Underwater robot battery life prediction method and device and storage medium

Abstract

The invention discloses an underwater robot battery life prediction method and device and a storage medium, and belongs to the technical field of battery health state monitoring. The method comprises the following steps: obtaining current and voltage data in the charging and discharging process of an underwater robot battery through an ampere-hour integral method; calculating to obtain electric quantity data of the battery in a set charging and discharging period; according to the voltage data and the electric quantity data of the battery, generating an electric quantity voltage curve image, a differential voltage curve image and a differential electric quantity curve image of the battery in a set discharge period, and performing three-channel image superposition; inputting the superposed image into a pre-trained and improved RepVGG neural network prediction model to obtain an underwater robot battery life prediction result; according to the invention, on one hand, the problem of insufficient feature extraction of the collected data in the battery life prediction process is solved; and on the other hand, the problems of low efficiency and low accuracy of the existing battery life prediction method are solved.

Description

Technical Field

[0001] The present invention relates to a method, device and storage medium for predicting the battery life of an underwater robot, and belongs to the technical field of battery health status monitoring. Background Art

[0002] Underwater robots are widely used in fields such as marine resource exploration, environmental monitoring, military reconnaissance, and underwater rescue and engineering operations. The battery endurance of an underwater robot is crucial, and the stability of battery performance is directly related to whether the underwater robot can complete long-term operation tasks in complex environments. However, directly judging the battery life will bring adverse effects. For example, premature battery replacement will lead to waste of resources, and incorrect prediction of battery life will cause the device to suddenly lose power at a critical moment. Therefore, accurate battery life prediction can achieve preventive maintenance, accurately evaluate the health status, and reasonably arrange battery replacement or repair, thereby reducing failures, lowering costs and increasing benefits, and avoiding safety accidents caused by battery damage.

[0003] Traditional battery life prediction can be divided into experience-based methods and performance-based methods, but they often have the following limitations: long test cycles, frequent and cumbersome data collection, large influence of environmental factors on results, limited prediction accuracy, etc. Moreover, currently, the battery life prediction methods developed using traditional feature extraction methods and data-driven methods have problems such as large computational amounts, large memory occupancy, poor accuracy and efficiency, and may be limited by hardware performance. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, device and storage medium for predicting the battery life of an underwater robot. On the one hand, it solves the problem of insufficient feature extraction of the collected data during the battery life prediction process; on the other hand, it solves the problems of low efficiency and low accuracy of the existing battery life prediction methods.

[0005] To achieve the above object, the present invention is implemented by the following technical solutions:

[0006] In the first aspect, the present invention provides a method for predicting the battery life of an underwater robot, including:

[0007] Calculating the power data of the battery during charge and discharge through the ampere-hour integration method for the current and voltage data obtained during the charge and discharge process of the underwater robot battery;

[0008] Generating a power-voltage curve image, a differential voltage curve image, and a differential power curve image within a set discharge period of the battery based on the voltage data and power data of the battery;

[0009] Perform three-channel image superposition on the power-voltage curve image, differential voltage curve image, and differential charge curve image;

[0010] Input the superimposed image into a pre-trained and improved RepVGG neural network prediction model to obtain the underwater robot battery life prediction result.

[0011] Furthermore, the current and voltage data during the battery discharge process of the underwater robot are obtained, and the charge data of the battery within a set charge-discharge cycle is calculated through the ampere-hour integration method, including:

[0012] Perform charge-discharge cycles on the battery until the rated capacity of the battery is reduced to a set percentage range of the initial capacity;

[0013] Use a data collector to read the current and voltage data of the current sensor and voltage sensor, where the current sensor and voltage sensor are connected in parallel or series in the underwater robot battery circuit;

[0014] Calculate the charge data during the battery charge-discharge process based on the current and voltage data according to the ampere-hour integration method, and store the current, voltage, and charge data in a Secure Digital card.

[0015] Furthermore, the current and voltage data are transmitted through low-frequency electromagnetic waves or ultrasonic technology.

