Deep sea sampling intelligent control method and system

Abstract

The invention relates to a deep sea sampling intelligent control method and system, and the method comprises the steps: obtaining a sampling demand, and presetting a sampling task script according to the sampling demand; according to a sampling task script, a sampling task queue is constructed, and the sampling task script comprises sampling depth, priority, sampling number and sensor configuration; the method comprises the following steps: periodically collecting marine environment data, and preprocessing the environment data to obtain preprocessed multi-parameter monitoring data; inputting the preprocessed multi-parameter monitoring data into a pre-trained environment perception model for analysis, outputting a sampling task triggering judgment result, and dynamically adjusting a sampling task queue according to the sampling task triggering judgment result; and in the sampling execution process, recording operation parameters, environment parameters and timestamp information of each sampling task, and generating environment labels corresponding to the collected samples. The deep sea sampling device has the effect of improving the stability of deep sea sampling operation.

Description

Technical Field

[0001] This application relates to the technical field of marine exploration and intelligent control, and in particular to an intelligent control method and system for deep-sea sampling. Background Technology

[0002] Currently, with the continuous deepening of deep-sea resource surveys and marine ecological environment research, automated sampling technology is increasingly widely used in deep-sea scientific exploration, especially in acquiring target data such as seawater physicochemical parameters, microbial samples, and pollutant residues. Sampling platforms are typically equipped with multiple sensors and sampling devices to perform periodic or responsive sampling operations under different depths, locations, and environmental conditions to obtain representative samples for subsequent analysis.

[0003] Existing deep-sea sampling control schemes are mainly based on static task scheduling and preset sampling logic. That is, a fixed sampling time, depth and sequence are preset before the task is issued, and sampling instructions are executed according to a fixed process. They lack the ability to dynamically perceive changes in the actual environment and are difficult to respond to sudden changes in the marine environment or prominent target events in a timely manner.

[0004] The existing technical solutions mentioned above have the following drawbacks: the existing deep-sea sampling control methods rely on static rules and lack dynamic decision-making and intelligent scheduling capabilities based on multi-source sensor data. They cannot adaptively adjust the task queue and sampling strategy according to real-time ocean environmental fluctuations, which affects the stability of deep-sea sampling operations. Therefore, there is room for improvement. Summary of the Invention

[0005] To improve the stability of deep-sea sampling operations, this application provides an intelligent control method and system for deep-sea sampling.

[0006] The above-mentioned objective of this application is achieved through the following technical solution: A deep-sea sampling intelligent control method, the deep-sea sampling intelligent control method comprising: Obtain sampling requirements and preset a sampling task script based on the sampling requirements; Based on the sampling task script, a sampling task queue is constructed, wherein the sampling task script includes sampling depth, priority, sampling quantity and sensor configuration; Marine environmental data is periodically collected and preprocessed to obtain preprocessed multi-parameter monitoring data. The preprocessed multi-parameter monitoring data is then input into a pre-trained environmental perception model for analysis, and a sampling task trigger determination result is output. The sampling task queue is dynamically adjusted based on the sampling task trigger determination result. During the sampling process, the operation parameters, environmental parameters, and timestamp information of each sampling task are recorded, and an environmental tag corresponding to the collected sample is generated. The environmental tag includes the original sensor value at the time of sampling, the sampling depth, and the device status information. The communication link status of the sampling platform is monitored in real time, and a control mode switching operation is performed based on the communication link status. When all communication channels are detected to be unavailable, local emergency control measures are triggered. The local emergency control measures include interrupting the sampling task, retracting the robotic arm, sealing the data, and putting the platform on standby.

[0007] By adopting the above technical solutions, and by acquiring sampling requirements and pre-setting sampling task scripts, the target area, depth range, sample quantity, and parameter requirements for sampling can be clearly defined before task execution, thus ensuring that the sampling task has a clear target orientation and operational basis. By constructing a sampling task queue and setting priorities and sensor configurations, orderly scheduling and precise execution of multiple tasks can be achieved, thereby improving the resource utilization efficiency and task response capability of deep-sea sampling. By periodically collecting and preprocessing marine environmental data, changes in the deep-sea environment can be perceived in a timely manner, and the quality of input data can be ensured, thereby enhancing the accuracy of subsequent model analysis and judgment. By inputting preprocessed data to the environmental perception model to output sampling task trigger judgment results, and dynamically adjusting the task queue accordingly, intelligent decisions on whether to execute sampling tasks can be made based on real-time environmental characteristics, thereby effectively responding to dynamic environmental changes and improving sample representativeness. By recording the operational parameters and environmental parameters of the sampling task to generate environmental tags, complete binding between samples and sampling background information can be achieved, thereby improving the scientific research value of sample data. By monitoring the status of the communication link in real time and triggering local emergency control measures when interrupted, the safety and task integrity of the sampling platform in extreme environments can be ensured, thereby improving the robustness and self-recovery capability of the overall system.

[0008] In one example, this application can be further configured such that: the step of obtaining sampling requirements and pre-setting a sampling task script based on the sampling requirements includes: Receive user input, which includes the target sampling area, sampling depth range, number of samples, and monitoring parameter requirements; A sampling task script is generated based on the input content. The sampling task script includes a task identifier, target sampling depth, task priority, sampling method, and sensor configuration.

[0009] By adopting the above technical solution, and by receiving user input and generating sampling task scripts, the specific requirements of the sampling task can be flexibly defined according to the user's intention, including the target area, sampling depth, sample quantity and monitoring indicators, thereby enhancing the controllability and personalization of the sampling task configuration and ensuring that the system has the ability to customize tasks to meet different scientific research needs.

[0010] In one example, this application can be further configured such that constructing the sampling task queue according to the sampling task script includes: The sampling task script contains multiple sampling task items, and the sampling depth, priority level and execution time window corresponding to each sampling task item are extracted. The sampling tasks are sorted according to the priority level to generate an initial task queue; Based on the similarity of the sampling depth and the overlap of the execution time window, the initial task queue is optimized and adjusted to obtain the sampling task queue. The optimization and adjustment include merging duplicate tasks and rearranging the task order.

