Intelligent terminal operation scene optimization method and system

Abstract

The invention discloses an intelligent terminal operation scene optimization method and system, and relates to the technical field of intelligent terminal operation scene optimization and computing resource management, and the method comprises the steps: obtaining environment state data and user interaction data; predicting resource gap information based on the state data; predicting a scene task to be triggered based on the interaction data, wherein the second time period is longer than the first time period; generating a resource allocation instruction and a preloading instruction according to the resource gap information and the scene task prediction result; dynamically distributing the computing resources and executing preloading operation; and collecting running state data when the scene task is executed, and performing feedback optimization on the prediction operation based on the running state data. According to the method, passive response of resource configuration and software preparation is converted into prospective execution through dual-time-domain prediction cooperation, and prediction parameters are continuously optimized in combination with execution feedback, so that scene triggering time delay and system jitter are remarkably reduced in a dynamic environment, and the stability of a control center, resource utilization efficiency and user experience consistency are improved.

Description

Technical Field

[0001] This invention relates to the field of operating scenario optimization and computing resource management technology for smart home control centers, specifically to a method and system for optimizing the operating scenario of smart terminals. Background Technology

[0002] In recent years, smart home control centers have gradually evolved from traditional device linkage hubs into intelligent terminals with edge computing and multimodal interaction capabilities. The services they support, such as voice recognition, video decoding, environmental perception, device orchestration, and local inference, continue to increase. With the heterogeneity of terminal-side computing power (coexistence of processors, graphics, or inference acceleration units) and the dynamic fluctuations of the home network environment, operating scenarios exhibit characteristics of multi-task concurrency, resource contention, and service quality sensitivity. To improve response speed and stability, the industry has developed a series of optimization paths in task scheduling, resource isolation, caching and preloading, and scene recognition, and is gradually introducing data-driven prediction mechanisms to support more proactive resource management and service orchestration.

[0003] Despite progress in related research and engineering practices, existing methods for optimizing smart terminal operating scenarios still have significant limitations. Firstly, existing solutions primarily rely on "reallocating resources after task triggering" or make coarse-grained predictions based solely on historical statistics. This makes it difficult to capture transient resource pressures caused by network jitter, concentrated reporting from sub-devices, or sudden environmental events on short timescales of milliseconds to seconds. Consequently, they cannot identify and quantify computing resource gaps within short future windows, leading to queuing, thread starvation, or service jitter during high concurrency. Secondly, existing scenario prediction and resource management are often disconnected: while some systems can infer scenarios that users may trigger, they lack a mapping mechanism to resource dimensions, failing to translate mid-term task predictions into executable resource allocation and preloading instructions. When multiple scenarios compete for the same resource, they typically still rely on static priorities, lacking conflict resolution strategies that dynamically adjust with resource pressure and task urgency. Third, preloading often remains at the level of simple caching or rule-based warm-up, lacking intensity control based on "trigger probability and expected trigger time," which can easily lead to ineffective warm-up consuming resources or insufficient preparation resulting in a degraded user experience. Fourth, existing closed-loop optimizations generally lack traceable data links: it is difficult to achieve spatiotemporal alignment and attribution verification between runtime status data and prediction results; model or rule updates lack clear adjustment criteria for the smallest unit of action (parameters, weights, thresholds), making it difficult to continuously suppress prediction drift caused by changes in user habits and the environment. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing methods for optimizing and scheduling intelligent terminal operation scenarios have the following problems: it is difficult to predict and reserve computing resource gaps in advance on a short time scale; there is a lack of executable instruction-based linkage between scenario task prediction, resource allocation, and preloading; there is a lack of traceable verification and attribution and continuous feedback optimization mechanisms based on operation status data; and there is a lack of how to achieve low latency, stability, and adaptive scenario operation optimization under conditions of multi-task concurrency and resource competition.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a method for optimizing the operation scenario of a smart terminal, characterized by comprising the following steps: Acquire the status data of the environment in which the smart home control center is located, as well as the user's interaction data; Based on the aforementioned status data, predict the computing resource gap information for the first future time period; Based on the interaction data, predict at least one scenario task that will be triggered in a second time period, which is longer than the first time period; Based on the computing resource gap information and the at least one scenario task, generate resource allocation instructions and preloading instructions; Based on the resource allocation instruction and the preloading instruction, the computing resources of the smart home control center are dynamically allocated and a preloading operation is performed. During the execution of the at least one scenario task, runtime status data is collected, and the prediction operation is optimized based on the runtime status data.

[0007] As a preferred embodiment of the intelligent terminal operation scenario optimization method of the present invention, the status data includes at least one of physical environment status data, network status data, and sub-device operation-related status data, and further includes an operation status summary of the smart home control center. The interactive data includes at least one of the following: voice interaction, graphical interface interaction, remote control, or system-triggered interactive events.

[0008] As a preferred embodiment of the intelligent terminal operation scenario optimization method described in this invention, the predicted computing resource gap information for the first future time period includes: Statistical and change features are extracted from the state data within a sliding time window to generate state features for prediction. Based on the state characteristics, resource demand prediction is performed to obtain the predicted resource demand in the first future time period. Based on the resource occupancy of the background resident tasks of the smart home control center and the system reserved resource threshold, the predicted available resources in the first future time period are estimated. The predicted resource demand is compared with the predicted available resources to determine the degree of the gap and the corresponding resource dimension, so as to generate the computing resource gap information; The computing resource gap information is associated with the timestamp of the information and stored in the prediction result cache.

[0009] As a preferred embodiment of the intelligent terminal operation scenario optimization method of the present invention, the prediction of at least one scenario task to be triggered in the second future time period includes: The interaction data is structured and the interaction elements are extracted. The interaction elements include at least one of intent information, control object identifier and action parameters. Interaction features are constructed within the sliding window based on the interaction elements. Based on the interaction features, scene task prediction is performed to obtain at least one candidate scene task and associated prediction information, wherein the prediction information includes at least the trigger probability and the expected trigger time range. The at least one candidate scene task, its associated prediction information, and corresponding interaction features are stored in the prediction result cache along with a timestamp.

[0010] As a preferred embodiment of the intelligent terminal operation scenario optimization method described in this invention, the generation of resource allocation instructions and preloading instructions includes: Based on the prediction information of the at least one scenario task, determine its task priority or urgency; Based on the aforementioned computational resource gap information, the resource pressure level and the corresponding resource dimension are determined; Based on the task priority or urgency, the resource pressure level, and the resource dimension, the resource allocation instruction is generated. The resource allocation instruction is used to configure one or more of the following for the at least one scenario task: computational resource quota, scheduling priority, resource isolation marker, and resource reclamation conditions. Based on the at least one scene task and its prediction information, the preloading instruction is generated. The preloading instruction is used to indicate the software components, model parameters or session context required for preloading the scene task. The preloading intensity is determined based on the trigger probability and expected trigger time range in the prediction information. When multiple scenario tasks are predicted, conflict resolution is performed between the resource allocation instructions and the preloading instructions based on task priority or urgency, and validity period or cancellation conditions are set for the generated instructions.