[0016] Furthermore, the method further includes: preprocessing the current, voltage, and charge data, specifically including:

[0017] Use the Gaussian filtering method to remove the noise components of the current, voltage, and charge data, and then reduce the high-frequency noise and fluctuations in the data through the weighted average method;

[0018] Perform normalization processing according to the distribution of the current, voltage, and charge data, and scale the values of the data to a set range to complete the preprocessing.

[0019] Furthermore, the three-channel image superposition of the power-voltage curve image, differential voltage curve image, and differential charge curve image includes:

[0020] Assign the power-voltage curve image, differential voltage curve image, and differential charge curve image to the red, green, and blue channels of the RGB image, and then superimpose them into a new image.

[0021] Furthermore, the improvement method of the RepVGG neural network prediction model is: increase the number of residual branches and small convolutional kernels in the original RepVGG neural network, introduce data augmentation and regularization techniques, and use a loss function that combines mean absolute error and mean square error.

[0022] In a second aspect, the present invention provides an underwater robot battery life prediction device, comprising:

[0023] An acquisition module, configured to calculate the power data of the battery during charge and discharge through the ampere-hour integration method based on the acquired current and voltage data during the discharge process of the underwater robot battery;

[0024] A generation module, configured to generate a power-voltage curve image, a differential voltage curve image, and a differential power curve image within a set discharge period of the battery based on the voltage data and power data of the battery;

[0025] An overlay module, configured to perform three-channel image overlay on the power-voltage curve image, the differential voltage curve image, and the differential power curve image;

[0026] A prediction module, configured to input the overlaid image into a pre-trained and improved RepVGG neural network prediction model to obtain the prediction result of the underwater robot battery life.

[0027] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in any one of the foregoing are implemented.

[0028] In a fourth aspect, the present invention provides a computer device, comprising:

[0029] A memory, configured to store computer programs / instructions;

[0030] A processor, configured to execute the computer programs / instructions to implement the steps of the method described in any one of the foregoing.

[0031] In a fifth aspect, the present invention provides a computer program product, comprising computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the method described in any one of the foregoing are implemented.

[0032] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0033] 1. The present invention provides a method, device, and storage medium for predicting the battery life of an underwater robot. By using a current sensor and a voltage sensor to obtain the current data and voltage data during the early discharge process of the battery, and calculating the power data according to the ampere-hour integration method, and then obtaining the power-voltage curve, differential voltage curve, and differential power curve of the battery. By performing multi-channel overlay on the three groups of curves of the battery and training the RepVGG neural network prediction model to predict the battery life, the efficiency and accuracy of battery life prediction are improved;

[0034] 2. Compared with the traditional data-driven battery life prediction method, the present invention uses a multi-channel superposition method to construct image features, provides comprehensive feature inputs by fusing multi-dimensional features, and provides more information for subsequent further processing using an improved deep learning model.

[0035] 3. Compared with the traditional experience-based or model-based battery life prediction methods, the RepVGG neural network prediction model of the present invention can better process complex data features and improve the accuracy of life prediction. During the model training process, an effective new loss function is used. This new loss function combines the mean absolute error and the mean square error, can more flexibly handle large errors and small errors, and uses the root mean square error and the coefficient of determination to evaluate the performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flowchart of a method for predicting the battery life of an underwater robot provided by an embodiment of the present invention;

[0037] Figure 2 is a structural diagram of a device for predicting the battery life of an underwater robot provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and cannot be used to limit the protection scope of the present invention.

[0039] Embodiment 1. This embodiment introduces a method for predicting the battery life of an underwater robot, including:

[0040] The current and voltage data during the discharge process of the underwater robot battery are obtained, and the battery power data during the charge and discharge process are calculated by the ampere-hour integration method;

[0041] According to the voltage data and power data of the battery, a power-voltage curve image, a differential voltage curve image, and a differential power curve image within the set discharge period of the battery are generated;

[0042] The power-voltage curve image, the differential voltage curve image, and the differential power curve image are subjected to three-channel image superposition;

[0043] The superimposed image is input into a pre-trained and improved RepVGG neural network prediction model to obtain the prediction result of the battery life of the underwater robot.

