Memory optimization system for intelligent health care accompanying robot

Abstract

The invention relates to the technical field of health care accompanying robots, and particularly discloses a memory optimization system for an intelligent health care accompanying robot, which comprises the following steps: acquiring state data of a first application program and a time sequence of memory resource usage amount; embedding coding of state data of the first application program and time sequence association of the memory resource usage amount of the first application program are carried out by utilizing a data analysis and coding mode based on deep learning; therefore, whether memory cleaning operation is executed on the first application program or not is intelligently judged according to semantic alignment fusion representation between the application program state embedded coding feature and the application program memory resource usage amount time sequence implicit association feature. In this way, the condition of the application program memory can be accurately recognized, and misjudgment possibly caused by a traditional static rule is effectively avoided. Moreover, the optimization strategy can be automatically adjusted according to different equipment and user behavior modes, and the method does not need to depend on a fixed rule set, so as to better adapt to diversified application scenes.

Description

Technical Field

[0001] The present application relates to the technical field of health care companion robots, and more specifically, to a memory optimization system for an intelligent health care companion robot. Background Art

[0002] As the functions of smart health companion robots become increasingly complex, they need to run multiple applications simultaneously to provide various services, such as health monitoring, entertainment interaction, social connection, etc. Such devices usually have limited computing resources and memory capacity. In particular, when the user no longer interacts with an application, the application should automatically go to the background and release unnecessary resources. However, the traditional memory management mechanism triggers resource recycling when memory is insufficient, resulting in resource waste and performance degradation.

[0003] In this regard, patent CN104375828B proposes a memory optimization method and device, which identifies the application that needs to release free memory by receiving a code containing a program identifier and a memory release function. This code is used to detect whether the application is switched from the foreground to the background. Once the switch is confirmed, the memory release function is automatically called to reclaim the free memory occupied by the application when it is running in the foreground but no longer needed in the background, thereby improving the terminal processing capacity and memory usage efficiency.

[0004] After detecting that a specific application has switched from the foreground to the background, the corresponding action will be triggered, which will notify and start the memory optimization logic to clean up the memory. However, the traditional method is that the memory optimization logic statically evaluates the received information based on a pre-configured set of rules, which may not be able to adapt to all application scenarios or changes in device status. For example, some rules may be suitable for situations where memory resources are tight, but may result in unnecessary memory release in an environment with sufficient memory. In addition, static evaluations are often based on periodic checks or specific event triggers, which may cause the system to have performance problems when it detects that memory optimization is needed, thereby affecting the user experience.

[0005] Therefore, an optimized memory optimization solution for intelligent health care companion robots is desired. Summary of the invention

[0006] This application provides a memory optimization system for an intelligent health care companion robot, which can accurately identify the application memory situation through intelligent analysis, effectively avoiding the misjudgment that may be caused by traditional static rules. In addition, it can automatically adjust its optimization strategy according to different devices and user behavior patterns, without relying on a fixed set of rules, to better adapt to diverse application scenarios, thereby providing a personalized memory management solution.

[0007] According to one aspect of the present application, a memory optimization system for an intelligent health care companion robot is provided, including: a code receiving and loading module, for receiving a memory optimization code carrying a program identifier and a memory release function, and loading the memory optimization code into the intelligent health care companion robot; a program monitoring module, for using the memory optimization code to monitor each application in the intelligent health care companion robot in real time, and detect whether each application is switched from the foreground to the background; a cleaning operation judgment module, for responding to a first application being switched from the foreground to the background, using memory optimization logic to generate a memory optimization judgment result to judge whether to perform a memory cleaning operation on the first application; a release module, for responding to the memory optimization judgment result that a memory cleaning operation is performed on the first application, calling the memory release function to release the memory of the first application Resources; wherein, the cleaning operation judgment module includes: a data acquisition unit, used to obtain the state data of the first application and obtain the time series of the memory resource usage of the first application; a data encoding unit, used to embed and time-series encode the state data of the first application and the time series of the memory resource usage of the first application respectively to obtain the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage; a data semantic alignment fusion unit, used to perform semantic information-guided feature alignment and fusion on the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage to obtain the fine-grained semantic alignment fusion features of the application state-resource usage, and generate the memory optimization judgment result based on the fine-grained semantic alignment fusion features of the application state-resource usage.

[0008] The present application provides a memory optimization system for an intelligent health care companion robot, which obtains the time series of the state data and memory resource usage of a first application, and uses data analysis and encoding methods based on deep learning to perform embedded coding of the state data of the first application and temporal association of the memory resource usage of the first application, thereby intelligently judging whether to perform a memory cleanup operation on the first application based on the semantic alignment fusion representation between the embedded coding features of the application state and the implicit association features of the temporal association of the application memory resource usage. Through intelligent analysis, the present application can accurately identify the status of the application memory, effectively avoiding the misjudgment that may be caused by traditional static rules. In addition, it can automatically adjust its optimization strategy according to different devices and user behavior patterns, without relying on a fixed set of rules, to better adapt to diverse application scenarios, thereby providing a personalized memory management solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0010] Figure 1 This is a schematic block diagram of a memory optimization system for an intelligent healthcare companion robot according to an embodiment of the present application.

[0011] Figure 2 This is a schematic block diagram of a cleaning operation judgment module in a memory optimization system for an intelligent health care companion robot according to an embodiment of the present application.