[0011] By adopting the above technical solution, and by parsing the sampling task script and extracting the sampling depth, priority level, and execution time window from the task items, the parameterized representation and task granularity of the sampling task can be realized, which facilitates the system's structured management of task content. By sorting and optimizing the task items, redundant operations can be avoided and execution efficiency can be improved, thereby achieving time resource coordination and spatial coverage optimization for deep-sea sampling operations.

[0012] In one example, this application can be further configured as follows: the periodic collection of marine environmental data and the preprocessing of the environmental data to obtain preprocessed multi-parameter monitoring data include: Environmental sensor data is collected at a preset time period to construct a multidimensional time series. The environmental sensor data includes dissolved oxygen, temperature, pH, conductivity, and pressure. The multidimensional time series is processed based on a sliding window mechanism, and outlier removal and noise filtering algorithms are used for data cleaning. Normalization is then performed to obtain the preprocessed multi-parameter monitoring data.

[0013] By adopting the above technical solutions, and by collecting environmental sensor data at a preset period and constructing a multidimensional time series, it is possible to ensure that environmental information is continuous and comprehensive, thereby providing basic support for subsequent trend analysis and model input. Through sliding window processing, multidimensional anomaly removal, and normalization operations, data quality can be improved and noise caused by sensor fluctuations or external interference can be eliminated, thereby improving the stability and accuracy of subsequent analysis models.

[0014] In one example, this application can be further configured such that: the deep-sea sampling intelligent control method also includes: Historical environmental sensor data and corresponding sampling task execution records are acquired to construct a training sample set. Each set of samples in the training sample set includes a multi-cycle sensor parameter sequence and a corresponding sampling trigger marker. A multi-layer neural network structure is used to construct a model framework, which includes a feature extraction layer, a state analysis layer, and a decision output layer. The training sample set is input into the model framework and trained based on a supervised learning algorithm, so that the model can predict whether to trigger a sampling task based on the current environmental data, thereby obtaining the pre-trained environmental perception model.

[0015] By adopting the above technical solutions, a training sample set can be constructed by acquiring historical environmental data and sampling task records, forming a data foundation with real labels and improving the credibility and applicability of model training. By constructing a multi-layer neural network model that includes feature extraction, state analysis and judgment output and training it based on supervised learning algorithms, the model can grasp the deep-level relationship between the environment and task triggering, thereby achieving high-precision perception and trigger judgment of the real-time environment and providing intelligent support for sampling decisions.

[0016] In one example, this application can be further configured as follows: inputting the preprocessed multi-parameter monitoring data into a pre-trained environmental perception model for analysis, outputting a sampling task trigger determination result, and dynamically adjusting the sampling task queue based on the sampling task trigger determination result includes: The preprocessed multi-parameter monitoring data is constructed into a sequence of sensor feature vectors for multiple consecutive periods, and input into the pre-trained environmental perception model for inference analysis to obtain the sampling task trigger determination result. The sampling task trigger determination result includes a classification result of whether the sampling task should be triggered and the corresponding trigger probability score. When the trigger probability score is higher than the preset trigger threshold, a sampling task of the corresponding depth is inserted into the sampling task queue; when the trigger probability score is lower than the preset trigger threshold and the trigger condition is not met for multiple consecutive cycles, the target task in the original queue is marked as skippable or delayed for execution, so as to dynamically optimize the resource configuration of the sampling task.

[0017] By adopting the above technical solutions, and constructing a sensor feature vector sequence from preprocessed multi-parameter monitoring data and inputting it into the model for analysis, the model's temporal understanding and judgment accuracy of environmental states can be improved, thereby enhancing the scientific nature of sampling decisions. By setting a trigger probability score and comparing it with a preset trigger threshold, it is possible to accurately determine whether a sampling task should be triggered, thereby achieving responsive sampling under critical environmental events. By skipping or postponing tasks when the score is low and there are multiple consecutive periods without triggering, platform resources can be saved, unnecessary sampling costs can be reduced, thereby improving the overall execution efficiency of the task queue and the representativeness of the samples.

[0018] In one example, this application can be further configured such that: the deep-sea sampling intelligent control method also includes: During the execution of the sampling task, the environmental fluctuation index is calculated based on the trend of environmental parameter changes within a continuous sampling period; when the environmental fluctuation index exceeds a preset stability threshold, the preset trigger threshold is reduced. When the environmental fluctuation index is lower than the preset stability threshold, the preset trigger threshold is increased.

[0019] By adopting the above technical solution, the environmental fluctuation index can be calculated based on the continuous periodic environmental parameter change trend during the sampling process, thereby quantifying the stability of the current environment and providing a dynamic basis for threshold adjustment. By lowering the trigger threshold when the fluctuation index is high and raising the threshold when the fluctuation index is low, the trigger sensitivity of the model can be adaptively adjusted, thereby effectively balancing the timeliness of sampling and resource utilization, and improving the system's intelligent control capability under complex sea conditions.

[0020] The second objective of this invention is achieved through the following technical solution: A deep-sea sampling intelligent control system, comprising: The sampling task generation module is used to obtain sampling requirements and preset sampling task scripts based on the sampling requirements; The queue construction module is used to construct a sampling task queue according to the sampling task script, wherein the sampling task script includes sampling depth, priority, sampling quantity and sensor configuration; An environmental data processing module is used to periodically collect marine environmental data and preprocess the environmental data to obtain preprocessed multi-parameter monitoring data. The trigger analysis module is used to input the preprocessed multi-parameter monitoring data into a pre-trained environmental perception model for analysis, output the sampling task trigger determination result, and dynamically adjust the sampling task queue according to the sampling task trigger determination result. The tag recording module is used to record the operation parameters, environmental parameters and timestamp information of each sampling task during the sampling process, and generate environmental tags corresponding to the collected samples. The environmental tags include the original sensor values ​​at the time of sampling, sampling depth and device status information. The emergency control module is used to monitor the communication link status of the sampling platform in real time and perform control mode switching operations based on the communication link status. When it is detected that all communication channels are unavailable, local emergency control measures are triggered, including sampling task interruption, robotic arm retraction, data sealing and platform standby.