[0044] Before introducing the embodiments of the present invention, the following interpretations are made for the relevant terms involved in the embodiments of the present invention:

[0045] The Ampere-hour integration method is a method for estimating the battery capacity by integrating the current flowing through the battery during the charging and discharging process. The core idea of this method is that according to the discharge current (in amperes, A) of the battery within a specific time period, the cumulative discharge amount of the battery is calculated through time integration, usually expressed in ampere-hours (Ah). The basic formula of the Ampere-hour integration method is: , where: is the cumulative discharge amount (Ah), is the discharge current (A) at the th measurement, is the time interval (h) between two measurements. Specifically, during the battery discharge process, the change of current with time is recorded. The Ampere-hour integration method accumulates the product of current and time to obtain the battery capacity at different time nodes.

[0046] RepVGG (Reparameterized VGG) is an improved convolutional neural network that improves the inference efficiency by using different structures in the training and inference stages. During the training stage, RepVGG uses multiple branches and complex convolutional operations to learn rich features. However, in the inference stage, these complex structures are "reparameterized" into a simple convolutional layer, thus significantly improving the inference speed. This method enables RepVGG to maintain high accuracy while significantly reducing the computational amount, especially suitable for devices with limited resources.

[0047] The Mean Absolute Error (MAE for short) is a commonly used evaluation metric for regression models, which is used to measure the difference between the predicted values and the true values of the model. It is defined as the average of the absolute differences between all predicted values and the true values. The formula for calculating the Mean Absolute Error is , where is the true value of the th observation, is the predicted value, and the size of the dataset is .

[0048] The Mean Squared Error (MSE for short) is used to measure the difference between the predicted values and the true values of the model. It reflects the prediction accuracy of the model by calculating the square of the difference between the predicted value and the actual value and averaging over all samples. The formula for calculating the Mean Squared Error is , where is the true value of the th observation, is the predicted value, and the size of the dataset is .

[0049] To solve the above problems, the present invention provides an underwater robot battery life prediction method and an underwater monitoring system based on multi-channel image features and an improved deep learning network. A current sensor is installed in series on the output loop of the battery pack, and a voltage sensor is installed in parallel between the positive and negative electrodes of each battery or battery pack to capture voltage data and current data during the charging and discharging process of the battery. Then, the ampere-hour integration method is used to calculate the power data during the charging and discharging process, and the Gaussian filtering method is used to remove noise components to ensure the smoothness and reliability of the data. Then, according to the distribution of the data, normalization processing is performed to eliminate the differences between different dimensions and ensure the consistency of the scales of each data. Then, the power-voltage curve, differential voltage curve, and differential power curve of each battery are generated, and a dataset is generated by multi-channel superimposition of the three sets of curve images of each battery. The dataset is input into the RepVGG neural network, and more local features of the image are obtained by increasing the number of residual branches and small convolution kernels. At the same time, data augmentation and regularization techniques are introduced to improve the generalization ability of the model, and the accuracy of battery life prediction is improved through an improved loss function. Through a set of matching underwater monitoring systems, real-time life prediction of the battery is achieved. Next, it will be described in conjunction with the accompanying drawings.

[0050] The underwater robot battery life prediction method provided in this embodiment is as Figure 1 shown, and its application process specifically involves the following steps:

[0051] Step 1: Sensor installation. All sensors and connection components are designed to be waterproof to ensure that the device can operate stably in the underwater environment. The selected current sensor model is Waveshare High Precision Current Sensor Module, which can provide a sampling rate of up to 1.1k SPS and can be flexibly adjusted according to different application requirements. It is installed in series on the output loop of the battery pack through an electrical connection method; the selected voltage sensor model is Adafruit ADS1115, which can provide a sampling rate of up to 860 SPS and can be flexibly adjusted according to different application requirements. It is installed in parallel between the positive and negative electrodes of each battery or battery pack through a terminal connector. The shells of the current sensor and the voltage sensor are sealed with high-strength waterproof materials (such as stainless steel, aluminum alloy, or corrosion-resistant plastic) to prevent water from entering the inside of the sensor and causing damage. At the same time, all electrical interfaces are sealed with waterproof gaskets, rubber gaskets, or silicone sealing materials to ensure that even in a high-pressure underwater environment, current short circuits or sensor failures can be avoided.

