Power state control method and related equipment

Abstract

The invention relates to the technical field of image forming devices, in particular to a power state control method and related equipment. The method comprises the following steps: in response to a trigger signal of a target option, obtaining operation information corresponding to each subsystem of the image forming device within a preset time; respectively inputting the operation information into M sub-models in a pre-trained deep learning model; m is a positive integer greater than or equal to 1; wherein each sub-model is used for outputting a target prediction usage rate of each subsystem in a future target duration, and the target durations corresponding to the sub-models are different; acquiring M groups of predicted utilization rate sequences output by the deep learning model; and controlling the power supply state of the image forming device according to the M groups of predicted utilization rate sequences. The power supply state of the image forming device is correspondingly adjusted by predicting the use condition of the image forming device in a future period of time through the deep learning model, so that the use experience of a user is improved while the energy consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image forming devices, and in particular to a power state control method and related equipment. Background Art

[0002] Existing image forming devices, such as printers and copiers, often automatically enter a sleep state after a period of inactivity when controlling their power state. This duration can be set by the user, known as the sleep duration. In sleep mode, the image forming device shuts down some subsystems to conserve energy. When a user initiates a job during sleep mode, the image forming device wakes up, and any shut-down subsystems reinitialize and process the job.

[0003] In some specific scenarios, some periods are used frequently, while others are used less frequently. If the sleep time is set to 4 hours, the image forming device will remain idle for 4 hours after the last use during the period of low usage, resulting in high energy consumption. If the sleep time is set to 15 minutes to save energy, users will have to wait for the image forming device to wake up almost every time they issue a print job, resulting in a poor user experience. Summary of the Invention

[0004] In view of this, the present invention provides a power state control method and related equipment, which uses a deep learning model to predict the usage of an image forming device in a period of time in the future and adjust the power state of the image forming device accordingly, thereby reducing energy consumption while improving the user experience.

[0005] In a first aspect, an embodiment of the present invention provides a power state control method applied to an image forming apparatus, wherein the image forming apparatus includes multiple subsystems, including: In response to a trigger signal of a target option, obtaining operation information corresponding to each of the subsystems of the image forming apparatus within a preset time; Input the operation information into M sub-models of a pre-trained deep learning model, respectively; M is a positive integer greater than or equal to 1; wherein each sub-model is used to output a target predicted usage rate of each subsystem within a target duration in the future, and the target duration corresponding to each sub-model is different; Obtaining M groups of predicted usage rate sequences output by the deep learning model; wherein each group of the predicted usage rate sequences includes a target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M submodels; The power state of the image forming apparatus is controlled according to the M sets of predicted usage rate sequences.

[0006] In a possible implementation, controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences includes: If all of the target predicted usage rates in the M groups of predicted usage rate sequences are less than a target threshold, setting each of the subsystems to a minimum power consumption state; The power state of the image forming apparatus is controlled to enter a deep sleep state.

[0007] In a possible implementation, controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences includes: If there is a target predicted usage rate greater than the target threshold value in the first set of predicted usage rate sequences, setting each of the subsystems to a ready state; Controlling the power state of the image forming device to enter a ready state; The target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models.

[0008] In a possible implementation, controlling the power state of the image forming apparatus to enter a ready state specifically includes: detecting whether the current power state of the image forming device is a ready state; If the current power state of the image forming apparatus is a ready state, controlling the subsystem corresponding to the target predicted usage rate greater than the target threshold in the first group of predicted usage rate sequences to perform advance job preparation; If the current power state of the image forming apparatus is a non-ready state, the power state of the image forming apparatus is switched to the ready state.

[0009] In a possible implementation, M is a positive integer greater than or equal to 2, and controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences includes: If all target predicted usage rates in the first group of predicted usage rate sequences are less than a target threshold, and there is a target predicted usage rate greater than or equal to the target threshold in the second group of predicted usage rate sequences, then setting the subsystems corresponding to the target predicted usage rates greater than or equal to the target threshold in the second group of predicted usage rate sequences to a ready state, and setting the subsystems corresponding to the target predicted usage rates less than the target threshold in the second group of predicted usage rate sequences to a low power consumption state; Controlling the power state of the image forming device to enter a standby state; Among them, the target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models, and the target duration corresponding to the second group of predicted usage rate sequences is greater than the target duration corresponding to the first group of predicted usage rate sequences.

[0010] In a possible implementation, M is a positive integer greater than or equal to 3, and controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences includes: If all target predicted usage rates in the first group of predicted usage rate sequences and the second group of predicted usage rate sequences are less than a target threshold, and there is a target predicted usage rate greater than or equal to the target threshold in the third group of predicted usage rate sequences, then setting the subsystems corresponding to the target predicted usage rates greater than or equal to the target threshold in the third group of predicted usage rate sequences to a low power consumption state, and setting the subsystems corresponding to the target predicted usage rates less than the target threshold in the third group of predicted usage rate sequences to a minimum power consumption state; controlling the power state of the image forming device to enter a light sleep state; Among them, the target duration corresponding to the first group of predicted usage sequences is the shortest target duration among the different target durations corresponding to the M sub-models, the target duration corresponding to the second group of predicted usage sequences is greater than the target duration corresponding to the first group of predicted usage sequences, and the target duration corresponding to the third group of predicted usage sequences is greater than the target duration corresponding to the second group of predicted usage sequences.

[0011] In a possible implementation, after controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences, the method further includes: Acquiring actual usage rates of each subsystem in the image forming apparatus within different target time periods to obtain an actual usage rate sequence; Comparing the predicted usage rate sequence output by the deep learning model with the actual usage rate sequence to obtain a comparison result; Determining the prediction accuracy of the deep learning model according to the comparison result; Determining whether the prediction accuracy is less than a preset prediction accuracy threshold; If the prediction accuracy is less than the preset prediction accuracy threshold, the deep learning model is disabled.

[0012] In one possible implementation, the training process of the sub-model in the deep learning model includes: Acquiring historical operation information of the image forming device; Dividing the historical operation information into time windows to generate a time series of the historical operation information; Preprocessing the historical operation information of each time window in the time series to obtain feature information corresponding to each time window; Obtaining the actual usage rate of each subsystem from the first time window to the second time window based on the ratio between the actual usage time of each subsystem from the first time window to the second time window and the target time corresponding to the submodel; wherein the interval between the first time window and the second time window is the target time corresponding to the submodel; Determining the actual usage rate from the first time window to the second time window as the marking information corresponding to the characteristic information of the third time window; wherein the third time window is a previous time window adjacent to the first time window; Inputting the feature information of multiple time windows and the corresponding label information into the initial model for iterative training; When it is detected that the initial model meets the preset conditions, the iterative training is stopped to obtain the sub-model.

[0013] In a possible implementation, the image forming apparatus is provided with a display module, and the display module displays a power status setting interface; the target option is displayed in the power status setting interface.

