A method and device for debugging compatibility of application software for heterogeneous devices

Abstract

The application provides a kind of application software development compatibility debugging method and device for heterogeneous equipment, and is related to application software debugging.The application constructs equipment dynamic characteristic map by collecting and integrating equipment hardware, system, resource, interface and running state characteristics through multi-modal feature fusion technology.Secondly, the real-time iterative optimization of compatibility calculation weight is realized by introducing reinforcement learning algorithm, and the feature adaptation accuracy is improved by combining the improved normalization model.Through abnormal fingerprint extraction and tracing algorithm, the precise positioning and root cause analysis of incompatibility problem are realized.Finally, based on dynamic confidence threshold, hierarchical adaptation, intelligent repair and cross-device debugging are executed.The application breaks through the limitations of traditional fixed strategy debugging, realizes the intelligentization, self-adaptation and precision of compatibility debugging, greatly improves the debugging efficiency and reduces the adaptation cost, and is suitable for various heterogeneous equipment platforms.

Description

Technical Field

[0001] This invention relates to the field of application software debugging, and in particular to a method and apparatus for compatibility debugging of application software development for heterogeneous devices. Background Technology

[0002] With the rapid iteration of IoT, edge computing, and industrial internet technologies, heterogeneous device clusters have become the mainstream scenario for application software deployment, encompassing servers, embedded devices, mobile terminals, edge nodes, and IoT devices with different architectures such as x86, ARM, and RISC-V. These devices exhibit significant differences in hardware configuration, instruction set type, operating system version, peripheral interface specifications, and communication protocol standards, leading to frequent compatibility issues when deploying application software across devices. These issues include startup failures, functional abnormalities, lag, crashes, and data synchronization errors, severely impacting user experience and stability in scenarios such as industrial production and public services.

[0003] For compatibility debugging of application software on heterogeneous devices, existing technologies generally suffer from problems such as single feature collection dimensions and rigid adaptation rules, failing to take into account dynamic operation and interface interaction characteristics; compatibility calculation uses fixed weights, lacking adaptability and failing to fit actual debugging scenarios; anomaly tracing only counts frequency without extracting feature fingerprints, resulting in low localization efficiency; there is a lack of cross-device debugging data sharing and experience reuse mechanisms, increasing adaptation costs; there is no self-learning capability, making it difficult to adapt to new heterogeneous devices and software versions; and the confidence decision threshold is fixed, failing to consider device complexity and scenario differences, easily leading to misjudgments and omissions, and overall failing to meet the requirements of efficient, accurate, and adaptive debugging.

[0004] Therefore, there is a need to provide a method and apparatus for application software development compatibility debugging for heterogeneous devices to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for compatibility debugging of application software development for heterogeneous devices. It breaks through the limitations of traditional fixed-strategy debugging, realizes intelligent, adaptive and precise compatibility debugging, improves debugging efficiency and reduces adaptation costs, enhances the universality and scalability of the solution, and effectively solves the defects of compatibility debugging of heterogeneous devices.

[0006] To achieve the above objectives, the present invention provides a method for compatibility debugging of application software development for heterogeneous devices, comprising the following steps: S1: Multimodal device feature acquisition and fusion, which collects static hardware features, dynamic operation features and interface interaction features of heterogeneous devices, and constructs a dynamic feature map of the devices through feature cleaning and fusion algorithms; S2: An improved normalization formula is used to normalize each feature in the equipment dynamic feature map, eliminating differences in feature dimensions and obtaining normalized feature values; the improved normalization formula is: ; In the formula, This represents the original characteristic value of the k-th device. A minimum threshold is preset for the k-th feature; A maximum threshold is preset for the k-th feature; This is the normalized value of the k-th feature; This is the feature importance correction coefficient, with a value range of [0.8, 1.2]. This is a deviation correction term; S3: Calculate the compatibility of reinforcement learning-driven systems. Introduce reinforcement learning algorithms and dynamically update the weight coefficients of the normalized features in S2 based on historical debugging data and real-time running status. Calculate the overall compatibility between the application software and the current device through weighted summation. S4: Extract the feature fingerprints of runtime abnormal events through the abnormal behavior capture module, and calculate the total runtime abnormal index based on the severity and scope of the abnormality. S5: Perform dynamic confidence decision-making and hierarchical debugging. Based on the overall compatibility obtained in S3, the total runtime anomaly index calculated in S4, and the device dynamic feature map constructed in S1, construct a dynamic confidence decision-making model, set adaptive thresholds, and perform normal operation, intelligent dynamic adaptation, precise breakpoint debugging, and cross-device collaborative repair according to the confidence interval. S6: Feed back the debugging results, root causes of anomalies, and adaptation solutions to the reinforcement learning model, and update the weight iteration strategy and anomaly fingerprint database.