[0052] Step 2: Acquisition of current signal and voltage signal. The battery is subjected to charge and discharge cycles until the rated capacity of the battery is reduced to 80% of the initial capacity. To cope with signal attenuation and noise interference in the underwater environment, a data collector of model Arduino Mega2560 is used to read the data of the current sensor and voltage sensor, and data transmission is carried out through low-frequency electromagnetic wave or ultrasonic technology. According to the ampere-hour integration method, the electricity quantity data during the charge and discharge process of the battery is calculated, and the collected data is stored in a Secure Digital (SD) card of model Lexar 633x 128GB SDXC UHS-I.

[0053] Step 3: Data preprocessing. The collected current and voltage data and the electricity quantity data obtained according to the ampere-hour integration method are smoothed; the Gaussian filtering method is used to remove the noise components, and through the weighted average method, the high-frequency noise and fluctuations in the signal are reduced, so as to ensure the smoothness and reliability of the data. Then, according to the distribution of the data, normalization processing is carried out to scale the values of the data to a unified range to eliminate the differences between different dimensions and ranges, thereby further reducing the influence caused by the differences in different feature scales.

[0054] Step 4: Feature extraction. After data preprocessing, first, according to the voltage data and electricity quantity data of each battery, the electricity quantity-voltage curve images, differential voltage curve images, and differential electricity quantity curve images of the first 100 discharge cycles of each battery are generated respectively; then, the three images of each battery are assigned to the red, green, and blue channels of the RGB image and superimposed into a new image to generate a dataset for battery life prediction. In this way, the data information of the voltage-electricity quantity curve image, differential voltage curve image, and differential electricity quantity curve image is effectively embedded into a multi-channel color image, so that the features of each image are presented in a visual image. In this way, the original single-channel images are combined into an RGB image, so that their relative changes can be intuitively displayed by colors, which helps to visually display multiple data features at the same time.

[0055] Step 5: Train the model. Based on the dataset generated from the already extracted image features, select the RepVGG neural network for training; add residual branches to the RepVGG neural network. By introducing short connections between multiple layers in the network, information can bypass some layers during forward propagation and be directly passed to subsequent layers, thus retaining more useful feature information and alleviating the vanishing gradient problem; by increasing the number of small convolutional kernels in the neural network, more local details of the input image features can be obtained, such as the tiny voltage fluctuations during the battery charging and discharging process; at the same time, introduce data augmentation and regularization techniques to improve the generalization ability of the model; use a loss function that combines mean absolute error and mean squared error to flexibly handle large and small errors and improve the accuracy of battery life prediction.

[0056] Step 6: Deploy and integrate the model. Save the trained model in an appropriate format. The saving process ensures that the structure and weights of the model are correctly stored for easy invocation during the inference stage; to cope with the underwater environment, the edge device is connected to the underwater robot in real time through a wireless communication module (such as Wi-Fi, LoRa, Bluetooth, etc.) to continuously monitor the operating state of the battery. The STM32 MCU can collect data such as current and voltage in real time and perform inference and analysis through the embedded battery life prediction model to predict the battery life. The prediction results and real-time data are then uploaded to the cloud or remote control platform through the wireless communication module. Through the cloud platform, the system can centrally manage and analyze the operating data of the underwater robot battery, not only storing historical data but also regularly optimizing the battery life prediction model. The computing power of the cloud enables the system to process a large amount of historical data and continuously adjust and refine the prediction algorithm based on this data, thereby improving the accuracy and reliability of battery life prediction. Over time, through continuous data updates and model training, the system can effectively address the challenges of battery life prediction under different environments and usage conditions and ensure the continuous improvement of prediction accuracy. In addition, based on the feedback from real-time monitoring and cloud analysis, the system can also trigger an alarm in a timely manner when the battery health status is abnormal, reminding the operator of possible battery failure risks or performance degradation, avoiding mission interruptions of the underwater robot caused by battery failures, and improving its work efficiency and reliability.