[0014] In a second aspect, an embodiment of the present invention provides an image forming apparatus, wherein the image forming apparatus includes multiple subsystems, including: A first acquisition module is configured to acquire operation information corresponding to each of the subsystems of the image forming apparatus within a preset time in response to a trigger signal of a target option; a processing module, configured to input the operation information into M sub-models of a pre-trained deep learning model, wherein M is a positive integer greater than or equal to 1; wherein each sub-model is configured to output a target predicted utilization rate of each subsystem within a target future duration, and the target duration corresponding to each sub-model is different; A second acquisition module is configured to acquire M groups of predicted usage rate sequences output by the deep learning model; wherein each group of the predicted usage rate sequences includes a target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M sub-models; A control module is configured to control a power state of the image forming apparatus according to the M groups of predicted usage rate sequences.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, including: at least one processor; and at least one memory in communication with the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method described in the first aspect.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the method described in the first aspect.

[0017] In this embodiment of the present invention, multiple independent sub-models are used to predict the target predicted usage rates of each subsystem in an image forming device over different time periods, thereby adjusting the power state of the image forming device. This allows for prediction of user usage of the image forming device and allows for waking up the image forming device before the user issues a job. By putting the image forming device into sleep mode when it is predicted that the user will not have any jobs to issue in the coming period, the image forming device can be put into sleep mode, reducing energy consumption while improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A flowchart of a power state control method provided by an embodiment of the present invention; Figure 2 A schematic diagram of a sleep time setting interface provided by an embodiment of the present invention; Figure 3 A schematic diagram of a model structure of a sub-model provided in an embodiment of the present invention; Figure 4 A schematic structural diagram of an image forming device provided by an embodiment of the present invention; Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0021] It should be clear that the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0022] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "an", "the" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0023] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. Furthermore, the character " / " in this document generally indicates an "or" relationship between the associated objects.

[0024] The printing control method of the embodiment of the present application is mainly applied to image forming devices, and the device forms of the above-mentioned image forming devices include but are not limited to: inkjet printers, laser printers, light emitting diode (LED) printers, copiers or multifunction machines, and multifunction peripherals (MFPs) that perform the above functions in a single device. The current power management strategy of image forming devices is: the user sets the sleep time, for example, 1 minute. 5 minutes, half an hour, 1 hour, 4 hours, etc. The image forming device enters a sleep state if no user operation is received within the set sleep time. In the sleep state, subsystems such as the print engine and the scanning engine panel backlight are turned off. This method cannot match the actual usage of the image forming device in various scenarios, which will result in poor user experience or low energy efficiency.

[0025] To address the above issues, embodiments of the present invention provide a power state control method. This method collects operational information from an image forming device and inputs it into a pre-trained deep learning model, enabling the deep learning model to predict the image forming device's usage over a period of time. The method then adjusts the power state based on the predicted usage. Figure 1 Flowchart of a power state control method provided by an embodiment of the present invention. Figure 1 As shown in , the method includes: Step 101 : In response to a trigger signal of a target option, obtaining operation information corresponding to each subsystem of an image forming apparatus within a preset time.

[0026] The subsystems in the image forming device may include but are not limited to: touch panel backlight, WIFI / Bluetooth, scanner motor, scanner CIS, scan engine, print engine and heating roller, hard disk, FPGA image processing unit, PCIE peripherals, atmosphere light and other internal functional components of the image forming device.

[0027] The image forming apparatus includes a display module that displays a power state setting interface. The power state setting interface displays target options. When a user wishes to use the power state control method provided in an embodiment of the present invention, the target option can be triggered by a click or other operation. The power state setting interface can be specifically implemented as a sleep time setting interface.

[0028] Figure 2 Schematic diagram of a sleep time setting interface provided by an embodiment of the present invention. Figure 2 As shown in the , the interface displays different sleep time options, namely 1 minute, 5 minutes, 15 minutes, 30 minutes, 1 hour, 8 hours and AI intelligent adjustment. Figure 2 The AI intelligent adjustment shown in the figure is the target option described above. If the user selects an option other than AI intelligent adjustment, the image forming apparatus automatically enters a sleep state after waiting for the selected sleep time following the last user operation. If the user selects the AI intelligent adjustment option, the power state control method provided in this embodiment of the present invention is executed.

[0029] The above-mentioned operation information may include but is not limited to: time information, date information, network information, universal serial bus information, consumables information, user operation information and print queue information.

[0030] Specifically, the time information may include but is not limited to: the current hardware real-time clock (RTC) time of the image forming apparatus, the standby time, the power-on time, the network time, etc.

[0031] The date information may include, but is not limited to, year, month, day, week, and whether the current date is a holiday.

[0032] The aforementioned network information mainly includes wired and wireless network information, including Address Resolution Protocol (ARP), http / https connections, mDNS, TTL time, IPP, Bonjour, SMB, Wi-Fi, Bluetooth, and other data packets that can be received by the printer's wired / wireless network card / Bluetooth device, connection status, user information, etc.

[0033] Universal Serial Bus (USB) information may include, but is not limited to, USB plug-in and unplug-out, enumeration, print / scan commands, USB device access, and other events or data packets.

[0034] The consumables information may include, but is not limited to, the status and remaining amount of a toner cartridge and a paper cartridge in the image forming apparatus.

[0035] User operation information may include, but is not limited to, panel clicks, panel page switching, physical key presses, opening and closing a cover to place paper, and other user operations on the image forming apparatus.

[0036] The print queue information may include a status of a queue of the image forming apparatus.

[0037] In some embodiments, the operation information may further include information that the image forming device is searched, information that the image forming device is queried, and information that the printing software is used.

[0038] Specifically, the image forming device search information refers to: when a user opens an app on a mobile terminal, searching for an image forming device via Bluetooth, Wi-Fi, or other means, thereby establishing a connection with the image forming device. Alternatively, when a PC terminal is powered on, the PC's operating system automatically connects to the installed image forming device. The image forming device can detect which devices are searching for it in the above situations. The device forms of the above-mentioned mobile terminals include, but are not limited to, mobile phones, smart bracelets, smart watches, tablet computers, etc.

[0039] Image forming device query information specifically refers to the fact that when an image forming device is online, the currently connected terminal will periodically query its status. For example, the device may be queried to see if it is out of paper or toner. The image forming device can detect which software on which device is querying it.

[0040] Printing software usage information specifically refers to when a user opens editing software (such as Word, PRD, or scanning software) on a terminal and performs certain image forming device-related operations (such as selecting an image forming device within the software, previewing a print, setting print properties, or setting scan settings). These operations invoke or load the image forming device's print driver. During this process, the editing software on the terminal sends characteristic information to the image forming device, thereby notifying the image forming device that it may be about to be invoked. Such terminal devices include, but are not limited to, mobile phones, tablets, computers, and servers.

[0041] In step 102, the operational information is input into M sub-models of a pre-trained deep learning model. M is a positive integer greater than or equal to 1. Each sub-model is configured to output a target predicted utilization rate for each subsystem within a target future duration, with each sub-model corresponding to a different target duration.

[0042] Among them, the target predicted usage rate refers to the ratio between the predicted usage time of the subsystem within the target time in the future and the target time. For example, if the predicted usage rate of the subsystem is 0.2 and the target time is 30 minutes, it means that the predicted usage time of the subsystem in the next 30 minutes is 6 minutes. One submodel is used to predict the target predicted usage rate of all subsystems within the target time. For example, the image forming device includes 3 subsystems, and the deep learning model includes two submodels. The target time corresponding to the two submodels is 5 minutes and 30 minutes, then one of the submodels is used to predict the target predicted usage rate of the three subsystems within 5 minutes based on the input operation information. The other submodel is used to predict the target predicted usage rate of the three subsystems within 30 minutes based on the input operation information.