[0012] Figure 3 This is a schematic diagram of the data flow of the cleaning operation judgment module in the memory optimization system for the intelligent health care companion robot in an embodiment of the present application.

[0013] Figure 4 This is a schematic block diagram of a data encoding unit in a memory optimization system for an intelligent health care companion robot according to an embodiment of the present application.

[0014] Figure 5 This is a schematic block diagram of a data semantic alignment and fusion unit in a memory optimization system for an intelligent health care companion robot according to an embodiment of the present application.

[0015] Figure 6 This is a schematic block diagram of an application state-resource usage fusion subunit in a memory optimization system for an intelligent healthcare companion robot according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0017] Based on this, this application proposes a memory optimization system for an intelligent health care companion robot, such as Figure 1As shown, the memory optimization system for the intelligent health care companion robot includes: a code receiving and loading module 10, which is used to receive the memory optimization code carrying the program identifier and the memory release function, and load the memory optimization code to the intelligent health care companion robot; a program monitoring module 20, which is used to use the memory optimization code to monitor the various applications in the intelligent health care companion robot in real time, and detect whether the various applications are switched from the foreground to the background; a cleaning operation judgment module 30, which is used to generate a memory optimization judgment result using the memory optimization logic in response to the first application switching from the foreground to the background to judge whether to perform a memory cleaning operation on the first application; a release module 40, which is used to call the memory release function to release the memory resources of the first application in response to the memory optimization judgment result that the memory cleaning operation is performed on the first application.

[0018] Exemplarily, in the code receiving and loading module 10, a memory optimization code carrying a program identifier and a memory release function is received, and the memory optimization code is loaded into the intelligent health care companion robot. It should be understood that the code receiving and loading module is not only responsible for receiving the memory optimization code carrying a specific program identifier and a memory release function, but also ensures that these codes can be correctly loaded into the robot's operating system environment, so as to achieve the timely release of the application's free memory and improve the overall performance of the system. Considering that the intelligent health care companion robot usually needs to run multiple applications at the same time to provide services such as health monitoring, entertainment interaction and social connection, and such devices often have limited computing resources and memory capacity, therefore, when the application switches from the foreground to the background, it becomes particularly important to release the memory resources that are no longer used in time. This can not only improve the processing power of the terminal, but also avoid the problem of slow system response or crash caused by insufficient memory. The code receiving and loading module receives and loads the memory optimization code containing the program identifier and the memory release function, so that the system can respond quickly after detecting the change of the application state and perform the necessary memory cleanup operation. Specifically, the working principle of the code receiving and loading module is based on the parsing and integration of the memory optimization code. It first receives this code from an external source (such as a server update or local storage), which embeds a program identifier for identifying the target application and a memory release function responsible for the actual memory release operation. During the reception process, the module will perform a series of verification measures, such as checksum checks and digital signature authentication, to ensure the authenticity and integrity of the code and prevent potential security threats. Once confirmed to be correct, the code receiving and loading module begins to load this optimized code into the memory environment of the smart health care companion robot.

[0019] In one embodiment, in order to ensure the security and compatibility of the code loading process, the code receiving and loading module will adopt dynamic loading technology, allowing real-time updating or adding of new memory optimization logic without affecting the currently running application. In addition, the module needs to work closely with the operating system to ensure that the memory optimization code can be smoothly connected to the system call interface and can trigger the memory release function at the appropriate time. This design not only improves the flexibility of the system, but also enhances its ability to cope with different application scenarios. In this process, the code receiving and loading module is not just a simple code transmission tool, but an important link in the entire memory optimization process. It provides a basis for the subsequent state detection module, enabling the latter to accurately monitor the foreground / background switching of the application; it also provides necessary support for the cleaning operation judgment module and the release module, ensuring that they can intelligently decide when and how to perform memory cleaning tasks according to preset conditions. In the end, this series of closely related operations work together to realize the effective management and optimization of memory resources by the intelligent health care companion robot, and improve the user experience and service quality.

[0020] Exemplarily, in the program monitoring module 20, the memory optimization code is used to monitor each application in the intelligent health care companion robot in real time, and detect whether each application is switched from the foreground to the background. It should be understood that using the memory optimization code to monitor each application in the intelligent health care companion robot in real time and detect whether these applications are switched from the foreground to the background, not only helps to release the memory that is no longer used in time, thereby avoiding the waste of memory resources, but also can significantly improve the response speed and stability of the system. When applications switch from the foreground state with frequent user interaction to the relatively static background state, they often no longer need to occupy a large amount of memory resources. However, in the absence of an appropriate management mechanism, these applications may continue to hold the memory originally used for foreground operations, resulting in other more important tasks unable to obtain sufficient resources, thereby affecting the overall performance. Therefore, the problem can be effectively solved by monitoring the changes in the application state in real time through the memory optimization code and taking immediate action when it is detected that the application switches from the foreground to the background. Specifically, the memory optimization code is designed to be able to run continuously in the operating system environment of the intelligent health care companion robot. It uses built-in state detection logic to track the life cycle events of each application, especially focusing on those transition points from the foreground to the background. To achieve this, the memory optimization code works closely with the operating system's process management module and subscribes to relevant event notifications, such as window focus changes, activity status updates, etc. Whenever an application changes state, the memory optimization code can quickly capture this information. At the same time, the memory optimization code also checks the specific identification of the application (i.e., the program identification) to determine which applications are the targets of memory cleanup. This is because not all applications running in the background need to release memory immediately; some service-type applications may still need to maintain a certain level of activity to provide background services. Therefore, the subsequent memory release process will only be triggered when it is confirmed that an application has indeed been completely switched from the foreground to the background and meets predefined conditions (for example, non-critical or low-priority applications).