[0021] By adopting the above technical solutions, and by acquiring sampling requirements and pre-setting sampling task scripts, the target area, depth range, sample quantity, and parameter requirements for sampling can be clearly defined before task execution, thus ensuring that the sampling task has a clear target orientation and operational basis. By constructing a sampling task queue and setting priorities and sensor configurations, orderly scheduling and precise execution of multiple tasks can be achieved, thereby improving the resource utilization efficiency and task response capability of deep-sea sampling. By periodically collecting and preprocessing marine environmental data, changes in the deep-sea environment can be perceived in a timely manner, and the quality of input data can be ensured, thereby enhancing the accuracy of subsequent model analysis and judgment. By inputting preprocessed data to the environmental perception model to output sampling task trigger judgment results, and dynamically adjusting the task queue accordingly, intelligent decisions on whether to execute sampling tasks can be made based on real-time environmental characteristics, thereby effectively responding to dynamic environmental changes and improving sample representativeness. By recording the operational parameters and environmental parameters of the sampling task to generate environmental tags, complete binding between samples and sampling background information can be achieved, thereby improving the scientific research value of sample data. By monitoring the status of the communication link in real time and triggering local emergency control measures when interrupted, the safety and task integrity of the sampling platform in extreme environments can be ensured, thereby improving the robustness and self-recovery capability of the overall system.

[0022] In summary, this application includes the following beneficial technical effects: 1. By acquiring sampling requirements and pre-setting sampling task scripts, the target area, depth range, number of samples, and parameter requirements for sampling can be clearly defined before the task is executed, thereby ensuring that the sampling task has a clear target orientation and operational basis; by constructing a sampling task queue and setting priorities and sensor configurations, the orderly scheduling and fine execution of multiple tasks can be achieved, thereby improving the resource utilization efficiency and task response capability of deep-sea sampling. 2. By periodically collecting and preprocessing marine environmental data, we can promptly perceive changes in the deep-sea environment and ensure the quality of input data, thereby enhancing the accuracy of subsequent model analysis and judgment. By inputting preprocessed data into the environmental perception model to trigger the sampling task judgment result and dynamically adjusting the task queue accordingly, we can intelligently decide whether to execute the sampling task based on real-time environmental characteristics, thus effectively responding to dynamic environmental changes and improving sample representativeness. By recording the operational parameters and environmental parameters of the sampling task to generate environmental tags, we can achieve complete binding between the sample and the sampling background information, thereby improving the scientific research value of the sample data. 3. By monitoring the status of the communication link in real time and triggering local emergency control measures when interrupted, the safety and mission integrity of the sampling platform can be guaranteed in extreme environments, thereby improving the robustness and self-recovery capability of the overall system. Attached Figure Description

[0023] Figure 1This is a flowchart of a deep-sea sampling intelligent control method according to one embodiment of this application; Figure 2 This is a flowchart illustrating the implementation of step S10 in a deep-sea sampling intelligent control method according to an embodiment of this application. Figure 3 This is a flowchart illustrating the implementation of step S20 in a deep-sea sampling intelligent control method according to an embodiment of this application; Figure 4 This is a flowchart illustrating the implementation of step S30 in a deep-sea sampling intelligent control method according to an embodiment of this application; Figure 5 This is a flowchart of step S40 in a deep-sea sampling intelligent control method according to an embodiment of this application; Figure 6 This is another implementation flowchart of step S40 in a deep-sea sampling intelligent control method according to one embodiment of this application; Figure 7 This is a flowchart illustrating step S43 of a deep-sea sampling intelligent control method according to an embodiment of this application; Figure 8 This is a principle block diagram of a deep-sea sampling intelligent control system according to one embodiment of this application. Detailed Implementation

[0024] The present application will be further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, such as Figure 1 As shown, this application discloses a deep-sea sampling intelligent control method, which specifically includes the following steps: S10: Obtain sampling requirements and preset sampling task scripts based on those requirements.

[0026] Specifically, the system receives sampling requirements from users through an interactive interface, including the target sea area location, expected sampling depth range, required sample quantity, environmental monitoring indicators, and priority preferences. Based on these requirements, the system automatically constructs a sampling task script. The script represents the task content in structured data form and sets basic parameters and execution constraints for each task. For example, when a user specifies that a certain area requires high-frequency sampling, the task script automatically sets a short-cycle revisit sampling plan and the corresponding sensor type, thereby ensuring that the task script has adaptability and flexibility for different sampling purposes.

[0027] S20: Based on the sampling task script, construct a sampling task queue. The sampling task script includes sampling depth, priority, number of samples, and sensor configuration.

[0028] Specifically, the field information in the sampling task script is parsed to extract the sampling depth value, task priority label, number of samples to be sampled, and corresponding sensor activation configuration for each task item. Based on priority and execution window order, the task items are added to the task scheduling queue in sequence. If some tasks have high timeliness requirements, they will be prioritized. For example, in a joint scientific expedition, it is necessary to prioritize the collection of salinity anomaly samples in the nearshore 5-20 meter range. In this case, the relevant tasks will be placed at the front of the queue and marked with a high priority so that the scheduling module can execute them first.

[0029] S30: Periodically collect marine environmental data and preprocess the environmental data to obtain preprocessed multi-parameter monitoring data.

[0030] Specifically, during the environmental data acquisition and preprocessing process, the latest multi-channel sensor data package is retrieved from the environmental monitoring module at fixed time intervals. This data package contains the raw output values ​​of sensors such as dissolved oxygen, temperature, pH, conductivity, and pressure collected within a continuous time period. The system uniformly marks the acquisition timestamps and sampling device status, and constructs the multi-dimensional sensor values ​​into a time series format. Then, a composite anomaly detection algorithm based on Z-score and IQR is used to remove outliers that exceed the reasonable range. Next, noise suppression is performed through a combination of wavelet filtering and moving average algorithms. Finally, normalization and standardization transformation operations are uniformly performed, and the output is a structured monitoring data vector sequence suitable for model inference.

[0031] S40: Input the pre-processed multi-parameter monitoring data into the pre-trained environmental perception model for analysis, output the sampling task trigger judgment result, and dynamically adjust the sampling task queue according to the sampling task trigger judgment result.