[0043] In some embodiments, the number of sub-models in the deep learning model can be adjusted according to actual needs. For example, it is necessary to predict the usage efficiency of each subsystem in the next 1 minute, 15 minutes and half an hour. Then the number of sub-models is 3, namely the first sub-model, the second sub-model and the third sub-model. Among them, the target duration corresponding to the first sub-model is 1 minute, the target duration corresponding to the second sub-model is 15 minutes, and the target duration corresponding to the third sub-model is half an hour. After the operation information is input into the deep learning model, the deep learning model uses the operation information as input data and inputs it into the three sub-models respectively. Afterwards, the first sub-model outputs the predicted usage rate of each sub-system in the next 1 minute. The second sub-model outputs the usage rate of each sub-system in the next 15 minutes. The third sub-model outputs the usage rate of each sub-system in the next half an hour. Among them, the target duration corresponding to each sub-model can be adjusted accordingly according to actual needs, and the embodiments of the present invention are not limited thereto.

[0044] Step 103: Obtain M groups of predicted usage rate sequences output by the deep learning model, wherein each group of predicted usage rate sequences includes the target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M sub-models.

[0045] For example, if you need to predict the usage of each subsystem within 5 minutes, 15 minutes, and 1 hour, the value of M is 3. The first submodel outputs the first set of predicted usage rate sequences. This set of predicted usage rate sequences is the set of target predicted usage rates for each subsystem within 5 minutes. The second submodel outputs the second set of predicted usage rate sequences. This set of predicted usage rate sequences is the set of target predicted usage rates for each subsystem within 15 minutes. The third submodel outputs the third set of predicted usage rate sequences. This set of predicted usage rate sequences is the set of target predicted usage rates for each subsystem within 1 hour.

[0046] Step 104 : Control the power state of the image forming apparatus according to the M groups of predicted usage rate sequences.

[0047] The power state may include: ready state, standby state, light sleep state and deep sleep state. The energy consumption of each power state is ranked from high to low.

[0048] In some embodiments, if all target predicted usage rates within the M sets of predicted usage rate sequences are less than the target threshold, this indicates that the likelihood of each subsystem being called upon in the near future is low. Therefore, each subsystem can be set to its lowest power consumption state. The image forming apparatus can also be controlled to enter a deep sleep state. Some subsystems can also be powered off to further reduce energy consumption.

[0049] In some embodiments, if a target predicted usage rate greater than a target threshold is present in the first set of predicted usage rate sequences, each subsystem is set to a ready state. The image forming apparatus is also controlled to enter a ready state. The target duration corresponding to the first set of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M submodels.

[0050] For example, there are three sub-models, each used to predict the target predicted usage rate of each subsystem within 1 minute, 15 minutes, and 1 hour. The shortest target duration is 1 minute. The first set of predicted usage rate sequences is the set of target predicted usage rates for each subsystem within 1 minute. If any of the target predicted usage rates for each subsystem within 1 minute exceeds the target threshold, it indicates that the image forming device is about to be activated. At this point, the image forming device can be controlled to switch to the ready state.

[0051] When controlling the power state of an image forming device to enter a ready state, it is possible to first detect whether the current power state of the image forming device is in a ready state. If the current power state of the image forming device is in a ready state, the subsystem corresponding to the target predicted usage rate in the first set of predicted usage rate sequences that is greater than the target threshold value can be controlled to perform advance job preparation. If the current power state of the image forming device is in a non-ready state, the power state of the image forming device is switched to a ready state, and then the subsystem corresponding to the target predicted usage rate in the first set of predicted usage rate sequences that is greater than the target threshold value can be controlled to perform advance job preparation. For example, the transfer roller can be preheated for printing and copying jobs, and scan calibration can be performed for scanning jobs.

[0052] In some embodiments, if all target predicted usage rates in the first group of predicted usage rate sequences are less than the target threshold, and there is a target predicted usage rate greater than or equal to the target threshold in the second group of predicted usage rate sequences. This indicates that the image forming device will not be called in a short period of time, but will be called soon. Therefore, the subsystem corresponding to the target predicted usage rate greater than or equal to the target threshold in the second group of predicted usage rate sequences can be set to a ready state, and the subsystem corresponding to the target predicted usage rate less than the target threshold in the second group of predicted usage rate sequences can be set to a low power consumption state, and the power state of the image forming device can be controlled to enter a standby state. Among them, the target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models, and the target duration corresponding to the second group of predicted usage rate sequences is greater than the target duration corresponding to the first group of predicted usage rate sequences.

[0053] For example, there are three sub-models, which are used to predict the usage rate of each subsystem within 1 minute, 5 minutes and 1 hour respectively. The first group of predicted usage rate sequences can be the target predicted usage rate of each subsystem within 1 minute, and the second group can be the target predicted usage rate of each subsystem within 15 minutes. When the target predicted usage rate of each subsystem in the first group of predicted usage rate sequences is lower than the target threshold, and there is a subsystem with a target predicted usage rate greater than or equal to the target threshold in the second group of predicted usage rate sequences, it means that the image forming device may be called after a short time. Therefore, the image forming device enters the standby state. At this time, only the subsystems with high usage rates in the second group of predicted usage rate sequences are kept in the ready state so that they can respond to the user's operations on the image forming device in a timely manner.

[0054] In some embodiments, for the case where the deep learning model includes three or more sub-models, if all target predicted usage rates in the first group of predicted usage rate sequences and the second group of predicted usage rate sequences are less than the target threshold, and there is a target usage rate greater than or equal to the target threshold in the third group of predicted usage rate sequences, then the subsystem corresponding to the target predicted usage rate greater than or equal to the target threshold in the third group of predicted usage rate sequences is set to a low power consumption state, and the subsystem corresponding to the target predicted usage rate less than the target threshold in the third group of predicted usage rate sequences is set to a minimum power consumption state, and the power state of the image forming apparatus is controlled to enter a light sleep state.

[0055] Among them, the target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models, the target duration corresponding to the second group of predicted usage rate sequences is greater than the target duration corresponding to the first group of predicted usage rate sequences, and the target duration corresponding to the third group of predicted usage rate sequences is greater than the target duration corresponding to the second group of predicted usage rate sequences.

[0056] For example, a deep learning system with three submodels is used to predict the usage rate of each subsystem within 1 minute, 15 minutes, and 1 hour, respectively. The first set of predicted usage rate sequences is a collection of target predicted usage rates for each subsystem within 1 minute, and the second set of predicted usage rate sequences is a collection of target predicted usage rates for each subsystem within 15 minutes. The third set is a collection of target predicted usage rates for each subsystem within 1 hour. When the target predicted usage rates of each subsystem within 1 minute and 15 minutes are both lower than the target threshold, while the target predicted usage rate within 1 hour is higher than the target threshold, it indicates that the image forming device has a low probability of being called within 15 minutes and a high probability of being called within 1 hour. Therefore, the image forming device controls the subsystems with high usage rates to be set to a low-power standby state, while the remaining subsystems enter the lowest power consumption state or the power-off state. At the same time, the power state of the image forming device is controlled to enter a light sleep state to reduce energy consumption.