[0021] Exemplarily, in the cleaning operation judgment module 30, in response to the first application switching from the foreground to the background, the memory optimization logic is used to generate a memory optimization judgment result to determine whether to perform a memory cleaning operation on the first application. It should be understood that the cleaning operation judgment module is responsible for generating a judgment result through the memory optimization logic in response to the event that the first application switches from the foreground to the background to determine whether a memory cleaning operation needs to be performed on the application. This process not only involves technical implementation details, but also considers resource management and user experience balance in actual application scenarios. When the first application switches from the foreground to the background, it means that the user no longer interacts directly with the application. At this time, the application usually no longer needs to occupy a large amount of memory resources required for the foreground operation. However, not all applications running from the background need to release their memory immediately; some service-type or background task processing applications may still need to maintain a certain degree of activity to continue to provide necessary functions. Therefore, the cleaning operation judgment module must be able to intelligently distinguish these situations to ensure that memory cleaning is triggered only at the appropriate time to avoid unnecessary performance loss or affect user experience.

[0022] Exemplarily, in the release module 40, in response to the memory optimization judgment result that the memory cleanup operation is performed on the first application, the memory release function is called to release the memory resources of the first application. It should be understood that the release module calls the memory release function to release the memory resources occupied by the application when receiving the memory optimization judgment result that the memory cleanup operation is performed on the first application. This process is not only to ensure that the system can efficiently utilize limited memory resources, but also to avoid unnecessary performance overhead and improve the response speed and stability of the intelligent health care companion robot in a multi-tasking environment. When the cleaning operation judgment module generates a memory optimization judgment result and confirms that a memory cleanup operation needs to be performed on the first application, the release module intervenes immediately. At this point, the main responsibility of the release module is to accurately call the memory release function to ensure that the specified application can release the memory resources that are no longer used in time. This immediate response mechanism is crucial to maintaining the overall performance of the system, especially on intelligent health care companion robots with limited resources, which helps prevent system freezes or crashes caused by insufficient memory.

[0023] In one embodiment, the release module first receives clear instructions from the cleaning operation judgment module, indicating which specific applications (identified by the program) and the corresponding memory areas should be cleaned. This step requires the release module to have high-precision task scheduling capabilities to ensure that no misoperation or omission of any key information will occur. At the same time, considering security and compatibility, the release module will also verify whether the received instructions are legal and complete, such as checking digital signatures or verification codes to prevent the injection of malicious code or the execution of erroneous commands. Once confirmed to be correct, the release module will start the calling process of the memory release function. If the memory release function is a public function, it can be called directly; for those private functions that are in a prohibited editing state, it is necessary to first obtain the absolute address of the authorized public function called by the application, and calculate the actual address of the memory release function, and then indirectly call the memory release function through this absolute address. This method ensures that even when facing private functions that cannot be directly edited, the memory cleaning operation can be performed safely and effectively. In the actual calling process, the release module will send the absolute address of the memory release function to the target application, so that the application can call the memory release function under the indication of the address. In addition, the release module needs to monitor the memory release process to ensure that all specified memory resources are properly reclaimed without any unreleased fragments remaining.

[0024] Correspondingly, in the cleanup operation judgment module, the technical concept of the present application is to obtain the time series of the state data and memory resource usage of the first application, and use data analysis and encoding methods based on deep learning to perform embedded coding of the state data of the first application and temporal association of the memory resource usage of the first application, so as to intelligently judge whether to perform a memory cleanup operation on the first application based on the semantic alignment fusion representation between the embedded coding features of the application state and the implicit association features of the temporal series of the application memory resource usage. Through intelligent analysis, the present application can accurately identify the status of the application memory, effectively avoiding the misjudgment that may be caused by traditional static rules. In addition, it can automatically adjust its optimization strategy according to different devices and user behavior patterns, without relying on a fixed set of rules, to better adapt to diverse application scenarios, thereby providing a personalized memory management solution.

[0025] In one embodiment, Figure 2 and Figure 3As shown, the cleaning operation judgment module 30 includes: a data acquisition unit 31, which is used to obtain the state data of the first application and obtain the time series of the memory resource usage of the first application; a data encoding unit 32, which is used to embed and time-series encode the state data of the first application and the time series of the memory resource usage of the first application respectively to obtain the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage; a data semantic alignment fusion unit 33, which is used to perform semantic information-guided feature alignment and fusion on the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage to obtain the application state-resource usage fine-grained semantic alignment fusion features, and generate the memory optimization judgment result based on the application state-resource usage fine-grained semantic alignment fusion features.

