An intelligent television system and control method supporting remote storage management

Abstract

The present application relates to the technical field of television storage management, in particular to an intelligent television system and control method supporting remote storage management, which comprises a user behavior portrait modeling unit, a multi-source collaborative decision unit and a remote configuration interface unit, and the control method comprises the following steps: collecting user viewing behavior data, generating behavior portrait parameters through a pre-trained long short-term memory network, calculating content access probability and marking high-priority backup objects accordingly, constructing a multi-dimensional decision matrix in combination with storage and bandwidth, generating an optimal backup path through a weighted entropy method and Dijkstra algorithm, performing block incremental backup, performing hash comparison to locate the breakpoint for resuming transmission after transmission interruption, and receiving mobile terminal parameter update decision logic, so as to solve the problems of backup resource waste and poor path adaptation, and improve the efficiency and reliability of intelligent television remote storage management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of television storage management, in particular to an intelligent television system supporting remote storage management and a control method. BACKGROUND

[0002] Television storage management is an important technology. Under the background of the continuous enrichment of the functions of intelligent televisions, the content volume of films, variety shows and series watched by users continues to grow, and the local storage capacity of intelligent televisions gradually becomes difficult to meet the long-term content retention needs. Remote storage becomes an important supplement. The core role of this technology is to relieve the local storage pressure of intelligent televisions through systematic remote storage management, avoid the problem that users need to frequently delete content due to insufficient storage, optimize content backup efficiency, ensure that the content preferred by users can be accessed at any time, balance storage costs, network overhead and device reliability, promote the upgrade of intelligent televisions from local storage dependence to local-remote collaborative storage, improve the long-term user experience, and widely adapt to intelligent television devices in home entertainment scenarios.

[0003] The existing intelligent television remote storage management technology faces the core problems of rigid backup strategy, poor path adaptability and inefficient interruption processing in actual application. In the selection of backup content, the existing technology mostly uses fixed rules to mark backup objects without dynamically judging the content value in combination with user viewing behavior. For example, only the content storage duration or type is generally backed up without considering the user's habit of following the updates of specific series and the characteristics of high-frequency viewing period, resulting in a large amount of content that is rarely accessed by users being backed up, wasting remote storage space and home network bandwidth, and the content that the user really needs to retain may be missed due to not being marked. In the selection of backup path, the existing technology often relies on network bandwidth or remote device distance alone without comprehensively considering storage cost, network stability and device reliability. For example, only the path with the highest bandwidth is selected without considering that the remote device corresponding to the path has a high failure rate, which easily leads to interruption or data damage during the backup process, or a low-cost remote storage is selected due to too many path hops, resulting in serious transmission delay. In the interruption processing during transmission, the existing technology mostly uses full-amount retransmission, i.e., the entire content file needs to be sent again after interruption without dividing the content into blocks and marking them. If the interruption occurs in the later stage of transmission, most of the data that has been transmitted needs to be sent repeatedly, which not only wastes bandwidth but also greatly prolongs the backup time, further reducing the user experience. These problems are superimposed on each other, resulting in low efficiency of remote storage management, waste of storage and bandwidth resources caused by backup redundancy, increased risk of backup failure due to improper path, prolonged backup period due to interruption retransmission, and ultimately the remote storage cannot effectively play a supplementary role, and even the user may abandon it due to poor experience. In order to solve this problem, we provide an intelligent television system supporting remote storage management and a control method. SUMMARY

[0004] The application aims to provide an intelligent television system and a control method supporting remote storage management to solve the problems in the background art.

[0005] Due to the rigid backup strategy, the resources are wasted or missed without combining the user behavior to select the backup content, therefore, the case generates portrait parameters by collecting watching data through the user behavior portrait modeling unit, and the multi-source collaborative decision unit calculates the access probability according to the portrait parameters and dynamically adjusts the threshold and marks the high-priority objects, which can accurately backup and save resources.

[0006] Due to the waste of bandwidth and long time consumption caused by full retransmission after transmission interruption, the case generates hash identification by blocking the high-priority objects, locates the breakpoint by comparing the hash when interrupted, uploads the subsequent and verifies the integrity, which can reduce redundant transmission and shorten the backup time consumption.

[0007] To achieve the above-mentioned purpose, an intelligent television system supporting remote storage management is provided, which comprises a user behavior portrait modeling unit for collecting user watching behavior data in real time and inputting the data into a pre-trained long short-term memory network to generate user behavior portrait parameters, and a multi-source collaborative decision unit for calculating content access probability value based on the user behavior portrait parameters, local storage remaining space, real-time bandwidth of home network and response delay of remote storage device, and performing the following operations:

[0008] (a) marking the content with a high-priority backup object if the content access probability value is higher than a preset threshold based on the user behavior portrait parameters;

[0009] (b) constructing a multi-dimensional decision matrix containing storage cost, network overhead and device reliability according to the local storage remaining space, real-time bandwidth of home network and response delay of remote storage device, and dynamically generating an optimal backup path by combining the multi-dimensional decision matrix with a weighted entropy value method;

[0010] (c) triggering the block incremental backup of the high-priority backup object when the real-time bandwidth of home network is higher than a preset network bandwidth for a fixed period of time and the remote storage device is idle;

[0011] A remote configuration interface unit is configured to provide an encrypted communication channel to a mobile terminal, receive a backup strategy weight parameter adjusted manually by a user and synchronize the parameter to the multi-source collaborative decision unit.

[0012] The second purpose of the application is to provide a method for implementing an intelligent television system supporting remote storage management comprising any one of the above, comprising the following steps:

[0013] S1, capture single viewing duration, single day viewing time period distribution, content type switching frequency and continuous episode skipping interval through the time sequence data collector, input the behavior data into the pre-trained long short-term memory network according to the time window sequence, extract the time period correlation features from the network hidden layer and fuse the historical portrait baseline, and dynamically generate the user behavior portrait parameters;

[0014] S2, calculate the content access probability value based on the user behavior portrait parameters, mark the high priority backup object combined with the dynamic baseline adjustment mechanism, construct a multi-dimensional decision matrix according to the local storage remaining space, the real-time bandwidth of the home network and the response delay of the remote storage device, calculate the information entropy weight by the weighted entropy value method and drive the Dijkstra optimization algorithm to generate the optimal backup path;

[0015] S3, the high priority backup object is segmented into fixed size data blocks and a unique hash identifier is generated, when it is detected that the real-time bandwidth of the home network is continuously stable and the remote storage device is idle, the incremental data block in the local hash identifier library is transmitted, if the transmission is interrupted, the breakpoint is located by the hash identifier difference and the resuming mechanism is triggered, and the partial write integrity of the remote data block is verified before resuming;

[0016] S4, receive the backup strategy weight parameter sent by the mobile terminal through the encrypted communication channel, and update the entropy value method weight distribution and dynamic baseline adjustment logic in the multi-source collaborative decision unit in real time.