[0032] Specifically, during the environmental data analysis and task queue adjustment process, the standardized multi-parameter monitoring data is first reorganized into feature vector windows of multiple sampling period lengths and input into the trained environmental perception model in a preset order. The model uses a neural network structure to perform feature encoding, state evaluation, and trigger classification reasoning on the input sequence, and finally outputs a classification label indicating whether a sampling task should be triggered and its corresponding trigger probability score. Based on the judgment result, the system performs a dynamic rearrangement operation on the sampling tasks in the task queue. When the reasoning result indicates that the triggering condition of a certain depth task is met and the trigger probability score is higher than the task scheduling threshold, the task is inserted into the first position of the execution queue in advance. Conversely, if the trigger scores of multiple consecutive periods are lower than the threshold, the task is marked as skippable or scheduled to be postponed, thus optimizing the real-time performance and responsiveness of sampling resource allocation.

[0033] S50: During the sampling process, record the operation parameters, environmental parameters and timestamp information of each sampling task, and generate an environmental label corresponding to the collected sample. The environmental label includes the original sensor value at the time of sampling, the sampling depth and the device status information.

[0034] Specifically, for each executed sampling task, its scheduling parameters, such as task number, execution start and end time, sampling depth control value, and sensor activation status, are automatically captured. Simultaneously, raw marine environmental sensor readings, such as temperature, conductivity, and turbidity values, are collected at the time of execution, and the operating status of the sampling equipment during execution is recorded, including the presence of abnormal currents or signal jitter. Finally, this information is summarized and encapsulated into a sample environment tag, uniquely associated with the physical sample collected, facilitating subsequent data backtracking and cross-analysis. S60: Monitors the communication link status of the sampling platform in real time and performs control mode switching operations based on the communication link status. When all communication channels are detected to be unavailable, local emergency control measures are triggered, including sampling task interruption, robotic arm retraction, data sealing and platform standby.

[0035] Specifically, during the communication link status monitoring and control mode switching process, the communication management module polls the connection status of all communication channels in real time, including satellite links, acoustic communication links, and optical fiber communication links. It obtains the availability flag and signal quality index of each link through the status confirmation mechanism. If it is determined that at least one communication link is stable and available, the normal communication control mode is maintained. If the monitoring results show that all links are unavailable, the local independent operation mode is immediately triggered and the emergency handling logic is called. The currently executing sampling task is automatically interrupted, the robotic arm is retracted to a safe position, the currently collected data buffer is sealed and the data write permission is locked. At the same time, the platform status is set to standby frozen state to prevent the task from continuing to be executed. After communication is restored, the management logic restores the normal control state.

[0036] By adopting the above technical solutions, and by acquiring sampling requirements and pre-setting sampling task scripts, the target area, depth range, sample quantity, and parameter requirements for sampling can be clearly defined before task execution, thus ensuring that the sampling task has a clear target orientation and operational basis. By constructing a sampling task queue and setting priorities and sensor configurations, orderly scheduling and precise execution of multiple tasks can be achieved, thereby improving the resource utilization efficiency and task response capability of deep-sea sampling. By periodically collecting and preprocessing marine environmental data, changes in the deep-sea environment can be perceived in a timely manner, and the quality of input data can be ensured, thereby enhancing the accuracy of subsequent model analysis and judgment. By inputting preprocessed data to the environmental perception model to output sampling task trigger judgment results, and dynamically adjusting the task queue accordingly, intelligent decisions on whether to execute sampling tasks can be made based on real-time environmental characteristics, thereby effectively responding to dynamic environmental changes and improving sample representativeness. By recording the operational parameters and environmental parameters of the sampling task to generate environmental tags, complete binding between samples and sampling background information can be achieved, thereby improving the scientific research value of sample data. By monitoring the status of the communication link in real time and triggering local emergency control measures when interrupted, the safety and task integrity of the sampling platform in extreme environments can be ensured, thereby improving the robustness and self-recovery capability of the overall system.

[0037] In one embodiment, such as Figure 2 As shown, in step S10, the sampling requirements are obtained, and a sampling task script is preset according to the sampling requirements, specifically including: S11: Receive user input, including target sampling area, sampling depth range, number of samples, and monitoring parameter requirements.

[0038] Specifically, during the process of receiving user input, the system receives task configuration parameters input by the operator through an interactive user interface. The input includes the geographical coordinates or latitude and longitude boundaries of the target sampling area, the start and end values ​​of the sampling depth range, the set value of the target sample quantity, and the types of multi-parameter indicators to be monitored, such as temperature, salinity, nutrients, or microbial components. During the interface interaction, the system performs syntax validation on the input format and issues prompts or restricts input for parameters that exceed the physical capabilities of the equipment. For example, when the sampling depth range is set to 0 meters to 12,000 meters and the target area is located in the Mariana Trench in the western Pacific Ocean, the system will automatically mark it as exceeding the maximum pressure resistance range of the equipment and prompt for modification of the depth parameters. After all parameter contents pass the validation, the set of configuration data is cached as a structured input record and passed to the task script generation module for subsequent processing.

[0039] S12: Generate a sampling task script based on the input content. The sampling task script includes a task identifier, target sampling depth, task priority, sampling method, and sensor configuration.

[0040] Specifically, during the process of generating the sampling task script, the system first parses the parameter values ​​in the structured input record, breaks down the target sampling depth into multiple discrete depth points, and constructs an independent task unit for each depth point. Each task unit is assigned a unique task identifier as an index for subsequent scheduling and log tracking. At the same time, priority numerical labels for each task are set according to user-specified or default rules. The sampling method is selected based on the task objective, choosing fixed-point sampling, stratified mixed sampling, or time-series sampling mode. The sensor configuration items are matched with the corresponding configuration combination from the sensor modules supported by the device according to the required monitoring indicators and the parameters are set. For example, when the user selects temperature and salinity as the monitoring parameters, the system will automatically load the temperature and salinity sensor template of the CTD module for configuration. Finally, all task units are encapsulated into a complete sampling task script structure and stored in the task queue construction module for scheduling preparation.

[0041] In one embodiment, such as Figure 3 As shown, in step S20, which involves constructing a sampling task queue based on the sampling task script, the specific steps include: S21: Parse the multiple sampling task items contained in the sampling task script, and extract the sampling depth, priority level and execution time window corresponding to each sampling task item.