[0057] In some embodiments, the accuracy of the target predicted usage rate of each sub-model in deep learning is automatically evaluated during the use of the image forming device, and the prediction accuracy is obtained by comparing the target predicted usage rate output by the sub-model and the actual usage rate. Specifically, the actual usage rate of each subsystem in the image forming device within different target time lengths can be obtained to obtain an actual usage rate sequence. The predicted usage rate sequence within different target time lengths output by the deep learning model is compared with the actual usage rate sequence within different target time lengths to obtain a comparison result. And the prediction accuracy of the deep learning model is determined based on the comparison result. Afterwards, it is determined whether the prediction accuracy is less than the preset prediction accuracy threshold. If the prediction accuracy is less than the preset prediction accuracy threshold, the deep learning model is disabled, and the deep learning model is enabled again after the accuracy of the deep learning model trained in the background meets the enabling requirements.

[0058] In the above embodiment, the sub-models in the deep learning model are independent of each other. The training process for each sub-model may include: first, obtaining historical operating information of the image forming device. The historical operating information contains the same data types as the aforementioned operating information, and may include time information, date information, network information, universal serial bus information, consumables information, user operation information, and print queue information.

[0059] However, the above historical operation information cannot be used directly for model training. It needs to be converted into a time-based sequence and then the data is normalized. Convert various historical operation information into numerical values in the range of 0-1. Therefore, the historical operation information can be divided according to the time window as a unit to generate a time series of historical operation information. The time window is the smallest unit of the time series. For example, there are 3 subsystems in total, and the corresponding target durations are 1 minute, 5 minutes, and half an hour respectively. The time window can be set to 1 minute, so that the historical operation information is divided into 1-minute units, and the historical operation information of multiple time windows with a time step of 1 minute is obtained.

[0060] Then, the historical operation information of each time window in the time series is preprocessed to obtain the feature information corresponding to each time window. The preprocessing step can be implemented as a normalization process.

[0061] Specifically, in order to enable the sub-model to learn the periodic changes in time and identify the different characteristics between morning, noon and evening times of the day, the time information in the historical operation information can be converted into a function with a value range of [0,1]. For example, the time information in the historical operation information can be expressed using formulas (1) and (2).

[0062] Formula (1):

[0063] Formula (2):

[0064] Where t is the number of minutes between the current time and 00:00 of the current day. To avoid being unable to distinguish the same value of the sin function at 0 and π, it is necessary to use the sin function of formula (1) and the cos function of formula (2) as two independent feature information to represent the time information, thereby normalizing the time information and converting the time information into feature information with a value range of [0,1].

[0065] The day of the week information in the date information can be normalized using the functions shown in formula (3) and formula (4), thereby converting the day of the week information into feature information with a value range of [0,1].

[0066] Formula (3):

[0067] Formula (4):

[0068] Where t is a positive integer from 0 to 6, representing Sunday, Monday, Tuesday, and Saturday respectively.

[0069] For statutory holidays in date information, a value of 1 may be used to indicate that the current date is a statutory holiday, and a value of 0 may be used to indicate that the current date is not a statutory holiday.

[0070] The Address Resolution Protocol (ARP) information in the network information reflects the number of devices in the local area network where the image forming apparatus is located, and its value range is 0 to 256. Therefore, the original value of the ARP information can be divided by 256 to convert the ARP information into feature information with a value range of [0, 1].

[0071] Similarly, in the http connection information in the network information, assuming that the number of simultaneous http sessions that the image forming device can support is x, the value of the collected http connection information can be divided by x, thereby converting the http connection information into feature information with a value range of [0,1].

[0072] Assuming that the maximum number of (remote printing) airprint jobs that the image forming device can process per unit time is y, the actual number of airprint jobs in the current unit time can be divided by y to convert it into feature information with a value range of [0,1].

[0073] Assuming that the maximum number of times the image forming device is queried (including network, USB, etc.) per unit time is z, the actual number of times the image forming device is queried per unit time can be divided by z to convert it into feature information with a value range of [0,1].

[0074] For printer search information, if the maximum preset number of searches per unit time (e.g., 1 minute) is q, the actual number of searches per unit time can be divided by q to convert it into feature information with a value range of [0,1].

[0075] In the historical operation information, the most frequently used clients (network, USB, app, etc.) are considered online. If they are online, the value is 1; if they are not online, the value is 0. This is converted into feature information with a value range of [0,1].

[0076] The online ratio of users who have used the image forming device in the historical operation information is calculated by dividing the current number of online users by the total number of users who have used the image forming device in the past. This is then converted into feature information with a value range of [0, 1].

[0077] For user operations such as whether the user inserts a USB flash drive, opens print preview, enters the XX panel page, presses the XX button, or places the original document, these user operation information can be converted into feature information with a value range of [0,1] in the form of 1 for yes and 0 for no.

[0078] In some embodiments, when the image forming device is in the shutdown state, the image forming device cannot store the operating information during that period. Therefore, the missing historical operating information is supplemented. In addition, shutdown, as a user habit, also affects the training of the deep model. For example, through the historical operating values, it can be known that the user turns off the image forming device in the afternoon every day, but forgets to turn it off one day. If the deep learning model learns the user's usage habit, it will predict that the image forming device will not be used after the afternoon and can enter the sleep state in advance to reduce energy consumption. Therefore, all feature information of the image forming device during the shutdown period can be padded with 0.

[0079] Next, the feature information for each time window needs to be labeled. Since the model needs to be trained to predict usage rates for a future period based on current operating data, the actual usage rates for later time windows need to be used as labeling information for earlier time windows. This allows the trained model to output the target predicted usage rate for the subsystem within a target future duration based on the current operating data.

[0080] Specifically, the actual usage rate of each subsystem from the first time window to the second time window can be obtained based on the ratio of the actual usage time of each subsystem from the first time window to the second time window to the target time corresponding to the submodel, where the interval between the first time window and the second time window is the target time corresponding to the submodel.

[0081] Next, the actual usage rate from the first to the second time windows is determined as the labeling information corresponding to the feature information of the third time window. The third time window is the preceding time window adjacent to the first time window. For example, if the target duration for the submodel is 5 minutes and the time window T is 1 minute, then for the third time window, the actual usage rate for the 5 minutes following the third time window is used as the labeled data. To do this, the ratio of the image forming device's actual usage duration from the fourth to the ninth time windows to the target duration (i.e., 5 minutes) is calculated to obtain the actual usage rate for the fourth to ninth time windows. This actual usage rate is then used as the labeling information for the third time window. For example, if the image forming device is in the target time period of 8:00 to 12:00 (this target time period can represent any historical moment up to the latest moment and cannot represent future moments, as future printer usage data cannot be collected for model training), and the device was only used for 1 minute at 9:00), then the actual usage rate for the image forming device at 8:59 for the next minute is 1 / 1 = 1. At 8:45, the actual usage rate of the image forming device for the next 30 minutes is 1 / 30 = 0.0333. At 8:30, the actual usage rate of the image forming device for the next hour is 1 / 60 = 0.0167. At 9:30, the actual usage rate of the image forming device for the next 30 minutes is 0 / 30 = 0.

[0082] Table 1-1 is an example of some characteristic information and marking information provided by an embodiment of the present invention.