[0017] Compared with the prior art, the beneficial effects of the present application are:

[0018] The user behavior portrait modeling unit collects the data of single viewing duration and time period distribution, inputs the pre-trained long short-term memory network to generate dynamic portrait parameters, and the multi-source collaborative decision unit calculates the content access probability and dynamically adjusts the threshold to mark the high priority backup object, so that the technical effect of accurately selecting the backup content is achieved, and the problems of resource waste or omission caused by the rigid backup strategy of the prior art are solved, which has the advantages of saving remote storage and network bandwidth resources and avoiding omission of user preference content.

[0019] The multi-source collaborative decision unit constructs a multi-dimensional decision matrix combined with the storage cost, network overhead and device reliability, calculates the information entropy weight by the weighted entropy value method, drives the Dijkstra algorithm to generate the optimal path, and enables the path redundancy mechanism to switch the standby path when the bandwidth fluctuates, so that the technical effect of dynamically matching the optimal backup path is achieved, and the problems of single path selection and backup failure or delay are solved, which has the advantages of balancing the backup cost and reliability and guaranteeing the continuity of backup.

[0020] The high-priority object is divided into blocks to generate a unique hash identification, the breakpoint is located by comparing the local and remote metadata hash during transmission interruption, the breakpoint is continued based on the breakpoint after network recovery, and data integrity is verified, technical effects of efficiently processing transmission interruption are achieved, the problems of bandwidth waste and long time consumption caused by full retransmission after interruption are solved, and the advantages of reducing redundant transmission, shortening backup time consumption and guaranteeing data integrity are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 It is a whole block diagram of the application;

[0022] Figure 2 It is a whole flow chart of the application.

[0023] The meanings of various labels in the figure are as follows:

[0024] 1, user behavior portrait modeling unit; 2, multi-source collaborative decision unit; 3, remote configuration interface unit. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0026] The application provides an intelligent television system supporting remote storage management, please refer to Figure 1 As shown in the figure, it comprises a user behavior portrait modeling unit 1 for collecting user viewing behavior data in real time and inputting the same into a pre-trained long short-term memory network to generate user behavior portrait parameters, the user behavior portrait modeling unit 1 captures four types of behavior data of single viewing time length, single-day viewing time period distribution, content type switching frequency and continuous episode skipping interval in real time through a built-in time sequence data collector, inputs the behavior data into a pre-trained long short-term memory network according to time window sequence, the hidden layer of the network extracts time period correlation features through a sliding time window mechanism, and the output layer fuses the time period correlation features and user historical portrait baseline to generate dynamically updated user behavior portrait parameters, wherein the historical portrait baseline is stored in a local encrypted cache area. The multi-source collaborative decision unit 2 performs the following operations according to the user behavior portrait parameters, local storage remaining space, real-time bandwidth of home network and response delay of remote storage device:

[0027] (a) calculating the content access probability value based on the user behavior portrait parameters, in operation (a), the multi-source collaborative decision unit 2 calculates the content access probability value based on the content type preference weight, the single-day viewing time period distribution density and the continuous episode chasing urgency in the user behavior portrait parameters through the behavior weight factor distribution module, wherein the behavior weight factor distribution module generates the content access probability value by performing convolution operation on the content type preference weight and the single-day viewing time period distribution density and superimposing the nonlinear correction amount of the continuous episode chasing urgency, the preset threshold value of the content access probability value adopts a dynamic baseline adjustment mechanism, the mechanism automatically adjusts the threshold interval according to the ratio of the local storage remaining space to the remote storage device response delay, when the content access probability value is higher than the upper limit of the dynamically adjusted threshold value, it is marked as a high-priority backup object, and at the same time, the probability value decay counter is triggered to reduce the frequency of repeated backup of the same content.

[0028] It needs to be further explained that, in order to realize the accurate backup management of the user preferred content of the intelligent television system, firstly, a dynamic user behavior portrait needs to be constructed through the user behavior portrait modeling unit 1 to provide a core basis for subsequent backup decision, and the specific implementation manner is as follows:

[0029] The user behavior portrait modeling unit 1 captures four types of key behavior data in real time through an internal time series data collector, which is a module specially designed to record the changes of user behavior over time, with millisecond-level timing accuracy and data encryption transmission function, which can avoid behavior data loss or leakage. Among them, the single viewing duration is calculated by monitoring the playback state of the smart TV, starting from the user clicking the play button, and ending when clicking stop or switching content. If the pause exceeds 5 minutes, it is determined to be temporarily away, and the duration before the pause is counted into the single viewing duration. For example, a user clicks to play at 19:00, pauses at 19:45 and exceeds 5 minutes, the single viewing duration is recorded as 45 minutes. The single-day viewing period distribution is divided into 12 two-hour periods in 24 hours, and the total viewing duration of the user in each period is counted to form the period distribution data. For example, a user watches 90 minutes from 19-21, watches 30 minutes from 21-23, and watches 0 in other periods, which reflects the user's viewing period preference. The content type switching frequency is calculated by recording the number of times the user switches content types within 1 hour, such as from TV series to variety shows, movies. If it switches 3 times within 1 hour, the frequency is 3 times / hour. The lower the frequency, the higher the user's focus on the current content type. The continuous episode jump interval is calculated for the user's behavior when watching continuous episodes. The time interval from the end of the current episode to the click of the next episode is calculated, such as clicking the second episode 10 seconds after watching the first episode. The interval is recorded as 10 seconds. The shorter the interval, the stronger the user's desire to continue. After capturing the four types of behavior data, the behavior data needs to be input into the pre-trained long short-term memory network in time window sequence. The time window sequence is to divide the continuous 7-day behavior data into 24 1-hour time windows per day, forming 168 (7x24) time window data blocks. Each data block contains four types of behavior data in the time window, such as single viewing duration, period distribution proportion, switching frequency, and jump interval from 8-9 on October 1. The sequence is formed in chronological order to ensure the time sequence correlation of the data. The pre-trained long short-term memory network (LSTM) is a model trained in advance with 100,000 groups of user behavior data, which can effectively handle the long-term dependence relationship in time series data and avoid the forgetting of early behavior information by traditional neural networks. When the sequence is input, the network first normalizes the data, mapping each type of data to the range of 0-1, such as normalizing 45 minutes to 0 when the longest single viewing duration is 180 minutes.25, the hidden layer of the network extracts period correlation features through a sliding time window mechanism. The sliding time window mechanism sets a sliding window containing three consecutive time windows in the hidden layer. The window slides along the sequence with a step of one time window. The correlation degree of three types of behavior data in the window is calculated each time the window slides. For example, the negative correlation between the content type switching frequency and the single viewing time is calculated (the longer the viewing time, the higher the correlation degree), the positive correlation between the interval of continuous episode jumps and the period distribution is calculated (the shorter the interval, the higher the correlation degree), and these correlation degrees are integrated into period correlation features. For example, if the average correlation degree in a sliding window is 0.8, it indicates that the user behavior in this period is highly correlated and the preference is stable. The output layer then integrates the period correlation features and the user historical profile baseline to generate dynamically updated user behavior profile parameters. The user historical profile baseline is a stable parameter of the user's behavior profile in the past 30 days, including the user's long-term content type preference and fixed viewing period basic information, such as the user's preference weight for TV series being 0.7 and the fixed viewing period being 19-21. The baseline is stored in a local encrypted cache area. The local encrypted cache area is an encrypted storage area built-in the smart TV, which uses AES-256 encryption algorithm to protect user profile data from being stolen or tampered by third parties. In the fusion of the output layer, the period correlation features and the historical profile baseline are first weighted and fused. For example, the baseline has a variety of preference weight of 0.3, and the period correlation features have a variety of correlation degree of 0.6. After fusion, the variety of preference weight becomes 0.3 x 0.6 + 0.6 x 0.4 = 0.42. Finally, the user behavior profile parameters including content type preference weight, single-day viewing period distribution density, and continuous episode chasing urgency are generated, realizing dynamic updating of the profile to ensure that the latest user behavior habits are reflected. After the user behavior profile parameters are generated, the multi-source collaborative decision-making unit 2 needs to perform operation (a) based on these parameters to calculate the content access probability value to determine the high-priority backup object. This is because only by accurately judging the content that the user is likely to access again can the waste of backup resources be avoided. The specific implementation is as follows.