[0026] Exemplarily, in the data acquisition unit 31, the state data of the first application is acquired, and the time series of the memory resource usage of the first application is acquired. In particular, the state data includes the application type and the user interaction behavior record. It should be understood that acquiring the time series of the state data and the memory resource usage can provide the system with a dynamic and multi-dimensional application operation portrait. Specifically, the application type information helps the system identify the functional attributes and service levels of the application. For example, some background service applications need to maintain a certain degree of activity even when switched to the background, while other types of entertainment or tool applications may no longer require a large amount of memory after losing user focus. The user interaction behavior record reflects the details such as the frequency, method and duration of interaction between the user and the application, which helps to predict the usage trend of the application in the future. By combining these static and dynamic information, the actual needs of each application can be more accurately evaluated, thereby avoiding unnecessary memory cleanup operations. In addition, the time series of memory resource usage provides a detailed view of the changes in application memory consumption over time. This is crucial to understanding the memory usage characteristics of applications at different stages (such as startup, running, pause, etc.). By analyzing this time series, the regularity and abnormality of application memory usage can be captured, and a more reasonable memory management strategy can be formulated. For example, if an app's memory usage drops significantly when it runs in the background, it may be a good candidate for cleanup; conversely, if an app maintains a high memory usage even in the background, it indicates that the app may be performing important background tasks and its memory should not be cleaned up easily.

[0027] In one embodiment, for state data, application type information and user interaction behavior records can be collected by monitoring APIs or event monitoring mechanisms. For example, the operating system usually records the timestamp of each application startup, shutdown, entry into the foreground or background, as well as the user's click, slide and other operation logs. This information can be directly used to build a historical archive of application usage, providing basic data support for subsequent intelligent decision-making. As for the time series of memory resource usage, the present application can use performance counters or the reporting function of the memory management subsystem to regularly sample and record the memory usage of each application. This type of data usually exists in the form of periodic snapshots, and each snapshot contains the memory allocation details of the application at a specific moment, such as stack size, cache occupancy, file mapping, etc. Over time, these snapshots are combined into a complete timeline, revealing the changing trend of application memory usage.

[0028] In one embodiment, Figure 4 As shown, the data encoding unit 32 includes: an application state embedding encoding subunit 321, which is used to use a program state embedding matrix to embed the state data of the first application to obtain an application state embedding encoding vector as the application state embedding encoding feature; an application memory resource timing encoding subunit 322, which is used to pass the time series of the memory resource usage of the first application through a sequence encoder based on the RNN-LSTM hybrid model to obtain an application memory resource usage timing implicit correlation feature vector as the application memory resource usage timing implicit correlation feature.

[0029] Exemplarily, in the application state embedding coding subunit 321, the state data of the first application is embedded and encoded using a program state embedding matrix to obtain an application state embedding coding vector as the application state embedding coding feature. It should be understood that, considering that the state data of the first application often contains a variety of different types of information, such as application type, user interaction behavior records, etc., these original data dimensions are relatively high and complex, and also contain the most critical and representative feature information of the state data. Therefore, in order to map high-dimensional state data to a low-dimensional space, realize data dimensionality reduction, and extract the most representative features that are important for memory cleanup judgment, in the technical solution of the present application, the state data of the first application is embedded and encoded using a program state embedding matrix to convert these discrete features into dense vector representations in a continuous space, while capturing the complex relationship between different states to obtain an application state embedding coding vector.

[0030] Exemplarily, in the application memory resource timing encoding subunit 322, the time series of the memory resource usage of the first application is passed through a sequence encoder based on an RNN-LSTM hybrid model to obtain an implicit correlation feature vector of the application memory resource usage timing as the implicit correlation feature of the application memory resource usage timing. It should be understood that, considering that the time series data of the memory resource usage of the first application has obvious timing characteristics, its value changes over time and there is often a dependency relationship between the data of the previous and next moments, that is, there are different degrees of timing correlation between different local time scales. However, it is difficult for traditional models to effectively capture this long-term dependency and the dynamic change law of the timing in the sequence. Based on this, the present application encodes the time series of the memory resource usage of the first application through a sequence encoder based on an RNN-LSTM hybrid model to obtain an implicit correlation feature vector of the application memory resource usage timing. It is worth mentioning that RNN is a neural network specifically used for processing sequence data, and its characteristic is that there is a cyclic connection in the network structure, so that information can be transmitted between different time steps of the sequence. For time series data, such as the time series of memory resource usage of the first application, RNN can use the hidden state of the previous moment and the input of the current moment to calculate the hidden state of the current moment, thereby learning the short-term time series fluctuation information in the sequence. LSTM is a variant of RNN, which is mainly designed to solve the gradient vanishing or gradient exploding problems that RNN is prone to when processing long sequence data. LSTM controls the flow of information by introducing special gate structures (input gate, forget gate, and output gate). The forget gate determines what information to discard from the cell state, the input gate determines how much new information to allow into the cell state, and the output gate determines how much cell state to output as the current output. This structure enables LSTM to better capture long-distance dependencies and process long time series data more effectively. Therefore, the RNN-LSTM hybrid model can not only use RNN to effectively capture the dynamic changes in the short-term time series of memory resource usage and quickly respond to immediate adjustments in memory usage patterns, but also use LSTM to ensure that information can be accurately remembered and forgotten even in the face of long-term data spans, while identifying the complex patterns hidden behind the data, which is crucial to understanding the memory behavior of the application.

[0031] In one embodiment, Figure 5As shown, the data semantic alignment fusion unit 33 includes: an application state-resource usage fusion sub-unit 331, which is used to perform semantic information-guided feature alignment fusion on the application state embedded coding vector and the application memory resource usage time series implicit correlation feature vector to obtain an application state-resource usage fine-grained semantic alignment fusion feature vector as the application state-resource usage fine-grained semantic alignment fusion feature; a judgment result generation sub-unit 332, which is used to generate the memory optimization judgment result based on the application state-resource usage fine-grained semantic alignment fusion feature vector.