[0042] Specifically, when performing the sampling task item parsing operation, the task units in the sampling task script are first read item by item, and the sampling depth value, priority level label, and executable time window interval are extracted. The sampling depth is in meters, the priority level is an integer value representing the urgency of task scheduling or the importance of scientific research, and the time window includes the earliest executable time and the latest deadline of the task. For example, a task item may be set with a sampling depth of 500 meters, a priority level of 2, and an execution time window of 8:00 to 10:30. All fields of the task item will be parsed into structured parameter entities and stored in the task parsing result set for subsequent sorting and optimization processing.

[0043] S22: Sort the sampling tasks according to their priority levels to generate an initial task queue.

[0044] Specifically, during the initial task queue sorting operation, all task items are sorted in descending order based on the parsed priority level field. Tasks with higher priority levels are ranked higher in the queue. If the priorities are the same, the tasks are further sorted in ascending order according to their earliest executable time to ensure the rationality of execution time. For example, if there are three tasks: Task A (priority 3, time window 9:00-11:00), Task B (priority 2, time window 8:00-10:00), and Task C (priority 3, time window 10:00-12:00), the sorting result is Task A, Task C, and Task B. After sorting, an initial task queue is generated and the queue index is marked for use by the scheduling module.

[0045] S23: Based on the similarity of sampling depth and the overlap of execution time windows, the initial task queue is optimized and adjusted to obtain a sampling task queue. The optimization and adjustment include merging duplicate tasks and rearranging the task order.

[0046] Specifically, when optimizing and adjusting the task queue based on depth similarity and time window overlap, clustering merging and local reordering are used to improve sampling efficiency. First, the depth difference between adjacent task items is calculated and a merging threshold is set. If the depth difference is less than 20 meters within the same window and the time windows overlap, the two tasks are considered as mergeable tasks. For example, task A (500 meters, 8:30 9:30) and task B (515 meters, 8:45 9:45) meet the merging conditions and are merged into a single sampling operation to be completed at 510 meters. At the same time, the execution time of the merged task is set to the intersection or extended range of the overlapping interval. Then, the merged task queue is reordered to reduce redundant scheduling, reduce energy consumption and improve task execution efficiency. Finally, an optimized sampling task queue is generated as the scheduling basis for subsequent execution.

[0047] In one embodiment, such as Figure 4 As shown, in step S30, marine environmental data is periodically collected and preprocessed to obtain preprocessed multi-parameter monitoring data, specifically including: S31: Collect environmental sensor data at a preset time period to construct a multi-dimensional time series. The environmental sensor data includes dissolved oxygen, temperature, pH, conductivity, and pressure.

[0048] Specifically, the system performs environmental sensor data acquisition within a set sampling period. The sampling period can be set to 30 seconds, 1 minute, or 5 minutes depending on the task requirements. The acquired raw data includes real-time readings from multiple sensor acquisition channels, where dissolved oxygen is expressed in mg / L, temperature in degrees Celsius, pH is dimensionless, conductivity in μS / cm, and pressure in kPa. The system aggregates the outputs of various sensors into an observation vector at a single time point according to timestamps and organizes them into a multi-dimensional time series in chronological order. For example, if the period is 1 minute, 10 consecutive minutes of data will form a time series matrix containing 10 sets of five-dimensional data, which serves as the basis for subsequent feature analysis and model input.

[0049] S32: Based on the sliding window mechanism, multidimensional time series are processed, outlier removal and noise filtering algorithms are used for data cleaning, and normalization and standardization operations are performed to obtain preprocessed multi-parameter monitoring data.

[0050] Specifically, when processing multidimensional time series, a sliding window mechanism is first used to divide the sequence into segments. The length of each window can be set to 5 periods, and the step size can be set to 1 period. Then, outlier removal is performed on each dimension of data within each window. For example, the IQR (interquartile range) removal method is used to replace outliers that exceed the upper and lower bounds with local means. At the same time, low-pass filtering or SG filtering methods are used to smooth sharp fluctuations and suppress high-frequency noise. For example, abrupt values ​​caused by mechanical fluctuations are removed from pressure sensor signals. After cleaning, the data of each sensor dimension is normalized. For example, the Z-score normalization method is used to map all data to a distribution with a mean of 0 and a standard deviation of 1, thereby eliminating the influence between different units and improving the numerical stability of the model during inference. Finally, a multi-parameter monitoring data sequence with a unified structure is formed as the input for model analysis.

[0051] In one embodiment, such as Figure 5 As shown, this intelligent control method for deep-sea sampling also includes: S401: Acquire historical environmental sensor data and corresponding sampling task execution records, construct a training sample set, and each sample in the training sample set includes a multi-cycle sensor parameter sequence and a corresponding sampling trigger flag.

[0052] Specifically, the execution records of several past sampling tasks and the corresponding environmental sensor data are extracted from the historical data storage module of the task platform to form a labeled dataset. Each sample consists of a sensor parameter sequence with a fixed period length and its corresponding sampling trigger marker. The sensor parameter sequence may include dissolved oxygen, temperature, pH, conductivity and pressure data within 10 consecutive minutes. A sliding window method is used to generate sample segments to increase the number of training samples. The sampling trigger marker is a binary label used to indicate whether the sample sequence ultimately triggered the sampling task. Positive samples indicate that sampling was indeed triggered, and negative samples indicate that it was not triggered. For example, if the environment fluctuates drastically within a certain period of time and causes the system to perform a sampling operation, the sensor sequence segment is marked as 1, otherwise it is marked as 0. In this way, a training sample set with a clear structure and accurate labels is constructed for model training.

[0053] S402: The model framework is constructed using a multi-layer neural network structure, which includes a feature extraction layer, a state analysis layer, and a decision output layer.

[0054] Specifically, in the model building stage, a multi-layer neural network structure is adopted to improve the ability to model complex time-series signals. The input layer receives the formatted sensor parameter sequence, the feature extraction layer can perform local pattern recognition on the time series of each sensor dimension based on a one-dimensional convolutional neural network (1D-CNN), the state analysis layer uses a recurrent neural network (RNN) or long short-term memory network (LSTM) structure to model the trend changes between different time points, and the output layer is a fully connected layer connected to a softmax or sigmoid activation function to output the probability value of whether sampling is triggered. The entire model structure can effectively extract the temporal correlation and feature representation of environmental states, such as learning the correlation between continuously changing dissolved oxygen and temperature signals and sampling events.

