device exceeds a preset rate threshold, an alarm message is generated.

8. The method for predicting the remaining available time of the image acquisition device according to any one of claims 1-7, characterized in that, The training method for the available duration prediction model includes: Create at least one worker thread and train the available duration prediction model for the corresponding dimension through the worker thread; The number of worker threads is the same as the number of available duration prediction models to be trained, and the at least one worker thread is a parallel thread.

9. The method for predicting the remaining available time of the image acquisition device according to claim 1, wherein the available time prediction model includes a reward function and a policy function during training; The input to the reward function includes the actual remaining available time of the sample image acquisition device at each prediction time and the historical prediction impact factor sequence. The output of the reward function is used to characterize the accuracy evaluation result of the prediction impact factor of the sample image acquisition device at each prediction time in the corresponding dimension. The historical prediction impact factor sequence includes the prediction impact factors of the sample image acquisition device in the corresponding dimension at least one prediction time before each prediction time. The predicted impact factor is obtained by the strategy function based on the operational status information at the corresponding prediction time, which predicts the probability distribution of the impact factor of the sample image acquisition device in the corresponding dimension.

10. The method for predicting the remaining available time of the image acquisition device according to claim 9, characterized in that, The operational status information for each dimension is input into a pre-trained available duration prediction model for the corresponding dimension to obtain the influence factor of the target image acquisition device in the corresponding dimension, as output by the available duration prediction model, including: The runtime status information for each dimension is encoded to obtain a feature vector; The feature vector is input into the policy function to obtain the probability distribution of the influence factors of the sample image acquisition device in the corresponding dimension output by the policy function; The probability distribution is sampled, and the influence factor of the target image acquisition device in the corresponding dimension is obtained based on the sampling results.

11. The method for predicting the remaining available time of the image acquisition device according to claim 9, characterized in that, The available time prediction model is trained in the following manner: Obtain the operating status information of the sample image acquisition device in the corresponding dimension at each prediction time and the actual remaining available time at the corresponding prediction time; The operating status information of the sample image acquisition device at each prediction time in the corresponding dimension is input into the policy function of the available duration prediction model to be trained, and the probability distribution of the influence factor of the sample image acquisition device at the corresponding prediction time in the corresponding dimension is obtained from the output of the policy function. Based on the probability distribution, sampling is performed to obtain the remaining available time for the sample image acquisition device in the corresponding dimension at the corresponding prediction time, and a historical prediction influence factor sequence for the corresponding prediction time is constructed. The actual remaining available time at the corresponding prediction time and the historical prediction impact factor sequence are input into the reward function to obtain the reward value at the corresponding prediction time output by the reward function. Based on the reward value at each prediction time and the probability distribution of the influencing factors, the parameters of the policy function are trained, and the trained policy function is used as the trained available duration prediction model.

12. The method for predicting the remaining available time of the image acquisition device according to claim 11, characterized in that, The actual remaining available time at the corresponding prediction time and the historical prediction influence factor sequence are input into the reward function to obtain the reward value at the corresponding prediction time output by the reward function, including: The prediction impact factor of the sample image acquisition device at each prediction time and the actual remaining available time at the corresponding prediction time are input into the reward function to obtain the accuracy reward value of the corresponding prediction time output by the reward function; the accuracy reward value is used to characterize the accuracy of the prediction remaining impact factor at the corresponding prediction time. The predicted influence factor of the sample image acquisition device at each prediction time and the historical predicted influence factor sequence at the corresponding prediction time are input into the reward function to obtain the stationarity reward value of the corresponding prediction time output by the reward function; the stationarity reward value is used to characterize the degree of stationar change of the predicted influence factor at the corresponding prediction time relative to the historical predicted influence factor sequence. The reward value for each prediction time is obtained by combining the accuracy reward value and the stability reward value for each prediction time.

13. A device for predicting the remaining available time of an image acquisition device, characterized in that, include: The status information acquisition module is used to acquire the operating status information of the target image acquisition device in at least one dimension; The available time prediction module for each dimension is used to input the running status information of each dimension into a pre-trained available time prediction model for the corresponding dimension, and obtain the influence factor of the target image acquisition device in the corresponding dimension output by the available time prediction model. The influence factor is used to characterize the degree of influence on the remaining available time of the target image acquisition device. The available time summary module is used to obtain the remaining available time of the target image acquisition device based on the influence factor of the target image acquisition device in each dimension; Each available duration prediction model is trained based on the operational status information and influencing factors of the sample image acquisition device in the corresponding dimension.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a method for predicting the remaining available time of the image acquisition device as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to perform a method for predicting the remaining available time of the image acquisition device as described in any one of claims 1 to 12.

Images