[0032] Exemplarily, in the application state-resource usage fusion subunit 331, the application state embedding coding vector and the application memory resource usage temporal implicit association feature vector are subjected to feature alignment fusion guided by semantic information to obtain the application state-resource usage fine-grained semantic alignment fusion feature vector as the application state-resource usage fine-grained semantic alignment fusion feature. It should be understood that the application state embedding coding vector reflects the key semantic features of the application state such as the type and user interaction behavior, while the application memory resource usage temporal implicit association feature vector reflects the implicit association of memory changes over time, and there is a complex semantic connection between the two features. Therefore, in order to fully explore the semantic associations between them, build an environment that allows them to interact deeply, and find the potential connections between the two based on semantic information, and explore those deep semantic associations that are difficult to detect with a single vector, so as to more accurately grasp the overall situation of the application, in the technical solution of the present application, a feature alignment fusion mechanism guided by semantic information is introduced to fuse the application state embedded coding vector and the temporal implicit association feature vector of the application memory resource usage to obtain a fine-grained semantic alignment fusion feature vector of application state-resource usage.

[0033] In one embodiment, Figure 6As shown, the application state-resource usage fusion subunit 331 includes: an application feature modulation secondary subunit 3311, which is used to perform point convolution feature modulation on the application state embedded coding vector and the application memory resource usage time series implicit association feature vector to obtain the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector; an information field feature fine-grained alignment secondary subunit 3312, which is used to perform semantic information field fine-grained alignment on the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector to obtain a fine-grained aligned application state embedded coding vector and a fine-grained aligned application memory resource usage time series implicit association feature vector; a position weighted sum processing secondary subunit 3313, which is used to calculate the position weighted sum between the fine-grained aligned application state embedded coding vector and the fine-grained aligned application memory resource usage time series implicit association feature vector to obtain the application state-resource usage fine-grained semantic alignment fusion feature vector.

[0034] Exemplarily, in the application feature modulation secondary subunit 3311, the application state embedding coding vector and the application memory resource usage time series implicit correlation feature vector are subjected to point convolution feature modulation to obtain the modulated application state embedding coding vector and the modulated application memory resource usage time series implicit correlation feature vector. In one embodiment, the modulated application state embedding coding vector and the modulated application memory resource usage time series implicit correlation feature vector have the same feature dimension. Specifically, the process can be expressed by the formula: ;in, and are the application state embedding encoding vector and the application memory resource usage time series implicit correlation feature vector, is the point convolutional coding, is the activation function, is the modulated application state embedding code vector, is the implicit correlation feature vector of the time series of memory resource usage of the modulated application.

[0035] It should be understood that the state data of the application (such as application type, user interaction behavior record) and the time series data of memory resource usage have different feature dimensions and semantic meanings. For example, the state data may contain discrete category information, while the memory resource usage is expressed as a continuous time series change. In order to effectively combine these two types of features, it is first necessary to solve the dimensional difference problem between them. To this end, this application adopts a feature modulation method based on point convolution (i.e., 1x1 convolution). Point convolution is a special convolution operation, the core of which is to change the number of channels of the feature map without affecting the spatial resolution. Specifically in this scenario, point convolution is used to adjust the dimensions of the application state embedding encoding vector and the time series implicit association feature vector of the application memory resource usage, so that the two can be compared and fused in the same dimension. This step not only solves the problem of dimensional inconsistency between the original feature vectors, but also enhances the abstractness and expressiveness of the feature representation by introducing nonlinear transformations. The feature vector modulated by point convolution not only maintains the rich information of the original features, but also can be effectively fused in the new dimension, laying the foundation for subsequent fine-grained semantic alignment. Point convolution feature modulation can more accurately capture the complex relationship between the behavior patterns of applications and their memory usage. For example, for an application, if it takes up a lot of memory when running in the foreground, but quickly reduces memory consumption after switching to the background, the modulated feature vector can more clearly reflect this trend of change. Similarly, for background service applications that continue to maintain high memory usage even in the background, the modulated feature vector can better retain its characteristics and help the system make more reasonable memory management decisions.

[0036] In one embodiment, the information field feature fine-grained alignment secondary subunit 3312 is used to: after fine-grained association coding of the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector, the obtained application state-resource usage fine-grained semantic association matrix is ​​subjected to semantic information field convolution coding to obtain the application state-resource usage feature fine-grained semantic information field. Specifically, the process can be expressed as follows: ;in, yes The transposed vector of yes Length, is the application state-resource usage fine-grained semantic association matrix, is a convolutional code with a convolution kernel of 3×3. It is a fine-grained semantic information field between application state and resource usage characteristics. It's matrix multiplication.

[0037] The modulated application state embedding coding vector and the modulated application memory resource usage time series implicit association feature vector are respectively mapped to the fine-grained semantic information field between the application state and resource usage features to obtain the fine-grained aligned application state embedding coding vector and the fine-grained aligned application memory resource usage time series implicit association feature vector. Specifically, the process can be expressed as follows: ;in, is the fine-grained aligned application state embedding encoding vector, It is a fine-grained alignment of the implicit correlation feature vector of the application memory resource usage timing.