[0055] S403: Input the training sample set into the model framework and train it based on the supervised learning algorithm so that the model can predict whether to trigger the sampling task based on the current environmental data, so as to obtain a pre-trained environmental perception model.

[0056] Specifically, the constructed training sample set is input into the neural network model, and a supervised learning algorithm is used for training. The loss function can be a binary cross-entropy function to evaluate the difference between the model's predicted output and the actual trigger mark. The network parameters are continuously optimized through the backpropagation algorithm to minimize the error. During the training process, a batch training mechanism (e.g., a batch size of 32) is adopted, and an early stopping strategy and a validation set are set to prevent overfitting. The final trained model is the environment perception model, which can quickly determine whether to trigger the sampling task based on the current environmental sensor input data in practical applications, thereby realizing intelligent auxiliary decision-making for task execution.

[0057] In one embodiment, such as Figure 6 As shown, in step S40, the preprocessed multi-parameter monitoring data is input into a pre-trained environmental perception model for analysis, and the sampling task trigger determination result is output. The sampling task queue is then dynamically adjusted based on the sampling task trigger determination result, including: S41: The preprocessed multi-parameter monitoring data is constructed into a sequence of sensor feature vectors for multiple consecutive periods, and input into a pre-trained environmental perception model for inference analysis to obtain the sampling task trigger determination result. The sampling task trigger determination result includes the classification result of whether the sampling task should be triggered and the corresponding trigger probability score.

[0058] Specifically, the preprocessed multi-parameter monitoring data is organized into a multi-dimensional feature vector sequence arranged continuously over time using a sliding time window method. Each feature vector contains the values ​​of dissolved oxygen, temperature, pH, conductivity, and pressure at the same time point. Sequence blocks are constructed by advancing according to the window step size. This feature vector sequence is input into a trained environmental perception model for inference. The model performs joint analysis on the trends, fluctuation amplitudes, and interrelationships of each parameter in the sequence, and outputs a classification label indicating whether a sampling task should be triggered and a corresponding trigger probability score. The classification label indicates whether the current environmental state meets the sampling conditions, and the trigger probability score reflects the likelihood of the sampling event occurring. For example, if the environmental perception model outputs "should be triggered" and the probability score is 0.87, it means that the model is highly confident that the current environment has sampling value.

[0059] Furthermore, to achieve quantitative judgment on the triggerability of multi-parameter monitoring data, a classification model based on supervised learning training is used to perform inference analysis on the sensor feature vector sequence. The model output is a real value in the interval [0,1], defined as the trigger probability score. This score represents the confidence level that the current monitoring data should be judged as triggering a sampling task, denoted as P. trigger =f(X) t ), where X t Let f represent the multidimensional feature vector at time t, and f be the trained model function, such as a deep feedforward neural network or a bidirectional LSTM network structure. A higher trigger probability score indicates that the environmental change more closely matches the pattern of historical trigger samples. Further comparison is made with the set trigger threshold θ. If P... trigger If the value is greater than θ, a sampling task will be generated; otherwise, the task will be delayed, skipped, or re-evaluated.

[0060] S42: When the trigger probability score is higher than the preset trigger threshold, insert a sampling task of the corresponding depth into the sampling task queue.

[0061] Specifically, in the model inference results, when the trigger probability score is detected to be higher than the preset trigger threshold, such as exceeding 0.8, a new task item will be automatically inserted into the current sampling task queue. The target depth parameter of the new task item is calculated from the pressure value in the current environmental data and confirmed in combination with the user-defined range. The insertion position is arranged according to the priority sorting rules of the task queue. If a task at that depth already exists in the current queue, a secondary task will be merged or inserted as appropriate to strengthen redundant sampling. For example, when a sudden change in pH and dissolved oxygen is detected at the same time and the model score is 0.91, a new sampling task at the corresponding depth will be added immediately to obtain the target water layer sample.

[0062] S43: When the trigger probability score is lower than the preset trigger threshold and the trigger condition is not met for multiple consecutive cycles, the target task in the original queue is marked as skippable or delayed for execution, so as to dynamically optimize the resource configuration of the sampling task.

[0063] Specifically, if the trigger probability score given in the model inference result is lower than the trigger threshold, for example, if the score is lower than 0.3 for three consecutive sliding cycles, the system will review the resources of the target tasks in the current sampling task queue that are in the pending execution state and mark the task as "skippable" or "delayed execution". Dynamic resource reallocation is achieved by updating the status field in the task meta information. Such delayed or skipped tasks will be re-evaluated in future cycles to avoid repeated sampling of low-value water layers. For example, if a certain depth segment shows a stable state for a long time and the sampling model has never reached the trigger threshold, the task corresponding to that depth segment can be temporarily not executed to save execution cycle and energy resources.

[0064] In one embodiment, such as Figure 7 As shown, this intelligent control method for deep-sea sampling also includes: S4301: During the execution of the sampling task, calculate the environmental fluctuation index based on the changing trend of environmental parameters within the continuous sampling period.

[0065] Specifically, during the continuous execution of the sampling task, the environmental sensor parameter values ​​are recorded in real time over multiple consecutive sampling periods, and time series curves are constructed according to parameter type. By calculating the root mean square deviation, rate of change, and extreme value fluctuation range of each parameter within a preset window, and combining the coordinated change trend of multi-dimensional parameters, an environmental fluctuation index for the current sampling area is generated. The environmental fluctuation index serves as an important assessment measure reflecting the dynamic stability of the external environment. For example, if dissolved oxygen drops sharply by more than 1.5 mg / L in the last three periods, accompanied by a temperature increase of more than 2°C, it can be inferred that the environment is changing drastically, and the corresponding fluctuation index will increase significantly. This index is used as the basis for the dynamic adjustment of subsequent control strategies.

[0066] Furthermore, to quantify the stability trend of environmental parameters over multiple consecutive periods, an environmental fluctuation index is defined to represent the overall change intensity of each sensor data within the observation window. The sliding window size is set to n, and for each type of sensor data {x1, x2, ..., x...}, the index is used to represent the overall change intensity of each sensor data within the observation window. n Calculate the normalized fluctuation score V x =σx / μx, where σ x μ is the standard deviation. x The mean is used as the average, and then a weighted fusion is performed on all monitored parameters: Where m is the number of dimensions of the monitored parameters, w i The stability sensitivity weights for each parameter are: a higher EFI indicates an unstable environment, and the sampling triggering strategy will be dynamically adjusted when the set stability threshold τ is exceeded.