[0030] The multi-source collaborative decision unit 2 first extracts three types of core indicators in the user behavior portrait parameters in operation (a): content type preference weight, single-day viewing time period distribution density, and continuous episode chasing urgency. The content type preference weight is assigned according to the content type, with a value range of 0-1 and a total weight of 1. For example, if a user watches TV series for 60% of the time, watches variety shows for 30% of the time, and watches movies for 10% of the time in the past 7 days, the corresponding preference weights are 0.6, 0.3, and 0.1, respectively. The single-day viewing time period distribution density is calculated by dividing the user's viewing time in each 1-hour time window by the total viewing time in a single day. For example, if a user watches TV for a total of 120 minutes in a single day and watches for 48 minutes from 19-20, the time period density is 48 / 120=0.4. The continuous episode chasing urgency is determined based on the interval between the episode update time and the user's historical chasing interval. If the episode is updated on the same day and the user's past chasing interval is no more than 24 hours, the urgency is set to 0.9. If the update is more than 3 days apart and the user has not watched it, the urgency is set to 0.3, with a value range of 0-1. The content access probability value is calculated by the behavior weight factor distribution module, which is a calculation unit specifically designed to fuse multi-dimensional behavior indicators and generate probability values. The core logic is to first fuse the basic preference and time period characteristics, and then use the chasing urgency to correct them, ensuring that the probability value matches the user's actual access intention. Specifically, the module first performs convolution operation on the content type preference weight and the single-day viewing time period distribution density. Convolution operation is used here to extract the associated features of the two indicators. For example, if a user's preference weight for TV series is 0.6 and the viewing density at 19-20 is 0.4, the convolution operation will highlight the feature value of the user's access to TV series at 19-20, making the fused features better reflect the user's access tendency to specific content at specific times, avoiding feature fragmentation caused by simple addition. After convolution operation, the basic associated feature value is obtained, for example, in the above case, the basic associated feature value is 0.24. The non-linear correction amount of the continuous episode chasing urgency is added to generate the content access probability value. The non-linear correction amount is an adjustment value set according to the non-linear change rule of the chasing urgency, which is different from the fixed proportional adjustment of linear correction. It dynamically changes the correction amplitude according to the urgency interval. When the urgency is between 0-0.5, the correction amount is urgency x 0.2 (to avoid excessive influence of low urgency), and when the urgency is between 0.5-1, the correction amount is urgency x 0.5 (high urgency needs to be considered). For example, if the chasing urgency of a continuous episode is 0.9, the non-linear correction amount is 0.9 x 0.5 = 0.45. Adding this correction amount to the basic associated feature value 0.24, the final content access probability value is 0.24 + 0.45 = 0.69, the value is closer to 1, the higher the probability of future access to the content of the user, get content access probability value, need to be marked by a predetermined threshold value of high priority backup object, but the fixed threshold value will be due to the local storage remaining space, remote storage device response delay changes resulting in inadequate adaptability, for example, local storage sufficient, remote response fast, fixed threshold value is too high will miss part of the user may access the content, local storage is tight, remote response slow, the threshold value is too low will result in backup too much invalid content, therefore, need to use dynamic baseline adjustment mechanism, the specific implementation is as follows:

[0031] The preset threshold value of the content access probability value adopts a dynamic baseline adjustment mechanism. The dynamic baseline adjustment mechanism is a strategy for adjusting the threshold interval in real time according to the local storage state of the smart television and the performance of the remote storage device. The core is to quantify the adaptation state of the current storage and the remote device by the ratio of the local storage remaining space and the remote storage device response delay, avoiding the one-sidedness of single dimension judgment. The local storage remaining space is obtained in real time by the storage management module of the smart television. The remote storage device response delay is obtained by sending a test data packet to the remote device and recording the round trip time. The ratio of the two is calculated as follows: local storage remaining space ÷ remote storage device response delay. The larger the ratio, the more sufficient the local storage and the faster the remote response, and the threshold interval can be appropriately reduced to allow more potential high access probability content to enter the backup range. The smaller the ratio, the higher the threshold interval, and the high priority content is strictly screened. The specific rule for automatically adjusting the threshold interval is as follows:

[0032] When the ratio ≥ 0.3 GB / ms (local sufficient, remote fast), the threshold interval is set to 0.5-0.7, i.e. the probability value is higher than 0.7, and is marked as high priority; when the ratio is between 0.1-0.3 GB / ms (state medium), the threshold interval is set to 0.6-0.8; when the ratio < 0.1 GB / ms (local tight, remote slow), the threshold interval is set to 0.7-0.9, for example, at a certain moment, the local storage remaining space is 8 GB, the remote response delay is 100 ms, the ratio is 0.08 GB / ms, and the threshold interval is adjusted to 0.7-0.9; if the content access probability value is 0.85, which is lower than 0.9, it is not marked; if the probability value is 0.92, which is higher than 0.9, it is marked as a high-priority backup object. At the same time, in order to avoid the same content being repeatedly backed up due to the probability value being high for a long time, such as a user having backed up a certain episode, the subsequent probability value is still high, resulting in repeated backup occupying resources, the probability value decay counter needs to be triggered to reduce the frequency of repeatedly backing up the same content. The probability value decay counter is a counting module started for the content marked as high priority. The initial value of the counter is 0, and the counter is incremented by 1 every 24 hours. Each count corresponds to a probability value decay of 0.05. The decay amplitude is calibrated through experiments, which can avoid missing judgment caused by rapid decay, and can effectively reduce repeated backup. For example, when a certain episode is first marked as high priority, the probability value is 0.92, and the counter is 0. After 24 hours, the counter is 1, and the probability value decays to 0.87. If the upper threshold is still 0.9 at this time, the probability value 0.87 is lower than the upper threshold, and is no longer marked as high priority, thereby avoiding repeated backup. If the counter accumulates to 5 times (after 120 hours), the probability value decays to 0.92-5x0.05=0.67, further reducing the risk of repeated backup. At the same time, if the user subsequently accesses the content again, the behavior portrait parameter will be updated, and the probability value will be raised again, ensuring that the content that truly has access demand can still be marked.