[0007] Preferably, the overall compatibility in S3 is calculated using the following formula: ; In the formula, For overall compatibility; ~ The normalized characteristic values ​​are, in order: CPU architecture, instruction set, memory usage, power consumption, peripheral interfaces, and communication protocol. The dynamic weight coefficient of the k-th feature at time t is iteratively updated by the reinforcement learning model, satisfying... The iterative formula is: ; in For learning rate, The gradient of the objective function is optimized for the weights.

[0008] Preferably, the total runtime exception index in S4 is calculated using the following formula: ; In the formula, This represents the total runtime exception index. m This represents the number of types of runtime exceptions. For the first i The severity coefficient of the anomaly ranges from [1,5], with a larger coefficient for higher severity. Let be the frequency of occurrence of the i-th type of anomaly; For the first i The impact range coefficient of the exception on the operation of the application software is set to [0.1, 1.0].

[0009] Preferably, the dynamic confidence decision model in S5 is: ; In the formula, Let be the confidence level of the debugging decision at time t; This is a dynamic adaptation coefficient, determined by the complexity of the device feature map, with a value range of [0.9, 1.1]. The greater the difference in device features, the better. The smaller.

[0010] Preferably, the hierarchical debugging strategy in S5 is as follows: when When the system is deemed fully compatible, it controls the application software to run normally and monitors its running status in real time. when When the system is deemed basically compatible, the intelligent dynamic adaptation module is activated. Based on the device dynamic feature map built in S1, the interface calls, resource allocation and code execution logic are automatically adjusted, and the abnormal fingerprint database is updated synchronously. when When an incompatibility is detected, precise breakpoint debugging is triggered. The incompatible code segment and root cause are located by matching the anomaly fingerprint, a personalized repair plan is generated, and cross-device collaborative debugging is started to synchronize the repair experience of similar devices to the current debugging process.

[0011] Preferably, the optimization objective of the reinforcement learning model in S6 is to minimize the compatibility error and anomaly frequency of the application software on heterogeneous devices, and maximize debugging efficiency; the objective function is: ; In the formula, To predict compatibility, For actual compatibility, T The debugging process takes time. , , All are weighting coefficients.

[0012] A device for debugging application software compatibility for heterogeneous devices includes a multimodal feature acquisition and fusion module, an adaptive normalization module, a reinforcement learning and compatibility calculation module, an anomaly fingerprint extraction and monitoring module, a dynamic decision-making and hierarchical debugging module, and a data feedback and model optimization module. The modules work together to realize the steps of the method for debugging application software compatibility for heterogeneous devices.