[0038] That is, first, through fine-grained association coding, the subtle connection between the modulated application state embedding coding vector and the temporal implicit association feature vector of the application memory resource usage can be captured. This encoding method not only focuses on the direct correspondence between the two, but also deeply explores the implicit association between them, such as how certain user interaction behaviors affect the memory consumption pattern, or the memory usage pattern of a specific application type in different time periods. Through the careful characterization of these complex relationships, a fine-grained semantic association matrix of application state-resource usage containing rich semantic information can be constructed. This matrix not only reflects the application state and memory usage at each moment, but also reveals the potential dependencies and interactions between the two.

[0039] Next, in order to further enhance the expressiveness of these semantic information, this application inputs the obtained fine-grained semantic association matrix into a semantic information field encoding network based on convolution kernels. The convolution operation plays a key role in this process: the convolution kernel, as a local receptive field, can focus on specific areas in the feature distribution, thereby identifying local patterns and their semantic meanings. By stacking multiple convolutional layers, the mapping from low-level features to high-level semantic concepts is gradually learned, and finally a fine-grained semantic information field between application state and resource usage features is formed. This information field not only contains the semantic information of the original feature vectors, but also covers the position alignment and offset information between them, so that each feature point can be more accurately understood in its context.

[0040] With the construction of the semantic information field completed, the next step is to map the modulated application state embedding encoding vector and the temporal implicit association feature vector of application memory resource usage to this information field respectively. This step is actually a cross-domain feature conversion, similar to domain adaptation in transfer learning, but specifically refers to the semantic alignment between feature vectors. In this mapping process, the position of each feature vector is adjusted according to the semantic strength and relevance of the corresponding position in the semantic information field to ensure that they better match the semantic distribution in the information field. The result is a fine-grained aligned application state embedding encoding vector and temporal implicit association feature vector of application memory resource usage. These aligned feature vectors not only retain the original semantic information, but also enhance the fine-grained positional semantic consistency between each other, so that subsequent fusion operations can be carried out on a more unified and coordinated basis.

[0041] Exemplarily, in the position-based weighted sum processing secondary subunit 3313, the position-based weighted sum between the fine-grained aligned application state embedded coding vector and the fine-grained aligned application memory resource usage temporal implicit association feature vector is calculated to obtain the application state-resource usage fine-grained semantic alignment fusion feature vector. Specifically, the process can be expressed as follows: ;in, and is the weighting coefficient, The application state-resource fused feature vector is aligned using fine-grained semantics.

[0042] Specifically, after preliminary processing such as fine-grained association coding, semantic information field construction, and feature mapping, the modulated application state embedding coding vector and the temporal implicit association feature vector of application memory resource usage have been accurately aligned in the same semantic space. At this point, the elements in each feature vector not only contain the original application state or memory usage, but also reflect the semantic relationship and position alignment characteristics between them. In order to further integrate this information, this application uses position weighted sum to merge the two feature vectors into a comprehensive representation. The core idea of ​​the position weighted sum is to linearly combine the elements of the two feature vectors according to the predefined weight coefficients at each corresponding position. This weighted method takes into account the differences in the importance of features at different positions, so that certain key features can be given higher weights, thereby occupying a more important position in the final fusion result.

[0043] More importantly, the position-based weighted sum process ensures that the semantic consistency between the fine-grained aligned application state embedding encoding vector and the temporal implicit association feature vector of application memory resource usage is maintained. Since the two feature vectors have been precisely aligned in the semantic information field, the elements at each position actually have similar semantic meanings when performing the weighted sum. This ensures that the fused feature vector not only retains the rich information of the original features, but also achieves a deeper fusion in the new dimension, forming a comprehensive representation that can fully describe the application state and its memory usage - that is, the application state-resource usage fine-grained semantic alignment fusion feature vector.

[0044] Exemplarily, in the determination result generation subunit 332, the memory optimization determination result is generated based on the application state-resource usage fine-grained semantic alignment fusion feature vector. In one embodiment, the determination result generation subunit is used to: pass the application state-resource usage fine-grained semantic alignment fusion feature vector through a classifier-based memory optimization logic determiner to obtain the memory optimization determination result. The memory optimization determination result is used to indicate whether to perform a memory cleanup operation on the first application. That is, the application state-resource usage fine-grained semantic alignment fusion feature vector obtained by semantically aligning and fusion of the application state embedded coding vector and the application memory resource usage time series implicit association feature vector is classified and processed to utilize the powerful pattern recognition and classification functions of the classifier to intelligently determine whether to perform a memory cleanup operation on the first application. It should be understood that the classifier, as a supervised learning model, can learn the memory usage patterns corresponding to different types of applications under different operating states during the training process, and establish a prediction model based on these patterns. When facing a new application instance, the classifier can output a probability value or category label based on the input feature vector to indicate whether the application needs to perform memory cleanup. This data-driven approach not only improves the accuracy of decision-making, but also enhances the flexibility and adaptability of the system, enabling it to better cope with the ever-changing application environment.

[0045] In a specific embodiment, the classifier-based memory optimization logic judge uses a support vector machine (SVM), which divides sample points of different categories by finding an optimal hyperplane to maximize the interval between two categories. For the memory optimization problem, "need to clean up" and "no need to clean up" can be regarded as two different categories. In actual applications, when a new application state-resource fine-grained semantic alignment fusion feature vector is input into the SVM, it will evaluate the state of the application based on the previously learned knowledge and give a clear judgment result - whether the memory cleanup operation should be performed.