[0033] marking the content whose content access probability value is higher than a preset threshold as a high-priority backup object, (b) constructing a multi-dimensional decision matrix including storage cost, network overhead, and device reliability according to the local storage remaining space, the home network real-time bandwidth, and the remote storage device response delay, when the multi-source collaborative decision unit 2 constructs the multi-dimensional decision matrix in operation (b), the storage cost is quantified by the product of the local storage remaining space occupancy rate and the unit capacity cost of the remote storage device, the network overhead is calculated based on the weighted sum of the home network real-time bandwidth fluctuation coefficient and the transmission path hop count, and the device reliability is generated by a reliability score according to the stability index and the historical failure rate of the remote storage device response delay, the three groups of dimension data are standardized to form the multi-dimensional decision matrix, and the multi-dimensional decision matrix is combined with the weighted entropy value method to dynamically generate an optimal backup path, when the multi-dimensional decision matrix is combined with the weighted entropy value method to dynamically generate the optimal backup path, the information entropy weight of the storage cost, the network overhead, and the device reliability is calculated by the entropy value method, the decision matrix is then weighted and aggregated to generate a path decision value according to the information entropy weight, and finally the path decision value is imported into the Dijkstra optimization algorithm to output the optimal backup path connecting the local storage and the target remote storage device, in the path decision value aggregation process of the Dijkstra optimization algorithm, if the fluctuation amplitude of the home network real-time bandwidth exceeds a preset tolerance range, a path redundancy evaluation mechanism is automatically enabled, which specifically includes: generating a backup path decision value based on historical network bandwidth data, comparing the decision value difference coefficient of the primary path and the backup path with the device reliability score, switching to the backup path when the difference coefficient exceeds the reliability score, and feeding back the switching event to the entropy value weight updating module.

[0034] Further explanation is needed, after the marking of the high-priority backup object is completed, the multi-source collaborative decision unit 2 needs to determine the optimal backup path through operation (b), the selection of this path needs to consider the cost and reliability of storage, network, and device in an all-round way to avoid low backup efficiency or resource waste caused by single-dimensional decision, therefore, a multi-dimensional decision matrix needs to be constructed first, the specific implementation is as follows:

[0035] In operation (b), the multi-source collaborative decision unit 2 first quantifies the storage cost, which needs to be combined with the local storage pressure and the remote storage economy, and is calculated by the product of the local storage remaining space occupancy rate and the remote storage device unit capacity cost, wherein the local storage remaining space occupancy rate is an index reflecting the degree of local storage tension, which is calculated by the intelligent television storage management module in real time, that is, (local storage total capacity-local storage remaining space) ÷ local storage total capacity × 100%, for example, the local storage total capacity is 100GB, and the remaining space is 15GB, then the occupancy rate is (100-15) ÷ 100 × 100% = 85%, and the remote storage device unit capacity cost is the fee charged by the remote storage service provider per unit capacity, which is obtained by the remote configuration interface unit 3 from the remote storage device metadata, for example, the unit capacity cost of a certain remote device is 0.5 yuan per GB per month, and the product of the two is the storage cost, such as 85% × 0.5 yuan / GB = 0.425 yuan / GB, the higher the value, the more tense the local storage, the higher the remote storage cost, and the lower the decision priority of the storage dimension, then the network overhead is calculated, which needs to reflect the bandwidth stability and the transmission path complexity, and is determined based on the weighted sum of the real-time bandwidth fluctuation coefficient of the home network and the transmission path hop count.The family network real-time bandwidth fluctuation coefficient is an index for measuring the stability of bandwidth, and the real-time bandwidth data within 1 minute is collected by the network monitoring module of the smart TV, and the coefficient is calculated as (bandwidth maximum - bandwidth minimum) ÷ bandwidth average, for example, the maximum bandwidth within 1 minute is 120 Mbps, the minimum is 80 Mbps, and the average is 100 Mbps, then the fluctuation coefficient is (120-80) ÷ 100=0.4, the larger the coefficient, the more unstable the bandwidth, and the higher the transmission risk, the transmission path hop count refers to the number of network routing nodes from the smart TV to the remote storage device, which is detected by the traceroute command, for example, the TV needs to go through 5 nodes of TV-home router-cell gateway-operator node-remote storage gateway, and the hop count is 5, the more the hop count, the higher the data transmission delay and packet loss probability, in the weighted sum calculation, according to the network influence weight preset (the influence of bandwidth fluctuation on transmission is greater, and the weight is set to 0.6; the path hop count weight is set to 0.4), for example, the fluctuation coefficient is 0.4, and the hop count is 5, then the network overhead is 0.4×0.6+5×0.4=0.24+2=2.24, the higher the value, the higher the transmission cost and risk of network dimension, then the device reliability is generated, the device reliability directly determines the security of backup data, and the reliability score is generated according to the stability index of remote storage device response delay and historical failure rate, the stability index of remote storage device response delay is an index for evaluating the consistency of remote device response, and the response delay data of remote device within 30 minutes (1 time per 1 minute, a total of 30 data points) is collected, and the standard deviation ÷ delay average is calculated, for example, the average delay is 60 ms, and the standard deviation is 6 ms, then the stability index is 6 ÷ 60=0.1, the smaller the index, the more stable the response, and the historical failure rate is the failure frequency of remote device within the past 30 days, which is counted by the remote storage device management platform, that is, the total failure time ÷ total running time × 100%, for example, the past 30 days have 2 failures, the cumulative failure time is 1 hour, and the total running time is 720 hours, then the historical failure rate is 1 ÷ 720 × 100%≈0.14%, the lower the failure rate, the more reliable the device.In the reliability score calculation, the stability index and the historical failure rate are first mapped to the range of 0-1 (index 0.1→0.1, failure rate 0.14%→0.14), and then calculated according to 1-(stability index*0.4+historical failure rate*0.6) (stability has a greater impact on real-time backup, weight 0.4; failure rate has a greater impact on long-term storage, weight 0.6). For example, 0.1*0.4+0.14*0.6=0.04+0.084=0.124, the reliability score is 1-0.124=0.876. The closer the score is to 1, the higher the device reliability. Finally, the three groups of dimension data are standardized. Standardization is to map the storage cost, network overhead, and device reliability data of different dimensions and different ranges to the interval of 0-1, avoiding the overemphasis of a certain dimension on the decision due to the difference in data range. In the specific processing, the storage cost and network overhead are mapped according to (original value-minimum value) / (maximum value-minimum value) (the larger the original value, the larger the standardized value, representing higher cost). The device reliability is mapped according to (original value-minimum value) / (maximum value-minimum value) (the larger the original value, the larger the standardized value, representing higher reliability). For example, the storage cost original value is 0.425 yuan / GB (minimum value 0.1, maximum value 0.8), and the standardized value is (0.425-0.1) / (0.8-0.1)≈0.46; the network overhead original value is 2.24 (minimum value 1, maximum value 5), and the standardized value is (2.24-1) / (5-1)≈0.31; the device reliability original value is 0.876 (minimum value 0.5, maximum value 1), and the standardized value is (0.876-0.5) / (1-0.5)≈0.75. The three groups of standardized data are arranged in the order of storage cost-network overhead-device reliability to form a vector such as [0.46, 0.31, 0.75]. If there are multiple candidate remote storage devices, each device corresponds to a vector. All vector combinations form a multi-dimensional decision matrix, which provides basic data for subsequent optimal path calculation. After the multi-dimensional decision matrix is constructed, a reasonable weight is assigned to each dimension by the weighted entropy value method, and then the optimal backup path is generated by combining the optimization algorithm. Because the importance of each dimension in the matrix to path selection is different, the weight is calculated objectively by the entropy value method to avoid decision bias caused by subjective weight setting. The specific implementation is as follows.