[0013] Preferably, the multimodal feature acquisition and fusion module is used to traverse the heterogeneous device cluster and collect the static hardware features, dynamic operating features, and interface interaction features of the devices. The static hardware features include CPU architecture and instruction set type; the dynamic operating features include memory usage, power consumption, and frame rate; the interface interaction features include peripheral interface type and communication protocol version; the built-in feature cleaning unit removes abnormal feature data and fills in missing feature values; the built-in feature fusion unit uses a weighted fusion algorithm to integrate the static hardware features, dynamic operating features, and interface interaction features to construct a device dynamic feature map, thereby achieving unified and visualized management of device features. The adaptive normalization module communicates with the multimodal feature acquisition and fusion module, receiving the original feature values ​​from the device's dynamic feature map; it has a built-in improved normalization calculation unit that calls the improved normalization formula. The system normalizes each original feature to eliminate dimensional differences between features and outputs the normalized feature values ​​to the reinforcement learning and compatibility calculation module. Simultaneously, it receives feature importance correction coefficients from the reinforcement learning and compatibility calculation module. Dynamically adjust the normalized calculation parameters; The reinforcement learning and compatibility calculation modules are communicatively connected to the adaptive normalization module and the data feedback and model optimization module, respectively, receiving normalized feature values; the built-in reinforcement learning unit, based on historical debugging data and real-time running status, uses iterative formulas... Dynamically iteratively update the weight coefficients of each feature This ensures that the weighting coefficients fit the current debugging scenario; it also includes a built-in compatibility calculation unit that calls the overall compatibility formula. Calculate the overall compatibility between the application software and the current device, and output the feature importance correction coefficient. The overall compatibility is then output to the adaptive normalization module and then to the dynamic decision-making and hierarchical debugging module. The anomaly fingerprint extraction and monitoring module monitors the application software's running status on heterogeneous devices in real time, capturing various runtime anomalies such as crashes, freezes, API call failures, and memory overflows. It includes a built-in anomaly fingerprint extraction unit to extract unique feature fingerprints for each type of anomaly and build an anomaly fingerprint database; and a built-in anomaly total index calculation unit to call the anomaly total index formula. Based on the severity coefficient of the anomaly Frequency of occurrence and range influence coefficient Calculate the total anomaly index during runtime, and output the anomaly fingerprint information and the total anomaly index to the dynamic decision-making and hierarchical debugging module; The dynamic decision-making and hierarchical debugging module communicates with the reinforcement learning and compatibility calculation module, the anomaly fingerprint extraction and monitoring module, and the multimodal feature acquisition and fusion module, respectively, to receive overall compatibility, runtime anomaly index, anomaly fingerprint information, and device dynamic feature maps; it also has a built-in dynamic confidence calculation unit that calls the dynamic confidence decision model. The dynamic adaptation coefficient is determined by combining the complexity of the device feature map. Calculate the confidence level of the debugging decision. Built-in hierarchical decision-making unit, which executes corresponding debugging strategies based on confidence intervals; The data feedback and model optimization module communicates with the dynamic decision-making and hierarchical debugging module, the reinforcement learning and compatibility calculation module, and the anomaly fingerprint extraction and monitoring module, respectively, to collect complete data from this debugging session, including overall compatibility, root causes of anomalies, repair plans, debugging time, and anomaly fingerprint information. It also has a built-in data processing unit to classify, organize, and validate the debugging data. Furthermore, it has a built-in model update unit that feeds the processed debugging data back to the reinforcement learning model, updates the weight iteration strategy, and supplements and improves the anomaly fingerprint database, enabling continuous optimization of both the reinforcement learning model and the anomaly fingerprint database.

[0014] Preferably, it also includes a storage module and a communication module; the storage module is used to store the device's dynamic feature map, historical debugging data, abnormal fingerprint database, weight coefficient iteration records and debugging reports; the communication module realizes data interaction between all modules, as well as communication with heterogeneous devices and external debugging terminals, and supports cross-device debugging data sharing and collaborative repair command transmission.

[0015] Therefore, the present invention employs the above-described method and apparatus for application software development compatibility debugging for heterogeneous devices, and the technical effects are as follows: (1) The present invention adopts multimodal feature fusion technology to simultaneously collect static hardware features, dynamic operation features and interface interaction features of the device, and constructs a dynamic feature map of the device, which makes up for the defect of incomplete feature dimensions in the existing technology. At the same time, through reinforcement learning, the feature importance correction coefficient is dynamically output and combined with the improved normalization formula to realize the dynamic adjustment of feature adaptation, which can flexibly cover complex heterogeneous device scenarios and adapt to the differences and operating states of different devices.

[0016] (2) The present invention introduces a reinforcement learning algorithm to realize the real-time iterative update of the compatibility calculation weight. The weight coefficient can be dynamically adjusted according to historical debugging data and real-time running status, breaking the limitation of fixed weight in the existing technology, making the overall compatibility calculation more in line with the actual debugging scenario, accurately reflecting the real compatibility status of application software and equipment, and reducing compatibility misjudgment.

[0017] (3) This invention captures the unique feature fingerprints of various runtime anomalies through anomaly fingerprint extraction technology, establishes anomaly fingerprint database, and calculates the total anomaly index by combining the severity of the anomaly, the frequency of occurrence and the scope of impact, so as to achieve accurate differentiation of anomaly types and root cause location, solves the problems of vague anomaly tracing and long location time in the prior art, significantly shortens the time for debugging personnel to troubleshoot problems and improves debugging efficiency.