[0057] After obtaining the trained RepVGG neural network prediction model, the superimposed generated partial images are used as the input of the life prediction model, and the performance of the life prediction model is tested after training. The test set is divided into five groups to test the life prediction model, and the test results are shown in Table 1. It can be seen that by using the above battery life prediction method based on image features, the voltage and power curves, differential voltage curves, and differential power curves of each battery are superimposed in multiple channels, the number of residual branches and small convolutional kernels of the RepVGG neural network is increased, data augmentation and regularization techniques are introduced, and a loss function combining mean absolute error and mean square error is used to achieve the stability, accuracy, and real-time performance of battery life prediction. Then, multiple sets of comparative experiments are also carried out. In the first set of comparative experiments, different loss functions are used for comparison, and the results are shown in Table 2. In the second set of comparative experiments, the RepVGG neural network using the loss function combining mean absolute error and mean square error is compared with other convolutional neural networks, and the results are shown in Table 3.

[0058] Table 1 Neural Network Prediction Results

[0059]

[0060] Table 2 Prediction Results Using Different Loss Functions

[0061]

[0062] Table 3 Prediction Results of Different Neural Networks

[0063]

[0064] Specific embodiments can be combined with Figure 1 description, such as Figure 1The complete workflow of the implementation stage is shown in the figure. First, the current and voltage data of the first 100 cycles of the battery discharge process are obtained through the current sensor and voltage sensor. The data of the current sensor and voltage sensor are read using a data logger modeled as Arduino Mega 2560. The power data of each battery is obtained according to the ampere-hour integration method, and the collected data is stored in a secure digital (SD) card modeled as Lexar 633x 128GB SDXC UHS-I. Then, the Gaussian filtering method is used to remove the noise component, and the high-frequency noise and fluctuation in the signal are reduced by weighted averaging to ensure the smoothness and reliability of the data. Then, according to the distribution of the data, normalization is performed to scale the data values ​​to a uniform range to eliminate the differences between different dimensions and ranges, thereby further reducing the impact caused by the differences in different feature scales. Then, the voltage power curve image, differential voltage curve image, and differential power curve image of each battery are generated respectively; then, the three images of each battery are assigned to the red, green, and blue channels of the RGB image, superimposed into a new image, and a data set for battery life prediction is generated. Then, the generated data set is trained with a life prediction model using deep learning technology. The selected model is the RepVGG neural network, which has the advantages of high inference accuracy and high efficiency. The trained life prediction model is deployed on the edge device and restarted. In the running mode, the same preprocessing and image feature extraction methods as in the recording mode are used to perform real-time reasoning and analysis through the monitoring system to predict the battery life.

[0065] The above description mainly introduces the solution provided by the embodiment of the present invention from the perspective of the life prediction method process and the functions of each step. According to the above description, the implementation of the embodiment of the present invention involves hardware structure and software modules, and through the combination of the functions of each part, a method for predicting the battery life of an underwater robot is finally realized. For each specific application, technicians can flexibly choose different methods to complete the functions, but these implementation methods should all fall within the scope of the present invention.

[0066] Figure 2 A structural schematic diagram of a battery life prediction device for an underwater robot is given, and the device includes:

[0067] Sensor nodes are used to monitor the current and voltage data generated during battery discharge in real time. Sensor nodes need to have real-time data acquisition and processing capabilities to ensure that they can continuously and quickly respond to changes and process data; as well as low latency and high response characteristics to ensure that the system can promptly feedback changes in battery status; at the same time, they must have good anti-interference capabilities and clock synchronization functions to ensure that multi-sensor systems work together.

[0068] The data acquisition module is used to collect the current data and voltage data received from the sensor nodes in real time, providing input for the subsequent battery life prediction model. The main task of the data acquisition module is to collect key information about battery performance from various sensors and preprocess this information for accurate and timely analysis. Through this data acquisition module, the battery life prediction device can accurately monitor the battery status, providing strong data support to help predict and manage the battery health status.