[0011] As a preferred embodiment of the intelligent terminal operation scenario optimization method described in this invention, the step of dynamically allocating and performing preloading includes: The resource allocation instruction and the preloading instruction are parsed to generate an execution action sequence, which is then executed in the order of dynamic resource configuration followed by preloading. Before executing the dynamic resource configuration, a resource availability check is performed. The resource availability check is used to determine whether the current system resources meet the resource configuration requirements corresponding to the resource allocation instruction and the system reserved resource threshold. Based on the resource allocation instructions, at least one of the following can be dynamically configured and its scheduling parameters adjusted through the resource management interface of the smart home control center or the resource management mechanism of the operating system: processor computing resources, graphics or inference acceleration resources, memory resources, and storage input / output resources. According to the preloading instruction, load the software components, model parameters or session context corresponding to the at least one scene task, and perform the corresponding level of preloading operation according to the preloading intensity; After completing the dynamic configuration and preloading operations, the resource allocation results and the preloading completion status are written to the execution status cache. If the resource availability verification fails, or the dynamic configuration or preloading operation fails or times out, perform a downgraded preloading or revoke the effective dynamic resource configuration.

[0012] As a preferred embodiment of the intelligent terminal operation scenario optimization method described in this invention, the step of feedback optimization includes: During the execution of the at least one scenario task, running status data is collected. The running status data includes at least task performance data and resource status data, and a task identifier and a collection timestamp are associated with the running status data. The prediction record corresponding to the task identifier and collection timestamp in the prediction result cache is called. The prediction record includes at least a state feature snapshot, an interaction feature snapshot and a prediction output. The running state data is then timestamped and compared with the prediction record. Based on the comparison and verification results, at least one of the first prediction parameter set used to predict the computing resource gap information and the second prediction parameter set used to predict the at least one scenario task is updated. The first prediction parameter set and the second prediction parameter set respectively include at least one of parameters, weights or thresholds. At least one of the updated first prediction parameter set and the second prediction parameter set is stored as a parameter snapshot and called in subsequent prediction processes.

[0013] Secondly, embodiments of the present invention provide a smart terminal operation scenario optimization system, comprising: Data acquisition module: Acquires status data of the environment in which the smart home control center is located, as well as user interaction data; Resource gap prediction module: Based on the status data, predict the computing resource gap information in the first future time period; Scene task prediction module: Based on the interaction data, predict at least one scene task that will be triggered in a second time period in the future, where the second time period is longer than the first time period; Instruction generation module: Based on the computing resource gap information and the at least one scenario task, generate resource allocation instructions and preloading instructions; Execution control module: Based on the resource allocation instruction and the preloading instruction, dynamically allocates the computing resources of the smart home control center and performs a preloading operation; Feedback optimization module: Collects running status data when the at least one scenario task is executed, and performs feedback optimization on the prediction operation based on the running status data.

[0014] The beneficial effects of this invention are as follows: By simultaneously utilizing environmental state data and user interaction data, computational resource gap prediction and scenario task prediction are formed at different time scales, and resource allocation instructions and preloading instructions are generated accordingly, transforming resource configuration and software resource preparation from "passive response" to "proactive execution". At the same time, during task execution, running status data is collected and aligned with historical prediction records to achieve continuous updates to prediction parameters (such as thresholds, weights, etc.), thereby effectively reducing scenario triggering latency and running jitter in dynamic network and multi-device concurrent environments, and improving the stability, resource utilization efficiency and user experience consistency of the smart home control center. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is an overall flowchart of a smart terminal operation scenario optimization method provided in the first embodiment of the present invention; Figure 2 The module connection diagram is provided for a smart terminal operation scenario optimization system according to the third embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0017] Example 1, referring to Figure 1 This invention provides a method for optimizing the operation scenarios of smart terminals, as one embodiment of the present invention.

[0018] S1: Obtain the status data of the environment in which the smart home control center is located and the user's interaction data.

[0019] Step S1 provides the input basis for the subsequent step S2 of predicting the computing resource gap and step S3 of predicting the scene task. The smart home control center obtains the status data of the environment and the interaction data of the user through sensor interface, network interface, device access interface and interaction entry, and performs time consistency processing and structured organization on the obtained data so that it can be directly called in subsequent prediction operations.

[0020] The environmental status data is used to characterize the objective state of the physical and logical environment in which the smart home control center is located, as well as the operational constraints related to task execution latency fluctuations. This data can be obtained from one or more data sources and is not limited to specific hardware or communication protocols. Furthermore, the status data may include one or more of the following: physical environment-related status data, network-related status data, and sub-device operation-related status data. Physical environment-related status data can be acquired by built-in or external environmental sensors in the control center, including sensors for collecting data on light intensity, temperature and humidity, noise intensity, presence or movement of people, and the opening and closing of doors and windows. Network-related status data can be acquired by the network monitoring components of the control center, including link latency, latency fluctuations, packet loss, estimated available bandwidth, connection stability, changes in the number of connected devices, or changes in device connection quality. Sub-device operation-related status data can be reported by home sub-devices to the control center or obtained by the control center through active polling, including sub-device online status, working mode, data traffic level, task retry trend, or congestion trend. Furthermore, the control center can also collect its own operational status summary as supplementary input to the status data. The operational status summary is used to characterize the comprehensive trend of resource pressure changes and may include processor load trends, memory pressure levels, or storage input / output waiting trends.

[0021] The user's interaction data is used to characterize the user's operational behavior and intent signals towards the smart home control center, serving as the basis for scene task prediction in the subsequent second time period. Furthermore, the interaction data can originate from one or more of voice interaction, graphical interface interaction, remote control, or system triggers. Voice interaction data includes wake-up events, recognized command semantics or intent categories, and session stage identifiers; graphical interface interaction data includes events such as clicks, swipes, scene panel triggers, or device control operations on the control center's local touchscreen or supporting applications, and their associated objects; remote control or system trigger data includes mobile remote control requests, scheduled task trigger records, or preset rule trigger records. It should also be noted that, for ease of subsequent prediction, the control center can structure the interaction data to include at least one or more of intent information, control object identifiers, and action parameters.

[0022] Furthermore, since the sampling frequency and triggering mechanism of the state data and the interaction data are different, the control center can perform time consistency processing on the two types of data, enabling them to be retrieved and called under a unified time base. Even further, the control center can attach a timestamp to each data record and mark or process delayed or missing records according to the most recent valid value strategy. It should also be noted that the control center can organize the two types of data into searchable data structures and write them into a cache area. This cache area can be a circular buffer to store state data and interaction event sequences within a recent time window. The cache window length can cover at least the historical range required for prediction in the second time period to support subsequent scene task prediction and backtracking verification.

[0023] It should be noted that after step S1 is completed, two types of data results are generated that can be called. The first is the state data of the environment and its corresponding alignment or summary features, which are used to calculate the resource gap prediction in the first time period of step S2. The second is the user's interaction data and its event sequence or summary features, which are used to predict the scene task in the second time period of step S3, thereby providing a data input basis for subsequent instruction generation, resource scheduling and preloading, and feedback optimization.

[0024] S2: Based on the state data, predict the computing resource gap information in the first time period in the future.

[0025] Step S2 is used to make a forward-looking judgment on the potential resource shortage or congestion risks that the smart home control center may experience in the short term, based on the status data obtained in step S1, before the actual triggering of the scene task. This provides a quantifiable decision-making basis for the subsequent step S4 to generate resource allocation instructions and preloading instructions. It should also be noted that the first time period is a short-term prediction window, the length of which can be preset or adaptively adjusted according to changes in the operating load of the control center. The first time period is usually shorter than the second time period used for scene task prediction to ensure that the prediction results can be used for real-time resource scheduling preparation.