[0046] Here, considering that the application state embedded coding vector and the application memory resource usage time series implicit correlation feature vector respectively represent the embedded coding features of the state data of the first application and the short-range-long-range bidirectional temporal correlation features of the time series of the memory resource usage of the first application, when performing feature alignment interaction based on the semantic information field, the misalignment of the semantic information field under the feature distribution difference caused by the source data modality will also cause the alignment interaction distribution richness of the application state-resource usage fine-grained semantic alignment fusion feature vector, resulting in repetition of unstructured probability mapping, affecting the accuracy of the classification results.

[0047] Therefore, in a preferred example, when the application state-resource usage fine-grained semantic alignment fusion feature vector is passed through a classifier-based memory optimization logic judge to obtain the memory optimization judgment result, the application state-resource usage fine-grained semantic alignment fusion feature vector is optimized, specifically including the steps of: calculating the autocorrelation matrix of the application state-resource usage fine-grained semantic alignment fusion feature vector to obtain the application state-resource usage fine-grained semantic alignment fusion autocorrelation matrix ;in, Representing the application state-resource fused feature vector using fine-grained semantic alignment, represents matrix multiplication, represents a transpose operation, where the application state-resource fused feature vector is a column vector using fine-grained semantic alignment.

[0048] Align the application state-resources using fine-grained semantics to fuse the self-correlation matrix Respectively with the learnable weight matrix After matrix multiplication with its transpose, the matrix product is dotted to obtain the application state-resource fusion multi-dimensional structure matrix using fine-grained semantic alignment: ;in, represents vector addition, represents the transpose of a matrix, Representing application state-resources using fine-grained semantic alignment fused multi-dimensional structure matrices.

[0049] The application state-resources are fused into a multi-dimensional structure matrix using fine-grained semantic alignment Fusion of the self-association matrix with the application state-resource using fine-grained semantic alignment After point subtraction, the point subtraction matrix is ​​fused with the application state-resource feature vector using fine-grained semantic alignment Perform matrix multiplication to obtain application state-resources Approximate feature vectors using fine-grained semantic alignment fusion fields : ;in, Represents vector subtraction.

[0050] Compute the application state-resource feature vector using fine-grained semantically aligned fusion field approximation All eigenvalues ​​of The square root of the sum of the squares of the application state-resources uses fine-grained semantic alignment to fuse smooth measure values .

[0051] The smooth measure value is fused with the application state-resource using fine-grained semantic alignment The application state-resource fusion feature vector is calculated as an index using fine-grained semantic alignment Each eigenvalue of The power function , minus one and divided by the corresponding eigenvalue To obtain application state-resource feature vectors using fine-grained semantic alignment fusion embedding mapping : ;in, It means point multiplication by position. Representation of the application state-resource using fine-grained semantic alignment fused smooth metric values The application state-resource fusion feature vector is calculated as an index using fine-grained semantic alignment The power function of each eigenvalue of Represents the inverse of each eigenvalue of the application state-resource fused feature vector computed using fine-grained semantic alignment.

[0052] Compute the application state-resource feature vector using fine-grained semantically aligned fusion field approximation The feature vector is embedded in the application state-resource map using fine-grained semantic alignment fusion The weighted sum of is used to obtain the optimized application state-resource fused feature vector using fine-grained semantic alignment.

[0053] That is, based on the tensor field approximation principle of multi-dimensional topological structure, the discrete singularities in the application state-resource usage fine-grained semantic alignment fusion feature vector are used as the smooth dimension of the multi-dimensional topological structure, so as to fix the tensor field pointing in the normal gradient direction near the local coordinate system represented by the principal curvature of the application state-resource usage fine-grained semantic alignment fusion feature vector, thereby realizing the differentiable embedding mapping of the manifold embedding form of the application state-resource usage fine-grained semantic alignment fusion feature vector to the probability measure smooth domain, thereby avoiding the redundant mapping phenomenon in the differentiable embedding mapping process caused by topological heterogeneity, and improving the accuracy of the classification result obtained by inputting the application state-resource usage fine-grained semantic alignment fusion feature vector into the memory optimized logic judgement based on the classifier.

[0054] In summary, according to the embodiment of the present application, a memory optimization system for an intelligent health care companion robot is explained, which obtains the time series of the state data and memory resource usage of the first application, and uses data analysis and encoding methods based on deep learning to perform embedded coding of the state data of the first application and the temporal association of the memory resource usage of the first application, thereby intelligently judging whether to perform a memory cleanup operation on the first application based on the semantic alignment fusion representation between the embedded coding features of the application state and the implicit association features of the temporal association of the application memory resource usage. Through intelligent analysis, the present application can accurately identify the situation of the application memory, effectively avoiding the misjudgment that may be caused by traditional static rules. In addition, it can automatically adjust its optimization strategy according to different devices and user behavior patterns, without relying on a fixed set of rules, to better adapt to diverse application scenarios, thereby providing a personalized memory management solution.

[0055] An embodiment of the present application further provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer implements the methods in the above embodiments of the present application.

[0056] An embodiment of the present application further provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer implements the methods in the above embodiments of the present application.