[0036] When the multi-dimensional decision matrix is combined with the weighted entropy method to dynamically generate the optimal backup path, first, the information entropy weight of the three dimensions of storage cost, network overhead and device reliability is calculated by the entropy method. The entropy method is an objective method for determining the weight based on the degree of data dispersion. The higher the degree of data dispersion, that is, the greater the difference between different candidate paths in this dimension, the richer the decision information contained in this dimension, and the higher the corresponding weight. The information entropy weight is the weight value of each dimension calculated by the entropy method, and the total weight is 1. In specific calculation, first, the information entropy of each dimension data in the multi-dimensional decision matrix is calculated, such as the standardized value of the storage cost of all candidate paths. For example, there are 3 candidate path data in a certain dimension, which are 0.46, 0.52 and 0.61. First, calculate the proportion of each data in the total sum of the data in this dimension (such as 0.46 ÷ (0.46 + 0.52 + 0.61) ≈ 0.32), and then calculate the information entropy of this dimension by substituting it into the entropy formula. Then, according to the information entropy, the difference coefficient is calculated (difference coefficient = 1 - information entropy). The larger the difference coefficient, the higher the weight. Finally, the difference coefficients of each dimension are normalized to obtain the information entropy weight. For example, the difference coefficients of the storage cost dimension are 0.35, the network overhead is 0.42, and the device reliability is 0.23. After normalization, the weights are 0.35 ÷ (0.35 + 0.42 + 0.23) = 0.35, 0.42 and 0.23, respectively, which means that the network overhead has the greatest impact on path decision, and the device reliability has the least impact. Then, the decision matrix is aggregated by weighting to generate the path decision value. Weighted aggregation is to multiply the standardized dimension data of each candidate path by the corresponding information entropy weight, and then sum to obtain the comprehensive decision value of the path. The size of the decision value needs to be judged in combination with the dimension attribute. The storage cost and network overhead are cost indicators (the smaller the value, the better), and the device reliability is a benefit indicator (the larger the value, the better). Therefore, when aggregating, it is necessary to ensure that the decision value can comprehensively reflect the optimal state of low cost and high reliability. For example, the standardized data of a certain candidate path is [0.46 (storage cost), 0.31 (network overhead), 0.75 (device reliability)], and the information entropy weight is [0.35, 0.42, 0.23]. Then the path decision value is (1-0.46) × 0.35 + (1-0.31) × 0.42 + 0.75 × 0.23, wherein (1-cost indicator) is the conversion of the cost indicator to a benefit indicator (the larger the value, the better). The calculation result is 0.54 × 0.35 + 0.69 × 0.42 + 0.75 × 0.23 ≈ 0.189 + 0.2898 + 0.1725 ≈ 0.6513, the closer the decision value is to 1, the more optimal the path is, and finally the path decision value is imported into the Dijkstra optimization algorithm to output the optimal backup path connecting the local storage and the target remote storage device, the Dijkstra optimization algorithm is a classical shortest path algorithm, which is used here to find the path with the maximum decision value (optimal in general) in the candidate paths, the algorithm first takes the decision value of all candidate paths as the path weight, takes the local storage as the starting point and each candidate remote storage device as the terminal point to construct a path graph, then calculates the path weight (i.e. the path decision value) from the starting point to each terminal point and records the current optimal path, and when the path weight of all terminal points is calculated, the path with the maximum weight is selected as the optimal backup path, and the algorithm outputs the specific transmission nodes of the path, such as local storage→home router→operator backbone node→remote storage device A, to ensure that the backup data can be transmitted along the optimal path, balancing cost and reliability, during the path decision value aggregation process of the Dijkstra optimization algorithm, the sharp fluctuation of the real-time bandwidth of the home network may cause the selected primary path to suddenly become unreliable, for example, a sharp decrease in bandwidth may cause a sharp rise in transmission delay, and if not adjusted in time, it may cause backup interruption or data loss, therefore, a path redundancy evaluation mechanism needs to be automatically enabled, and the specific implementation is as follows.

[0037] During the path decision value aggregation process of the Dijkstra optimization algorithm, the fluctuation amplitude of the real-time bandwidth of the home network is first monitored in real time, the fluctuation amplitude refers to the difference between the maximum value and the minimum value of the real-time bandwidth within 10 seconds divided by the average bandwidth, and the preset tolerance range is the upper limit of the bandwidth fluctuation allowed by the system, which is experimentally calibrated to be 20%, i.e. when the fluctuation amplitude exceeds 20%, it is determined that the bandwidth is unstable, for example, the average bandwidth is 100 Mbps, the maximum value within 10 seconds is 110 Mbps, and the minimum value is 75 Mbps, the fluctuation amplitude is (110-75) ÷ 100 x 100% = 35%, which exceeds the preset tolerance range of 20%, at this time the algorithm automatically enables the path redundancy evaluation mechanism, which specifically includes:

[0038] The first step is to generate a backup path decision value based on historical network bandwidth data. The historical network bandwidth data is the bandwidth fluctuation coefficient and transmission path hop count of the same period in the past 7 days cached by the smart TV, such as 20:00 now. The bandwidth fluctuation coefficient and transmission path hop count of 20:00-20:10 in the past 7 days are cached. Combined with the storage cost of the backup path in the current multi-dimensional decision matrix, the device reliability data, the path decision value of the backup path is regenerated according to the weighting sum and reliability score calculation method in the foregoing, for example, the backup path corresponds to remote storage device B, the historical bandwidth fluctuation coefficient is 0.3, the hop count is 4, the storage cost standardization value is 0.51, the device reliability is 0.82, the information entropy weight is unchanged, and the backup path decision value is calculated as approximately 0.682. The second step is to compare the decision value difference coefficient of the main path and the backup path with the device reliability score. The decision value difference coefficient is an index for measuring the gap between the main path and the backup path decision value. The calculation method is |main path decision value-backup path decision value|÷main path decision value×100%, for example, the main path decision value is 0.715, the backup path is 0.682, the difference coefficient is |0.715-0.682|÷0.715×100%≈4.6%, and the device reliability score is the reliability score of the remote storage device corresponding to the current main path (such as 0.876). When comparing, if the difference coefficient≤device reliability score (such as 4.6%≤87.6%), it means that the gap between the main path and the backup path is small, and the main device reliability is high, so there is no need to switch. If the difference coefficient>device reliability score, such as the main path decision value is reduced to 0.62 due to bandwidth fluctuation, the backup path is 0.682, the difference coefficient is approximately 10%, which exceeds the device reliability score 8.76%, so it is determined that the main path cannot meet the optimal transmission requirement, and needs to be switched to the backup path. The third step is to switch to the backup path when the difference coefficient exceeds the reliability score, and feed back the switching event to the entropy value method weight updating module. When switching, the multi-source collaborative decision unit 2 immediately sends a path switching instruction to the data transmission module, stops the main path data transmission, and starts the backup path transmission, such as switching from local→device A to local→device B. At the same time, the switching time, main and backup path information, and switching reason log are recorded. The entropy value method weight updating module is a unit responsible for dynamically adjusting the information entropy weight. After receiving the switching event, it will record the path switching caused by the bandwidth fluctuation as a signal for increasing the influence of network overhead dimension, and appropriately increase the weight of the network overhead dimension when calculating the information entropy weight next time, to ensure that the subsequent path decision pays more attention to network stability and reduces frequent switching caused by bandwidth fluctuation, realizing adaptive optimization of path decision.

[0039] (c) When it is detected that the real-time bandwidth of the home network is higher than the preset network bandwidth for a fixed period of time and the remote storage device is idle, triggering the chunk incremental backup of the high-priority backup object. The execution process of the chunk incremental backup in operation (c) is as follows:

[0040] The high-priority backup object is divided into fixed-size data blocks according to the content type, a unique hash identifier is generated for each data block, when the remote storage device is detected to be idle, only the incremental data block not recorded in the local hash identifier library is transmitted, and after the transmission is completed, the new hash identifier is synchronized to the metadata database of the remote storage device, if the transmission is interrupted during the block incremental backup process, the remote storage management module locates the unfinished transmission data block by comparing the hash identifier difference between the local and remote metadata databases and records the breakpoint position, when the network bandwidth recovers to a continuous and stable state, the resume mechanism is triggered based on the breakpoint position, and the integrity of the partial writing of the target data block in the remote storage device is verified in priority before resuming.

[0041] It needs to be further explained that after the optimal backup path connecting the local storage and the remote storage device is determined by the Dijkstra optimization algorithm, the multi-source collaborative decision unit 2 starts the block incremental backup process of operation (c). Compared with the traditional full backup which needs to repeatedly transmit the already backed up data, the block incremental backup only transmits the newly added data block which has not been backed up, which can greatly reduce the network bandwidth occupation and the backup time, while ensuring the integrity of data transmission. The specific implementation manner is as follows:

[0042] The execution process of the block incremental backup in operation (c) is as follows:

[0043] First, the high-priority backup object is divided into fixed-size data blocks according to content type, which here covers specific data categories of the high-priority backup object, such as TV series, variety video content, and user-collected TXT, PDF document content. The structures and compression characteristics of different types of data differ greatly. Video data has large byte size and strong continuity, while document data has small byte size and fragmentation. Dividing by content type can adapt to different data characteristics to reduce transmission errors. The fixed size is dynamically set according to the content type. For video content, each data block is fixed at 100MB to balance the stability of single transmission and request frequency. For document content, each data block is fixed at 10MB to avoid frequent request overhead caused by small block transmission. The segmentation operation is performed by the file segmentation module built-in the smart TV. This module reads the binary data of the high-priority backup object in segments according to the preset byte number. For example, an 800MB TV series file is divided into 8 data blocks of 100MB each. A unique block sequence number and the file identification to which the block belongs are added at the end of each block to ensure accurate correspondence with the original file during subsequent data recombination and avoid block sequence disorder. Then, a unique hash identification is generated for each data block. The hash identification is a 64-bit string calculated by the SHA-256 hash algorithm on the complete binary content of the data block. Its core characteristics are that different data blocks correspond to different hash identifications, and the same data block corresponds to a unique hash identification, which can accurately judge whether the data block is repeated or tampered with. During the generation process, the file segmentation module outputs a data block, and then calls the local hash calculation unit to operate on the binary content of the block. At the same time, the hash identification is stored in association with the block sequence number, the file name to which the block belongs, and the generation time, forming a block-hash mapping relationship to provide a basis for subsequent incremental judgment. Then, when the remote storage device is detected to be idle, the multi-source collaborative decision unit 2 obtains the real-time state data of the device by sending periodic heartbeat packets (every 5 seconds) to the remote storage device. If the resource usage rate returned by the heartbeat packet for 3 consecutive times is lower than the threshold value, it is determined that the device is in an idle state. If the remote device is in a busy state, such as a CPU usage rate of 50%, the device is re-detected every 10 seconds until the device is idle to avoid packet loss or delay aggravation caused by data transmission when the device load is too high. Subsequently, only the incremental data blocks not recorded in the local hash identification library are transmitted. The local hash identification library is a SQLite database stored locally in the smart TV, which specifically records the hash identifications of all data blocks that have been successfully backed up to the remote storage device. The database table structure includes the backup time block sequence number field of the hash identification belonging to the file to ensure that the backed-up block information can be quickly queried. Before transmission, the multi-source collaborative decision unit 2 compares all hash identifications of the data blocks to be backed up this time with the records in the local hash identification library one by one. If the hash identification of a data block has no matching record in the library, i.e., the block has not been backed up,The data blocks to be backed up this time are marked as incremental data blocks, for example, 8 data blocks are to be backed up this time, of which 6 have been identified by hash in the local library, and 2 have not, so only the 2 incremental data blocks are transmitted, effectively reducing redundant transmission and saving network bandwidth. Finally, the new hash identifier is synchronized to the meta database of the remote storage device after the transmission is completed. The meta database of the remote storage device is a structured MySQL database used by the remote storage end to manage all backup data blocks, recording the hash identifier, storage path, associated high-priority backup object name, and reception time information of the data block, ensuring that the data block can be quickly located during subsequent data reading and verification. During synchronization, the smart TV sends the hash identifier, the file to which the data block belongs, and the block sequence number information of the newly transmitted data block to the remote meta database through a TLS 1.3 encrypted HTTP POST request. After receiving the information, the remote meta database queries the storage path of the corresponding data block to confirm that the file is complete, and then writes the new hash identifier and associated information into the database table to complete the synchronization, ensuring that the hash records on the local and remote sides are consistent, providing accurate reference for the next incremental backup. During the incremental backup process, fluctuations in the home network may cause transmission interruption. If all data blocks are directly retransmitted, it will cause resource waste, so a targeted resuming mechanism needs to be executed through the remote storage management module to ensure that the data can be completely backed up. The specific implementation is as follows:

[0044] If a transmission interruption occurs during the chunk incremental backup process, the smart TV first determines the interruption state. If the smart TV does not receive a successful reception confirmation packet or receives a failed reception packet (such as a data verification error) within 30 seconds after sending a data block to the remote storage device, the current data block transmission is immediately stopped, and the key information at the time of interruption is recorded, including the hash identifier of the data block being transmitted and the number of bytes that have been transmitted. Subsequently, the remote storage management module compares the hash identifiers in the local and remote meta databases. The remote storage management module is a functional unit built into the smart TV and is responsible for interacting with the remote storage device, including data transmission control, device state monitoring, and data verification. When comparing, the module first sends a hash list acquisition request to the remote storage device. The remote meta database returns a list of all data block hash identifiers that have been stored and are related to the current backup object, including a complete reception and an incomplete reception state marker. Then, the module compares the hash list returned by the remote device with the hash list of the local data to be backed up one by one, and filters out two types of difference data: one is the hash identifier that is recorded in the local device but not in the remote meta database, corresponding to the incremental block that has not started transmission, and the other is the hash identifier that is recorded in the local device and the remote meta database but marked as incomplete reception, corresponding to the incomplete block that has experienced a transmission interruption. Then, the incomplete transmission data block is located and the breakpoint position is recorded. When locating the incomplete block, the remote storage management module finds the corresponding incomplete data block based on the hash identifier marked as incomplete reception in the remote meta database. The breakpoint position refers to the number of bytes that have been successfully transmitted to the remote storage device for the incomplete data block. The remote meta database records the number of received bytes for each incomplete reception data block in real time. The remote storage management module obtains this number of bytes through a breakpoint query request and marks the breakpoint in the corresponding position of the local data to be backed up, such as marking the binary file of the local data block from the 65MB+1 byte as to-be-resumed starting from the 65MB+1 byte. This ensures that the subsequent resumption is only for the unreceived part. Then, when the network bandwidth recovers to a continuous and stable state, the network bandwidth recovery to a continuous and stable state is determined by the network monitoring module of the smart TV. The module collects real-time bandwidth data every 5 seconds, calculates the fluctuation amplitude of the maximum and minimum values of the bandwidth within 10 seconds, and determines that the network has recovered to a stable state if the fluctuation amplitude is ≤10% for 6 consecutive times (30 seconds). If the fluctuation amplitude still exceeds 10%, the monitoring continues to avoid starting resumption when the network is unstable, which may cause another interruption. Subsequently, the resumption mechanism is triggered based on the breakpoint position. When the resumption mechanism is started, the remote storage management module reads the binary data from the breakpoint (such as 65MB+1 byte) of the local data to be backed up based on the breakpoint position, generates a resumption data segment (such as a resumption data segment of 35MB for a 100MB data block from 65MB+1 byte to 100MB), and sends a resumption request to the remote storage device. The request includes the hash identifier of the incomplete data block, the breakpoint position, and the byte range of the resumption data segment. After receiving the request, the remote storage device prepares to receive the resumption data at the breakpoint of the corresponding data block.Ensure the continuity of the data block, avoid repeated transmission of the received part, and finally verify the partial write integrity of the target data block in the remote storage device before resuming transmission, the partial write integrity refers to verifying whether the received part of the data block in the remote storage device is consistent with the corresponding part of the local data to be backed up, to prevent the remote received data from being damaged due to transmission interruption, when verifying, the remote storage management module first sends a partial data hash request to the remote storage device, obtains the binary content of the remote received data block (such as 65MB part), and calculates the SHA-256 hash value of the content, at the same time, the local module reads the binary content of the corresponding byte range (0 to 65MB) in the data to be backed up, and also calculates the SHA-256 hash value, if the hash values calculated at the two places are consistent, it is determined that the partial write is complete, and the resuming transmission can be started, if the hash values are inconsistent, the remote storage management module sends a data cleaning request to the remote device to delete the received part of the data block in the remote device, and then transmits from the start position of the data block again, to ensure the accuracy of the resuming transmission data, avoid the influence of splicing damaged data blocks on subsequent use, finally connected with the remote configuration interface unit 3, the remote configuration interface unit 3 is used for providing an encrypted communication channel to the mobile terminal, receiving the backup strategy weight parameter adjusted manually by the user and synchronizing to the multi-source collaborative decision unit 2, and improving the effectiveness of remote storage management.

[0045] The application generates behavior portrait parameters through collecting user viewing behavior data, calculates content access probability and marks high-priority backup objects according to the behavior portrait parameters, constructs a multi-dimensional decision matrix combining storage and bandwidth, generates an optimal backup path through weighted entropy value method and Dijkstra algorithm, performs block incremental backup, performs breakpoint resuming transmission through hash comparison after transmission interruption, and receives mobile terminal parameter update decision logic.

[0046] The second purpose of the application is to provide a method for realizing the intelligent television system supporting remote storage management, comprising the following steps:

[0047] S1, capture single viewing duration, single day viewing time period distribution, content type switching frequency and continuous episode skipping interval through a time sequence data collector, input the behavior data according to the time window sequence into a pre-trained long short-term memory network, extract time period association features from the network hidden layer and fuse historical portrait baselines, and dynamically generate user behavior portrait parameters;

[0048] S2, calculate the content access probability value based on the user behavior portrait parameters, mark the high-priority backup objects in combination with the dynamic baseline adjustment mechanism, and construct a multi-dimensional decision matrix according to the local storage remaining space, the real-time bandwidth of the home network and the response delay of the remote storage device, calculate the information entropy weight by the weighted entropy method and drive the Dijkstra optimization algorithm to generate the optimal backup path;

[0049] S3, split the high-priority backup objects into fixed-size data blocks and generate unique hash identifiers, when it is detected that the real-time bandwidth of the home network is continuously stable and the remote storage device is idle, transmit the incremental data blocks in the local hash identifier library, if the transmission is interrupted, locate the breakpoint by the hash identifier difference and trigger the resume transmission mechanism, and verify the partial write integrity of the remote data block before resuming the transmission;

[0050] S4, receive the backup strategy weight parameters sent by the mobile terminal through the encrypted communication channel, and update the entropy method weight distribution and dynamic baseline adjustment logic in the multi-source collaborative decision unit in real time.