[0069] The data processing module is used to receive the current data, voltage data, and power data from the data acquisition module, remove the noise components using the Gaussian filtering method, and reduce the high-frequency noise and fluctuations in the signal through weighted averaging to ensure the smoothness and reliability of the data. Then, according to the data distribution, normalization processing is performed to scale the data values to a unified range to eliminate the differences between different dimensions and ranges, further reducing the impact caused by different feature scale differences. Voltage-power curve images, differential voltage curve images, and differential power curve images of each battery are generated respectively. The three images of each battery are assigned to the red, green, and blue channels of the RGB image and superimposed into a new image to generate a dataset for battery life prediction. The data processing module needs to include data preprocessing algorithms specifically for battery life prediction and image feature extraction methods.

[0070] The battery life prediction model is used to predict the battery life. The model is trained and optimized based on the images and corresponding life labels in the training set. The trained model can predict the battery life based on the images generated from the battery data. The battery life prediction model needs to be fully trained and verified to ensure the accurate prediction ability for different types of battery life.

[0071] The test and display module is used to receive the prediction results output by the battery life prediction model and display them to the user or other devices. The trained battery life prediction model is deployed on the edge device for real-time monitoring of the battery health status. At the same time, this module can also provide data recording, alarm, and remote communication functions to help users monitor the battery status in real time and perform corresponding maintenance and processing.

[0072] In this embodiment, current data and voltage data during the early discharge process of the battery are obtained by using a current sensor and a voltage sensor, and the power data is calculated according to the ampere-hour integration method. Furthermore, the power-voltage curve, differential voltage curve, and differential power curve of each battery are obtained. By performing multi-channel superposition on the three groups of curves of each battery and training the RepVGG neural network model, the efficiency and accuracy of battery life prediction are improved. On the one hand, compared with the traditional data-driven battery life prediction method, this embodiment uses the multi-channel superposition method to construct image features, provides comprehensive feature inputs by fusing multi-dimensional features, and provides more information for subsequent further processing using the improved deep learning model. On the other hand, compared with the traditional experience-based or model-based battery life prediction methods, the deep learning model can better process complex data features and improve the accuracy of life prediction. During the model training process, an effective new loss function is used. This new loss function combines the mean absolute error and the mean square error, can more flexibly handle larger errors and smaller errors, and uses the root mean square error and the coefficient of determination to evaluate the performance of the model. And various waterproof and anti-interference devices are equipped, enabling battery life prediction underwater. In addition, this embodiment is also equipped with a set of monitoring systems to achieve real-time life prediction of the battery. Compared with the traditional battery management system, this monitoring system can monitor the state of the battery in real time, issue early warnings in a timely manner, prevent potential failures, and can also be remotely monitored and maintained through the cloud platform, reducing the maintenance cost and difficulty. The trained deep learning model can be used as a life prediction model to perform life prediction based on the voltage and power data of the battery waiting to be predicted.

[0073] Embodiment 2. This embodiment provides an underwater robot battery life prediction device, including:

[0074] An acquisition module, configured to calculate the power data of the battery during charging and discharging through the ampere-hour integration method for the current and voltage data obtained during the discharge process of the underwater robot battery.

[0075] A generation module, configured to generate a power-voltage curve image, a differential voltage curve image, and a differential power curve image within a set discharge period of the battery according to the voltage data and power data of the battery.

[0076] A superposition module, configured to perform three-channel image superposition on the power-voltage curve image, the differential voltage curve image, and the differential power curve image.

[0077] A prediction module, configured to input the superimposed image into a pre-trained and improved RepVGG neural network prediction model to obtain the underwater robot battery life prediction result.

[0078] For the specific function implementations of the above-mentioned modules, refer to the relevant content in the method of Embodiment 1, which will not be elaborated here.

[0079] Embodiment 3 provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the method described in any one of Embodiment 1 are implemented.