[0026] The computing resource gap information is used to characterize the degree of mismatch between "predicted resource demand" and "predicted available resources" in the first future time period. It can be output in the form of levels, scores, risk markers, or multi-dimensional gap results, and is not limited to a single numerical form. It should also be noted that the computing resource gap information can be provided separately for different resource dimensions, including but not limited to one or more of processor computing resources, graphics or inference acceleration resources, memory resources, and storage input / output resources. When the control center does not have a specific hardware acceleration unit, gap results can be output only for the actual existing resource dimensions to ensure that the solution is adaptable to different hardware configurations.

[0027] It should be noted that, to form the input for prediction, the control center constructs state features for the state data. These state features include statistical and change characteristics of the state data within the nearest time range. Further, the state features may include one or more of the following: changes in network-related states, changes in connection topology stability, changes in the rate of change of sub-device reported load, and short-term trends in the control center's own operational summary state. Specifically, changes in network-related states may reflect short-term changes in link latency, latency fluctuations, packet loss, or available bandwidth; changes in connection topology stability may reflect changes in the number of connected devices or changes in connection quality; changes in the rate of change of sub-device reported load may reflect changes in the data reporting frequency, bitrate, or retry trend of high-load devices; and short-term trends in the control center's own operational summary state may reflect changes in processor usage, task queue backlog, or storage input / output waiting trends. Furthermore, the state features can be constructed using a sliding time window approach. When state data is missing, delayed, or subject to noise jitter, robust features can be formed by combining missing data markers, recent valid value retention, or smoothing processing results to reduce the impact of abnormal sampling on the prediction results.

[0028] Furthermore, the control center performs resource demand prediction and available resource estimation based on the aforementioned state characteristics, and generates the computing resource gap information accordingly. It should be noted that resource demand prediction characterizes the resource load intensity required by the control center to process tasks in the first future time period under the environmental and network conditions corresponding to the aforementioned state characteristics; available resource estimation characterizes the remaining resources available for scenario tasks in the first future time period under the constraint of ensuring the stable operation of the operating system and basic services. It should also be noted that resource demand prediction can be implemented by a prediction model, a set of rules, or a combination of both. The prediction model can be a time-series prediction model or a classification model, and the set of rules can be rules based on threshold and trend determination, but this invention is not limited to a specific model type; available resource estimation can be determined by combining the current background resident task occupancy, the system reserved resource threshold, and the control center's own operating summary status, wherein the system reserved resource threshold is used to ensure the minimum resource guarantee for basic services.

[0029] Furthermore, the control center calculates the gap result based on the resource demand forecast and the available resource estimate, and outputs the calculated resource gap information. It should be noted that the gap calculation can be implemented using difference, proportion, or hierarchical mapping methods to map the "degree to which demand exceeds available resources" into a gap output that can be used for scheduling decisions. When the gap result is below a preset threshold, a low gap or no gap marker can be output; when the gap result is above the preset threshold, a high gap marker can be output, indicating the corresponding resource dimension, so that subsequent step S4 can specifically determine the resource reservation level and scheduling bias strategy.

[0030] It should also be noted that the prediction in step S2 can be updated periodically and / or by event triggering to adapt to the rapid changes in state data. At the same time, the control center can record the corresponding state feature snapshot, prediction timestamp and gap output for each gap prediction result and write them into the prediction result cache area for subsequent step S6 to compare, verify and optimize after the scenario task is executed, so as to achieve continuous correction and stable improvement of short-term resource gap prediction.

[0031] S3: Based on the interaction data, predict at least one scenario task that will be triggered in a future second time period, the second time period being longer than the first time period.

[0032] Step S3 is used to make a forward-looking judgment on user intent and its potential business behaviors over a longer timescale than the first time period in step S2, in order to form a prediction result of "what kind of scenario task will be triggered in the future second time period". This prediction result, together with the computing resource gap information output in step S2, serves as a key input for step S4 to generate resource allocation instructions and preloading instructions. It should also be noted that the second time period is used to cover the typical lead time of user interaction-triggered scenario tasks. Its length can be preset or adaptively set, and it should meet the constraint of being longer than the first time period to ensure that the control center can complete the necessary resource preparation and software preloading before the scenario task is actually triggered.

[0033] The interaction data refers to the user interaction-related data obtained in step S1, which includes at least real-time interaction events and may further include one or more of the following: interaction context information and historical interaction logs, to enhance the ability to characterize task triggering trends within the second time period. It should be noted that the real-time interaction events may include voice wake-up events, intent categories and control objects obtained from voice command parsing, graphical interface or touch operation events, mobile or remote control requests, scheduled task trigger records, preset rule trigger records, etc.; the interaction context information may include session stage identifiers, interaction source identifiers, continuous interaction window states, etc.; the historical interaction logs may include summary information such as the trigger frequency, trigger time distribution, and commonly used device combinations of recent similar scenario tasks, but this invention is not limited to a specific log structure.

[0034] It should also be noted that, to improve the feasibility and stability of predictions, the control center can structure and preprocess the interactive data to ensure a unified expression of the prediction input. Furthermore, the control center can perform intent parsing and slot extraction on the voice interaction data to obtain interactive elements that can be used for task mapping. These interactive elements include one or more of intent information, control object identifiers, and action parameters. For graphical interfaces or touch events, the corresponding interface element identifiers, target device or target scene identifiers, and operation types can be recorded. For remote control or system trigger records, the trigger source, trigger condition category, and trigger target can be recorded. Moreover, when interactive events occur repeatedly within a short period or exhibit jitter, the control center can perform de-jittering or merging processing to avoid unstable prediction results caused by repeated events.

[0035] Furthermore, the control center constructs interaction features based on the interaction data. These features characterize the probability and urgency of scene task triggering within the second time period. It should be noted that the interaction features may include one or more of the following: interaction frequency, nearest neighbor interaction time interval, changes in interaction session stages, recent intent category distribution, distribution of recently involved device categories or device combinations, and nearest neighbor expiration information triggered by timed or rule-based mechanisms. Specifically, interaction frequency and nearest neighbor interaction time interval characterize user activity and continuity; intent category distribution characterizes the user's short-term goal tendency; device combination distribution characterizes possible combinations of linked tasks; and nearest neighbor expiration information characterizes the deterministic triggering trend of preset tasks within the second time period. It should also be noted that the above interaction features can be formed using a sliding window method, with the window length being less than or equal to the second time period, and can be adaptively adjusted according to the interaction event density to improve feature stability while ensuring real-time performance.

[0036] It should be noted that the control center performs scene task prediction based on the interaction features to obtain at least one scene task that will be triggered in the second future time period. Further, the scene task prediction can be implemented by a task prediction model, a rule set, or a combination of both. The task prediction model can be a sequence classification model, a time-series prediction model, or a probability evaluation model. The rule set can be discrimination rules based on triggering conditions, session state, or operation path, but this invention is not limited to a specific model type. Furthermore, in one embodiment, the control center can map interaction elements to candidate scene tasks through a mapping table or strategy library of interaction elements to scene tasks, and filter and sort the candidate tasks based on interaction features to output at least one candidate scene task. The candidate scene task may include, but is not limited to, one or more of the following: voice control tasks, security alarm handling tasks, automated linkage tasks, multimedia control tasks, or comfort adjustment tasks.