[0051] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent television system supporting remote storage management, characterized by, The user behavior portrait modeling unit (1) is used for real-time collection of user viewing behavior data and input of the user viewing behavior data into a pre-trained long short-term memory network to generate user behavior portrait parameters. (a) calculating a content access probability value based on the user behavior portrait parameters, and marking content with a content access probability value higher than a preset threshold as a high-priority backup object; (b) constructing a multi-dimensional decision matrix including storage cost, network overhead, and device reliability based on the local storage remaining space, the real-time bandwidth of the home network, and the response delay of the remote storage device, and dynamically generating an optimal backup path by combining the multi-dimensional decision matrix with a weighted entropy method; In operation (b), the storage cost is quantified by the product of the local storage remaining space occupancy rate and the unit capacity cost of the remote storage device, the network overhead is calculated based on the weighted sum of the real-time bandwidth fluctuation coefficient of the home network and the transmission path hop count, and the device reliability is generated based on the stability index and the historical failure rate of the response delay of the remote storage device to generate a reliability score, and the three groups of dimension data are standardized to form the multi-dimensional decision matrix; (c) triggering chunk incremental backup of the high-priority backup object when it is detected that the real-time bandwidth of the home network is higher than the preset network bandwidth for a fixed period of time and the remote storage device is idle; The remote configuration interface unit (3) is used for providing an encrypted communication channel to a mobile terminal, receiving a backup strategy weight parameter adjusted manually by a user, and synchronizing the backup strategy weight parameter to the multi-source collaborative decision unit (2).

2. The smart TV system supporting remote storage management according to claim 1, characterized by: The user behavior portrait modeling unit (1) captures four types of behavior data, including single viewing duration, single-day viewing time period distribution, content type switching frequency, and continuous episode skipping interval, in real time through an embedded time series data collector, inputs the behavior data into a pre-trained long short-term memory network according to a time window sequence, extracts period correlation features through a sliding time window mechanism in a hidden layer of the network, and fuses the period correlation features and a historical portrait baseline to generate dynamically updated user behavior portrait parameters in an output layer, wherein the historical portrait baseline is stored in a local encrypted cache area.

3. The smart TV system supporting remote storage management according to claim 1, wherein, In operation (a), the multi-source collaborative decision unit (2) calculates the content access probability value based on the content type preference weight, the single-day viewing time period distribution density, and the continuous episode chasing urgency in the user behavior portrait parameters through a behavior weight factor distribution module, wherein the behavior weight factor distribution module performs convolution operation on the content type preference weight and the single-day viewing time period distribution density, superimposes a nonlinear correction amount of the continuous episode chasing urgency to generate the content access probability value.

4. The smart TV system supporting remote storage management according to claim 3, characterized by: The preset threshold of the content access probability value adopts a dynamic baseline adjustment mechanism, which automatically adjusts the threshold interval according to the ratio of the local storage remaining space to the response delay of the remote storage device. When the content access probability value is higher than the upper limit of the dynamically adjusted threshold, it is marked as a high-priority backup object, and a probability value decay counter is triggered to reduce the frequency of repeatedly backing up the same content.

5. The smart TV system supporting remote storage management of claim 1, wherein: The multi-dimensional decision matrix is combined with the weighted entropy value method to dynamically generate an optimal backup path, wherein the information entropy weight of three dimensions of storage cost, network overhead and device reliability is calculated by the entropy value method, the decision matrix is aggregated by weighting according to the information entropy weight to generate a path decision value, and finally the path decision value is input into the Dijkstra optimization algorithm to output the optimal backup path connecting the local storage and the target remote storage device.

6. The smart TV system supporting remote storage management according to claim 5, characterized by: In the path decision value aggregation process of the Dijkstra optimization algorithm, if it is detected that the fluctuation amplitude of the real-time bandwidth of the home network exceeds the preset tolerance range, a path redundancy evaluation mechanism is automatically enabled, which specifically includes: generating a backup path decision value based on historical network bandwidth data, comparing the decision value difference coefficient of the primary path and the backup path with the device reliability score, switching to the backup path when the difference coefficient exceeds the reliability score, and feeding back the switching event to the entropy value method weight updating module.

7. The smart TV system supporting remote storage management of claim 1, wherein: The execution process of the block incremental backup in operation (c) is as follows: The high-priority backup objects are divided into fixed-size data blocks according to content types, a unique hash identifier is generated for each data block, when the remote storage device is detected to be idle, only the incremental data blocks not recorded in the local hash identifier library are transmitted, and after the transmission is completed, the new hash identifier is synchronized to the metadata database of the remote storage device.

8. The smart TV system supporting remote storage management according to claim 5, characterized by: If a transmission interruption occurs in the block incremental backup process, the remote storage management module locates the data blocks not completed for transmission and records the breakpoint position by comparing the hash identifier difference between the local and remote metadata databases, when the network bandwidth recovers to a continuous and stable state, the retransmission mechanism is triggered based on the breakpoint position, and the partial write integrity of the target data block in the remote storage device is verified before retransmission.

9. A control method for implementing an intelligent television system supporting remote storage management comprising the system of any of claims 1-8, characterized by: The method comprises the following steps: S1, capturing single viewing duration, single-day viewing time period distribution, content type switching frequency and continuous episode skipping interval through a time series data collector, inputting the behavior data into a pre-trained long short-term memory network according to a time window sequence, extracting time period association features from the network hidden layer and fusing historical portrait baselines to dynamically generate user behavior portrait parameters; S2, calculating content access probability values based on the user behavior portrait parameters, marking high-priority backup objects by combining a dynamic baseline adjustment mechanism, constructing a multi-dimensional decision matrix according to the local storage remaining space, the real-time bandwidth of the home network and the response delay of the remote storage device, calculating information entropy weights by the weighted entropy value method and driving the Dijkstra optimization algorithm to generate an optimal backup path; S3, dividing the high-priority backup objects into fixed-size data blocks and generating unique hash identifiers, when it is detected that the real-time bandwidth of the home network is continuous and stable and the remote storage device is idle, transmitting the incremental data blocks in the local hash identifier library, if a transmission interruption occurs, locating the breakpoint by the hash identifier difference and triggering the retransmission mechanism, and verifying the partial write integrity of the remote data block before retransmission; S4, receiving the backup strategy weight parameters sent by the mobile terminal through an encrypted communication channel, and updating the entropy value method weight distribution and dynamic baseline adjustment logic in the multi-source collaborative decision unit in real time.

Images