[0067] S4302: When the environmental fluctuation index exceeds the preset stability threshold, reduce the preset trigger threshold.

[0068] Specifically, when the calculated environmental fluctuation index exceeds the preset stability threshold, for example, above 0.75, it indicates that the current marine environment is in a highly dynamic state. At this time, the trigger threshold in the environmental perception model will be automatically lowered, for example, from the original setting of 0.85 to 0.7, in order to improve the model's response sensitivity to potential change events, thereby triggering more sample collection tasks in time before the abnormal changes have stabilized. For example, in a seawater upwelling event, a sudden change in dissolved oxygen caused the fluctuation index to reach 0.92. The system will trigger an early warning and appropriately relax the trigger criteria to strengthen the sampling density.

[0069] S4303: When the environmental fluctuation index is lower than the preset stability threshold, increase the preset trigger threshold.

[0070] Specifically, when the environmental fluctuation index is lower than the preset stability threshold for several consecutive periods, such as stabilizing below 0.3, it indicates that the environmental parameters in the current sampling area fluctuate less and the trend of change tends to be stable. In order to prevent redundant investment of sampling resources, the trigger threshold will be appropriately increased, for example from 0.75 to 0.88, to reduce unnecessary sampling requests for stable water bodies. For example, in the stable section of the deep sea layer, the pH, conductivity and pressure curves maintain narrow fluctuations for a long time. The system can suppress redundant task queuing by dynamically increasing the trigger standard, thereby improving sampling efficiency and platform operation resource utilization.

[0071] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0072] In one embodiment, a deep-sea sampling intelligent control system is provided, which corresponds one-to-one with the deep-sea sampling intelligent control method described in the above embodiments. For example... Figure 8 As shown, this deep-sea sampling intelligent control system includes a sampling task generation module, a queue construction module, an environmental data processing module, a trigger analysis module, a tag recording module, and an emergency control module. Detailed descriptions of each functional module are as follows: The sampling task generation module is used to obtain sampling requirements and preset sampling task scripts based on those requirements. The queue construction module is used to build a sampling task queue based on the sampling task script. The sampling task script includes sampling depth, priority, sampling quantity and sensor configuration. The environmental data processing module is used to periodically collect marine environmental data and preprocess the environmental data to obtain preprocessed multi-parameter monitoring data. The trigger analysis module is used to input the pre-processed multi-parameter monitoring data into the pre-trained environmental perception model for analysis, output the sampling task trigger judgment result, and dynamically adjust the sampling task queue according to the sampling task trigger judgment result; the tag recording module is used to record the operation parameters, environmental parameters and timestamp information of each sampling task during the sampling process, and generate environmental tags corresponding to the collected samples. The environmental tags include the original sensor values ​​at the time of sampling, sampling depth and device status information. The emergency control module is used to monitor the communication link status of the sampling platform in real time and perform control mode switching operations based on the communication link status. When it is detected that all communication channels are unavailable, local emergency control measures are triggered, including sampling task interruption, robotic arm retraction, data sealing and platform standby.

[0073] Optionally, the sampling task generation module includes: The user input processing submodule is used to receive user input, which includes the target sampling area, sampling depth range, number of samples, and monitoring parameter requirements. The task script generation submodule is used to generate sampling task scripts based on the input content. The sampling task scripts include task identifiers, target sampling depths, task priorities, sampling methods, and sensor configurations.

[0074] Optionally, queue building blocks include: The task parsing submodule is used to parse multiple sampling task items contained in the sampling task script and extract the sampling depth, priority level and execution time window corresponding to each sampling task item. The task sorting submodule is used to sort the sampled task items according to their priority level and generate an initial task queue; The queue optimization submodule is used to optimize and adjust the initial task queue based on the similarity of sampling depth and the overlap of execution time windows to obtain a sampled task queue. The optimization and adjustment include merging duplicate tasks and rearranging the task order.

[0075] Optionally, the environmental data processing module includes: The data acquisition submodule is used to collect environmental sensor data at preset time periods and construct multidimensional time series. The environmental sensor data includes dissolved oxygen, temperature, pH, conductivity and pressure. The data preprocessing submodule is used to process multidimensional time series data based on a sliding window mechanism. It employs outlier removal and noise filtering algorithms for data cleaning and performs normalization operations to obtain preprocessed multi-parameter monitoring data.

[0076] Optionally, this deep-sea sampling intelligent control system also includes: The training data construction module is used to acquire historical environmental sensor data and corresponding sampling task execution records to construct a training sample set. Each sample in the training sample set includes a multi-cycle sensor parameter sequence and a corresponding sampling trigger marker. The model structure construction module is used to construct a model framework using a multi-layer neural network structure. The model framework includes a feature extraction layer, a state analysis layer, and a decision output layer. The supervised training module is used to input the training sample set into the model framework and train it based on the supervised learning algorithm. This enables the model to predict whether to trigger the sampling task based on the current environmental data, so as to obtain a pre-trained environmental perception model.

[0077] Optionally, the trigger analysis module includes: The model inference submodule is used to construct a sequence of sensor feature vectors for multiple consecutive periods from the preprocessed multi-parameter monitoring data, and input it into the pre-trained environmental perception model for inference analysis to obtain the sampling task triggering judgment result. The sampling task triggering judgment result includes the classification result of whether the sampling task should be triggered and the corresponding triggering probability score. The task insertion submodule is used to insert a sampling task of the corresponding depth into the sampling task queue when the trigger probability score is higher than the preset trigger threshold. The task optimization submodule is used to mark the target task in the original queue as skippable or delayed when the trigger probability score is lower than the preset trigger threshold and the trigger condition has not been met for multiple consecutive cycles, so as to dynamically optimize the resource configuration of the sampled task.

[0078] Optionally, this deep-sea sampling intelligent control system also includes: The fluctuation calculation module is used to calculate the environmental fluctuation index based on the changing trend of environmental parameters within a continuous sampling period during the execution of the sampling task. The threshold reduction module is used to reduce the preset trigger threshold when the environmental fluctuation index exceeds the preset stability threshold. The threshold enhancement module is used to increase the preset trigger threshold when the environmental fluctuation index is lower than the preset stability threshold.