[0037] It should also be noted that the prediction result output in step S3 is a structured scenario task prediction result, which includes at least candidate scenario task identifiers and prediction information associated with the candidate scenario task. Further, the prediction information may include one or more of the following: trigger probability or confidence level, expected trigger time range, task priority or urgency marker, and a typical resource requirement profile or key parameter summary required for task execution; wherein, the typical resource requirement profile is used to characterize the historical consumption patterns, preset resource templates, or expected demand intensity of processor, graphics processor, memory, or input / output resources when executing the scenario task, so that step S4 can determine the focus of resource reservation and the scope of preloading objects accordingly. Furthermore, the control center can set a confidence threshold or an upper limit on the number of candidates. When the confidence level of a candidate task is lower than the threshold, it can choose not to output the candidate task, only output lightweight candidate tasks, or only output candidate tasks that do not involve heavy resource preloading, thereby reducing the risk of resource waste caused by misprediction.

[0038] Furthermore, to adapt to rapid changes in user intent, the prediction in step S3 can be updated using event triggering and / or periodic updates. It should be noted that when a new interaction event, a change in session stage, or a timed trigger is about to expire is detected, the control center can trigger a prediction update; when interaction events are sparse, the control center can also refresh the prediction results at fixed intervals to maintain the timeliness of the prediction results. It should also be noted that, to support the feedback optimization in step S6, the control center can write the interaction feature snapshot, prediction timestamp, candidate scene tasks, and their prediction information for each prediction into the prediction result cache, and after the scene task is actually executed, compare and verify whether it was triggered, when it was triggered, and whether the triggered task type is consistent, in order to correct the parameters or rule weights for subsequent task predictions.

[0039] It should also be noted that the output of step S3, together with the computational resource gap information of step S2, will be used in the instruction generation process of step S4. This will enable the control center to prioritize high-confidence or high-urgency candidate scenario tasks when resource shortages are high, and to adopt a more conservative resource reservation and preloading strategy for low-urgency candidate tasks, thereby improving the determinism of overall scenario task execution and the stability of system operation.

[0040] S4: Generate resource allocation instructions and preloading instructions based on the computing resource gap information and the at least one scenario task.

[0041] Step S4 is used to fuse and make decisions on the computing resource gap information output by step S2 and the scenario task prediction result output by step S3, forming two types of instructions that can be directly consumed by the system execution module, namely resource allocation instructions and preloading instructions, so as to complete the preparation on the resource side and software side before the scenario task is actually triggered. It should also be noted that the essence of step S4 is to establish a mapping relationship among "resource pressure - task requirements - execution actions": when it is predicted that there is a risk of resource tension in the first future period, key resources are reserved and scheduling is tilted through resource allocation instructions; when it is predicted that a specific scenario task will be triggered in the second future period, software resources and running contexts required for this task are prepared in advance through preloading instructions to reduce the cold start overhead and resource contention at the moment of triggering.

[0042] Furthermore, the control center first determines the task priority or urgency assessment of the at least one scenario task to be used as a weight factor for subsequent instruction generation. It should be noted that the priority determination can be comprehensively obtained based on the confidence level, expected trigger time range, task type, and task security attributes in the scenario task prediction result. For example, tasks involving security alarm handling, door lock control, etc. are set as high priority, and tasks involving audio-visual entertainment, non-critical synchronization, etc. are set as relatively low priority; at the same time, tasks with a closer expected trigger time can be given a higher urgency. Even further, when step S3 outputs a typical resource demand profile or a key parameter summary, the control center can identify the resource sensitivity of this scenario task, such as identifying that it is mainly affected by processor computing power, graphics or inference acceleration, memory bandwidth, or storage input / output, so as to determine the target dimension of resource tilt in the resource allocation instruction.

[0043] It should be noted that the control center then determines the resource pressure level and pressure dimension based on the computing resource gap information to decide the resource reservation intensity and scheduling strategy. Further, the resource pressure level can be mapped from the gap level, gap score, or risk mark, and the pressure dimension can be determined by the resource type indicated by the gap information. It should also be noted that when the gap information indicates a low gap or no gap, the control center can adopt a conservative resource reservation strategy, only making lightweight reservations and lightweight preloading for high-priority tasks; when the gap information indicates a high gap, the control center can adopt an enhanced resource reservation strategy, performing stronger resource isolation or scheduling tilt for high-priority tasks, and imposing speed limits, delays, or concurrency restrictions on low-priority background tasks to release the remaining key resources.

[0044] Furthermore, based on the aforementioned task-side and resource-side assessments, the control center generates the resource allocation instructions. It should be noted that these resource allocation instructions are used to perform executable configuration adjustments to the computing resources within the control center before the scenario task is triggered. They may include at least one or more of the following: resource quotas, scheduling priorities, concurrency limits, resource isolation flags, and resource reclamation conditions for candidate scenario tasks or their corresponding processes / threads. Furthermore, the resource allocation instructions can be generated for different resource dimensions. For example, for processor computing resources, instructions can be generated to increase the priority of critical task threads, bind processing cores, limit the concurrency of non-critical threads, or reduce their scheduling weights; for graphics or inference acceleration resources, instructions can be generated to adjust the inference session concurrency limit, increase the priority of critical inference task queues, or reduce the submission frequency of non-critical inference tasks; for memory and storage input / output resources, instructions can be generated to reserve critical task caches, limit background log compression or large file read / write bandwidth, and adjust the priority of critical task data channels. It should also be noted that the above instructions are not limited to implementation using a specific operating system interface; those skilled in the art can select appropriate resource control mechanisms for implementation based on the control center's operating environment.

[0045] It should also be noted that the control center generates the preloading instructions to prepare the software resources required to execute the scene task before it is triggered. Furthermore, the preloading instructions may include at least one or more of the following: loading functional modules or service components of the target scene task, initializing runtime dependencies, loading models or configurations, establishing a pre-connection with the target device, preparing session context and cached data, and warming up key data paths. Furthermore, the selection of preloading objects can be associated with the scene task type. For example, for voice control tasks, preloading the key modules and session context required for the voice processing link; for security alarm handling tasks, preloading analysis components and establishing a connection context with the camera data source; and for automated linkage tasks, preloading the rule engine and device control template. It should also be noted that to avoid resource waste due to misprediction, the control center can set the preloading intensity based on the confidence level of the scene task and the expected trigger time range: when the confidence level is high and the trigger time is near, full preloading is performed; when the confidence level is low or the trigger time is far, lightweight preloading or only quick-recoverable initialization operations are performed.

[0046] Furthermore, when multiple scenario tasks are predicted during the second time period, the control center can perform conflict resolution and instruction merging. It should be noted that conflict resolution includes at least two types: resource conflicts and action conflicts. Resource conflicts manifest as multiple candidate tasks requesting reservations for the same resource dimension, while action conflicts manifest as multiple candidate tasks requiring the loading or initialization of mutually exclusive components or causing a sudden increase in system load. Furthermore, the control center can arbitrate based on task priority, urgency, and gap pressure level, allocating resources and performing preloading preferentially for high-priority tasks; and employing strategies such as delayed preloading, reduced reservation intensity, or retaining only critical dependencies for low-priority tasks. Even further, the control center can merge the resource allocation instructions of multiple candidate tasks into a unified configuration change set, and set an effective time range and reclamation conditions for this set to avoid system instability caused by duplicate configurations.