[0018] (4) This invention introduces a cross-device collaborative repair mechanism, establishes a system for sharing debugging data and reusing experience among heterogeneous devices, eliminates the need for repeated debugging of compatibility issues of similar devices, allows for direct reuse of existing repair experience, significantly reduces the workload of repeated debugging, lowers the adaptation cost of application software deployment across heterogeneous devices, and shortens the product development cycle.

[0019] (5) This invention feeds back the results of each debugging, the root cause of the anomaly and the repair plan to the reinforcement learning model through a data feedback mechanism, continuously updates the weight iteration strategy and the anomaly fingerprint database, realizes the self-optimization of the model and debugging capabilities, and continuously improves the accuracy of compatibility prediction and the efficiency of anomaly location with the accumulation of debugging cases. It can flexibly adapt to new heterogeneous devices and software versions without frequent adjustment of debugging strategies.

[0020] (6) The present invention constructs a dynamic confidence decision model, sets a dynamic adaptation coefficient in combination with the complexity of the equipment feature map, and realizes the adaptive adjustment of the confidence threshold. This avoids the defects of the existing technology that fixes the threshold without considering the complexity of the equipment and the difference of the scenario, reduces the misjudgment and omission of debugging decisions, and improves the accuracy and scientificity of debugging decisions.

[0021] (7) The debugging device corresponding to this invention achieves full-process automation from feature acquisition, compatibility calculation, anomaly monitoring to hierarchical debugging and model optimization through the collaborative work of multiple modules. The addition of storage and communication modules ensures data storage security and smooth data interaction between modules and devices. The device can be adapted to various heterogeneous device clusters such as x86, ARM, and RISC-V without large-scale modification of existing equipment. The deployment cost is low and it can be widely used in various heterogeneous device application scenarios such as industrial Internet, Internet of Things, and edge computing. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for compatibility debugging of application software development for heterogeneous devices according to the present invention; Figure 2 This is a diagram of an application software development compatibility debugging device for heterogeneous devices according to the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0025] Example like Figure 1 As shown, this invention provides a method for compatibility debugging of application software development for heterogeneous devices. Using three types of heterogeneous devices—ARM-based embedded devices, x86-based terminal devices, and RISC-V-based edge nodes—as test objects, a cross-platform industrial control application software is debugged. The specific steps are as follows: The S1 multimodal device feature acquisition and fusion system collects three types of device features through its feature acquisition module: static hardware features (CPU architecture: ARM Cortex-A76, x86-64, RISC-V RV64G; instruction set: ARMv8, x86SSE4.2, RISC-V IMA), dynamic operating features (memory usage: 2GB / 8GB / 4GB; power consumption: 5W / 15W / 8W; frame rate: 30fps / 60fps / 45fps), and interface interaction features (peripheral interfaces: RS485 / USB3.0 / Ethernet; communication protocols: Modbus / TCP / IP / MQTT). Abnormal data is removed using a feature cleaning algorithm, and a weighted fusion algorithm is used to integrate the three types of features to construct a dynamic feature map of the device, achieving unified and visual management of device features.

[0026] S2 adaptive feature normalization processing takes the original values ​​of the six collected features (CPU architecture, instruction set, memory usage, power consumption, peripheral interface, and communication protocol) and substitutes them into the improved normalization formula: ; In the formula, This represents the original characteristic value of the k-th device. A minimum threshold is preset for the k-th feature; A maximum threshold is preset for the k-th feature; all thresholds are industry standard thresholds. This is the normalized value of the k-th feature; This is the feature importance correction coefficient, dynamically output by the reinforcement learning model based on feature importance, with a value range of [0.8, 1.2]. This is a deviation correction term. =0.05; finally, the normalized values ​​of the 6 features are obtained. ~ ; S3: Calculate the compatibility of reinforcement learning-driven learning and initialize the weight coefficients. =0.3、 =0.15、 =0.2、 =0.15、 =0.1、 =0.1, learning rate =0.01; Based on the objective function of the reinforcement learning model, combined with historical debugging data (in this embodiment, the historical debugging data consists of 100 sets of heterogeneous device debugging cases and real-time operating status), the weight coefficients are iteratively updated using the following iterative formula: After 50 iterations, stable weighting coefficients were obtained. =0.32、 =0.16、 =0.21、 =0.14、 =0.09、 =0.08; Substituting into the compatibility formula: The overall compatibility of the application software on the three types of devices was calculated to be 0.78 (ARM devices), 0.92 (x86 devices), and 0.65 (RISC-V devices).