[0080] Embodiment 4 provides a computer device, including:

[0081] A memory for storing computer programs / instructions;

[0082] A processor for executing the computer programs / instructions to implement the steps of the method described in any one of Embodiment 1.

[0083] Embodiment 5 provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the method described in any one of Embodiment 1 are implemented.

[0084] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

[0085] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, the present disclosure can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0086] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0087] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more processes and / or blocks Figure 1 in one or more processes and / or blocks Figure 1 specified in the function.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operation steps are performed on the computer or other programmable apparatus to generate a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more processes and / or blocks Figure 1 in one or more processes and / or blocks Figure 1 specified in the function.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than limit the scope of its protection. Although the present disclosure has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent replacements to the specific implementation manners of the invention, but these changes, modifications or equivalent replacements are all within the scope of protection of the pending claims of the disclosure.

Claims

1. A method for predicting the battery life of an underwater robot, characterized in that: include: The current and voltage data obtained during the discharge process of the underwater robot battery are used to calculate the battery power data during the charging and discharging process through the ampere-hour integration method; Generate a voltage curve image, a differential voltage curve image and a differential charge curve image of the battery in a set charge and discharge cycle according to the voltage data and charge data of the battery; Superimposing the three-channel images of the charge voltage curve image, the differential voltage curve image and the differential charge curve image; The superimposed image is input into the pre-trained and improved RepVGG neural network prediction model to obtain the underwater robot battery life prediction result.

2. The underwater robot battery life prediction method according to claim 1, characterized in that: The current and voltage data obtained during the discharge process of the underwater robot battery are calculated by the ampere-hour integration method to obtain the power data of the battery during the charging and discharging process, including: The battery is charged and discharged in cycles until the rated capacity of the battery is reduced to a set percentage range of the initial capacity; Using a data collector to read current and voltage data of a current sensor and a voltage sensor, wherein the current sensor and the voltage sensor are connected in parallel or in series in a battery circuit of the underwater robot; The current and voltage data are calculated according to the ampere-hour integration method to obtain the power data during the battery charging and discharging process, and the current, voltage and power data are stored in a secure digital card.

3. The underwater robot battery life prediction method according to claim 2, characterized in that: The current and voltage data are transmitted via low-frequency electromagnetic waves or ultrasonic technology.

4. The underwater robot battery life prediction method according to claim 1, characterized in that: The method further includes: preprocessing the current, voltage and power data, specifically including: Use Gaussian filtering to remove noise components from current, voltage and power data, and then use weighted averaging to reduce high-frequency noise and fluctuations in the data. According to the distribution of current, voltage and power data, normalization processing is performed to scale the data values ​​to the set range to complete the preprocessing.

5. The underwater robot battery life prediction method according to claim 1, characterized in that: The three-channel image superposition of the electric quantity voltage curve image, the differential voltage curve image and the differential electric quantity curve image comprises: The charge voltage curve image, the differential voltage curve image and the differential charge curve image are assigned to the red, green and blue channels of the RGB image, and then superimposed into a new image.

6. The underwater robot battery life prediction method according to claim 1, characterized in that: The improvement method of the RepVGG neural network prediction model is: increasing the number of residual branches and small convolution kernels in the original RepVGG neural network, introducing data enhancement and regularization techniques, and using a loss function that combines mean absolute error and mean square error.

7. A battery life prediction device for an underwater robot, characterized in that: include: An acquisition module is used to calculate the electric quantity data of the battery during the charging and discharging process by using the ampere-hour integration method to obtain the current and voltage data of the underwater robot battery during the discharging process; A generating module, used for generating a voltage curve image of the battery quantity, a differential voltage curve image and a differential charge curve image within a set discharge cycle of the battery according to the voltage data and charge data of the battery; A superposition module is used to perform three-channel image superposition of an electric quantity voltage curve image, a differential voltage curve image and a differential electric quantity curve image; The prediction module is used to input the superimposed image into the pre-trained and improved RepVGG neural network prediction model to obtain the underwater robot battery life prediction result.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: include: Memory, for storing computer programs / instructions; A processor, configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images