[0057] An embodiment of the present application also provides a chip, including a circuit, for executing the methods in the above embodiments of the present application.

[0058] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0059] In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is a kind of association relationship that describes associated objects, indicating that three relationships can exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0060] The prefixes such as "first" and "second" used in the embodiments of the present application are only used to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers used to distinguish description objects in the embodiments of the present application does not constitute a limitation on the described objects. For the description of the described objects, please refer to the description of the context in the embodiments, and the use of such prefixes should not constitute an unnecessary limitation.

[0061] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0062] In the various embodiments of the present application, unless otherwise specified or logically conflicting, the terms and / or descriptions between the various embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0063] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0064] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0065] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A memory optimization system for an intelligent health care companion robot, characterized in that: include: A code receiving and loading module, used for receiving a memory optimization code carrying a program identifier and a memory release function, and loading the memory optimization code into the intelligent health care companion robot; A program monitoring module, used to use the memory optimization code to monitor each application in the intelligent health care companion robot in real time, and detect whether each application is switched from the foreground to the background; a cleaning operation judgment module, used to generate a memory optimization judgment result using the memory optimization logic in response to the first application switching from the foreground to the background to judge whether to perform a memory cleaning operation on the first application; A release module, for executing a memory cleanup operation on the first application in response to the memory optimization judgment result, and calling the memory release function to release the memory resources of the first application; wherein the cleanup operation judgment module includes: a data acquisition unit, for acquiring the state data of the first application and acquiring the time series of the memory resource usage of the first application; a data encoding unit, for embedding and timing encoding the state data of the first application and the time series of the memory resource usage of the first application respectively to obtain the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage; a data semantic alignment fusion unit, for performing feature alignment fusion guided by semantic information on the embedded coding features of the application state and the time series implicit correlation features of the application memory resource usage to obtain the fine-grained semantic alignment fusion features of the application state and resource usage, and generating the memory optimization judgment result based on the fine-grained semantic alignment fusion features of the application state and resource usage.

2. The memory optimization system for the intelligent health care companion robot according to claim 1, characterized in that: The status data includes application type and user interaction behavior records.

3. The memory optimization system for the intelligent health care companion robot according to claim 2, characterized in that: The data encoding unit includes: an application state embedding encoding subunit, which is used to use a program state embedding matrix to embed the state data of the first application to obtain an application state embedding encoding vector as the application state embedding encoding feature; an application memory resource timing encoding subunit, which is used to pass the time series of the memory resource usage of the first application through a sequence encoder based on an RNN-LSTM hybrid model to obtain an application memory resource usage timing implicit correlation feature vector as the application memory resource usage timing implicit correlation feature.

4. The memory optimization system for the intelligent health care companion robot according to claim 3, characterized in that: The data semantic alignment fusion unit includes: an application state-resource usage fusion sub-unit, which is used to perform semantic information-guided feature alignment fusion on the application state embedded coding vector and the application memory resource usage time series implicit correlation feature vector to obtain an application state-resource usage fine-grained semantic alignment fusion feature vector as the application state-resource usage fine-grained semantic alignment fusion feature; and a judgment result generation sub-unit, which is used to generate the memory optimization judgment result based on the application state-resource usage fine-grained semantic alignment fusion feature vector.

5. The memory optimization system for the intelligent health care companion robot according to claim 4, characterized in that: The application state-resource usage fusion subunit includes: an application feature modulation secondary subunit, which is used to perform point convolution feature modulation on the application state embedded coding vector and the application memory resource usage time series implicit association feature vector to obtain the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector; an information field feature fine-grained alignment secondary subunit, which is used to perform semantic information field fine-grained alignment on the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector to obtain a fine-grained aligned application state embedded coding vector and a fine-grained aligned application memory resource usage time series implicit association feature vector; a position weighted sum processing secondary subunit, which is used to calculate the position weighted sum between the fine-grained aligned application state embedded coding vector and the fine-grained aligned application memory resource usage time series implicit association feature vector to obtain the application state-resource usage fine-grained semantic alignment fusion feature vector.

6. The memory optimization system for the intelligent health care companion robot according to claim 5, characterized in that: The modulated application state embedded coding vector and the modulated application memory resource usage temporal implicit correlation feature vector have the same feature dimension.

7. The memory optimization system for the intelligent health care companion robot according to claim 6, characterized in that: The information field feature fine-grained alignment secondary sub-unit is used to: after fine-grained association encoding of the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector, perform semantic information field convolution encoding on the obtained application state-resource usage fine-grained semantic association matrix to obtain the fine-grained semantic information field between application state and resource usage features; respectively map the modulated application state embedded coding vector and the modulated application memory resource usage time series implicit association feature vector to the fine-grained semantic information field between application state and resource usage features to obtain the fine-grained aligned application state embedded coding vector and the fine-grained aligned application memory resource usage time series implicit association feature vector.

8. The memory optimization system for the intelligent health care companion robot according to claim 7, characterized in that: The determination result generating subunit is used to: pass the application state-resource usage fine-grained semantic alignment fusion feature vector through a classifier-based memory optimization logic determiner to obtain the memory optimization determination result.

9. The memory optimization system for the intelligent health care companion robot according to claim 8, characterized in that: The memory optimization determination result is used to indicate whether to perform a memory cleanup operation on the first application.

Images