[0027] S4: Through the abnormal behavior capture module, abnormal events during application software runtime are captured in real time, including four types of exceptions: interface call failure, memory overflow, rendering lag, and protocol incompatibility; the characteristic fingerprint of each type of exception is extracted (such as the error code and call frequency of interface call failure, and the address range of memory overflow), and a severity coefficient is set. (API call failed) (Memory overflow) (Rendering lag) (Protocol incompatibility), scope of impact coefficient , ; Statistical analysis of the frequency of occurrence of various anomalies Substitute into the formula The calculated total anomaly indices for the three types of devices were 18.7 (ARM devices), 4.2 (x86 devices), and 11.3 (RISC-V devices).

[0028] S5: Determine the dynamic adaptation coefficients for the three types of equipment based on the complexity of the equipment feature map. ARM devices x86 devices RISC-V devices Substitute into the dynamic confidence formula The calculated confidence levels for debugging decisions for the three types of devices were 0.041 (ARM devices), 0.978 (x86 devices), and 0.052 (RISC-V devices), respectively.

[0029] Hierarchical debugging execution: - x86 devices: It was determined to be fully compatible, the application software ran normally, and the running status was monitored in real time with no abnormal feedback. - RISC-V devices: This is not true; in reality... The system was determined to be incompatible, triggering precise breakpoint debugging. By matching the anomaly fingerprint, the two root causes of protocol incompatibility and insufficient memory allocation were located, and a repair plan was generated (modifying the MQTT protocol adaptation code and optimizing the memory allocation algorithm). At the same time, the cross-device collaboration module was called, and the debugging experience of similar RISC-V devices was referenced to accelerate the repair process. - ARM devices: The system was determined to be incompatible. Through abnormal fingerprint extraction, the root cause of the interface call failure was located to be the incompatibility of the RS485 interface driver. A driver adaptation patch was generated, and after the repair was completed, the confidence level was recalculated, eventually reaching 0.88, which met the conditions for full compatibility.

[0030] S6: Feed the data such as compatibility, root cause of anomalies, repair plan, and debugging time of this debugging into the reinforcement learning model to update the weight iteration strategy and anomaly fingerprint database; after optimization, the model's compatibility prediction error for similar heterogeneous devices is reduced by 35%, the anomaly location time is shortened by 40%, and the subsequent debugging efficiency is significantly improved.

[0031] like Figure 2 As shown, an application software compatibility debugging device for heterogeneous devices includes a multimodal feature acquisition and fusion module, an adaptive normalization module, a reinforcement learning and compatibility calculation module, an anomaly fingerprint extraction and monitoring module, a dynamic decision-making and hierarchical debugging module, and a data feedback and model optimization module. The modules work together to realize the steps of the application software compatibility debugging method for heterogeneous devices.