[0047] To ensure the controllability and rollbackability of instruction execution, the resource allocation instruction and the preloading instruction can carry an expiration date or triggering condition, and can record the generation timestamp and associated task identifier. Furthermore, when a scenario task is not actually triggered within the expiration date or is predicted to fail, the control center can, based on the reclamation conditions, cancel resource reservations, release the cache occupied by preloading, and restore background task concurrency to reduce resource waste caused by long-term occupation. Even further, the control center can write the instructions generated in this step and their generation basis into the instruction cache area for subsequent step S6 to compare and verify whether the instructions are reasonable, whether the reservations are excessive or insufficient, and whether the preloading is effective after the task is actually executed, thereby supporting closed-loop feedback optimization.

[0048] It should be noted that the output of step S4 includes at least the resource allocation instruction and the preloading instruction. These two types of instructions serve as direct inputs for the subsequent step S5 to perform dynamic allocation and preloading operations, enabling the control center to complete resource-side preparation before the risk of resource shortage occurs and software-side preparation before the scene task is triggered, thereby reducing the risk of latency jitter and stuttering at the moment the scene task is triggered.

[0049] S5: Based on the resource allocation instruction and the preloading instruction, dynamically allocate the computing resources of the smart home control center and perform a preloading operation.

[0050] Step S5 is used to implement the resource allocation instructions and preloading instructions generated in step S4 into executable system operations, enabling the smart home control center to complete resource reservation and software preheating before the actual triggering of the scene task, thereby reducing resource contention and cold start overhead at the moment of triggering. It should also be noted that step S5 does not require the reconfiguration of all resources in the control center, but rather dynamically adjusts the key resource dimensions associated with the at least one scene task, and achieves "controllable and rollbackable" resource configuration changes and preloading execution under the premise of meeting system stability constraints.

[0051] Furthermore, the control center can be configured with an instruction execution module to parse the resource allocation instructions and the preloading instructions, and generate a specific sequence of execution actions accordingly. It should be noted that the sequence of execution actions may include two parts: resource configuration actions and preloading actions, and can be executed in the order of "resources first, preloading later" to avoid the preloading process being preempted by background tasks before resources are reserved, leading to preloading failure or delay. It should also be noted that when the resource allocation instructions and preloading instructions have an effective time range or triggering conditions, the control center can execute the corresponding actions only after the triggering conditions are met or the effective time range is entered, to reduce resource waste caused by invalid reservations and invalid preheating.

[0052] It should be noted that the dynamic allocation is used to adjust the runtime configuration of computing resources to ensure that the predicted candidate scenario tasks can obtain the necessary execution resources after being triggered. Furthermore, dynamic allocation may include configuring one or more of the following: processor computing resources, graphics or inference acceleration resources, memory resources, and storage input / output resources, and the configuration method is not limited to using a specific operating system interface. Furthermore, in the dynamic allocation of processor computing resources, the control center can adjust the scheduling parameters of the processes or threads corresponding to candidate scenario tasks, including increasing scheduling priority, adjusting available time shares, limiting the concurrency of non-critical background tasks, and setting higher preemption weights for critical tasks. In the dynamic allocation of graphics or inference acceleration resources, the control center can set higher scheduling weights or higher queue priorities for acceleration sessions of critical tasks, and set concurrency limits or submission frequency limits for non-critical tasks. In the dynamic allocation of memory resources, the control center can reserve cache space for critical tasks, increase the resident priority of critical data structures, or limit the memory expansion of background tasks. In the dynamic allocation of storage input / output resources, the control center can set higher service levels for read / write paths related to critical tasks, and apply rate limiting or delay strategies to non-critical input / output such as background log writing and large file synchronization.

[0053] It should also be noted that, to ensure system stability, the control center can perform resource availability verification before executing dynamic allocation to determine whether resource reservation will lead to instability of basic services. Furthermore, the resource availability verification may include checking whether the system's resource reservation threshold is met, whether background resident tasks are in a high-occupancy state, and whether critical system services are in an alarm state. When the verification fails, the control center can adopt a degraded execution strategy, such as only increasing the priority of critical tasks without strong isolation reservation, or only limiting the concurrency of some background tasks without implementing full rate limiting, thereby improving the availability of critical tasks while avoiding system instability.

[0054] Furthermore, the preloading operation is used to prepare the software resources and runtime context required to execute the scenario task before it is triggered, thereby reducing the initialization time at the moment of triggering. It should be noted that the preloading operation may include, but is not limited to: loading service components or functional modules corresponding to the scenario task, initializing runtime dependencies, loading models or configurations, establishing a pre-connection with the target device, pre-generating control instruction templates, restoring or creating session contexts, and loading key data into the cache. Furthermore, the preloading operation can be performed in stages according to preloading intensity: when the confidence level of the candidate scenario task is high and the expected trigger time is near, a full preloading can be performed, allowing the task to enter a directly executable state; when the confidence level is low or the expected trigger time is far away, a lightweight preloading can be performed, only completing the initialization and key dependency preparation that can be quickly recycled, to reduce resource consumption caused by misprediction.

[0055] It should also be noted that, to avoid putting reverse pressure on the system during preloading, the control center can set resource limits and interruption conditions for the preloading process. Furthermore, the control center can allocate independent resource budgets to preloading tasks. When resource shortage pressure increases during the first time period or a critical task is about to be triggered, the concurrency of non-critical preloading operations can be paused or reduced to prioritize the stability of critical links. Even further, when anomalies such as unreachable target devices, failed loading of dependent components, or timeouts are detected during preloading, the control center can record the cause of the anomaly and switch to a degraded preloading method, such as only retaining local component initialization, delaying device pre-connection, or performing compensatory loading only when the task is triggered, thereby ensuring the overall robustness of the system.

[0056] Furthermore, after executing resource configuration changes and preloading, the control center can write the current resource configuration status and preloading completion status into the execution status cache for subsequent step S6 to compare effects and provide feedback optimization. It should be noted that the execution status includes at least one or more of the following: the effective result of the resource allocation instruction, the completion marker of the preloading instruction, the preloading time, and a summary of occupied resources. It should also be noted that when a candidate scenario task is not actually triggered or is predicted to fail within the preset validity period, the control center can, according to the reclamation conditions in the instruction, cancel resource reservations, release the cache occupied by preloading, and restore background task concurrency to avoid resource waste caused by long-term occupation.

[0057] It should be noted that after completing step S5, the smart home control center enters the "executable ready state" for the at least one scene task. This means that when the scene task is actually triggered in the future, it can obtain a more stable supply of resources with lower initialization overhead, thereby reducing the risk of end-to-end latency jitter and stuttering during the execution of the scene task, and providing a comparable execution basis for the operation status collection and feedback optimization in step S6.

[0058] S6: Collect running status data when the at least one scenario task is executed, and optimize the prediction operation based on the running status data.

[0059] Step S6 constitutes the adaptive learning closed loop in the optimization method. Its core lies in using the running status data generated during the actual execution of the scenario task, which reflects both the user experience quality and the actual resource consumption, as a feedback signal to continuously correct and consolidate the preceding prediction operation, so as to reduce the prediction drift risk caused by environmental changes, network fluctuations, or user habit migration.