[0083]

[0084] Table 1-1 As shown in Table 1-1, Feature Information 1 through Feature Information N correspond to the aforementioned historical operation information. Subsystem 1 through Subsystem N specifically refer to information about whether each subsystem is in use. If the subsystem is in use, the value is 1; if not, the value is 0. The tag information represents the utilization rate of each subsystem.

[0085] The feature information and corresponding labeling information from multiple time windows are then fed into the initial model for iterative training. Iterative training is terminated when the initial model meets pre-set conditions, resulting in a trained sub-model. The pre-set conditions can include convergence of the initial model and an accuracy greater than a pre-set threshold.

[0086] The above-mentioned feature information and tag information are automatically generated by software running in the image forming device, without the need for manual intervention or manual labeling. The image forming device can collect all the above-mentioned feature information and can confirm which of its subsystems are used at what time.

[0087] In some embodiments, the feature information in the collected historical operation information may reach dozens or even hundreds of types through continuous iterative improvements. The types of feature information collected may vary depending on the model of the image forming device, and new feature information may be introduced as the image forming device firmware is updated.

[0088] Based on this situation, this embodiment provides a method for collecting historical operation information with compatibility and scalability to solve the above-mentioned problems. Specifically, by classifying the input feature information and setting the information type index, the type of feature information and the data index value correspond one to one, so that the input of the model has compatibility and scalability. For example, the input of the model will support a maximum of N types of feature information in the foreseeable future, that is, the input of each time step is a 1 x N tensor: {feature information 1, feature information 2...feature information N}. If the three types of feature information, feature information 1, feature information 2, and feature information 5, are currently supported, the other unsupported feature information can be padded with 0 to complete it into a 1 x N tensor: {feature information 1, feature information 2, 0, 0, feature information 5,..., 0}, and then subsequent processing can be performed after padded.

[0089] By padding uncollected feature information with zeros, we ensure that the model is consistent with historical operational information collected in real-world applications and that the trained sub-model is compatible and scalable across multiple models. If a new feature K is discovered in the future, simply replacing the zeros in the Kth feature in the tensor with the collected real information will complete the expansion of the new feature information.

[0090] In some embodiments, the initial model may select a deep learning model based on a long short-term memory (LSTM) network. Multi-variable multi-output prediction is achieved by employing N identical LSTM models. For example, three LSTM deep learning models are used to predict the usage rate of an image forming device in the next 1 minute, 15 minutes, and 1 hour, respectively. The three LSTM deep learning models share the feature information and corresponding tag information of the aforementioned multiple time windows. Although the three LSTM deep learning models have the same structure, their functions differ. The 1-minute model is suitable for predicting situations in which an image forming device is about to be used. The 15-minute model is suitable for predicting the usage of an image forming device in the near future. The 1-hour model has a longer time span, making it easier to capture the usage patterns of printers in different time periods throughout the day and suitable for predicting the usage of image forming devices over a longer period of time in the future.

[0091] All three LSTM models use the same feature information and corresponding label information as training data. The difference lies in the data resampling based on different time steps during training and inference. For example, the time step of the 1-minute model can be set to 1 minute or shorter, the time step of the 15-minute model can be set to 1 minute to tens of minutes, and the time step of the 1-hour model can be set to tens of minutes to 1 hour.

[0092] The time spans of the above three LSTM deep learning models are only examples and can be modified to any other time span combination, such as a 30-second model + a 30-minute model + a 2-hour model.

[0093] When training the initial model, a rolling prediction method can be used, and a supervised learning mechanism can be set up. Each time a step length of one unit is moved, the feature information set for a given period of time is used to predict the utilization rate of the image forming device for a future period of time. The error between the predicted utilization rate and the actual utilization rate collected is calculated using a loss function. The neural network weights are then updated through the back propagation through time (BPTT) method to minimize the loss function value. Depending on the final results of the model training, an appropriate loss function can be selected from loss functions such as mean square error, logarithm, and cross entropy.

[0094] The supervised learning mechanism consists of input data X (i.e., feature information) and output data Y (i.e., predicted usage rate). The model can learn how to predict the correct output Y from dataset X. The feature information in Table 1-1 can be used as X, and the labeled information can be used as Y.

[0095] Taking the prediction model for each subsystem's usage rate within the next hour (i.e., the aforementioned submodel) as an example, with 180 days of historical data, a total of 180 × 24 × 60 - 1 = 259,199 minutes of feature information can be obtained for model training. The final minus 1 operation is performed because the image forming device's usage within the next minute has not yet occurred, and the feature information for the last minute of the time window cannot be labeled. For example, if the target duration is one hour, each set of feature information used for model training can contain two hours (120 minutes) of feature information. This allows the trained model to output the predicted usage rate for the next hour based on the feature information from the past two hours.

[0096] Based on the 180 days of historical data, we obtain a total of (259,199 - 120 + 1 = 259,080) supervised learning X and Y datasets. In this dataset, the feature information for time windows 0 to 119 is the 0th set of X values, denoted as x0. The labeled data corresponding to the feature information for time window 119 is the 0th set of Y values, denoted as y0. The feature information for time windows 1 to 120 is the 1st set of X values, denoted as x1. The labeled data corresponding to the 120th time window is the 1st set of Y values, denoted as y1. ... The feature information for time windows 259,079 to 259,198 is the 259,079th set of X values, denoted as x259,079. The labeled data corresponding to time window 259,198 is the 259,079th set of Y values, denoted as y259,079.

[0097] Assuming each time window contains N features, the shape of the X dataset corresponding to the 180 days of historical operating information is (259080, 120, N). This is a three-dimensional array consisting of 259,080 120×N two-dimensional arrays. x0, x1…x259079 are all 120×N two-dimensional arrays. The resulting Y dataset has a shape of (259080, 1), which is 259,080 values. Finally, we obtain the X and Y datasets for supervised model training.

[0098] When training the initial model to obtain sub-models with different target durations, different time steps can be used. For example, for the prediction model corresponding to the target predicted usage rate of a subsystem within 1 minute (i.e., the sub-model described above), the past 16 1-minute time steps can be used. That is, 16 minutes of feature information can be used to predict the subsystem usage rate within the next 1 minute. For the 15-minute prediction model, 16 5-minute time steps, or 80 minutes of feature information, can be used to predict the subsystem usage rate within the next 15 minutes. For the 1-hour prediction model, 16 half-hour time steps, or 8 hours of feature information, can be used to predict the subsystem usage rate within the next 1 hour.

[0099] In some embodiments, the neural network used in the sub-model can use fully connected layers for feature extraction, with a multi-layer LSTM network capturing temporal dependencies. This can be combined with an attention mechanism to assign different weights to each time step, improving prediction accuracy. Dropout can be used to enhance the model's generalization capabilities and prevent overfitting.

[0100] Figure 3 Schematic diagram of a sub-model structure provided by an embodiment of the present invention. Figure 3As shown in the figure, the model consists of Input, Linear layer (fully connected layer 1), Relu (activation function), Linear layer (fully connected layer 2), Relu (activation function), Dropout, multiple LSTM layers, Attention (attention mechanism), Dropout, and Linear layer (fully connected output layer) and Output layer.