[0032] The multimodal feature acquisition and fusion module is used to traverse heterogeneous device clusters and collect static hardware features, dynamic operating features, and interface interaction features of the devices. Static hardware features include CPU architecture and instruction set type; dynamic operating features include memory usage, power consumption, and frame rate; interface interaction features include peripheral interface type and communication protocol version. The built-in feature cleaning unit removes abnormal feature data and fills in missing feature values; the built-in feature fusion unit uses a weighted fusion algorithm to integrate static hardware features, dynamic operating features, and interface interaction features to construct a device dynamic feature map, realizing unified and visualized management of device features. The adaptive normalization module communicates with the multimodal feature acquisition and fusion module, receiving the original feature values ​​from the device's dynamic feature map; it has a built-in improved normalization calculation unit that calls the improved normalization formula. The system normalizes each original feature to eliminate dimensional differences between features and outputs the normalized feature values ​​to the reinforcement learning and compatibility calculation module. Simultaneously, it receives feature importance correction coefficients from the reinforcement learning and compatibility calculation module. Dynamically adjust the normalized calculation parameters; The reinforcement learning and compatibility calculation modules are communicatively connected to the adaptive normalization module and the data feedback and model optimization module, respectively, receiving normalized feature values; the built-in reinforcement learning unit, based on historical debugging data and real-time running status, uses iterative formulas... Dynamically iteratively update the weight coefficients of each feature This ensures that the weighting coefficients fit the current debugging scenario; it also includes a built-in compatibility calculation unit that calls the overall compatibility formula. Calculate the overall compatibility between the application software and the current device, and output the feature importance correction coefficient. The overall compatibility is then output to the adaptive normalization module and then to the dynamic decision-making and hierarchical debugging module. The anomaly fingerprint extraction and monitoring module monitors the application software's running status on heterogeneous devices in real time, capturing various runtime anomalies such as crashes, freezes, API call failures, and memory overflows. It includes a built-in anomaly fingerprint extraction unit to extract unique feature fingerprints for each type of anomaly and build an anomaly fingerprint database; and a built-in anomaly total index calculation unit to call the anomaly total index formula. Based on the severity coefficient of the anomaly Frequency of occurrence and range influence coefficient Calculate the total anomaly index during runtime, and output the anomaly fingerprint information and the total anomaly index to the dynamic decision-making and hierarchical debugging module; The dynamic decision-making and hierarchical debugging module communicates with the reinforcement learning and compatibility calculation module, the anomaly fingerprint extraction and monitoring module, and the multimodal feature acquisition and fusion module, respectively, to receive overall compatibility, runtime anomaly index, anomaly fingerprint information, and device dynamic feature maps; it also has a built-in dynamic confidence calculation unit that calls the dynamic confidence decision model. The dynamic adaptation coefficient is determined by combining the complexity of the device feature map. Calculate the confidence level of the debugging decision. Built-in hierarchical decision-making unit, which executes corresponding debugging strategies based on confidence intervals; The data feedback and model optimization module communicates with the dynamic decision-making and hierarchical debugging module, the reinforcement learning and compatibility calculation module, and the anomaly fingerprint extraction and monitoring module, respectively, to collect complete data from this debugging session, including overall compatibility, root causes of anomalies, repair plans, debugging time, and anomaly fingerprint information. It also has a built-in data processing unit to classify, organize, and validate the debugging data. Furthermore, it has a built-in model update unit that feeds the processed debugging data back to the reinforcement learning model, updates the weight iteration strategy, and supplements and improves the anomaly fingerprint database, enabling continuous optimization of both the reinforcement learning model and the anomaly fingerprint database.

[0033] It also includes a storage module and a communication module; the storage module is used to store the device's dynamic feature map, historical debugging data, anomaly fingerprint database, weight coefficient iteration records and debugging reports; the communication module realizes data interaction between all modules, as well as communication with heterogeneous devices and external debugging terminals, and supports cross-device debugging data sharing and collaborative repair command transmission.