[0060] The operational status data is collected during the execution of at least one scenario task, and includes at least one or more of task performance data and resource status data. Further, the task performance data may include response latency from task triggering to effectiveness, task completion time, control command issuance success rate, number of retries, number of task timeouts, and the number of stutters or frame drops in audio / video tasks; the resource status data may include processor usage changes, task queue backlog, queuing status of graphics or inference acceleration sessions, memory pressure and recycling frequency, storage input / output waiting trends, and network transmit / receive buffer backlog or retransmission trends. It should also be noted that, to ensure the feedback data can be used for subsequent comparative analysis, the control center may attach at least one or more of the following to the operational status data: task identifier, collection timestamp, and resource dimension identifier.

[0061] The control center performs traceable verification and attribution by calling the state feature snapshots, interaction feature snapshots, and corresponding prediction outputs stored in the prediction result cache in steps S2 and S3, aligning them with the timestamps and task identifiers of the collected running status data. Further, the verification and attribution include, but are not limited to: comparing the computational resource gap information output in step S2 with the resource scarcity during the scene task execution to determine if the gap prediction is underestimated or overestimated; comparing the candidate scene tasks output in step S3 with the actual triggered scene task types and trigger time ranges to determine if the task prediction is accurate and if there is a trigger timing offset. Furthermore, the control center can also combine the resource configuration activation status and preloading completion status written in step S5 as auxiliary evidence to distinguish between "prediction deviation" and "execution failure," thereby avoiding misattributing resource allocation or preloading failure to the prediction model.

[0062] It should be noted that, based on the verification and attribution results, the control center performs feedback optimization on the prediction operation. The smallest unit of action for this feedback optimization can be set down to the model parameters, rule weights, or threshold settings used for prediction. Furthermore, for the gap prediction in step S2, the feedback optimization can be applied to one or more of the following: the internal parameters of the short-term resource prediction model, the threshold or hierarchical mapping rule for resource demand prediction, the retention threshold or background occupancy weight for available resource estimation; for the task prediction in step S3, the feedback optimization can be applied to one or more of the following: the internal parameters of the scene task prediction model, the confidence weight of the mapping relationship between interaction features and candidate tasks, the candidate selection threshold, or the upper limit strategy for the number of candidates. It should also be noted that the feedback optimization is not limited to using machine learning updates; it can also be implemented using adaptive adjustments to rule weights, thresholds, or mapping relationships.

[0063] Furthermore, when queuing, timeouts, or significant latency fluctuations are detected in the execution of critical scenario tasks, and resource status data indicates resource scarcity, it can be determined that the gap prediction is too low or the sudden change in resource pressure has not been fully captured. The control center can increase the gap risk weight corresponding to this status characteristic, increase the sensitivity of gap classification, or lower the available resource estimation result so that high gap information can be output earlier in similar situations. Conversely, when resource scarcity is detected for a long period of time but resource reservation or preloading occupancy is high, it can be determined that the gap prediction is too high or the strategy is excessive. The control center can reduce the corresponding risk weight, relax the gap classification threshold, or reduce the reservation trigger condition to reduce invalid reservations.

[0064] Furthermore, if it is detected that the candidate scenario task predicted in step S3 is not triggered within the second time period, or the actual triggered task is inconsistent with the predicted task type, it can be determined as a task prediction deviation. The control center can reduce the task confidence weight under the corresponding interaction feature, adjust the mapping relationship between interaction elements and scenario tasks, or update the candidate screening threshold. If it is detected that the prediction is correct but the trigger time range is significantly off, the control center can correct the generation strategy of the expected trigger time range to improve the fit of the trigger time sequence prediction. Furthermore, if step S3 outputs a typical resource demand profile that is inconsistent with the actual resource consumption, the control center can correct the typical resource demand profile based on the running status data to make it more accurately represent the resource consumption intensity of the task under different environmental conditions, thereby improving the targeting of the instruction generation in step S4.

[0065] It should also be noted that, to reflect the completeness of adaptive learning, the control center can perform positive reinforcement in addition to negative correction. Furthermore, when a prediction is detected as correct and the task execution is smooth, and resource consumption matches the gap assessment, the control center can strengthen the weights of the key state features or interaction features on which the prediction was based, or solidify the threshold and mapping strategy used this time, so that the confidence of future predictions can be maintained or improved under similar conditions, and unnecessary policy oscillations can be reduced.

[0066] Furthermore, the feedback optimization can be implemented using online updates or batch updates. It should be noted that the control center can cache and aggregate the feedback data, triggering an update after meeting preset sample size, time span, or confidence level conditions to avoid excessive disturbance to the prediction strategy from a single abnormal task. When persistent performance degradation or resource shortage alarms are detected, a fast update path can also be triggered to correct deviations promptly. It should also be noted that to ensure system stability, the control center can set safety boundaries for updates, including limiting the maximum adjustment range of parameters, weights, or thresholds in a single update, and retaining a snapshot of the parameters before the update to support rollback, thereby restoring to a stable state in the event of an update anomaly.

[0067] It should be noted that through the above-mentioned collection of operational status data, spatiotemporal alignment and correlation verification, and feedback optimization and positive reinforcement of the prediction minimum action unit, the control center can continuously correct the resource gap prediction in the first time period and the scenario task prediction in the second time period, so that the resource allocation and preloading strategy remains effective under different home environments and different user habits, thereby achieving continuous optimization of the smart terminal operation scenario.

[0068] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0070] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0071] Example 3, referring to Figure 2 As an embodiment of the present invention, a smart terminal operation scenario optimization system is provided, which includes a data acquisition module, a resource gap prediction module, a scenario task prediction module, an instruction generation module, an execution control module and a feedback optimization module; Data acquisition module: Acquires status data of the environment in which the smart home control center is located, as well as user interaction data; Resource gap prediction module: Based on the status data, predict the computing resource gap information in the first future time period; Scene task prediction module: Based on the interaction data, predict at least one scene task that will be triggered in a second time period in the future, where the second time period is longer than the first time period; Instruction generation module: Based on the computing resource gap information and the at least one scenario task, generate resource allocation instructions and preloading instructions; Execution control module: Based on the resource allocation instruction and the preloading instruction, dynamically allocates the computing resources of the smart home control center and performs a preloading operation; Feedback optimization module: Collects running status data when the at least one scenario task is executed, and performs feedback optimization on the prediction operation based on the running status data.

[0072] Example 4 is an embodiment of the present invention, which provides a method for optimizing the operation scenario of a smart terminal. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0073] This embodiment verifies the engineering feasibility and performance benefits of an intelligent terminal operation scenario optimization method in a smart home control center. The test platform uses a mass-produced smart home control center with a multi-core processor, graphics or inference acceleration capabilities, limited memory and local storage, and runs a home-specific operating system. The test scenario is set up as a typical home topology: lighting, air conditioning, fresh air systems, curtains, door locks, TV boxes, speakers, cameras, and other devices are arranged in areas such as the living room, bedroom, entryway, and kitchen. Sensors for temperature, humidity, light, sound, and human presence are also deployed, totaling approximately 15 devices. The network side uses home Wi-Fi access, and controllable link disturbance conditions are set to simulate latency fluctuations, increased jitter, and intermittent packet loss common in home networks during peak evening hours. On the load side, multiple high-concurrency triggers are set, such as camera bitrate adjustment, security event reporting, continuous voice commands, and centralized distribution of remote control commands from mobile devices, to enhance the realism of resource contention.