[0101] The shape of the data input to Input is (batch_size, time_steps, features). Batch_size is the data batch. When model training is performed, the batch size is 16, 32, and other values to improve training efficiency. When the model training is completed and used for actual prediction, the value of batch_size can be 1. Time_steps is the number of time steps, that is, how many time steps in the past are used to predict the next time step. For example, a 1-minute model can use the feature information of the past 16 minutes of time steps as input to obtain the predicted usage rate of the subsystem in the next 1 minute. Features is the number of feature information supported by the image forming device, indicating how many types of feature information there are.

[0102] Fully connected layer 1: This layer first flattens the input data of (batch_size, time_steps, features) shape into (batch_size time_steps, features), so that each time step passes through the fully connected layer independently, and then uses the linear fully connected layer to extract the feature information. The output shape is (batch_size time_steps, N1) contains N1 features.

[0103] Fully connected layer 2: (batch_size The data of time_steps, N1 is then feature extracted and processed into (batch_size time_steps, N2) contains N2 features, and then reshaped into (batch_size, time_steps, N2) shape data.

[0104] Multi-layer LSTM: Each LSTM layer processes and outputs the corresponding hidden state for each time step. For example, the first LSTM layer processes data of the shape (batch_size, time_steps, N²) and outputs (batch_size, time_steps, hidden_size), where hidden_size indicates the number of hidden states in the LSTM. Stacking multiple LSTM layers allows for the extraction of more layers of time series features.

[0105] Attention mechanism: A fully connected layer calculates the importance weight (attention score) of each time step and processes the hidden state of each time step through weighted summation to obtain a context vector. This means that the output of the LSTM layer (batch_size, time_steps, hidden_size) is processed into (batch_size, hidden_size), and multiple hidden states of multiple input time steps are converted into multiple hidden states of one output time step.

[0106] Fully connected output layer: This layer processes the (batch_size, hidden_size) data into (batch_size, functions), where functions represents the number of subsystems that the model needs to predict. For example, if the model predicts the four subsystems {engine preheating, panel backlight, scanner calibration, and hard drive power supply}, the number of functions is 4.

[0107] Output: The model outputs data of shape (batch_size, functions). When the trained model is used for actual prediction, the batch_size value is 1.

[0108] In some embodiments, the different power states of the image forming device may affect the collection of operating information, and thus affect the prediction of usage. When the power state of the image forming device is ready, standby, or light sleep, the main SoC system operates normally, the image forming device can collect operating information normally, and can predict the subsystem usage. When the image forming device is in deep sleep, the memory DDR is in self-refresh state, and the main operating system is in frozen state. Simple tasks are performed by the small core inside the SoC, and the image forming device can only receive wake-up signals from USB, specific network packets, or IO signals. It is unable to collect operating information, and thus is unable to predict usage.

[0109] Therefore, when the image forming device's power state is deep sleep, the main system can be unfrozen through a scheduled wakeup. Network and USB data collected and cached in RAM during deep sleep can be stored, and subsystem usage can be predicted and the subsequent power state determined. If the determination is to continue deep sleep, the main system is immediately put back into deep sleep. This process only wakes up the main system for a short period of time and has little impact on overall power consumption.

[0110] In some embodiments, because a deep learning model needs to collect data for a period of time and be trained before it can be used, when the user selects the AI intelligent adjustment option, if the image forming device does not store sufficient historical operating data or the deep learning model has not yet been trained to a usable level, the power state control method provided by the embodiment of the present invention will be executed according to the user's last sleep time setting or the default sleep time until the image forming device has been used for a period of time and the historical operating information collected has trained the model to a usable level.

[0111] If the user does not select the AI intelligent adjustment option, the image forming device will still collect historical operating information, but will not train and predict the deep learning model. This historical operating information will be rolled over to the memory for a period of time (e.g., three months). This allows the deep learning model to be immediately usable when the user subsequently enables this option.

[0112] In some embodiments, the image forming device uses two processing paths for storing real-time collected operating information in memory. One path involves rolling collection and labeling, compressing and storing the data in memory for model training. The other path involves directly inputting the data into a trained deep learning model for prediction after the power state control method provided by an embodiment of the present invention is activated.

[0113] A currently trained deep learning model can be retrained. A trained deep learning model can also be switched to the currently running model. These two models run simultaneously or in a time-sharing fashion on processing units such as the CPU, GPU, and NPU. When hot-switching a deep learning model, the current power management policy is first set to a preset value (e.g., a sleep time of 4 hours), the currently running deep learning model is stopped, and the newly trained deep learning model is then loaded and run.

[0114] The training of the deep learning model is performed using an idle CPU or NPU at a low priority and can be scheduled or preempted without affecting the normal operation of the image forming device.

[0115] In an embodiment of the present invention, the image forming device automatically adapts to the scene by collecting data, and adopts the best power management strategy after predicting the subsystems and frequency of use of the image forming device in the future through a deep learning model, thereby reducing user waiting time and energy waste, improving user experience and overall energy efficiency of the printer. Assume that a certain image forming device has a standby power of 20W, a sleep power of 3W, and a sleep time of 4 hours. After adopting this solution, the standby time of 2 hours at noon and 4 hours in the evening is reduced every weekday, and the monthly power consumption will be reduced by approximately (20-3) / 1000 (2+4) 20 (days) = 2.04 kWh. This shows that the embodiment of the present invention has a significant effect of reducing energy consumption.

[0116] In some embodiments, the above method requires predicting the target predicted usage rate of each subsystem and performing differentiated control on each subsystem, thus placing high demands on the image forming device's processor. To address this issue, embodiments of the present invention provide another power state control method that only predicts the usage rate of the image forming device, thereby reducing processing pressure and increasing prediction speed.

[0117] Step 201 : In response to a trigger signal of a target option, obtaining operation information of an image forming apparatus within a preset time.

[0118] The process of obtaining the operation information in this step is the same as that in step 101 and will not be repeated here.

[0119] In step 202, the operating information is input into each of N sub-models in the pre-trained deep learning model. N is a positive integer greater than or equal to 1. Each sub-model is configured to output a target predicted utilization rate of the image forming device within a target future duration. Each sub-model corresponds to a different target duration.

[0120] This step differs from step 102 above in that the sub-model outputs only the predicted usage rate of the image forming device itself, and does not include the predicted usage rates of the individual subsystems. For example, if there are three sub-models, corresponding to target durations of 5 minutes, 15 minutes, and 1 hour, the output of the deep learning model is the target predicted usage rate of the image forming device within the next 5 minutes, the next 15 minutes, and the next hour.

[0121] Step 203: Obtain N target predicted usage rates output by the deep learning model.

[0122] Among them, the N predicted usage rates output by the deep learning model are the set of target predicted usage rates corresponding to different target durations output by the N sub-models.

[0123] Step 104 : determining the power state of the image forming apparatus according to the N target predicted usage rates.

[0124] If the N target predicted usage rates output by the deep model are all less than the target threshold, it indicates that the predicted usage rate of the image forming device in the future will be low, so the power state of the image forming device can be controlled to enter a deep sleep state. In the deep sleep state, most subsystems in the image forming device are in the lowest power consumption state or are powered off.

[0125] For example, a deep learning model includes three sub-models, each used to predict the image forming device's usage within 1 minute, 15 minutes, and half an hour. If the target usage predictions output by all three sub-models are below the target threshold, the image forming device will be used less frequently within the next half hour, and can therefore be switched to a deep sleep state to save energy.