[0034] Therefore, this invention employs the aforementioned method and apparatus for application software development compatibility debugging for heterogeneous devices. It utilizes multimodal feature fusion and improved normalization technology to completely solve the problems of single feature acquisition and rigid adaptation, significantly improving the comprehensiveness and flexibility of feature adaptation. It introduces reinforcement learning algorithms to achieve dynamic iteration of compatibility weights, combined with a dynamic confidence decision model, greatly improving the scientific rigor and accuracy of compatibility status determination and debugging decisions. It achieves precise source tracing through anomaly fingerprint extraction, coupled with a cross-device collaborative repair mechanism, efficiently improving debugging efficiency, reducing repetitive workload, and lowering adaptation costs and development cycles. It leverages data feedback to achieve autonomous model learning and iterative optimization, enhancing the versatility and scalability of the solution, adapting to new heterogeneous devices and software versions. The supporting apparatus achieves full-process automation through multi-module collaboration, and the addition of storage and communication modules enhances practicality.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for compatibility debugging of application software development for heterogeneous devices, characterized in that, Includes the following steps: S1: Multimodal device feature acquisition and fusion, which collects static hardware features, dynamic operation features and interface interaction features of heterogeneous devices, and constructs a dynamic feature map of the devices through feature cleaning and fusion algorithms; S2: An improved normalization formula is used to normalize each feature in the equipment dynamic feature map, eliminating differences in feature dimensions and obtaining normalized feature values; the improved normalization formula is: ； In the formula, This represents the original characteristic value of the k-th device. A minimum threshold is preset for the k-th feature; A maximum threshold is preset for the k-th feature; This is the normalized value of the k-th feature; This is the feature importance correction coefficient, with a value range of [0.8, 1.2]. This is a deviation correction term; S3: Calculate the compatibility of reinforcement learning-driven systems. Introduce reinforcement learning algorithms and dynamically update the weight coefficients of the normalized features in S2 based on historical debugging data and real-time running status. Calculate the overall compatibility between the application software and the current device through weighted summation. S4: Extract the feature fingerprints of runtime abnormal events through the abnormal behavior capture module, and calculate the total runtime abnormal index based on the severity and scope of the abnormality. S5: Perform dynamic confidence decision-making and hierarchical debugging. Based on the overall compatibility obtained in S3, the total runtime anomaly index calculated in S4, and the device dynamic feature map constructed in S1, construct a dynamic confidence decision-making model, set adaptive thresholds, and perform normal operation, intelligent dynamic adaptation, precise breakpoint debugging, and cross-device collaborative repair according to the confidence interval. S6: Feed back the debugging results, root causes of anomalies, and adaptation solutions to the reinforcement learning model, and update the weight iteration strategy and anomaly fingerprint database.

2. The application software development compatibility debugging method for heterogeneous devices according to claim 1, characterized in that, The overall compatibility in S3 is calculated using the following formula: ； In the formula, For overall compatibility; ~ The normalized characteristic values ​​are, in order: CPU architecture, instruction set, memory usage, power consumption, peripheral interfaces, and communication protocol. The dynamic weight coefficient of the k-th feature at time t is iteratively updated by the reinforcement learning model, satisfying... The iterative formula is: ； in For learning rate, The gradient of the objective function is optimized for the weights.

3. The application software development compatibility debugging method for heterogeneous devices according to claim 2, characterized in that, The total runtime anomaly index in S4 is calculated using the following formula: ； In the formula, This represents the total runtime exception index. m This represents the number of types of runtime exceptions. For the first i The severity coefficient of the anomaly ranges from [1,5], with a larger coefficient for higher severity. Let be the frequency of occurrence of the i-th type of anomaly; For the first i The impact range coefficient of the exception on the operation of the application software is set to [0.1, 1.0].

4. The application software development compatibility debugging method for heterogeneous devices according to claim 3, characterized in that, The dynamic confidence decision model in S5 is as follows: ; In the formula, Let be the confidence level of the debugging decision at time t; This is a dynamic adaptation coefficient, determined by the complexity of the device feature map, with a value range of [0.9, 1.1]. The greater the difference in device features, the better. The smaller.

5. The application software development compatibility debugging method for heterogeneous devices according to claim 4, characterized in that, The specific hierarchical debugging strategy in S5 is as follows: when When the system is deemed fully compatible, it controls the application software to run normally and monitors its running status in real time. when When the system is deemed basically compatible, the intelligent dynamic adaptation module is activated. Based on the device dynamic feature map built in S1, the interface calls, resource allocation and code execution logic are automatically adjusted, and the abnormal fingerprint database is updated synchronously. when When an incompatibility is detected, precise breakpoint debugging is triggered. The incompatible code segment and root cause are located by matching the anomaly fingerprint, a personalized repair plan is generated, and cross-device collaborative debugging is started to synchronize the repair experience of similar devices to the current debugging process.

6. The application software development compatibility debugging method for heterogeneous devices according to claim 3, characterized in that, The optimization objective of the reinforcement learning model in S6 is to minimize the compatibility error and anomaly frequency of the application software on heterogeneous devices, and maximize debugging efficiency; the objective function is: ； In the formula, To predict compatibility, For actual compatibility, T The debugging process takes time. , , All are weighting coefficients.

7. A software compatibility debugging device for heterogeneous devices, characterized in that, It includes a multimodal feature acquisition and fusion module, an adaptive normalization module, a reinforcement learning and compatibility calculation module, an anomaly fingerprint extraction and monitoring module, a dynamic decision-making and hierarchical debugging module, and a data feedback and model optimization module. The modules work together to implement the steps of the application software compatibility debugging method for heterogeneous devices as described in any one of claims 1 to 6.