[0074] The experiment included six comparative subjects to demonstrate the synergistic value of the invention's "short-term resource gap prediction + medium-term scenario task prediction + instruction-based resource allocation and preloading + runtime feedback self-optimization": Scheme A is trigger-based scheduling (traditional responsive); Scheme B is historical static allocation (based on historical averages or fixed configurations); Scheme C is task prediction and preloading only (no gap reservation); Scheme D is gap prediction and reservation only (resource reservation is made but preloading is weakened); Scheme E is dual prediction but without feedback (prediction parameters are fixed); Scheme F is dual prediction with a feedback loop (this invention). All schemes are run under the same equipment topology, the same network disturbance intensity, and the same task set to ensure fair comparison.

[0075] The implementation process strictly follows steps S1–S6 of Example 1 in the instruction manual. In phase S1, the control center collects environmental status data and user interaction data at fixed intervals, combined with event-triggered methods. Environmental status data includes at least: physical environment readings (temperature, humidity, light intensity, ambient noise intensity, etc.), network status (round-trip latency, jitter, packet loss rate, instantaneous throughput changes, etc.), sub-device operational status (online changes, reporting frequency, bitrate, etc.), and further includes the control center's operational summary status (processor utilization trend, ready queue length, memory pressure, and storage I / O wait times, etc.). Interaction data includes at least voice interaction events, graphical interface or mobile terminal control events, and remote control and rule-triggered events. All the above data is uniformly timestamped and written to a cache for subsequent sliding window feature construction.

[0076] In phase S2, using the first time period as a short-term prediction window (second-level, adaptive to the system scheduling cycle and disturbance duration), the state data is subjected to sliding time window statistical and change feature extraction, and the state features used for prediction are output. Based on this, resource demand prediction is performed, and the available resources are estimated by combining the background resident task occupancy and the system reserved resource threshold. This is then compared to obtain computing resource gap information (including gap degree and corresponding resource dimensions, such as processors, memory, and storage input / output). The gap information is associated with the generation timestamp and written to the prediction result cache. In phase S3, using the second time period as a longer prediction window (minute-level, longer than the first time period), the interaction data is structured, interaction elements are extracted, and interaction features are constructed. Candidate scenario tasks that may be triggered in the second time period and their prediction information (at least including trigger probability and expected trigger time range) are predicted and written to the prediction result cache as well. To improve engineering rationality, although scheme D does not perform strong preloading, it still retains lightweight preparation actions (such as session context initialization and dependency checks) to ensure a reasonable non-zero preloading hit rate for some tasks during high-pressure periods.

[0077] In Phase S4, information on computing resource gaps and scenario task predictions are integrated to determine task priority or urgency, resource pressure level, and resource dimensions, generating resource allocation instructions and preloading instructions. Resource allocation instructions configure resource quotas, scheduling priorities, resource isolation flags, and resource reclamation conditions; preloading instructions instruct the preloading of software components, model parameters, or session contexts, setting the preloading intensity based on trigger probability and expected trigger time range. When multiple candidate tasks are predicted, conflict resolution is performed, and validity periods or cancellation conditions are set for the instructions. In Phase S5, the two types of instructions are parsed to generate execution action sequences, executed in the order of "dynamic resource configuration first, then preloading." Before executing dynamic resource configuration, resource availability is checked to ensure that the resource configuration requirements corresponding to the instructions are met and the system's reserved resource threshold is not exceeded. After execution, the resource allocation results and preloading completion status are written to the execution status cache. If the verification fails or execution fails / timeouts, downgraded preloading or cancellation of effective dynamic resource configuration is performed to avoid triggering a system avalanche. In the S6 stage, runtime status data (task performance data and resource status data) is collected during the execution of scenario tasks. The data is then aligned with the prediction records in the prediction result cache according to the task identifier and timestamp. Comparison and attribution analysis are performed, prediction parameters (parameters, weights, thresholds, etc.) are updated, and parameter snapshots are persisted for subsequent rolling predictions, thereby suppressing model drift caused by changes in environment and habits.

[0078] The test lasted for several consecutive days, focusing on the metrics during peak evening load periods, and summarizing the performance and resource efficiency of different solutions. Key data is recorded in Table 1. Table 1: Experimental Data Recording Table

[0079] Table 1 shows that Scheme F (dual prediction + feedback closed loop, this invention) simultaneously improves "experience stability, resource bottleneck suppression, preloading effectiveness, and resource efficiency," and the improvement direction forms a verifiable causal link with S1–S6. First, in terms of response latency, the P95 response latency is reduced from 479.1 ms in Scheme A to 193.4 ms, the average response latency is reduced from 218.6 ms to 97.6 ms, and the cold start overhead is reduced from 146.7 ms to 41.3 ms, indicating that this invention has significant suppression capabilities for both "tail latency" and "cold start." This phenomenon can be explained by a dual-instruction mechanism: Scheme C, which only performs task prediction and preloading, significantly reduces the cold start overhead to 72.4 ms, but its P95 is still 338.6 ms. This indicates that under intense resource contention, even if the software is prepared in advance, the trigger moment may still be hampered by processor scheduling and I / O waiting. Scheme D, which only performs gap prediction and reservation, is better than C in P95 (311.7 ms), but its cold start overhead is still 98.6 ms, indicating that resource reservation cannot eliminate the loading and session construction costs of task initialization. This invention outputs resource allocation instructions and preloading instructions simultaneously in S4, and executes them in S5 according to the "resource first, preload later" principle. This ensures that critical tasks obtain resource guarantees and necessary software preparation before triggering, thereby statistically reducing both "waiting latency" and "initialization latency." This is the fundamental reason why Scheme F is superior in both P95 and cold start metrics.

[0080] Secondly, in terms of stability and reliability, the number of stuttering / frame dropping events decreased from 6.5 times / hour in Scheme A to 1.3 times / hour in Scheme F, the timeout rate of high-priority tasks decreased from 4.8% to 0.7%, accompanied by a decrease in peak CPU utilization from 92.9% to 77.6% and I / O wait time P95 from 37.8 ms to 20.6 ms. These indicators collectively point to the fact that this invention, through the forward-looking identification of short-term resource gaps in S2 and the resource availability verification in S5, significantly reduces the probability of the system entering the saturation range, thereby reducing timeouts and stuttering caused by queue backlog, thread starvation, and I / O congestion. It is worth noting that while Scheme E (dual prediction without feedback) is superior to the single prediction scheme in terms of latency and stuttering, it is still significantly inferior to Scheme F (e.g., P95 of 252.9 ms, stuttering events of 2.3 times / hour, and gap prediction error MAPE of 15.2%). This indicates that relying solely on "dual prediction" is insufficient to maintain high accuracy and stability in the long term. It is essential to depend on S6's verification attribution and parameter snapshot updates to suppress prediction biases caused by network environment, sub-device firmware updates, and user habit shifts. The gap prediction error MAPE of Scheme F decreased to 10.3%, accompanied by an improvement in the Top-1 accuracy of task prediction to 83.5%. This is a result of simultaneously correcting for both "short-term resource pressure" and "medium-term task trends," demonstrating that closed-loop feedback in this invention is not merely an additional function, but a necessary structure for continuous optimization.