[0126] In some embodiments, if the first target predicted usage rate is greater than or equal to the target threshold, the power state of the image forming device is controlled to enter the ready state. The target duration corresponding to the first target predicted usage rate is the shortest target duration among N different target durations. For example, the deep learning model contains 3 sub-models. They are used to predict the usage rate of the image forming device within 1 minute, 1 hour, and 2 hours respectively. The target durations corresponding to the 3 sub-models are 1 minute, 1 hour, and 2 hours respectively. The first target predicted usage rate is the target predicted usage rate corresponding to the shortest 1 minute. When the first target predicted usage rate is greater than or equal to the target threshold, it means that the probability of the image forming device being called in a short time is high, and the image forming device can be controlled to enter the ready state. In the ready state, each subsystem in the image forming device is in the ready state.

[0127] In some embodiments, if the first target predicted usage rate is less than a target threshold and the second target predicted usage rate is greater than or equal to the target threshold, the power state of the image forming apparatus is controlled to enter a standby state, wherein the target duration corresponding to the first target predicted usage rate is the shortest target duration among N different target durations, and the target duration corresponding to the second target predicted usage rate is greater than the target duration corresponding to the first target predicted usage rate.

[0128] For example, if the first target duration is 5 minutes and the second target duration is 30 minutes, and the target predicted usage rate within 5 minutes output by the deep learning model is less than the target threshold, but the target predicted usage rate within 30 minutes is greater than or equal to the target threshold, then the image forming device has a low probability of being used within 5 minutes and a high probability of being used within 30 minutes. Therefore, the image forming device can be controlled to enter a standby state. In the standby state, only some subsystems of the image forming device remain in a ready state, while other subsystems enter a low-power state or are powered off.

[0129] In some embodiments, if the first target predicted usage rate and the second target predicted usage rate are both less than the target threshold, and the third target predicted usage rate is greater than or equal to the target threshold, the image forming device is controlled to enter a light sleep state. The target duration corresponding to the first target predicted usage rate is the shortest target duration among N different target durations, and the target duration corresponding to the second target predicted usage rate is greater than the target duration corresponding to the first target predicted usage rate. The target duration corresponding to the third target predicted usage rate is greater than the target duration corresponding to the second target predicted usage rate. In the light sleep state, some subsystems of the image forming device remain in a low-power standby state, while the remaining subsystems enter a minimum power consumption state or are powered off.

[0130] For example, the first target duration is 1 minute, the second target duration is 15 minutes, and the third target duration is 1 hour. If the target predicted usage rates for both 1 minute and 15 minutes output by the deep learning model are less than the target threshold, but the target predicted usage rate for 1 hour is greater than or equal to the target threshold, this indicates that the image forming device is less likely to be used in the short term and more likely to be used after a longer period. Therefore, the image forming device can be controlled to enter a light sleep state to reduce energy consumption.

[0131] In the above embodiment, the deep learning model is composed of multiple sub-models. The multiple sub-models are independent of each other. The training process of each sub-model specifically includes: obtaining historical operation information of the image forming device.

[0132] The training steps of the sub-model in this embodiment are similar to Figure 1 The main difference between the training steps of the sub-models of the methods shown in

[15] lies in the difference in labeling information. Figure 1 The sub-model in requires the actual usage rate of all sub-systems as label data, while this embodiment only requires the actual usage rate of the image forming apparatus itself.

[0133] Table 1-2 shows the feature information and corresponding marking information of some time windows.

[0134]

[0135] Table 1-2 As shown in Table 1-2, a total of N types of feature information are included, with three sub-models corresponding to target durations of 1 minute, 15 minutes, and 1 hour, respectively. The time window numbers in Table 1-2 represent the number of minutes from 00:00 on the current day. An image forming device is considered in use when it is processing a task (such as a print task, a copy task, or a scan task). An image forming device is considered unused when it is not executing any task.

[0136] Finally, the feature information of multiple time windows and the corresponding label information are input into the initial model for iterative training. When it is detected that the initial model meets the preset conditions, the iterative training is stopped and a sub-model is obtained. This sub-model can only predict the utilization rate of the image forming device itself, and cannot predict the utilization rate of each subsystem. Although the method of this embodiment cannot perform fine control of each subsystem, the method of this embodiment requires less operating information, the target utilization rate that needs to be predicted is less, and the requirement for processing resources is lower than Figure 1 The method of the illustrated embodiment can be implemented in image forming devices with relatively poor processing capabilities, thereby lowering the application threshold.

[0137] correspond Figure 1 The power state control method shown in the figure, an embodiment of the present invention provides an image forming device. Figure 4 FIG. 1 is a schematic structural diagram of an image forming device provided by an embodiment of the present invention. Figure 4 As shown in , the image forming apparatus includes: a first acquisition module 401 , a processing module 402 , a second acquisition module 403 and a control module 404 .

[0138] The first acquisition module 401 is configured to acquire operation information corresponding to each subsystem of the image forming apparatus within a preset time in response to a trigger signal of a target option.

[0139] Processing module 402 is configured to input the operational information into M sub-models of a pre-trained deep learning model. M is a positive integer greater than or equal to 1. Each sub-model is configured to output a target predicted utilization rate for each subsystem within a future target duration, with each sub-model corresponding to a different target duration.

[0140] The second acquisition module 403 is used to acquire M groups of predicted usage rate sequences output by the deep learning model, wherein each group of predicted usage rate sequences includes the target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M sub-models.

[0141] The control module 404 is configured to control the power state of the image forming apparatus according to the M groups of predicted usage rate sequences.

[0142] Figure 5A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the electronic device may include at least one processor and at least one memory in communication with the processor, wherein the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the instructions in this specification. Figure 1-Figure 3 The illustrated embodiment provides a power state control method.

[0143] like Figure 5 As shown, the electronic device is implemented as a general-purpose computing device. Components of the electronic device may include, but are not limited to, one or more processors 510, a communication interface 520, and a memory 530, and a communication bus 540 connecting different system components (including the memory 530, the communication interface 520, and the processor 510).

[0144] Communication bus 540 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnection (PCI) bus.

[0145] Electronic devices typically include a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, removable and non-removable media.

[0146] Memory 530 may include computer-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer storage media. Memory 530 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of this specification.

[0147] A program / utility having a set (at least one) of program modules may be stored in memory 530. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules generally implement the functions and / or methods of the embodiments described herein.

[0148] The processor 510 executes various functional applications and data processing by running the programs stored in the memory 530, such as implementing the Figure 1-Figure 3 The illustrated embodiment provides a power state control method.

[0149] The embodiment of this specification provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program performs the following operations: Figure 1-Figure 3 The illustrated embodiment provides a power state control method.

[0150] The embodiment of this specification provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to execute the present specification. Figure 1-Figure 3 The illustrated embodiment provides a power state control method.

[0151] The computer-readable storage medium described above may take the form of any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0152] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0153] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this specification. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0154] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. Throughout this specification, "plurality" means at least two, such as two or three, unless otherwise specifically defined.

[0155] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of this specification includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of this specification belong.