8. The application software compatibility debugging device for heterogeneous devices according to claim 7, characterized in that, The multimodal feature acquisition and fusion module is used to traverse heterogeneous device clusters and acquire the static hardware features, dynamic operating features, and interface interaction features of the devices. Static hardware characteristics include CPU architecture and instruction set type; dynamic operational characteristics include memory usage, power consumption, and frame rate; interface interaction characteristics include peripheral interface type and communication protocol version. The built-in feature cleaning unit removes abnormal feature data and fills in missing feature values; The built-in feature fusion unit uses a weighted fusion algorithm to integrate static hardware features, dynamic operation features, and interface interaction features to construct a dynamic feature map of the device, thereby achieving unified and visualized management of device features. The adaptive normalization module communicates with the multimodal feature acquisition and fusion module to receive various original feature values ​​from the device's dynamic feature map; Built-in improved normalization calculation unit, calling the improved normalization formula The system normalizes each original feature to eliminate dimensional differences between features and outputs the normalized feature values ​​to the reinforcement learning and compatibility calculation module. Simultaneously, it receives feature importance correction coefficients from the reinforcement learning and compatibility calculation module. Dynamically adjust the normalized calculation parameters; The reinforcement learning and compatibility calculation module is connected to the adaptive normalization module and the data feedback and model optimization module, respectively, and receives normalized feature values. The built-in reinforcement learning unit, based on historical debugging data and real-time running status, uses iterative formulas... Dynamically iteratively update the weight coefficients of each feature Ensure that the weighting coefficients are appropriate for the current debugging scenario; Built-in compatibility calculation unit, calling the overall compatibility formula Calculate the overall compatibility between the application software and the current device, and output the feature importance correction coefficient. The overall compatibility is then output to the adaptive normalization module and then to the dynamic decision-making and hierarchical debugging module. The anomaly fingerprint extraction and monitoring module monitors the running status of application software on heterogeneous devices in real time, capturing various runtime anomalies such as crashes, freezes, interface call failures, and memory overflows; it also has a built-in anomaly fingerprint extraction unit that extracts the unique feature fingerprints of each type of anomaly event and establishes an anomaly fingerprint database. Built-in abnormal total index calculation unit, calling the abnormal total index formula ; Based on the severity coefficient of anomalies Frequency of occurrence and range influence coefficient Calculate the total anomaly index during runtime, and output the anomaly fingerprint information and the total anomaly index to the dynamic decision-making and hierarchical debugging module; The dynamic decision-making and hierarchical debugging module communicates with the reinforcement learning and compatibility calculation module, the anomaly fingerprint extraction and monitoring module, and the multimodal feature acquisition and fusion module, respectively, to receive overall compatibility, runtime anomaly index, anomaly fingerprint information, and device dynamic feature maps; it also has a built-in dynamic confidence calculation unit that calls the dynamic confidence decision model. The dynamic adaptation coefficient is determined by combining the complexity of the device feature map. Calculate the confidence level of the debugging decision. Built-in hierarchical decision-making unit, which executes corresponding debugging strategies based on confidence intervals; The data feedback and model optimization module communicates with the dynamic decision-making and hierarchical debugging module, the reinforcement learning and compatibility calculation module, and the anomaly fingerprint extraction and monitoring module, respectively, to collect complete data of this debugging, including overall compatibility, root cause of anomalies, repair plan, debugging time and anomaly fingerprint information; The built-in data processing unit categorizes, organizes, and verifies the validity of debugging data; The built-in model update unit feeds back the organized debugging data to the reinforcement learning model, updates the weight iteration strategy, and supplements and improves the anomaly fingerprint database, enabling continuous optimization of the reinforcement learning model and the anomaly fingerprint database.

9. The application software compatibility debugging device for heterogeneous devices according to claim 7, characterized in that: It also includes a storage module and a communication module; the storage module is used to store the device's dynamic feature map, historical debugging data, anomaly fingerprint database, weight coefficient iteration records, and debugging reports; The communication module enables data interaction between all modules, as well as communication with heterogeneous devices and external debugging terminals, supporting cross-device debugging data sharing and collaborative repair command transmission.

Images