[0079] For specific limitations regarding the intelligent control system for deep-sea sampling, please refer to the limitations of the intelligent control method for deep-sea sampling mentioned above, which will not be repeated here. Each module in the aforementioned intelligent control system for deep-sea sampling can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0080] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0081] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A deep-sea sampling intelligent control method, characterized in that, The intelligent control method for deep-sea sampling includes: Obtain sampling requirements and preset a sampling task script based on the sampling requirements; Based on the sampling task script, a sampling task queue is constructed, wherein the sampling task script includes sampling depth, priority, sampling quantity and sensor configuration; Marine environmental data is collected periodically and preprocessed to obtain preprocessed multi-parameter monitoring data. The preprocessed multi-parameter monitoring data is input into a pre-trained environmental perception model for analysis, and the sampling task trigger determination result is output. The sampling task queue is dynamically adjusted according to the sampling task trigger determination result. During the sampling process, the operation parameters, environmental parameters and timestamp information of each sampling task are recorded, and an environmental label corresponding to the collected sample is generated. The environmental label includes the original sensor value at the time of sampling, the sampling depth and the device status information. The system monitors the communication link status of the sampling platform in real time and performs a control mode switching operation based on the communication link status. When all communication channels are detected to be unavailable, local emergency control measures are triggered, including sampling task interruption, robotic arm retraction, data sealing, and platform standby.

2. The intelligent control method for deep-sea sampling according to claim 1, characterized in that, The process of obtaining sampling requirements, and pre-setting a sampling task script based on those requirements, includes: Receive user input, which includes the target sampling area, sampling depth range, number of samples, and monitoring parameter requirements; A sampling task script is generated based on the input content. The sampling task script includes a task identifier, target sampling depth, task priority, sampling method, and sensor configuration.

3. The intelligent control method for deep-sea sampling according to claim 1, characterized in that, The step of constructing the sampling task queue according to the sampling task script includes: The sampling task script contains multiple sampling task items, and the sampling depth, priority level and execution time window corresponding to each sampling task item are extracted. The sampling tasks are sorted according to the priority level to generate an initial task queue; Based on the similarity of the sampling depth and the overlap of the execution time window, the initial task queue is optimized and adjusted to obtain the sampling task queue. The optimization and adjustment include merging duplicate tasks and rearranging the task order.

4. The intelligent control method for deep-sea sampling according to claim 1, characterized in that, The periodic collection of marine environmental data, followed by preprocessing of the environmental data to obtain preprocessed multi-parameter monitoring data, includes: Environmental sensor data is collected at a preset time period to construct a multidimensional time series. The environmental sensor data includes dissolved oxygen, temperature, pH, conductivity, and pressure. The multidimensional time series is processed based on a sliding window mechanism, and outlier removal and noise filtering algorithms are used for data cleaning. Normalization is then performed to obtain the preprocessed multi-parameter monitoring data.

5. The intelligent control method for deep-sea sampling according to claim 1, characterized in that, The intelligent control method for deep-sea sampling also includes: Historical environmental sensor data and corresponding sampling task execution records are acquired to construct a training sample set. Each set of samples in the training sample set includes a multi-cycle sensor parameter sequence and a corresponding sampling trigger marker. A multi-layer neural network structure is used to construct a model framework, which includes a feature extraction layer, a state analysis layer, and a decision output layer. The training sample set is input into the model framework and trained based on a supervised learning algorithm, so that the model can predict whether to trigger a sampling task based on the current environmental data, thereby obtaining the pre-trained environmental perception model.

6. The intelligent control method for deep-sea sampling according to claim 1, characterized in that, The step of inputting the preprocessed multi-parameter monitoring data into a pre-trained environmental perception model for analysis, outputting a sampling task trigger determination result, and dynamically adjusting the sampling task queue based on the sampling task trigger determination result includes: The preprocessed multi-parameter monitoring data is constructed into a sequence of sensor feature vectors for multiple consecutive periods, and input into the pre-trained environmental perception model for inference analysis to obtain the sampling task trigger determination result. The sampling task trigger determination result includes a classification result of whether the sampling task should be triggered and the corresponding trigger probability score. When the trigger probability score is higher than the preset trigger threshold, a sampling task of the corresponding depth is inserted into the sampling task queue; When the trigger probability score is lower than the preset trigger threshold and the trigger condition is not met for multiple consecutive cycles, the target task in the original queue is marked as skippable or delayed for execution, so as to dynamically optimize the resource configuration of the sampling task.

7. The intelligent control method for deep-sea sampling according to claim 6, characterized in that, The intelligent control method for deep-sea sampling also includes: During the sampling task, the environmental fluctuation index is calculated based on the changing trends of environmental parameters within a continuous sampling period. When the environmental fluctuation index exceeds a preset stability threshold, the preset trigger threshold is lowered; When the environmental fluctuation index is lower than the preset stability threshold, the preset trigger threshold is increased.

8. A deep-sea sampling intelligent control system, characterized in that, The deep-sea sampling intelligent control system includes: The sampling task generation module is used to obtain sampling requirements and preset sampling task scripts based on the sampling requirements. The queue construction module is used to construct a sampling task queue according to the sampling task script, wherein the sampling task script includes sampling depth, priority, sampling quantity and sensor configuration; An environmental data processing module is used to periodically collect marine environmental data and preprocess the environmental data to obtain preprocessed multi-parameter monitoring data. The trigger analysis module is used to input the preprocessed multi-parameter monitoring data into a pre-trained environmental perception model for analysis, output the sampling task trigger determination result, and dynamically adjust the sampling task queue according to the sampling task trigger determination result. The tag recording module is used to record the operation parameters, environmental parameters and timestamp information of each sampling task during the sampling process, and generate environmental tags corresponding to the collected samples. The environmental tags include the original sensor values ​​at the time of sampling, sampling depth and device status information. The emergency control module is used to monitor the communication link status of the sampling platform in real time and perform control mode switching operations based on the communication link status. When it is detected that all communication channels are unavailable, local emergency control measures are triggered, including sampling task interruption, robotic arm retraction, data sealing and platform standby.

Images