[0081] Furthermore, considering the balance between preloading effectiveness and resource efficiency, Scheme D achieves a preloading hit rate of 38.5%, consistent with its engineering intuition that "gap prediction can indirectly point to high-load tasks and trigger lightweight preparation actions," avoiding the unreasonable phenomenon of "only gap prediction but almost no preloading hit." Meanwhile, Scheme F of this invention, with a preloading hit rate of 78.6%, reduces resource waste to 6.3%, superior to Scheme E's 8.7% and Scheme C's 10.8%. This demonstrates that in S4, this invention adjusts the preloading intensity through "task urgency / probability / expected trigger time range," and combines resource pressure level and resource dimension for conflict resolution and validity period control, thereby improving the hit rate while suppressing invalid reservations and invalid preheating. In contrast, Scheme B (historical static allocation) has a resource waste rate as high as 13.4%, indicating that static strategies tend to reserve redundancy for uncertainty, thus sacrificing resource efficiency. This invention, through a combination of "prediction-driven + instruction-based execution + feedback correction," dynamically converges redundancy, demonstrating a synergistic effect that is difficult to achieve simultaneously with existing technologies. In summary, Table 1 provides mutually corroborating evidence from multiple dimensions, including latency, stuttering, timeout, prediction error, hit rate, and waste rate, supporting the technical effects of this invention in achieving lower tail latency, higher stability, stronger adaptability, and better resource efficiency in complex home environments, and can be reasonably attributed to the key structural design of S1–S6.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing the operation scenario of a smart terminal, characterized in that, include: Acquire the status data of the environment in which the smart home control center is located, as well as the user's interaction data; Based on the aforementioned status data, predict the computing resource gap information for the first future time period; Based on the interaction data, predict at least one scenario task that will be triggered in a second time period, which is longer than the first time period; Based on the computing resource gap information and the at least one scenario task, generate resource allocation instructions and preloading instructions; Based on the resource allocation instruction and the preloading instruction, the computing resources of the smart home control center are dynamically allocated and a preloading operation is performed. During the execution of the at least one scenario task, runtime status data is collected, and the prediction operation is optimized based on the runtime status data.

2. The method for optimizing the operation scenario of a smart terminal as described in claim 1, characterized in that, The status data includes at least one of the following: physical environment status data, network status data, and sub-device operation-related status data, and further includes a summary of the operation status of the smart home control center; The interactive data includes at least one of the following: voice interaction, graphical interface interaction, remote control, or system-triggered interactive events.

3. The method for optimizing the operation scenario of a smart terminal as described in claim 2, characterized in that, The predicted computing resource gap information for the first future time period includes: Statistical and change features are extracted from the state data within a sliding time window to generate state features for prediction. Based on the state characteristics, resource demand prediction is performed to obtain the predicted resource demand in the first future time period. Based on the resource occupancy of the background resident tasks of the smart home control center and the system reserved resource threshold, the predicted available resources in the first future time period are estimated. The predicted resource demand is compared with the predicted available resources to determine the degree of the gap and the corresponding resource dimension, so as to generate the computing resource gap information; The computing resource gap information is associated with the timestamp of the information and stored in the prediction result cache.

4. The method for optimizing the operating scenarios of intelligent terminals as described in claim 3, characterized in that, The prediction of at least one scenario task to be triggered in the second future time period includes: The interaction data is structured and the interaction elements are extracted. The interaction elements include at least one of intent information, control object identifier and action parameters. Interaction features are constructed within the sliding window based on the interaction elements. Based on the interaction features, scene task prediction is performed to obtain at least one candidate scene task and associated prediction information, wherein the prediction information includes at least the trigger probability and the expected trigger time range. The at least one candidate scene task, its associated prediction information, and corresponding interaction features are stored in the prediction result cache along with a timestamp.

5. The method for optimizing the operating scenarios of intelligent terminals as described in claim 4, characterized in that, The generated resource allocation instructions and preloading instructions include: Based on the prediction information of the at least one scenario task, determine its task priority or urgency; Based on the aforementioned computational resource gap information, the resource pressure level and the corresponding resource dimension are determined; Based on the task priority or urgency, the resource pressure level, and the resource dimension, the resource allocation instruction is generated. The resource allocation instruction is used to configure one or more of the following for the at least one scenario task: computational resource quota, scheduling priority, resource isolation marker, and resource reclamation conditions. Based on the at least one scene task and its prediction information, the preloading instruction is generated. The preloading instruction is used to indicate the software components, model parameters or session context required for preloading the scene task. The preloading intensity is determined based on the trigger probability and expected trigger time range in the prediction information. When multiple scenario tasks are predicted, conflict resolution is performed between the resource allocation instructions and the preloading instructions based on task priority or urgency, and validity period or cancellation conditions are set for the generated instructions.

6. The method for optimizing the operating scenarios of intelligent terminals as described in claim 5, characterized in that, The dynamic allocation and preloading operation includes: The resource allocation instruction and the preloading instruction are parsed to generate an execution action sequence, which is then executed in the order of dynamic resource configuration followed by preloading. Before executing the dynamic resource configuration, a resource availability check is performed. The resource availability check is used to determine whether the current system resources meet the resource configuration requirements corresponding to the resource allocation instruction and the system reserved resource threshold. Based on the resource allocation instructions, at least one of the following can be dynamically configured and its scheduling parameters adjusted through the resource management interface of the smart home control center or the resource management mechanism of the operating system: processor computing resources, graphics or inference acceleration resources, memory resources, and storage input / output resources. According to the preloading instruction, load the software components, model parameters or session context corresponding to the at least one scene task, and perform the corresponding level of preloading operation according to the preloading intensity; After completing the dynamic configuration and preloading operations, the resource allocation results and the preloading completion status are written to the execution status cache. If the resource availability verification fails, or the dynamic configuration or preloading operation fails or times out, perform a downgraded preloading or revoke the effective dynamic resource configuration.

7. The method for optimizing the operation scenario of a smart terminal as described in claim 6, characterized in that, The feedback optimization includes: During the execution of the at least one scenario task, running status data is collected. The running status data includes at least task performance data and resource status data, and a task identifier and a collection timestamp are associated with the running status data. The prediction record corresponding to the task identifier and collection timestamp in the prediction result cache is called. The prediction record includes at least a state feature snapshot, an interaction feature snapshot and a prediction output. The running state data is then timestamped and compared with the prediction record. Based on the comparison and verification results, at least one of the first prediction parameter set used to predict the computing resource gap information and the second prediction parameter set used to predict the at least one scenario task is updated. The first prediction parameter set and the second prediction parameter set respectively include at least one of parameters, weights or thresholds. At least one of the updated first prediction parameter set and the second prediction parameter set is stored as a parameter snapshot and called in subsequent prediction processes.

8. A smart terminal operation scenario optimization system, used to implement the smart terminal operation scenario optimization method as described in any one of claims 1 to 7, characterized in that, include: Data acquisition module: Acquires status data of the environment in which the smart home control center is located, as well as user interaction data; Resource gap prediction module: Based on the status data, predict the computing resource gap information in the first future time period; Scene task prediction module: Based on the interaction data, predict at least one scene task that will be triggered in a second time period in the future, where the second time period is longer than the first time period; Instruction generation module: Based on the computing resource gap information and the at least one scenario task, generate resource allocation instructions and preloading instructions; Execution control module: Based on the resource allocation instruction and the preloading instruction, dynamically allocates the computing resources of the smart home control center and performs a preloading operation; Feedback optimization module: Collects running status data when the at least one scenario task is executed, and performs feedback optimization on the prediction operation based on the running status data.

Images