[0156] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0157] It should be noted that the devices involved in the embodiments of this specification may include but are not limited to personal computers (Personal Computer; hereinafter referred to as: PC), personal digital assistants (Personal Digital Assistant; hereinafter referred to as: PDA), wireless handheld devices, tablet computers (Tablet Computer), mobile phones, MP3 displays, MP4 displays, etc.

[0158] In the several embodiments provided in this specification, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0159] In addition, the functional units in the various embodiments of this specification may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional units.

[0160] The aforementioned integrated unit implemented as a software functional unit can be stored in a computer-readable storage medium. The software functional unit is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a connector, or a network device, etc.) or a processor to execute portions of the method steps described in various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0161] The above description is only a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this specification should be included in the scope of protection of this specification.

[0162] In this specification, reference can be made to the same or similar parts between the various embodiments. In particular, for the device embodiment and the terminal embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiment.

Claims

1. A power state control method, characterized in that: Applied to an image forming apparatus, the image forming apparatus includes multiple subsystems, including: In response to a trigger signal of a target option, obtaining operation information corresponding to each of the subsystems of the image forming apparatus within a preset time; Input the operation information into M sub-models of a pre-trained deep learning model, respectively; M is a positive integer greater than or equal to 1; wherein each sub-model is used to output a target predicted usage rate of each subsystem within a future target duration, and the target duration corresponding to each sub-model is different; Obtaining M groups of predicted usage rate sequences output by the deep learning model; wherein each group of the predicted usage rate sequences includes a target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M sub-models; The power state of the image forming apparatus is controlled according to the M sets of predicted usage rate sequences.

2. The method according to claim 1, characterized in that The controlling the power state of the image forming device according to the M groups of predicted usage rate sequences includes: If all of the target predicted usage rates in the M groups of predicted usage rate sequences are less than a target threshold, setting each of the subsystems to a minimum power consumption state; The power state of the image forming apparatus is controlled to enter a deep sleep state.

3. The method according to claim 1, characterized in that The controlling the power state of the image forming device according to the M groups of predicted usage rate sequences includes: If there is a target predicted usage rate greater than the target threshold value in the first set of predicted usage rate sequences, setting each of the subsystems to a ready state; Controlling the power state of the image forming device to enter a ready state; The target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models.

4. The method according to claim 3, characterized in that The controlling the power state of the image forming apparatus to enter a ready state specifically includes: detecting whether the current power state of the image forming device is a ready state; If the current power state of the image forming apparatus is a ready state, controlling the subsystem corresponding to the target predicted usage rate greater than the target threshold in the first group of predicted usage rate sequences to perform advance job preparation; If the current power state of the image forming apparatus is a non-ready state, the power state of the image forming apparatus is switched to the ready state.

5. The method according to claim 1, wherein M is a positive integer greater than or equal to 2, and controlling the power state of the image forming device according to the M groups of predicted usage rate sequences includes: If all target predicted usage rates in the first group of predicted usage rate sequences are less than a target threshold, and there is a target predicted usage rate greater than or equal to the target threshold in the second group of predicted usage rate sequences, then setting the subsystems corresponding to the target predicted usage rates greater than or equal to the target threshold in the second group of predicted usage rate sequences to a ready state, and setting the subsystems corresponding to the target predicted usage rates less than the target threshold in the second group of predicted usage rate sequences to a low power consumption state; Controlling the power state of the image forming device to enter a standby state; Among them, the target duration corresponding to the first group of predicted usage rate sequences is the shortest target duration among the different target durations corresponding to the M sub-models, and the target duration corresponding to the second group of predicted usage rate sequences is greater than the target duration corresponding to the first group of predicted usage rate sequences.

6. The method according to claim 1, characterized in that M is a positive integer greater than or equal to 3, and controlling the power state of the image forming device according to the M groups of predicted usage rate sequences includes: If all target predicted usage rates in the first group of predicted usage rate sequences and the second group of predicted usage rate sequences are less than a target threshold, and there is a target predicted usage rate greater than or equal to the target threshold in the third group of predicted usage rate sequences, then setting the subsystems corresponding to the target predicted usage rates greater than or equal to the target threshold in the third group of predicted usage rate sequences to a low power consumption state, and setting the subsystems corresponding to the target predicted usage rates less than the target threshold in the third group of predicted usage rate sequences to a minimum power consumption state; controlling the power state of the image forming device to enter a light sleep state; Among them, the target duration corresponding to the first group of predicted usage sequences is the shortest target duration among the different target durations corresponding to the M sub-models, the target duration corresponding to the second group of predicted usage sequences is greater than the target duration corresponding to the first group of predicted usage sequences, and the target duration corresponding to the third group of predicted usage sequences is greater than the target duration corresponding to the second group of predicted usage sequences.

7. The method according to claim 1, characterized in that After controlling the power state of the image forming apparatus according to the M groups of predicted usage rate sequences, the method further includes: Acquiring actual usage rates of each subsystem in the image forming apparatus within different target time periods to obtain an actual usage rate sequence; Comparing the predicted usage rate sequence output by the deep learning model with the actual usage rate sequence to obtain a comparison result; Determining the prediction accuracy of the deep learning model according to the comparison result; Determining whether the prediction accuracy is less than a preset prediction accuracy threshold; If the prediction accuracy is less than the preset prediction accuracy threshold, the deep learning model is disabled.

8. The method according to claim 1, characterized in that The training process of the sub-model in the deep learning model includes: Acquiring historical operation information of the image forming device; Dividing the historical operation information into time windows to generate a time series of the historical operation information; Preprocessing the historical operation information of each time window in the time series to obtain feature information corresponding to each time window; Obtaining the actual usage rate of each subsystem from the first time window to the second time window based on the ratio between the actual usage time of each subsystem from the first time window to the second time window and the target time corresponding to the submodel; wherein the interval between the first time window and the second time window is the target time corresponding to the submodel; Determining the actual usage rate from the first time window to the second time window as the marking information corresponding to the characteristic information of the third time window; wherein the third time window is a previous time window adjacent to the first time window; Inputting the feature information of multiple time windows and the corresponding label information into the initial model for iterative training; When it is detected that the initial model meets the preset conditions, the iterative training is stopped to obtain the sub-model.

9. The method according to claim 1, characterized in that The image forming apparatus is provided with a display module, and the display module displays a power state setting interface; the target option is displayed in the power state setting interface.

10. An image forming apparatus, comprising a plurality of subsystems, characterized in that: include: A first acquisition module is configured to acquire operation information corresponding to each of the subsystems of the image forming apparatus within a preset time in response to a trigger signal of a target option; A processing module, configured to input the operation information into M sub-models of a pre-trained deep learning model respectively; M is a positive integer greater than or equal to 1; wherein each sub-model is used to output a target predicted usage rate of each sub-system within a future target duration, and the target duration corresponding to each sub-model is different; A second acquisition module is configured to acquire M groups of predicted usage rate sequences output by the deep learning model; wherein each group of the predicted usage rate sequences includes a target predicted usage rate of each subsystem, and the M groups of predicted usage rate sequences correspond one-to-one to the M sub-models; A control module is configured to control a power state of the image forming apparatus according to the M groups of predicted usage rate sequences.

11. An electronic device, characterized in that: include: at least one processor; as well as at least one memory in communication with the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor can execute the method according to any one of claims 1 to 9 by calling the program instructions.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the method according to any one of claims 1 to 9.

Images